Energy and Power Engineering - scirp.orgProf. Qingquan Chen Prof. Yusheng Xue Prof. Ryuichi Yokoyama...

Energy and Power Engineering, 2010, 2, 137-211 Published Online August 2010 in SciRes (http://www.SciRP.org/journal/epe/)

Copyright © 2010 SciRes. EPE

TABLE OF CONTENTS

Volume 2 Number 3 August 2010

Draw of Infinite Energy from Space and Negations of Two Important Laws

F. S. Liu…………………………………………………………………………………………………………………………137

Migratory Behavior of Franklin’s Gulls (Larus pipixcan) in Peru

J. Burger, M. Gochfeld, R. Ridgely………………………………………………………………………………………………143

Regional Coordination for under Frequency Load Shedding

M. A. Anuar, H. Bevrani, T. Hiyama……………………………………………………………………………………………148

Classification of Power Quality Disturbances Using Wavelet Packet Energy Entropy and LS-SVM

M. Zhang, K. C. Li, Y. S. Hu……………………………………………………………………………………………………154

Constraints Based Decision Support for Site-Specific Preliminary Design of Wind Turbines

A. Arbaoui, M. Asbik……………………………………………………………………………………………………………161

Experimental Investigation of Solar Panel Cooling by a Novel Micro Heat Pipe Array

X. Tang, Z. H. Quan,Y. H. Zhao…………………………………………………………………………………………………171

A Burning Experiment Study of an Integral Medical Waste Incinerator

R. Xie, J. D. Lu, J. Li, J. Q. Yin……………………………………………………………………………………………………175

Chaotic Optimal Operation of Hydropower Station with Ecology Consideration

X. F. Huang, G. H. Fang, Y. Q. Gao, Q. J. Dong………………………………………………………………………………182

Simulation on SO2 and NOX Emission from Coal–Fired Power Plants in North-Eastern North America

S. C. Ma…………………………………………………………………………………………………………………………190

Multi-Bias Model for Power Diode Using a Very High Description Language

H. Mnif, M. Najari, H. Samet, N. Masmoudi……………………………………………………………………………………196

Insulation State On-Line Monitoring and Running Management of Large Generator

Q. D. Sun, Z. X. Zhou, W. Q. Guo………………………………………………………………………………………………203

The Design of New Sensorless BLDCM Control System for Electric Vehicle

Z. B. Ren, X. P. Liu……………………………………………………………………………………………………………208

Energy and Power Engineering (EPE)

Journal Information

SUBSCRIPTIONS

The Energy and Power Engineering (Online at Scientific Research Publishing, www.SciRP.org) is published quarterly by Scientific

Research Publishing, Inc., USA.

Subscription rates: Print: $50 per issue.

To subscribe, please contact Journals Subscriptions Department, E-mail: [email protected]

SERVICES

Advertisements

Advertisement Sales Department, E-mail: [email protected]

Reprints (minimum quantity 100 copies)

Reprints Co-ordinator, Scientific Research Publishing, Inc., USA.

E-mail: [email protected]

COPYRIGHT

Copyright©2010 Scientific Research Publishing, Inc.

All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by

any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except as described below, without the

permission in writing of the Publisher.

Copying of articles is not permitted except for personal and internal use, to the extent permitted by national copyright law, or under

the terms of a license issued by the national Reproduction Rights Organization.

Requests for permission for other kinds of copying, such as copying for general distribution, for advertising or promotional purposes,

for creating new collective works or for resale, and other enquiries should be addressed to the Publisher.

Statements and opinions expressed in the articles and communications are those of the individual contributors and not the statements

and opinion of Scientific Research Publishing, Inc. We assumes no responsibility or liability for any damage or injury to persons or

property arising out of the use of any materials, instructions, methods or ideas contained herein. We expressly disclaim any implied

warranties of merchantability or fitness for a particular purpose. If expert assistance is required, the services of a competent

professional person should be sought.

PRODUCTION INFORMATION

For manuscripts that have been accepted for publication, please contact:

E-mail: [email protected]

Energy and Power Engineering, 2010, 2, 137-142 doi:10.4236/epe.2010.23020 Published Online August 2010 (http://www.SciRP.org/journal/epe)


Draw of Infinite Energy from Space and Negations of Two Important Laws

Fusui Liu Department of Physics, Beijing University, Beijing, China

E-mail: [email protected] Received February 25, 2010; revised April 15, 2010; accepted June 12, 2010

Abstract This paper shows that energy of 105 ton of oil can be obtained from space by fs (fermtosecond) eletromag-netic pulse technique in one second and one cm3 without any loss. This paper shows that the energy conser-vation law and Fermi golden rule should be negatived in some cases. The negation of Fermi golden rule has important influences on many fields based on quantum mechanics. For example, the present knowledge on the charge distribution in atomic nucleus might be wrong completely. This paper emphasizes that the propo-sition on introducing the concept of the energy support ability in space will cause a series of unimaginable discoveries, and, therefore is of epoch-making significance. This paper gives indirect experimental verifica-tions for the necessity of introducing the concept of energy support ability in space, and suggests a very sim-ple experiment to show directly that the energy conservation law and Fermi golden rule should be negatived in some cases. Keywords: Energy Conservation Law, Fermi Golden Rule, Fermtosecond Technique

1. Introduction In the calculations for the probability of transition to con-tinuous spectrum all textbooks of quantum mechanics make the following four assumptions [1-11]. 1) The tran-sition matrix element and the density of states are an energy constant, and the transition rate does not depend on time, which is called Fermi golden rule; 2) The transi-tion probability is determined only by the height of the first peak in curve of the energy density of transition probability; 3) The width of the first peak is determined by the energy uncertainty principle; 4) It is easy to see that the first peak is of property of energy conservation, and the second peak is not of property of energy conser-vation. However, considering that (height of second peak) /(height of second peak) = 0.06, the second peak and more higher order peaks are neglected i.e., the energy nonconservation in transition process does not been con-sidered.

We strongly doubt the correctness of the above four assumptions, based on the following considerations. 1) It is obvious that there is no verification for the correctness of neglecting the energy variations of the transition ma-trix element and density of states in any cases. However, they are strongly energy-dependent in some cases; 2)

There is no verification for the correctness of the time independence of transition rate in any cases; 3) Actually, the formula of transition probability is derived without using the energy uncertainty principle. Therefore, the explanation for the width of the first peak in terms of energy uncertainty principle is utterly unjustifiable; 4) The second peak and others of energy nonconservation should not been neglected in any cases. We should try to increase the height of the second peak and others. If we can, then mankind can have infinite energy without any loss in terms of energy nonconservation.

This paper makes exact calculations for the transition probability, and obtains many important discoveries, which are stated in Sections 2, 3 and 4. Some discussions are given in Section 5. Our conclusions are listed in Se-ction 6. 2. Draw of Infinite Energy from Space and

Negation of Energy Conservation Law in Some Cases

For the convenience of statement, at first we study the elementary theory of photoeffect. Let us consider a hydrogen atom in ground state. The Hamiltonian H is H = H0 + H'. |m > is the state vector of discrete spectrum of H =

F. S. LIU


138

H0. |k > is the state vector of continuous spectrum of H = H0. B is the domain of |k >. H' is the Hamiltonian of electron of hydrogen atom in electromagnetic field. H' is [1]

H' = ercos(θ) E0 (eiωt' +e iωt')

= H" (e iωt' +e iωt'), (1)

where r is the position vector of electron, θ is the angle between field and r. Assume that the duration time of field is between 0 and t0. The probability of transition from state |m > to one state |k >, Wm→k, at t ≥ t0 due to absorption of a photon is [1]

2 20

2

14 | | '' | | sin ( )

2

( )2

k m

m k

k m

k H m tW

h

(2)

where )2/()( 2 mkk and m is the mass of electron.

For simplicity, we consider the boundary absorption of hydrogen atom in ground state. In this case m

6.13 eV. The energy density of transition probabi-lity from state |m> to any state |k > with energy between

kE and kk dEE per unit solid angle at 0tt due

to absorption of a photon , kdEmW , is [1]

)(||''||4

2

2

kdEm EmHk

Wk

kk

k

dEt

2

02

2

1sin

, (3)

where )( kE is the density of states, and it is [1]

ddELm

Ek

k sin8

21

)(33

2

132

32

1

, (4)

where ddsin is solid angle. <r|m> = 2/10 )( a exp

)/( 0ar , and 0a is Bohr radius [1]. <r|k> = exp2/3L

(ik.r) [1]. After simple derivation we have

kdm dWk

42

5

2

1

22

570

20

2

2

cos128

maEe

.2sin

)2

1(2

02

2

1

620

kk

k

kk

k d

t

ma

(5)

From Equation (5) we know that kdmW is propor-

tional to I .

2

02

2

1

617

2sin

)10396.21( k

k

kk

k

t

I

IIkk

k .)10396.21(

2

1

617

IIIII. . (6)

In the now available calculations the energy-dependent factor III is taken to be an energy constant which is equal to the energy determined by the center of the first peak in the curve of factor II versus k , all en-

ergy variations come from the factor II , and, therefore, the energy variation of

kdmW comes only from the

factor II [1-11]. If we use the fs electromagnetic pulse technique, then 15

0 10t second. Figure 1 gives the

curve of II versus f, where khf 15102 . From

Figure 1 we see that (height of the first peak)/(height of second peak) = 0.06. The previous general points of view are that the second peak of energy nonconservation can be neglected because it is too small. That the center of the first peak is at 0k means energy conservation,

and the width of the first peak comes from energy uncer-tainty principle [1,2]. However, if we take 0k ,

which is determined by the center of the first peak in Fi-gure 1, then 0III , which means 0 kdmW i.e.,

the transition of boundary absorption of electron of hy-drogen atom is exactly prohibited. It is obvious that this prohibition does not fit the experiments, and is com-pletely wrong. Therefore, we have to consider the energy variation of the factor III . The curve of )( IIIIII

versus f is shown in Figure 2. From Figure 2 we see that the transition of boundary absorption can happen because

IIIII does not always equal to zero. The curve in Fi-gure 2 does not have any connection with energy uncer-tainty principle (By the way, here we should mention

f (1015 Hz) 0 1 2 3

1.0

= 0.5

0.0

Figure 1. Theoretical curve of II versus f (1015 Hz). If the energy variation of III is neglected, as usually be done in all quantum mechanics books, then II is proportional to

frequency density of transition probability kdωm

W .

F. S. LIU


139

that the proof in [3] for the energy uncertainty principle in terms of our Figure 1 is completely wrong), and the heights of the second and third peaks are nearly equal to the height of the first peak. Our numerical calculations show that (height of the eleventh peak)/(height of first peak) = 0.05, which is nearly equal to the value 0.06 in the last paragraph. From Figure 2 we see that this transi-tion is seriously energy-nonconservative, and the general energy conservation law should be negatived in this case. The energy variation of the factor III is written in real space, depends on the space properties such as dimen-sions, and supports the energy nonconversation transition. Therefore, we name the factor in (3), |<k|H"|m>|2 )( kE ,

as energy support ability in space. The energy noncon-servation comes from the 2/3

k -dependence of the en-

ergy support ability in space. From Figure 2 and our additional calculations we can

take that the transition probability of emission of an electron with energy 15105 h /second by a hydrogen atom in ground state under boundary absorption of fs electromagnetic pulse is 0.03. The number of hydrogen atoms in one 3cm is 2.7 19106 . If we can take the electron energy larger than 15105 h /second, then the energy obtained by transitions of energy nonconservation in one 3cm of hydrogen atoms and in one second is

2210414.0 erg, which corresponds to the energy 510 ton of oil. The electron with energy between 0 and

1510h /second can emit a photon with energy 13.6 eV into electromagnetic field, and goes back to the ground state [1]. Thus, we can actually obtain infinite energy from space without any loss in principle. 3. Negation of Fermi Golden Rule If we assume that the term III i.e., the energy support

f (1015 Hz) 0 1 2 3 4

－

0

1

Figure 2. Theoretical curve of I versus f (1015 Hz). I is exactly proportional to frequency density of transition

kdωmW . Figure 2 shows that energy can be seriously non-

conservative.

ability in space, is an energy constant, then the above transition rate per solid angle after the fs electromagnetic pulse is

00

1kdEm dW

tw

k

kk

k

k dt

tE

mk

02

02

0

2

2

2

1sin1

)(

||''||4

, (7)

which is independent of time, and is called Fermi golden rule [1-11]. However, if we consider the energy variation of the term III , then Figure 3 shows that w is strongly time-dependent, and the Fermi golden rule should be negatived completely.

Let us give some other examples to show that Fermi-golden rule should be negatived in many cases. First example is elastic scattering of an electron by an atomic nucleus [2]. The transition rate per solid angle is [2]

d

t

EVt

wi

i

if 2

2

2

)(2

)(sin

)(||||4

d

t

sFs

e

t i

i

2

22

2

2

)(2

)(sin

)()(44

, (8)

where E is the energy of the final state after scattering between electron and nucleus pcE (

)2222 cMs , M is the mass of nucleus, p is the mo-

mentum vector of electron after scattering, 0E is

the energy of initial state of electron and nucleus, 0E

)( 0 Mcpc , p0 is the momentum of electron before

log10t -15 -10 -5 0 5 10 15

20

15

10

5

0

–5

–10

log 1

0W

Figure 3. w is the transition probability in ionization process per second and solid angle. The unit of t is second. Figure 3 shows that Fermi golden rule should be negatived.

F. S. LIU


140

scattering, s = p0 – p, and )(sF is called form factor,

which is the Fourier-transformed charge distribution and reflects the deviation of the nuclear charge distribution from point structure [2]. If we neglect the energy de-pendence of energy support ability in space, then (8) is called Rutherford scattering formula which was derived by classical mechanics [3], and was confirmed without considering the energy variation of the energy support ability in space by quantum mechanics [3]. Based on Rutherford formula, Robert Hofstadter made systemati-cal measurements, got the form factor )(sF , obtained

charge distribution of atomic nuclei, and was awarded the Nobel Prize in 1961 [2]. However, if we consider the energy dependence of the energy support ability in space, try that one only considers the second moment of the charge distribution, then )(sF is proportional to 2s ,

and make an exact calculations for w , then we have Figure 4. Figure 4 shows that Fermi golden rule should be negative. From Figure 4 we see that w is strongly time-dependent, Fermi golden rule should be negatived, and the charge distribution of atomic nuclei measured by Robert Hofstadter, which was based on Fermi golden rule, should be wrong. The correct method to measure the structure factor )(sF is as follows. First, we calcu-

late a theoretical curve of the frequency density of transi-tion probability in (8) without )(sF . Second, we measure

the experimental data of frequency density of transition probability, and from the width of the first peak, , we can know the value of duration time t of scattering

./1 t It should be interesting that the duration time can be measured by experiment. The differences between the theoretical curve and experimental data come from factor )(sF . From this )(sF one definitely can obtain

a new charge distribution which is different more or less from that gotten by Robert Hofstadter. Here we should point out that all the now available theories on elastic

log10t

-15 -10 -5

-5

-10

log 1

0W

-15

-20

Figure 4. w is the transition probability in scattering process per second and solid angle. The unit of t is second.

and inelastic scattering, which are important part of quantum mechanics, assume that the energy support ability in space is an energy constant, use Fermi golden rule to discuss scattering problem [1-11], and are more or less wrong definitely. Therefore, all the now available theo-ries on scattering before this paper should be reformed.

Actually, we can give many examples which show that Fermi golden rule should be negatived, and the energy dependence of the energy support ability in space should be considered. For example, let us look at the quantum transitions under the influence of time-independent interactions. This is a very width research field which contains: 1) Internal conversion, that is, the process in which an excited nucleus transfers its energy to the atomic elec-trons. 2) Auger effect, that is, the readjustment of the electron shells of atom with several electrons, accompa-nied by the ejection of one electron from the atom. We shell consider internal conversion. [4] gives already that the energy support ability of space is strongly dependent on energy (See (100.9) of [4]). However, A. S. Davidov, [4] does not consider this strong dependence of energy, and still uses Fermi golden rule. Therefore, the result is definitely wrong. If one considers the influence of energy support ability in space, then one can obtains correct conclusion definitely. 4. An Epoch-Making New Concept—The

Energy Support Ability in Space The introduction of concept of the energy support ability in space will cause more and more significant discover-ies. For example, if the energy support ability in space for the boundary ionization by fs electromagnetic pulse technique is proportional to 6303 )101/( kk instead of the above 6175.1 )101/( kk , then the curve of energy density of transition probability is much different from Figure 2. If the electron in ground state mE = –13.6 eV absorbs a photon with energy E = 13.6 eV and emits an ionized electron, then the emitted electron with energy 301022 fE kk = 3 6.131015 eV has largest transition probability. This kE corre-sponds to the energy of 64 ton of oil i.e., the energy is strongly nonconservative. The Equation (100.9) in [4] already gave an example which is of strong energy de-pendence of the energy support ability in space.

The Fermi golden rule has been used in many fields such as atomic physics, nucleus physics, particle physics, condensed matter state physics. The negation of Fermi golden rule will cause a series of new discoveries and corrections in these fields. The energy conservation law was a never wavering and natural law before the publica-tion of this paper. The negation of the energy conserva-tion law in some cases in this paper will cause a series of new unimaginable discoveries definitely.

F. S. LIU


141

5. Discussion Although all results in Sections 2, 3 and 4 come just from exact calculations of transition probability, many readers still do not believe their correctness. Let us give some indirect experimental verifications for the correct-ness of considering the energy dependence of the energy support ability in space. First, let us consider the study on relaxation process which is an ancient project more than 100 years. The KWW empirical law is that the re-laxation function is KWW_exp( t/ ) . 1KWW is only

for a few materials, and for 90% of materials 0

KWW < 1. [14-17] show that if we consider the energy

dependence of the energy support ability in space, then 1KWW . Second, [18-20] show that if we consider the

energy dependence of the energy support ability in space, then cold fusion can occur, and the result of experimental observations for the cold fusion is true.

Reference [3] and Section 3 of this paper point out that the Rutherford scattering formula, which was based on classical mechanics, and the Mott-Gorden scattering formula, which was based on quantum mechanics and neglecting the energy variation of the energy support ability in space, are the same. This fact tells us that the energy variation of the energy support ability in space includes quantum effect, and, therefore, it can not been neglected. If we neglect it, then classical and quantum mechanics give the same result.

A simple direct experimental verification on the ne-cessity of introducing the concept of energy support abil-ity in space is to obtain the experimental data corre-sponding to Figure 2. 6. Conclusions From our exact derivations and numerical calculations in Sections 2, 3 and 4 we obtain the following conclusions. 1) It is absolutely necessary to consider the energy de-pendences of the transition matrix element and the den-sity of states in transition and scattering processes. However, all the now available theories on the transition and scattering processes do not consider these energy dependences, and, therefore, should be revised. 2) The general energy conservation law should be negatived in some cases. 3) It is possible to obtain infinite energy from space without any loss. 4) The Fermi golden rule should be negatived in some cases because that the ap-proximation of neglecting the energy dependence of the energy support ability in space is reasonable only in a few cases. 5) The transition process does not have any connection with energy uncertainty principle. 6) The concept on the energy support ability in space will be-come an important new concept. 7) Section 3 points out that the duration time of scattering between electron and

nucleus can be measured by experiment. 8) The current standard model of cosmology, or Big Bang model, has been receiving wider and wider attention since the dis-covery of cosmic background radiation at 2.73 K. The observable facts upon which the standard model is based are, in fact, very few [10]. This paper shows that the en-ergy support ability in space is only determined by the structure of space, and, therefore, it can always supply energy without any loss i.e., the energy is infinite in cosmology. Because energy can become mass, the mass in the cosmology is also infinite. The cosmology being of infinite energy and mass can not collapse, should have infinite lifetime, and the Big Bang model can not be cor-rect. 9) The present theory to estimate the energy in cosmology is as follows. If all the energy in cosmology is 1, then the energy of galaxy is 4/100, the energy of dark mass is 23/100, and the energy of dark energy is 73/100. It is obvious that this estimation is based on the energy finiteness of cosmology. This paper concludes that the above estimation for the energy distribution in cosmology is wrong because the energy in cosmology is infinite. 10) G. Amelino-Camelia [21] pointed out that combing general relativity with quantum mechanics is the last hundle to be overcome in the “quantum revolu-tion”. One of the most exciting approaches to the unifi-cation of general relativity and quantum mechanics is the idea of a space-time that is itself quantized, for example, replacing the space-time continuum with a collection of isolated points. This paper shows that the energy support ability in space depends on the structure of space. Therefore, the energy support ability in space can be used to judge any proposed model of space structure. 11) B. R. Martin [13] pointed out that the observable quanti-ties in nuclear and particle physics are cross-sections and decay rates. However, we should note that the formulas to calculate the two quantities are used Fermi golden rule. This paper shows that Fermi golden rule should be nega-tived, especially, in the calculations of cross-sections. Therefore, many conclusions coming from the two quan-tities might be wrong. 7. References [1] L. I. Schiff, “Quantum Mechanics,” McGraw-Hill Book

Company, New York, 1968.

[2] W. Greiner, “Quantum Mechanics: An Introduction,” 3rd Edition, Springer-Verlag Belin Heidelberg, New York, 1994.

[3] L. D. Landau and E. M. Lifshitz, “Quantum Mechanics: Non-Relativistic theory,” Pergamon Press, Oxford, 1958.

[4] A. S. Davidov, “Quantum Mechanics,” 2nd Edition, Per-gamon Press, Oxford, 1976.

[5] B. R. Desai, “Quantum Mechanics,” Cambridge Univer-sity, Cambridge, 2010.

[6] E. R. Bittner, “Quantum Dynamics: Applications in Bio-

F. S. LIU


142

logical and Materials Systems,” Taylor and Francis/CRC Press, New York, 2010.

[7] R. L. Liboff, “Introductory Quantum Mechanics,” Addi-son-Wesley, New York, 2003.

[8] K. T. Hecht, “Quantum Mechanics,” Springer-Verlag, New York, 2000.

[9] A. I. M. Rae, “Quantum Mechanics,” Taylor and Francis, New York, 2008.

[10] E. Elbaz, “Quantum: The Quantum Theory of Particles, Fields, and Cosmology,” Springer, New York, 1998.

[11] J. J. Sakurai and S. F. Tuan, “Modern Quantum Mechani-cs,” Addison-Wesley, New York, 1994.

[12] S. L. Kakani and S. Kakani, “Nuclear and Particle Phy-sics,” Anshan, Kent, 2008.

[13] B. R. Martin, “Nuclear and Particle Physics,” 2nd Edition, John Wisley and Sons Ltd, West Eussex, 2009.

[14] F. S. Liu and C. Wen, “Dynamics of Continuous-Time Random Walk, Fractional Time Dispersion, and Frac-tional Exponential Time Relaxation,” Physical Review B (Condensed Matter), Vol. 40, No. 10, 1989, pp. 7091- 7095.

[15] F. S. Liu and W. F. Chen, “A New Universal Theory of Non-Exponential Relaxation,” Journal Physics D: App-

lied Physics, Vol. 27, No. 4, 1994, pp. 845-847.

[16] F. S. Liu, K. D. Peng and W. F. Chen, “Departure from Fermi Golden Rule,” International Journal of Theoretical Physics, Vol. 40, No. 11, 2001, pp. 2037-2043.

[17] F. S. Liu and W. F. Chen, “Necessity of Exact Calcula-tion for Transition Probability,” Communications in Theoretical Physics, Beijing, Vol. 39, No. 2, 2003, pp. 209-211.

[18] F. S. Liu and W. F. Chen, “Phonon-Induced Hopping Rate Enhancement in the Pd-D System,” Journal of Phy- sics: Condensed Matter, Vol. 15, No. 29, 2003, pp. 4995- 5000.

[19] F. S. Liu, Y. M. Hou and W. F. Chen, “Theory of Fusion during Acoustic Cavitation in ODC 63 Liquid,” Journal

of Condensed Matter Nuclear Science, Vol. 1, No. 1, 2007, pp. 142-147.

[20] F. S. Liu and W. F. Chen, “The Effect of Phonon-Induced Hopping Enhancement and Exact Theory of Cold Fu-sion,” Communications in Theoretical Physics. Vol. 23, No. 2, 1995, pp. 241-244.

[21] G. Amelino-Camelia, “Quantums Theory’s Last Chal-lenge,” Nature (London), Vol. 408, No. 6813, 2000, pp. 661-664.



Migratory Behavior of Franklin’s Gulls (Larus pipixcan) in Peru

Joanna Burger1, Michael Gochfeld2, Robert Ridgely3 1Division of Life Sciences, Rutgers University, Piscataway, USA

2Environmental and Occupational Medicine, UMDNJ-Robert Wood Johnson Medical School, Piscataway, USA 3N. Sandwich, New Hampshire, UK

E-mail: [email protected], [email protected], [email protected] Received March 29, 2010; revised May 10, 2010; accepted June 20, 2010

Abstract Information on the migratory pathways for birds is essential to the future citing of wind power facilities, par-ticularly in off-shore waters. Yet, relatively little is known about the coastal or offshore migratory behavior of most birds, including Franklin’s gulls (Larus pipixcan), a long-distant migrant. We report observations along the coast of Peru made in November 2008 to determine where birds concentrated. Wind facilities can not avoid regions of high avian activity without knowing where that activity occurs. Migrant flocks of 250 to 50,000 were observed on coastal farmfields, dumps and estuaries, on beaches and mudflats, and up to 45 km offshore. Bathing and foraging flocks ranged in size from 20 to 500 birds, and most flocks were monospeci-fic, with occasional grey-headed (Larus cirrocephalus) and band-tailed (L. belcheri) on the periphery. While previous notes report Franklin’s gulls foraging coastally, we found flocks feeding up to 45 km offshore by diving for prey or feeding on the water. The relative percentage of birds of the year varied in migrant flocks from zero to 14%, with lower numbers of young foraging aerially on insects (only 1%). The percentage of young feeding over the ocean decreased with increasing distance from shore; no young of the year were re-corded at 36-44 km offshore. While there were large flocks of Franklin’s gulls resting on the water inshore, the number of gulls foraging offshore did not decline up to 45 km offshore. The presence of foraging flocks of Franklin’s gulls out to 45 km offshore, and occupying space from 0 to 20 m above the water, suggests that they would be vulnerable to offshore anthropogenic activities, such as offshore drilling and wind facilities. Keywords: Migration, Larus pipixcan, Franklin’s Gulls, Gulls, Migrants, Young of the Year, Habitat Use,

Flock Associations, Wind Farms, Offshore Drilling

1. Introduction The siting of wind facilities has become an important topic as governments and industry consider the possibil-ity of large-scale offshore facilities. Yet little is known of the ecology and behavior of species, such as marine mammals, fish, and birds, in offshore regions where wind facilities might be sited. Before siting many such facilities, it is essential to understand whether the loca-tions would impact ecological resources in these sites.

The migratory behavior of birds is an important, but often little studied aspect of their life cycle, mainly be-cause long-distance migrants are difficult to study. They often migrate at night, at high altitudes, or at unpredict-able places and times. Further, scientists often focus on the breeding season, or on native species, or on the rare

migrants, making information on abundant migrants par-ticularly lacking. Yet, for many species, migration is one of the most risky life stages, because of predation, weather conditions, obstacles (such as buildings or tow-ers [1,2]), or lack of foraging habitats [3-8].

Information on the locations, habitats, and timing of migration is needed to understand both the vulnerability of a species to natural forces, as well as to potential an-thropogenic activities, such as wind facilities. While sci-entists have long recognized the threats to migrants of anthropogenic terrestrial threats, such as buildings and towers [1,9,10], little attention has been devoted to coastal and offshore migrants. With the recent focus on renewable energy, many countries are turning to offshore wind farms, and the question of risk to avian populations that migrate offshore is coming to the fore, with the re-

J. BURGER ET AL.


144

alization that there is a lack of information on the spe-cific locations of common and abundant migrants along coasts, and out to the edge of the continental shelf.

In this paper we report on observations of migrant Franklin’s (Larus pipixcan) Gulls in coastal Peru. We were particularly interested in flock locations (distribu-tion along the coast), habitats, and in the percentage of young of the year present in flocks (an indication of re-productive success). Little quantitative information is available on migrant Franklin’s Gulls in South America [11]. They were believed to migrate mainly offshore over the ocean, and in Peru to migrate low over the deserts [12], but timing, flock size, habitat use, and flock asso-ciations were poorly known [11].

2. Methods All observations were made in Peru from 1 to 23 No-vember 2008. We visited freshwater marshes, coastal marshes, and beaches from south of Lima to northern Peru, recording the numbers of adult and young of the year Franklin’s Gulls, along with other species of gulls that were present. Counts of adults and young of the year were made at each location. Photographs were also taken of flocks both on the ground and in the air, and these were enlarged digitally to confirm counts and the ratio of

adults to young of the year. On 5 November 2008 we travelled 44 km offshore

from Lima. 3. Results Migrant flocks of 250 to 50,000 Franklin’s Gulls were observed on coastal farmfields, dumps and estuaries, on beaches and mudflats, and up to 45 km offshore (Table 1). Most flocks were either migrating or coming in to roost or preen, but some flocks were feeding aerially on insects, and these contained very few young (1%). At several locations (Villa, Ventanilla) we observed flocks descending from high altitudes (out of range of binocu-lars) to the beach or marsh locations to drink, bath, and preen vigorously.

Bathing and foraging flocks ranged in size from 20 to 500 birds, but resting flocks ranged up to 50,000. A high percentage of gulls in resting flocks were engaged in vigorous preening (up to 60%). While previous notes report Franklin’s gulls foraging coastally, we found flocks feeding up to 45 km offshore by diving for prey or feed-ing on the water (Table 2). Although there were large flocks of Franklin’s Gulls resting on the water inshore, the number of gulls foraging offshore did not decline up to 45 km offshore.

Table 1. Observations of Franklin’s Gulls (Larus pipixcan) from Peru (November 2008). Young of the year accounted for ??% of the gulls (where counts could be made visually and from photographs).

Date Location Habitat Number of Franklin’s Gulls

(% young of the year) Presence of other species

1 November Bayovar, N. Peru Beach 50,000 + (not recorded) 2 November Villa, near Lima Freshwater pool 800 (10) Band-tailed gulls

Beach dunes 5000 (12) Grey-headed gulls at edge of flock Nearby saltwater 5000 (12) None In air, hawking insects 200 (1) None

5 November Lima harbor out to 44 km Coastal/ocean 14,260 (2) Mainly monospecific

6 November 80 km north of Lima Aerial migrants, 5 km from coast

300 (8) None

Ventanilla Freshwater marsh near coast 273 (10) None Ventanilla Aerial migrants above town 2,100 (not recorded) None

8 November Pimentel Pimentel beach,

sewage outfall, and mudflat310 (11)

Dense monospecific with flocks of 20 grey-headed and 30 kelp gulls on

edge, and 3 elegant terns within Franklin’s Gull flock

Santa Rosa Santa Rosa Beach 1325 (11)

Dense monospecific flocks with grey-headed and band-tailed at

edges; a short distance away was a dense flock of 2,500

grey-headed gulls Garbage dump 254 (18) None Wet marshes and farmfields 1513 (15) None

9 November Abra de Porcuya

(east side) Flying over Andes 1 (adult) None

23 November Villa, Lima Beach, marshes and ocean 10,000 (14+) Kelp, Band-tailed and Grey-headed

in nearby flocks, with some at edges of Franklin’s Gull flocks

Note: Percent of young based on visual and photographic counts except 1 and 23 November, and for the aerial flock at Ventanilla (where the light prevented aging of the gulls).

J. BURGER ET AL.


145

Table 2. Number of Franklin’s Gulls in a coastal transect out to 44 km (Lima, Peru, November 5, 2008). Such information is directly relevant to offshore activities, such as shipping, oil drilling and wind farm construction.

Distance from shore (km) Number of Franklin’s Gulls in air (additional gulls rafting on water)

Percent of young of the year in flying or feeding flocks

0-4 2316 (4,200) 8 4.1-8 803 (1000) 4

8.1-12 918 (110) 5 12.1-16 74 (210) 0 16.1-20 330 1 20.1-24 300 (156) 0 24.1-28 279 3 28.1-32 804 (408) 0 32.1-36 1028 (76) 2 36.1-40 577 (180) 0 40.1-44 491 0

The relative percentage of birds of the year varied in

migrant flocks from zero to 14%, with lower numbers of young foraging aerially on insects (only 1%, Table 1). The percentage of young feeding over the ocean de-creased with increasing distance from shore; no young of the year were recorded at 36-44 km offshore. The gulls we observed were mainly occupying the vertical space from the water to 20 m above the water (although mi-grants were much higher), but were concentrated below 10 m.

Most flocks were monospecific, with occasional Grey- headed (Larus cirrocephalus) and Band-tailed (L. belcheri) Gulls on the periphery (Table 1). At some beaches, there were discrete and dense flocks of these two species, along with discrete flocks of kelp gulls (La-rus dominicanus) a few meters or hundreds of meters from the Franklin’s Gulls. Franklin’s Gulls resting or roosting on beaches often stood in very dense flocks, nearly touching one another.

Even in dense migrant flocks, Franklin’s Gulls are vulnerable to predators. On 23 November, two Franklin’s Gulls were killed by two different Peregrine Falcons (Falco peregrinus) visible at the same time. In one case an immature Peregrine flew up to a Franklin’s Gull flock swirling over land and flipped upside down to snatch a gull’s breast, riding with it to the ground. Five minutes later, a second immature Peregrine rose higher than a different gull flock, and dove into it in the classic manner. Although the gull flock scattered, the Peregrine pursued one bird until it slammed into the gull, exploding the gull and forcing it to the ground.

Two additional observations bear mention: 1) In late October 2007, several flocks of 600-1000 birds flew high overhead (at the limit of binocular vision) at the La Ven-tosa area of the Isthmus of Tehuantepec in Mexico; other flocks (100-1000) flew low and close to shore moving south and east (A. Farnsworth, pers. comm.). In 2003, Franklin’s gulls had only just begun to reach the northern beaches of Chile (Valparaiso to Astero Lampa Santiago de Pacifica): from 9-10 November fewer than 20 gulls

were observed at each of several different beaches, but by 10-12 November the number had built up to 100 at several locations (F. Lesser, pers. comm.). 4. Discussion With the world-wide development of renewable energy resources, such as wind power, it is essential to deter-mine before facilities are built whether there are conflicts with wildlife that would provide an ecological threat that would impact operations. Many of the initial sitings of wind facilities were within migratory or overwintering ranges of birds, and resulted in high avian mortality, and some curtailing of operations [13,14]. This paper pro-vides data that can be used in considering the offshore patterns of migratory gulls, particularly Franklin’s Gulls.

The Franklin’s Gulls observed in this report were likely migrants just arriving in Peru, as judged by the large dense flocks engaged in vigorous preening, and their descent in large and continuous flocks from high altitudes. That is, when we scanned the sky with binocu-lars in areas where birds were descending, we could just make out birds at the limit of binocular vision still de-scending. The presence of relatively large flocks of 5,000 to 50,000 birds suggests that they were arriving, and had not spread out along the coast.

Like other authors [12,15,16] we found them mainly along the coast, but one was in the Andes. Birds found in the high Andes may well be either lost, or merely on a different migration route.

While many different foraging and migratory habitats have been reported for Franklin’s Gulls in North Amer-ica, few have been recorded for South America [11]. Habitats recorded in South America include fishmeal plants, rivers, coasts, and behind trawlers [17,18]. We found them resting, bathing and foraging on beaches, saltwater and freshwater marshes, sewage outfalls, farm-fields, and garbage dumps. While these habitats are not unexpected, given their use of them in North America, it

J. BURGER ET AL.


146

requires documentation. For most flocks, about 10-12% of the gulls were

young of the year, although far fewer young were in flocks offshore and almost none engaged in aerial hawk-ing for insects. This is not surprising, since both aerial foraging and foraging offshore on fish are more difficult foraging tasks than feeding on invertebrates along the shore or on garbage [19-22]. That 10-12% of the flocks are young of the year indicates successful reproduction and migration over thousands of km; there are no previ-ous data on percentages of young in migrant flocks in the southern US, Central America or South America.

The presence of flocks of foraging and resting gulls out to 45 km indicates that this species would be vulner-able to any human activity on the continental shelf. While it has previously been reported that Franklin’s Gulls may migrate over the ocean, there were no quanti-tative data on numbers or distances from shore. Further, reporting that gulls migrate over the ocean does not in-dicate the location of these birds (either longitudinally or horizontally). In this study we report birds resting and feeding on the water, and flying above the water at ele-vations that would put them at risk from anthropogenic activities on the water. That is, there were gulls in every 4 km block from 0 to 44 km offshore, and there were gulls flying from the water level to 20 m above the water. As governments and companies strive to diversify energy, there is a need to have both qualitative and quantitative information on the spatial envelope birds occupy at dif-ferent times of the year. The data in this paper indicate that migrating (and potentially overwintering) Franklin’s Gulls in Peru occupy an envelope of space from 0 to 45 km offshore (and likely further out) and from 0 to 20 m from the water’s surface. Migrants descending from the sky came through space from the limit of binocular vi-sion directly to the water or land. 5. Acknowledgements We thank L. Navarette and A. Farnsworth for field com-panionship and data, F. Lesser for data from Chile, G. Engblom for insights on the species in pelagic waters, and Lelis Navarette for logistical help and field observa-tions while we were in Peru. 6. References [1] W. P. Erickson, G. D. Johnson and D. P. Young, Jr., “A

Summary and Comparison of Bird Mortality from An-thropogenic Causes with an Emphasis on Collisions,” Forest Service General Technical Report, 2005, pp. 1029- 1042.

[2] T. Longcore, C. Rich and S. A. Gauthreaux, Jr., “Height, Guy Wires, and Steady-Burning Lights Increase Hazard of Communication Towers to Nocturnal Migrants: A Re-view and Meta-analysis,” The Auk, Vol. 125, No. 2, 2008,

pp. 485-492.

[3] J. D. Goss-Custard, “The Ecology of the Wash. III. Den-sity Related Behaviour and the Possible Effects of a Loss of Feeding Grounds on Wading Birds (Charadrii),” Jour-nal of Applied Ecology, Vol. 14, 1977, pp. 721-739.

[4] J. Burger, “Shorebirds as Marine Animals,” In: J. Burger and B. L. Olla, Eds., Behavior of Marine Animals, Shore-birds: Breeding Behavior and Populations, Plenum Press, New York, Vol. 6, 1984, pp. 17-81.

[5] J. Burger, “The Effect of Human Activity on Shorebirds in Two Coastal Bays in Northeastern United States,” En-vironmental Conservation, Vol. 13. No. 2, 1986, pp. 123- 130.

[6] P. Kerlinger, “Showdown at Delaware Bay,” Natural History Magazine, Vol. 107, No. 4, 1998, pp. 56-58.

[7] N. Warnock, C. Elphck and M. A. Rubega, “Shorebirds in the Marine Environment,” In: E. A. Shreiber and J. Burger, Eds., Biology of Marine Birds, CRC Press, Boca Raton, 2001, pp. 581-615.

[8] Fish & Wildlife Service, “Piping Plover: Atlantic Coast Population Recovery Plan,” 2008. http://www/fws/gov/ northeast/pipingplover/recplan/ecology/html

[9] S. R. Morris, A. R. Clark, L. H. Bhatti and J. L. Glasgow, “Television Tower Mortality of Migrant Birds in Western New York and Youngstown, Ohio,” Northeastern Natu-ralist, Vol. 10, No. 1, 2003, pp. 67-76.

[10] D. Klem, Jr., “Glass, a Deadly Conservation Issue for Birds,” Bird Observer, Vol. 34, No. 2, 2006, pp. 73-81.

[11] J. Burger and M. Gochfeld, “Franklin’s Gull (Larus pipixcan),” In: A. Poole and F. Gills, Eds., The Birds of North America, The Academy of Natural Sciences, Phi- ladelphia; The American Ornithologists’ Union, Wash-ington, D. C., Vol. 3, No. 116, 1994. (2009 e-published update on American Ornithological website)

[12] R. C. Murphy, “Oceanic Birds of South America,” Ame- rican Museum of Natural History, New York, Vol. 2, 1936.

[13] K. S. Smallwood and C. G. Thelander, “Bird Mortality in the Altamont Pass Wind Resourc Area, California,” Jour- nal of Wildlife Management, Vol. 72, No. 1, 2008, pp. 215-223.

[14] K. S. Smallwood, L. Ruggeb and M. L. Morrison, “In-fluence of Behavior on Bird Mortality in Wind Energy Developments,” Journal of Wildlife Management, Vol. 73, No. 7, 2009, pp. 1082-1098.

[15] S. L. Hilty, W. L. Brown and G. Tudor, “A Guide to the Birds of Columbia,” Princeton University Press, Prince-ton, 1986.

[16] R. S. Ridgely and P. J. Greenfield, “The Birds of Ecua-dor,” Cornell University Press, Ithaca, 2001.

[17] M. A. Plenge, “Notes on Some Birds in West-Central Peru,” Condor, Vol. 76, 1974, pp. 326-330.

[18] T. S. Weichler, S. Garthe, G. Luna-Jorquera and J. Mo-raga, “Seabird Distribution on the Humboldt Current in Northern Chile in Relation to Hydrography, Productivity, and Fisheries,” ICES Journal of Marine Sciences, Vol. 61,

J. BURGER ET AL.


147

No. 1, 2004, pp. 148-154.

[19] D. C. Duffy, “The Foraging Ecology of Peruvian Sea-birds,” The Auk, Vol. 100, No. 4, 1983, pp. 800-810.

[20] J. Burger, “Foraging Efficiency in Gulls: A Congeneric Comparison of Age Differences in Efficiency and Age of Maturity,” Studies in Avian Biology, Vol. 10, No. 225, 1987, pp. 83-89.

[21] J. Burger, “Foraging Behavior in Gulls: Differences in Method, Prey, and Habitat,” Colonial Waterbirds, Vol. 11, No. 1, 1988, pp. 9-23.

[22] D. A. Shealer, “Foraging Behavior and Food of Sea-birds,” In: E. A. Shreiber and J. Burger, Eds., Biology of Marine Birds, CRC Press, Boca Raton, 2001, pp. 137- 178.



Regional Coordination for under Frequency Load Shedding

Mohamad Ahmad Anuar, Hassan Bevrani, Takashi Hiyama Department of Computer Science and Electrical Engineering, Kumamoto University, Kumamoto, Japan

E-mail: [email protected] Received April 12, 2010; revised May 25, 2010; accepted July 2, 2010

Abstract Frequency deviation can be used as an indicator of imbalance between supply and demand. When generation is insufficient, it can cause frequency decline in a power system operation. Implementing under frequency load shedding (UFLS) is one of the common methods to overcome this problem. This paper proposes a novel approach for adaptive load shedding. The concept is an extension of shared and targeted load shedding using reserve margin. The optimal system configuration is then selected from those candidates to fulfill operational objectives. Operational constraints related to system parameters, threshold frequency, total of load shed and control area including line capacity are considered. An example using four sub-areas connected to an exter-nal system shows that the proposed regional coordination as an adaptive UFLS is feasible. Keywords: Load Shedding, Multi-Area Power Systems, Load Frequency Control, Emergency Control

1. Introduction Frequency deviation can be used as an indicator of imbalance between generation and demand. At the same time, it is needed to make sure that the system frequency is in allowable range. Transmission operator or balancing authority should ensure that the transmission system is operated so that instability, uncontrolled separation or cascading outages will not occur as result of the most severe single contingency and specified multiple contingencies [1]. Practically, transmission operator or balancing authority has the capability and authority to shed load rather than the risk of an uncontrolled failure in the intercon- nection when generation or transmission capacity is in-sufficient. The operators of large scale electrical power systems must be constantly alert of possibilities of a sys-tem failure. This was one of the reasons of cascading problem which occurred in North America blackout on August 14, 2003 [2]. The system experienced asynchro-nous oscillation which lasted for about 1 min 40 s, and no out-of-step relays acted to island the asynchronous system and settle the oscillation. When asynchronous oscillation exists for such a long time, surely the power system will experience cascading tripping of generators, and the system blackout will happen because of load- generation imbalance.

Previous studies on the load shedding scheme can be categorized into static and adaptive schemes [3]. In static

scheme, a certain amount of load is shed when the sys-tem frequency falls below certain threshold. This scheme is the most simple and used by most utilities. Whereas, adaptive methods are used to consider the characteristics of the power system, generator dynamic behavior under large disturbance and nonlinear interacting generators [4-6]. Almost, both above described methods are based on frequency threshold and/or frequency gradient. The under frequency load shedding is triggered/initiated when the frequency drops below the frequency threshold. Frequency gradient provide an important slope as an index to predict the contingency and manage an appro-priate emergency control plan.

This paper proposes a method using adaptive under frequency load shedding. The methodology adopted in this method incorporating frequency response analysis, sys-tem parameters, frequency threshold, total of load shed and control area including line capacity in transmission lines. This paper is organized to describe the methodology in Section 2, a test case in Section 3, results and discussion in Section 4, and finally conclusions in Section 5.

2. Methodology 2.1. Frequency Response Analysis Figure 1 shows simplified frequency response model wh-

M. A. ANUAR ET AL.


149

ere ∆PL, ∆PC, Rsys, Msys(s) are the system load change, supplementary control, drooping characteristic, and gov-ernor-turbine dynamic model, respectively [4]. The sys-tem frequency deviation ∆f, equivalent inertia H, and equivalent load damping coefficient D are defined as follows:

N

i iN

i i

N

i i

N

i ii

DD HH

HfHf

11

11

,,

,/)( (1)

Since, the supplementary control dynamic is usually slower than emergency control dynamics, ∆PC, can be ignored in an emergency condition analysis. According to Figure 1, frequency deviation can be written as

)]()([2

1)( sPsP

DHssf Lm

(2)

or, taking the inverse Laplace transform,

)()(

)(2)()()( tfD

td

tfdHtPtPtP DLm

(3)

∆PD(t) shows the load-generation imbalance is propor-tional to the total load change. The magnitude of total load-generation imbalance immediately after the occur-rence of disturbance at t = 0+ s can be expressed as fol-lows:

)(

)(2

td

tfdHPD

(4)

where dtfd / is the frequency gradient in a power

system and is proportional to the magnitude of total load-generation imbalance. For initial rate of frequency change, from (2) with no speed governing, at t = 0+ s and ∆Pm = 0, can be reduced to,

DHs

sPsf L

2

)()( (5)

for a step change in the load by ∆PL, the Laplace trans-form of the load change is

SPsP LL /)( (6)

and rearrange Equation (5),

]2

1[

/)(

sD

Hs

DPsf L

(7)

and taking the inverse Laplace transform,

D

H

t

LL eD

PH

D

Ptf

2

2)

2()( (8)

Hence, the initial rate of frequency change at t = 0+ s is proportional to ∆PL/D,

DPdt

tfdL

t

/)(

0

(9)

As mentioned before, the main factor and parameters that control the behavior of the frequency are the amount of disturbance, damping D, and inertia H parameters. The effect of the later two parameters should be consid-ered in load shedding planning. From (9) it can be seen that increase in D causes a decrease in frequency gradi-ent. Therefore, higher value of D gives a higher stability and the final system frequency will be stabilized at a higher level. Furthermore, H does not influence the ini-tial amount of frequency gradient, but influences the system dynamics, and higher H may improve the system stability under conditions of disturbance. 2.2. Frequency Threshold and Load Percentages In normal condition, for most existing networks allow-able frequency deviation range can be ± 1% whereas in emergency condition it is ± 4% from nominal frequency. The selection of frequency threshold and the number of load shedding steps depend on the system. In this paper three steps, 1%, 2%, 3% step increment, is considered with the frequency deviation from 59.4 Hz to 58.2 Hz as shown in Figure 2. The amount of load to be shed in each step is 10% of total system load. This is because large turbine-generators of the system are not rated for continuous operation below 59.4 Hz. A load shedding program starting at 59.4 Hz would be more effective in minimizing the depth of the under frequency for the large disturbance and the first shedding frequency should not be too close to the normal frequency [6].

If the frequency is still below 59.4 Hz even after the three steps of under frequency load shedding have oc-curred, all appropriate areas shall coordinate additional manual load shed amounts with their transmission op-erator or balancing authority. If frequency continues to decline below 58.2 Hz, transmission operator or balanc-ing authority shall take any necessary action to arrest the frequency decline except the opening of transmission tie lines.

PC-

+ sysM (s)

PL

+

-2Hs +

1 f

Pm

Rsys

1

Figure 1. Simplified frequency response model.

f

normal Under

frequencyHz

fmax fmin

Figure 2. Frequency operating range.

M. A. ANUAR ET AL.


150

2.3. Total of Load Shed In a load shedding scheme, total amount of load shed can be considered as

)( ,reserveLDLS PPP (10)

where, ΔPD is the load disturbance and ΔPL,reserve is the reserve (secondary control reserve) capacity of the sys-tem with maximum allowable change of frequency 0.6 Hz as mentioned in Subsection 2.2 [7].

From (8), the relation between Δf(t) and ΔPL can be represented as

LD

Ht

PeD

H

Dtf

2

2

21)( (11)

2.4. Control Area Load Shedding A control area is an electrical system bounded by inter-connected (tie-line) to control generation for maintaining power interchange schedule and contributing to frequency regulation [8]. A significant decline in frequency may require the shedding of load in order to avoid widespread system outages and to minimize the risk of damage to equipment.

The three frequency threshold values (59.4 Hz, 58.8 Hz, and 58.2 Hz) are the same for all the sub-areas so that all entities would participate during a region-wide or multi-region load shedding. During planning we can de-termine the generation reliability of the power system. One of parameters for generation reliability is reserve margin (RM) which is defined as [9]

Load

LoadCapacity InstalledRM N

% (12)

where N is the number of sub-areas. The weight or con-tribution factor for the RM for each area can be obtained using (12). A new sequence for load shedding can be created by ranking the RM from the smallest to the larg-est.

NRMRMRM ,,, 21 (13)

The sub-area with RM1 contributes the most to the system unreliability because of less RM. Negative value of the RM indicates negative reserve, or in other word. The load is greater than generation in that particular sub-area. For system stability the load shedding opera-tion can be targeted, sequentially, starting from the sub- area with the least RM until the system frequency is sta-bilized to a new steady state condition. 3. Test Case The study system is composed of four sub-areas con-

nected to an external system. The configuration of the study system is shown in Figure 3. Area I and II are in-terconnected through a 500 kV tie-line. Area I consists of four sub-areas A–D. The sub-areas A–D have eight, five, seven, and three thermal units, respectively. So, external system is considered as Area II. The power system pa-rameters are considered similar to the practical system, which is described in detail in [10-12]. 4. Results and Discussion Table 1 shows the system parameters for four sub-areas connected to an external system. It is observed that sub- area C has the highest value of D and H (i.e. D = 0.0576 and H = 0.384 respectively) which implies that it has the highest stability margin in comparison to other sub-areas.

With the given parameters of generation and load, the RM for all the sub-areas can be calculated by using Equation (12) and displayed in Table 2. As seen in the table, sub-areas A and C have the lowest (RM = –42) and the highest (RM = 284.85) RM value respectively.

SuB-AREA A

SuB-AREAB

External

SuB-AREAC

SuB-AREA D

Figure 3. Two control area power system.

Table 1. System parameters.

System Parameter

Sub-area A

Sub-area B

Sub-areaC

Sub-areaD

D (Load Damping factor)

0.0352 0.02 0.0576 0.0352

H (System Inertia) 0.2347 0.133 0.384 0.2347

Table 2. Generation and load parameters.

Sub-area

A Sub-area

B Sub-area

C Sub-area

D

Load (pu) 0.6269 0.2388 0.1492 0.4776

Generation(pu) 0.362 0.2537 0.5742 0.3026

Reserve Margin (%) –42 6.24 284.85 –36.64

M. A. ANUAR ET AL.


151

The maximum reserved power available in Area I is 1500 MW (10% of peak load demand). For load shed-ding scheme we consider two cases of extreme test sce-narios i.e. a large load disturbance of 3000 MW in sub- area A and C. The total area load demand is much higher than the reserved power, whilst the primary and supple-mentary controls cannot maintain the frequency at the nominal value. Under this condition the system is under emergency condition and the UFLS scheme should be implemented to recover the system frequency.

4.1. Disturbance at Sub-Area A As seen in Table 2, sub-area A has the least RM and there-

fore a large load disturbance of 0.3 pu occurred and the implementations of UFLS are considered in the same area. Figure 4 shows the amount of load shed, the fre-quency deviation and frequency gradient in all the sub-areas. In the Figure 4(a), only one step load shed-ding (10% of the load) was implemented. It is sufficient enough to bring the system frequency back to the near normal allowable region as in Figure 4(b). Figure 4(c) shows the detection of frequency gradient in emergency condition.

Figure 5 shows the tie line and trunk line power flows under the test condition. Furthermore, it may also be not- ed that the trunk line power flows into sub-area A except for sub-area D, which actually imports power from sub- area A.

40 50 60 70 80 90 100 110 120 130 140 1500

0.1

0.2

(a)

P

UF

LS (

pu)

40 50 60 70 80 90 100 110 120 130 140 150

-0.6-0.4-0.2

0

(b)

f

(H

z)

-0.5

0

0.5

d f/

dt (

Hz/

s)Δ

PU

FLS

(pu

) Δ

f (H

z)

dΔf/d

t (H

z/s)

(a)

(b)

(c)

0 -0.2 -0.4 -0.6

0.5

0

-0.5

40 50 60 70 80 90 100 110 120 130 140 150

Time (sec)

40 50 60 70 80 90 100 110 120 130 140 150

40 50 60 70 80 90 100 110 120 130 140 150

0.2

0.1

0

Figure 4. Load shedding plan in sub-area A, frequency deviation and frequency gradient in all sub-areas of Area I respec-tively.

40 50 60 70 80 90 100 110 120 130 140 15

40 50 60 70 80 90 100 110 120 130 140 15

40 50 60 70 80 90 100 110 120 130 140 15

40 50 60 70 80 90 100 110 120 130 140 15

Ptie

[M

W]

40 50 60 70 80 90 100 110 120 130 140 150

time (s)

0

-2000

-4000

P BA

[M

W]

P DA

[M

W]

P CA

[M

W]

1000500

0-500

6000

4000

2000

0 -1000

-2000

-3000

40 50 60 70 80 90 100 110 120 130 140 150

40 50 60 70 80 90 100 110 120 130 140 150

40 50 60 70 80 90 100 110 120 130 140 150

Figure 5. Tie line and trunk power fluctuation load change in sub-area A.

M. A. ANUAR ET AL.


152 4.2. Disturbance at Sub-Area C A large load disturbance of 0.3 pu is considered. Since sub-area C has the highest system parameters and RM, the load shedding does not implement in the same area. Instead, load shedding was implemented in some other area. Due to low RMs and identical system parameters, load could be shed in either sub-area A or D. Figure 6 shows the amount of load shed, the frequency deviation

and frequency gradient in all the sub-areas. As seen in Figure 6(a), one step load shedding is implemented at sub-area D. Since sub-area D imports power from A, it is better to shed load in D. It is sufficient enough to bring the system frequency back to the near normal allowable region in Figure 6(b). The detection of frequency gradi-ent in emergency condition is shown in Figure 6(c).

Figure 7 shows the tie line and trunk line power flows under the test condition to support Figure 6. Usually

40 50 60 70 80 90 100 110 120 130 140 150

2

(a)

40 50 60 70 80 90 100 110 120 130 140 15

6420

(b)

5

0

5

ΔP

UF

LS (

pu)

Δf (

Hz)

dΔ

f/dt (

Hz/

s)

0.2

0.1

0

(b)

(c)

0 -0.2 -0.4 -0.6

0.50

-0.5

40 50 60 70 80 90 100 110 120 130 140 150

Time (sec)

(a)

40 50 60 70 80 90 100 110 120 130 140 150

40 50 60 70 80 90 100 110 120 130 140 150

Figure 6. Load shedding plan in sub-area D, frequency deviation and frequency gradient in all sub-areas of Area I respec-tively.

40 50 60 70 80 90 100 110 120 130 140 150

40 50 60 70 80 90 100 110 120 130 140 150

40 50 60 70 80 90 100 110 120 130 140 150

40 50 60 70 80 90 100 110 120 130 140 150time[s]

Ptie

[M

W]

time (s)

2000

0

-2000

-4000

P BA

[M

W]

P DA

[M

W]

PCA

[M

W]

1000

0

-1000

6000

4000

2000

0

2000

0

-2000

40 50 60 70 80 90 100 110 120 130 140 150

40 50 60 70 80 90 100 110 120 130 140 150

40 50 60 70 80 90 100 110 120 130 140 150

40 50 60 70 80 90 100 110 120 130 140 150

Figure 7. Tie line and trunk power fluctuation load change in sub-area C.

M. A. ANUAR ET AL.


153 sub-area C will transfer power to sub-area A before the disturbance took place. But, in this case, 0.3 pu distur-bance is manageable to load-generation imbalance (gen-eration is greater than load) as discussed in Table 2. So, sub-area C reduced power transfer to sub-area A after the disturbance. Consequently sub-area A reduced power transfer to sub-area D. Hence, Figure 7 is the evidence of the amounts of power transfer from sub-area C to A and sub-area A to D in a drastic reduction manner. 5. Conclusions Regional coordination in emergency conditions is very important for power system operation and security. These regions are interconnected to each other for improving reliability and reducing cost. Practically, each region has different generation and load. This condition will affect reserve margin. The reserve margin is used to identify a sequential load shedding. This paper shows that regional coordination for four sub-areas connected to an external system using adaptive under frequency load shedding is feasible. 6. Acknowledgements The authors would like to thank Graduate School of Sci-ence and Technology (GSST), Kumamoto University, Japan and MARA University of Technology (UiTM), Malaysia for continuous support of this research. 7. References [1] R. Baldick, B. Chowdhury, I. Dobson, Z. Dong, B. Gou,

D. Hawkins, H. Huang, M. Joung, D. Kirschen, F. X. Li, J. Li, Z. Y. Li, C.-C. Liu, L. Mili, S. Miller, R. Podmore, K. Schneider, K. Sun, D. Wang, Z. G. Wu, P. Zhang, W. J. Zhang and X. P. Zhang, “Initial Review of Methods for Cascading Failure Analysis in Electric Power Transmis-sion Systems,” IEEE Power and Energy Society General Meeting, Pittsburgh, 20-24 July 2008, pp. 1-8.

[2] Y. Liu and Y. Liu, “Aspects on Power System Islanding

for Preventing Widespread Blackout,” Proceeding of the 2006 IEEE International Conference on Networking, Sensing and Control, Ft. Lauderdale, 23-25 April 2006, pp. 1090-1095.

[3] J. J. Ford, H. Bevrani and G. Ledwich, “Adaptive Load Shedding and Regional Protection,” International Jour-nal of Electrical Power & Energy Systems, Vol. 31, No. 10, 2009, pp. 611-618.

[4] H. Bevrani, G. Ledwich and J. J. Ford, “On the Use of df/dt in Power System Emergency Control,” Proceedings of IEEE Power Systems Conferences and Exposition, Se-attle, 15-18 March 2009, pp. 1-6.

[5] T. Tomsic, G. Verbic and F. Gubina, “Revision of the Underfrequency Load-Shedding Scheme of the Slovenian Power System,” Electric Power Systems Research, Vol. 77, No. 5-6, 2007, pp. 494-500.

[6] S.-J. Huang and C.-C. Huang, “Adaptive Load Shedding Method with Time-Based Design for Isolated Power Sys-tems,” International Journal of Electrical Power & En-ergy Systems, Vol. 22, No. 1, 2000, pp. 51-58.

[7] H. Bevrani, “Robust Power System Frequency Control,” Springer, New York, 2009.

[8] M. A. Anuar, U. Dorji and T. Hiyama, “Principle Areas for Islanding Operation Based on Distribution Factor Matrix,” The 15th International Conference on Intelligent Systems Applications to Power Systems, Curitiba, 8-12 November 2009, pp. 1-6.

[9] NERC, Assessments & Trends, Reliability Indicators, “ALR 1-3 Planning Reserve Margin”. http://www.nerc. com

[10] T. Hiyama, S. Koga and Y. Yoshimuta, “Fuzzy Logic Based Multi-Functional Load Frequency Control,” Pro-ceedings of the IEEE PES Winter Meeting, Singapore, 23-27 January 2000, Vol. 2, pp. 921-926.

[11] T. Hiyama and G. Okabe, “Coordinated Load Frequency Control between LFC Unit and Small Sized High Power Energy Capacitor System,” Proceedings of the Interna-tional Conference on Power System Technology, Singa-pore, 21-24 November 2004, pp. 1229-1233.

[12] H. Bevrani, G. Ledwich, Z. Y. Dong and J. J. Ford, “Re-gional Frequency Response Analysis under Normal and Emergency Conditions,” Electric Power System Research, Vol. 79, No. 5, May 2009, pp. 837-845.



Classification of Power Quality Disturbances Using Wavelet Packet Energy Entropy and LS-SVM

Ming Zhang, Kaicheng Li, Yisheng Hu College of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan, China

E-mail: [email protected] Received April 11, 2010; revised May 22, 2010; accepted June 27, 2010

Abstract The power quality (PQ) signals are traditionally analyzed in the time-domain by skilled engineers. However, PQ disturbances may not always be obvious in the original time-domain signal. Fourier analysis transforms signals into frequency domain, but has the disadvantage that time characteristics will become unobvious. Wavelet analysis, which provides both time and frequency information, can overcome this limitation. In this paper, there were two stages in analyzing PQ signals: feature extraction and disturbances classification. To extract features from PQ signals, wavelet packet transform (WPT) was first applied and feature vectors were constructed from wavelet packet log-energy entropy of different nodes. Least square support vector ma-chines (LS-SVM) was applied to these feature vectors to classify PQ disturbances. Simulation results show that the proposed method possesses high recognition rate, so it is suitable to the monitoring and classifying system for PQ disturbances. Keywords: Power Quality (PQ), Wavelet Packet Transform (WPT), Wavelet Packet Log-Energy Entropy,

Least Square Support Vector Machines (LS-SVM)

1. Introduction The deregulation polices in electric power systems re-sults in the absolute necessity to quantify power quality (PQ). This fact highlights the need for an effective rec-ognition technique capable of detecting and classifying the PQ disturbances. Traditionally PQ recordings are analyzed in the time-domain by skilled engineers. How-ever, PQ disturbances may not always be obvious in the original time-domain signal. One of the traditional signal processing techniques called Fourier transform provides information in frequency-domain but it does have limita-tions. One crucial limitation is that a Fourier coefficient represents a component that lasts for all time. This makes Fourier analysis less suitable for non-stationary signals. Wavelet analysis, which provides both time and fre-quency information, can overcome this limitation. Unlike the Fourier transforms, the wavelet transform has a fully scalable window, which allows a more accurate local description and separation of signal characteristics [1]. The wavelet transform has been applied to the wide range of PQ signals analysis: feature extraction [2], noise reduction [3], and data compression [4]. Recently, The identification of PQ disturbances is often based on artifi-

cial neural network (ANN) [5], fuzzy method (FL) [6], expert system (ES) [7], support vector machines (SVM) [8], and hidden Markov model (HMM) [9]. Many of the studies proposed in the literature present that these tech-niques can use feature vectors derived from disturbance waveforms to classify PQ disturbances.

The types of PQ disturbances include the sag, inter-ruption, swell, harmonic, notch, oscillatory transient (Osc. transient) and impulsive transient (Imp. transient) (see Figure 1) [10]. In this paper, the combined tech-nique of wavelet packet transform (WPT) and least square support vector machines (LS-SVM) for PQ dis-turbances recognition is presented. Decision making is performed in two stages: feature extraction and LS-SVM as a classifier. Figure 2 shows the block diagram of the classification system. The details of each stage are de-scribed in the next sections. High accuracies were achieved by using the LS-SVM trained on the wavelet packet log-energy entropy of different nodes.

The rest of this paper is organized as follows. In Sec-tion 2, the feature extraction by WPT is explained. In Section 3, brief review of the LS-SVM with the mini-mum output coding (MOC) technique is presented.

In Section 4, the results of classification of the LS- SVM trained on wavelet packet log-energy entropy to

M. ZHANG ET AL.


155

Figure 1. Power quality disturbance waveforms: (a) Normal signal; (b) Sag; (c) Interruption; (d) Swell; (e) Harmonic; (f) Notch; (g) Oscillatory transient; (h) Impulsive transient.

Figure 2. Block diagram of the classification system.

the studied PQ disturbance signals are presented. Finally, conclusions are given in Section 5.

2. Feature Extraction Using WPT The purpose of the feature extraction process is to select and retain relevant information from original signals. The WPT was first applied to decompose the original PQ signals into frequency bands. One of the advantages of the WPT is that it is able to decompose signals at various resolutions, which allows accurate feature extraction fro- m non-stationary signals like PQ disturbances. The fea-tures of signals, such as wavelet packet energy entropy, were then extracted from these decomposed signals as feature vectors.

The wavelet transform decomposes a signal into a set of basic functions called wavelets. These basic functions are obtained by dilations, contractions and shifts of a unique function called wavelet prototype. Continuous wavelets are functions generated functions generated from one single function by dilations and translations of a unique admissible mother wavelet )(tψ :

)(1)(, abt

atba

−= ψψ (1)

where 0,, ≠ℜ∈ aba are the scale and translation parameters, respectively, and t is the time. The func-tion set ( )(, tbaψ ) is called wavelet family. It is common to employ both wavelet and scaling functions in the transform representation. In general, the scale and shift parameters of the discrete wavelet family are given by

=a ja0 and jakbb 00= , where j and k are inte-gers. The function family with discretized parameters becomes:

)()( 02/

0, kbtaat jjkj −= −− ψψ (2)

where )(, tkjψ is called the discrete wavelet transform (DWT) basis.

DWT analyzes the signal at different frequency bands, with different resolutions by decomposing the signal into a coarse approximation and detail information. DWT employs two sets of functions called scaling functions )(tϕ and wavelet functions )(tψ , which associated with low- pass and high-pass filters, respectively. The original sig-nal )(tx can be decomposed to:

∑ ∑∑=

+=J

jjk

kjk

ktkdtkctx

jj1

)()()()()( ψϕ (3)

where j is the level number of the wavelet decomposi-tion, Jj ,,2,1 L= with J the time of the wavelet decomposition. jc and jd are the approximation coeffi-cients and detail coefficients of )(tx , respectively.

Because the information in higher frequency compo-nents is important, the frequency resolution of DWT may

M. ZHANG ET AL.


156

not be fine enough to extract pertinent frequency infor-mation about the signal. The necessary frequency resolu-tion may be achieved by using WPT, an extension of the DWT. In the WPT, the wavelet detail at each level is, in addition to decomposition of only the wavelet approxi-mation in the regular wavelet analysis, further decom-posed in to its own approximation and detail components. By this process, some lower frequency contents leaked in the wavelet details at the previous level can be further sifted out at the current level and also the frequency resolution for signal analysis increases. As a result, the WPT may provide better accuracy in both higher and lower frequency components of the signal.

Figure 3 shows the wavelet packet decomposition tree for three levels ( 3=J ). For each level of decomposition the signal is filtered into approximate information of the signals (lower frequency component) and detail informa-tion (higher frequency component). If this procedure is repeated J times, a filter bank is created with J filters.

To evaluate the importance of the wavelet packet components to a signal, the concept of entropy is often ap-plied in signal processing and there are various defini-tions of entropy in the literature. Among them, two rep-resentative ones are used in the present article, i.e. the energy entropy and the Shannon entropy. The wavelet packet energy entropy at a particular node n in the wave-let packet tree of a signal is a special case of p = 2 of the p-norm entropy, defined as

)1(, ≥= ∑ pwcEntp

kknn (4)

where knwc , denotes the wavelet packet coefficients cor-responding to node n at time k. It was demonstrated that the wavelet packet energy has more potential for use in signal classification as compared to the wavelet packet coefficients alone. The wavelet packet energy represents energy stored in a particular frequency band and is mainly used in this study to extract the dominant fre-quency components of the signal.

The Shannon energy entropy and relative Shannon en-ergy entropy are defined respectively as [11]

Figure 3. Wavelet packet decomposition tree.

∑−=k

knknn wcwcEnts )log( 2.

2. (5)

nnornn EntsEntsREnts _/= (6)

where nnorEnts _ is the Shannon energy entropy of the normal signal corresponding to node n.

In this paper, one of the commonly used entropy, log- energy entropy is also defined as

∑=k

knn wcEntl )log( 2. (7)

The relative log-energy entropy is proposed as

nnornn EntlEntlREntl _/= (8)

where nnorEntl _ is the log-energy entropy of the normal signal corresponding to node n. 3. LS-SVM The second stage is the disturbances classification. Sup-port vector machine (SVM) can avoid the problems of over learning, dimension disaster and local minimum in the classical study method, and is applied in many classi-fication problems successfully [8,11]. According to the practice, [12] advanced by J. A. K. Suyken can overcome the disadvantage of slow training velocity in the large scale problem, as LS-SVM algorithm translates the qua- dratic optimization problem into that of solving linear equation set. Although a wide range of classifiers are available, we use LS-SVM in this paper.

We consider a training set of N data points kk yx , , Nk ,,2,1 L= , where n

kx ℜ∈ is the input data, ℜ∈ky is the thk − output data, the SVM constructs a deci-sion function that is represented by:

bxwxy T +=)( (9)

where the dimension of w is not specified. It means that it can be infinitely dimensional. The separating hyper-plane that creates the maximum distance between the plane and the nearest data is called as the optimal sepa-rating hyperplane as shown in Figure 4.

In LS-SVM for the function estimation the following optimization problem can be given

∑=

+=N

kk

TLSebw

eCwwebwJ1

221

21

,,),,(min (10)

subject to the equality constraints Nkebxwy kk

Tk ,...,1, =++= (11)

where ke are slack variables and C is a positive real constant. One defines the Lagrangian

∑=

−++−=N

kkkk

TkLS yebxwJebwL

1)();,,( αα (12)

M. ZHANG ET AL.


157

wm 2

=

1wx b+ = +0wx b+ =

1wx b+ = −

w

m

Figure 4. Optimal separating hyper plane.

with Lagrange multipliers kα . The conditions for opti-mality are

1

1

0

0 0

0

0 0

k

k

N

k kk

N

kk

k k

Tk k k

L w xw

Lb

L eeL w x b e y

α

α

α γ

α

=

=

∂ = → =∂∂ = → =∂ ∂ = → =∂

∂ = → + + − =∂

∑

∑ (13)

for Nk ,,2,1 L= . It can be written immediately as the solution to the following set of linear equations:

0 0 0

0 0 0

0 0

0

T

w

b

γ

−

− = −

I X

I 1

I I I e

α YX 1 I

r

r

(14)

with ],...,[X 1 Nxx= , ],...,[Y 1 Nyy= , ]1,...,1[1 =r

, =e

1[ ,..., ]Ne e and ],...,[α 1 Nαα= . The solution is finally given by

=

+

−− YαIXX1

1 001

bT

T

γr

r (15)

with kk

k xw ∑= α , Ce kk /α= . The support values

kα are proportional now to the errors at the data points. So far we explained the linear case. SVM’s with

polynomials, splines, radial basis function networks, or multilayer perceptrons as kernels are obtained after map-ping the input data into a higher dimensional space by

)( kxφ , where )(⋅φ : hnn ℜ→ℜ . The number hn does not have to be specified because of the application of

Mercer’s condition, which means that )()(),( j

Tkjk xxxxK φφ= (16)

can be imposed for these kernels. Finally, the nonlinear function takes the form:

bxxKxyN

kkk += ∑

=1),()( α (17)

where the parameters kα , b follow from (15) after

replacing jT

k xx by ),( jk xxK . Multi-class classification was realized by the combi-

nation of LS-SVM classifiers with the minimum output coding (MOC) technique. In the MOC technique, up to

m2log (where m is the number of classes) LS-SVM clas-

sifiers were trained, and each of them aimed to separate a different combination of classes. There were eight classes (normal signal, sag, interruption, swell, harmonic, notch, oscillatory transient and impulsive transient) in this study, so three classifiers were necessary to differen-tiate them. The coding was defined by the codebook represented by a matrix, where the columns represent the different classes, and the rows indicate the results of the binary classifiers. The multi-class classifier output code for a pattern is a combination of targets of these three classifiers. In this study, the eight classes were encoded in the following codebook of minimum output coding:

T

codebook

CCCCCCCC

−−−−−−−−

−−−−=

111111111111111111111111

87654321

where 8,7,6,5,4,3,2,1 CandCCCCCCC are normal signal, sag, interruption, swell, harmonic, notch, oscilla-tory transient and impulsive transient, respectively. 4. Simulation Analysis To test classification results for PQ disturbances, the testing samples of these PQ disturbances have been gen-erated using algebraic equations [14]. The advantage of using algebraic equations for evaluation is the flexibility of adjusting signal noise contents as well as various waveform parameters such as the disturbance occurrence time, harmonic contents, sag depth, etc.

These disturbance waveforms are generated at a sam-pling rate of 256 samples/cycle for a total of 2560 points (10 cycles). In order to create different disturbance cases, some unique parameters such as starting time, magnitude, duration, frequency, and damping are allowed to change randomly. The random generation of signals is helpful for the testing of the classification more reliable since none of these attributes is fixed for real distribution sys-

M. ZHANG ET AL.


158

tem disturbances. Using wavelet packet decomposition, each signal shown

above was decomposed to level 3. The wavelet ‘Daub4’ was selected because it is more adequate for classifica-tion of PQ disturbances [13]. The wavelet packet energy entropy of different nodes of the decomposed signals were calculated, which could be used to identify the type of PQ disturbances. The performances of difference wavelet packet energy entropy for feature sets are shown in Figure 5. From above Figure 5, we can conclude that relative log-energy entropy is more effective than traditional relative Shannon energy entropy, which can am-plify the errors among the feature vectors. These features consist of 8-dimension feature space.

In this paper, we construct a LS-SVM by using radial basis function (RBF) as kernel function in LS-SVM pro-

posed above.

)2

exp(),( 2

2

σji

ji

xxxxK

−−= (18)

where σ is the width of the kernel. For training the SVMs with RBF kernel functions, one

has to predetermine the σ values. The optimal or near optimal σ values can only be ascertained after trying out several, or even many values. Beside this, the choice of C parameter in the SVM is very critical in order to have a properly trained SVM. The SVM has to be trained for different C values until to have the best result. From the Figure 6, It is found that the near optimal val-ues are 12 =σ and 4=C .

node (a)

×105

node (b)

×104

node (c)

node (d)

Figure 5. Performance comparison of difference wavelet energy entropy of the waveforms in Figure 1: (a) Wavelet packet Shannon energy entropy; (b) Relative wavelet packet Shannon energy entropy; (c) Wavelet packet log-energy entropy; (d) Relative wavelet packet log-energy entropy.

M. ZHANG ET AL.


159

Each decomposed signal now has eight features ( J 3= ). The feature vectors of PQ disturbances are fed to

the LS-SVM for classification. The LS-SVM topology used for classification is shown in Figure 7. We trained three different LS-SVMs (LS-SVM1, LS-SVM2, LSSV- M3) for seven different PQ disturbances (seven hundred samples of various PQ disturbances).The patterns to be distinguished from others are represented by +1 and the remaining patterns represented by -1 for both training and testing procedures.

The output of three different LS-SVMs constructs the code of the input PQ signals, which the type of a distur-bance or the normal signal will be identified. In the pre-sent work a standard feed-forward network with 8 input neurons, 12 hidden neurons, and 7 output neurons was compared to the LS-SVM implementation. Furthermore, our results indicate that solutions obtained by LS-SVM training seem to be more robust with a smaller standard

error compared to standard ANN training using the same features as inputs.

The other seven hundred PQ disturbances of various types have been generated for the testing. The classifica-tion results in a correct identification rate of 97.7% are shown in Table 1 using the proposed LS-SVM classifier. For comparison purposes, the total classification accura-cies on the same test sets and the CPU times of training of the two classifiers are presented in Table 2. It is found that the proposed LS-SVM classifier performed better than the standard ANN classifier.

To evaluate the performance of the kernel function, three LS-SVM classifiers were developed based on the linear kernel, the polynomial kernel, and the RBF kernel. The classification results with linear, polynomial and RBF kernel are shown in Table 3. The accuracy of clas-sification is high in RBF kernel in comparison with the polynomial and linear kernels.

Figure 6. Comparison of accuracy acquired with different C and 2σ values for RBF kernels.

Table 1. Classification results using the proposed LS-SVM classifier.

Type of PQ

disturbances

Number of disturbances


classified

Number of disturbances misclassified

Classification Accuracy

(%)

Sag 100 97 3 97

Interruption 100 97 3 97

Swell 100 99 1 99

Harmonic 100 98 2 98

Notch 100 99 1 99

Osc. transient 100 97 3 97

Imp. transient 100 96 4 96

Sum 700 684 16 97.7

Table 2. Comparison of the classification indices between the LS-SVM and ANN classifiers.

Classifier Training set samples

Testing set samples

Mean training time (s)

Mean testing ime (s)

Mean correct

ratios (%)

LS-SVM 700 700 9.968 1.922 97.7

ANN 700 700 101.523 1.993 95.2

Table 3. Classification accuracies for the different kernels used.

Kernel used

Number of disturbances in training


in testing

Number of disturbances misclassified

Classifica-tion

accuracy (%)

Linear 700 700 27 96.1

Polynomial 700 700 20 97.1

RBF 700 700 16 97.7

[ ]c3c2c1 YYYCodebook' =

C8C7C6C5C4C3C2C1Decision

Figure 7. Classification of PQ disturbances based on MOC (Codebook’ is one column of Codebook).

M. ZHANG ET AL.


160 5. Conclusions In this paper, an attempt has been made to extract efficient features of the PQ disturbances using WPT and to classify the disturbances using LS-SVM with the MOC technique. It is also found that relative wavelet packet log-energy entropy is considered as feature vectors, which are suitable for classification of PQ disturbances. For comparison different classifiers, the LS-SVM and ANN classifiers were implemented to deal with the same classification. The classification accuracies and the CPU times of training showed that the LS-SVM classifier produces considerably better performance than that of the ANN classifier. 6. Acknowledgements The authors would like to thank to the support of Wuhan Xinlian Science and Technology Ltd. 7. References [1] S. Mallat, “A Wavelet Tour of Signal Processing,” Aca-

demic Press, San Diego, California, 1998. [2] S. Santoso, E. J .Powers and P. Hofman, “Power Quality

Assessment via Wavelet Transform Analysis,” IEEE Transaction on Power Delivery, Vol. 11, No. 2, 1996, pp. 924-930.

[3] H. T. Yang and C. C. Liao, “A De-Noising Scheme for Enhancing Wavelet-Based Power Quality Monitoring System,” IEEE Transaction on Power Delivery, Vol. 16, No. 3, 2001, pp. 353-360.

[4] S. Santoso, E. J. Powers and W. M. Grady, “Power Qual-ity Disturbance Data Compression Using Wavelet Trans-form Methods,” IEEE Transaction on Power Delivery, Vol. 12, No. 3, 1997, pp. 1250-1257.

[5] A. K. Ghosh and D. L. Lubkeman, “The Classification of Power System Disturbance Waveforms Using a Neural

Network Approach,” IEEE Transaction on Power Deliv-ery, Vol. 10, No. 1, 1995, pp. 109-115.

[6] T. X. Zhu, S. K. Tso and K. L. Lo, “Wavelet-Based Fuzzy Reasoning Approach to Power Quality Distur-bance Recognition,” IEEE Transaction on Power Deliv-ery, Vol. 19, No. 4, 2004, pp. 1928-1935.

[7] M. B. I. Reaz, F. Choong, M. S. Sulaiman, F. Mohd- Yasin and M. Kamada, “Expert System for Power Qual-ity Disturbance Classifier,” IEEE Transaction on Power Delivery, Vol. 22, No. 3, 2007, pp. 1979-1988.

[8] P. Janik and T. Lobos, “Automated Classification of Power Quality Disturbances Using SVM and RBF Net-works,” IEEE Transaction on Power Delivery, Vol. 21, No. 3, 2006, pp. 1663-1669.

[9] J. Chung, E. J. Powers, W. M. Grady and S. C. Bhatt, “Power Disturbance Classifier Using a Rule-Based Method and Wavelet Packet-Based Hidden Markov Model,” IEEE Transaction on Power Delivery, Vol. 17, No. 1, 2002, pp. 233-241.

[10] IEEE Recommended Practice for Monitoring Electric Power Quality, IEEE Standards Description: 1159-1995, 2009.

[11] G. S. Hu, F. F. Zhu and Z. Ren, “Power Quality Distur-bance Identification Using Wavelet Packet Energy En-tropy and Weighted Support Vector Machines,” Expert Systems with Applications, Vol. 35, No. 1-2, 2008, pp. 143-149.

[12] J. A. K. Suykens and J. Vandewalle, “Least Squares Sup-port Vector Machine Classifiers,” Neural Processing Let-ter, Vol. 9, No. 3, 1999, pp. 293-300.

[13] N. S. D. Brito, B. A. Souza and F. A. C. Pires, “Daube-chies Wavelets in Quality of Electrical Power,” 8th In-ternational Conference on Harmonics and Quality of Power, Athens, 14-18 October 1998, pp. 511-515.

[14] T. K. Abdel-Galil, M. Kamel, A. M. Youssef, E. F. El-Saadany and M. M. A. Salama, “Power Quality Dis-turbance Classification Using the Inductive Inference Approach,” IEEE Transaction on Power Delivery, Vol. 19, No. 4, 2004, pp. 1812-1818.



Constraints Based Decision Support for Site-Specific Preliminary Design of Wind Turbines

Abdelaziz Arbaoui1, Mohamed Asbik2 1M2I - ENSAM Meknès Ismaïlia, Meknès, Maroc

2Laboratoire de Physique des Matériaux et Modélisation des Systèmes, Unité associée au CNRST-URAC, Faculté des Sciences, Zitoune, Meknès, Maroc

E-mail: abdelaziz_arbaoui, [email protected] Received April 7, 2010; revised May 21, 2010; accepted July 1, 2010

Abstract This study presents a decision-support tool for preliminary design of a horizontal wind turbine system. The function of this tool is to assist the various actors in making decisions about choices inherent to their activi-ties in the field of wind energy. Wind turbine cost and site characteristics are taken into account in the used models which are mainly based on the engineering knowledge. The present tool uses a constraint-modelling technique in combination with a CSP solver (numerical CSPs which are based on an arithmetic interval). In this way, it generates solutions and automatically performs the concept selection and costing of a given wind turbine. The data generated by the tool and required for decision making are: the quality index of solution (wind turbine), the amount of energy produced, the total cost of the wind turbine and the design variables which define the architecture of the wind turbine system. When applied to redesign a standard wind turbine in adequacy with a given site, the present tool proved both its ability to implement constraint modelling and its usefulness in conducting an appraisal. Keywords: Wind Turbine, Decision Support, Preliminary Design, Cost Modelling, Constraint Satisfaction

Problem (CSP), Digital CSP Solver

1. Introduction For the past fifteen years, horizontal axis wind turbine systems (HAWT) have developed at a fast pace. Because of the renewability and cleanliness of the energy pro-duced, incorporating such systems has become a key element in the new energy policies of many countries. Governments and non-trading companies show an im-portant interest in sustainable development through the extensive incorporation of wind energy into electricity generation systems. Distributors are interested in the viability and in the cost as well as the quality of the en-ergy produced. Aims of investors have been focalized on potential profits whereas designers, manufacturers and project managers define the architecture of the system and its fitness to the site.

Like all projects, a wind energy one is punctuated by successive phases with well-defined goals. In each phase, operations have to be performed and decisions have to be made by the various actors. Technical, economical, en-vironmental and political issues lead the actors to justify

their decision approach and search for decision-support means and tools. The main actors involved in the deci-sion making process in the preliminary design phase are investors and distributors. To make a decision, these ac-tors require external knowledge to their organisations. These are mainly within the competence of the project manager, manufacturer and scientist, and are needed to be translated into trends or estimations to be usable in the preliminary decision process. In addition, the character-istics of required data and models depend on the decision environment and inexpressible needs [1].

Various tools and software has been developed for wind energy systems. The objective of such tools is to maximise the performance and/or decrease the produced energy cost. Frequently, the strength properties and stresses of structures are all taken into account, with a finite-element and/or modal-analysis approach. Some tools use digital simulations to reproduce the aerody-namic characteristics of wind at the site. These ones fo-cus on designing and defining details of wind energy systems, and they are not designed to provide decision support during the preliminary design phase [2].

A. ARBAOUI ET AL.


162

This paper aims at presenting a knowledge base sys-tem for supporting decisions in the preliminary design of wind turbines. This tool is based on the development of a set of relations (called constraints) derived from engi-neering knowledge. Engineering knowledge has been related to the electrical energy production and the in-vestment costs of the wind turbine systems.

The development of the knowledge base system has been performed through three main steps (see Figure 1): Analysis and structuring of the design problem, Development of a model relating to the design

problem as a Constraint Satisfaction Problem, Implementation of the Constraint Satisfaction

Problem on a digital CSP solver based on interval analysis [3].

The knowledge base system aims at exploring the so-lution space of the design problem. This exploration process is not an optimisation process since every solution satisfying the whole set of constraints of the problem is regarded as a solution. Provided that the size of the solu-tion space is reasonably wide, the exploration process may be complete, namely, the solver delivers the com-plete set of solution of the problem. Therefore, decision- makers are able to select wind turbines among a list. During this selection process, they are able to take into consideration some preferences resulting from their knowledge, which may be out of the scope of the model. 2. Constraint Satisfaction Problem Solvers Digital processing tools of the Constraint Satisfaction Problem (CSP) solver type have recently been developed to cope with the difficulties presented by preliminary design. These tools are based on the notion of constraint, which converts the designer’s knowledge into the form of conditions of compatibility between the variables of a design problem. Specific requirements of the industry, criteria of functional specifications and physical behav-iour can all be described by the constraints. Generally, we call a Constraint Satisfaction Problem any problem that can be described in terms of a set of relationships

Figure 1. Applied approach.

called constraints “C”, variables “V” and domain values “D”. Values assigned to the variables must belong to their respective domains while still satisfying problem constraints [4].

CSP solvers deal with problems integrating a large number of variables with values that evolve in the con-tinuous space of real values. These variables represent dimensions, state variables (pressure, temperature, etc.) or performance criteria (costs, yields, etc.). The solvers are called digital CSP solvers. The sort of problems en-countered in preliminary design also integrates variables that evolve in discrete domains, such as lists of concept, components or materials. Using mixed CSP solvers, con-tinuous and discrete variables can be treated together.

The value domains assigned to the variables are inter-vals or unions of real intervals for the real value vari-ables and enumerated sets or unions of integer intervals for the discrete variables. These domains can be left fairly broad so that no potential solutions to the design problem are eliminated.

The constraints traditionally used to represent the de-signer’s input can be divided into three categories:

- Equal constraints (type “X = Y”) usually represent laws of physics or definitions of performance criteria,

- Unequal constraints (type “X < Y”) usually represent economic constraints (costs), required space, etc.

- Logical constraints (type “W → (X and Y) or Z”) represent conditional constraints such as technical skill rules, selection of components from catalogues, etc.

Constraints as used in constraint programming are re-lations which restrict the variable domains. The relation-ships that we take into account are algebraic ones which can integrate basic functions (trigonometric, logarithmic, etc.). A knowledge database is a file which indexes all the constraints and the domains assigned to the variables in a design problem. This database is digitally processed by a CSP solver which calculates the domain solutions for each variable that satisfies all the constraints in the problem.

In this study, we use the “Constraint Explorer®” soft-ware. This software was developed in the context of pro-ject CO2 (RNTL French Project “Conception par Con-traintes”). It processes the knowledge bases in a two- phase iterative and sequential alternant which gradually reduces and partitions the domains assigned to the CSP variables. The domains are gradually reduced until they satisfy the stipulations defined by the solver user. Figure 2 shows these phases in an example with two constraints and two variables. The variables take their values from intervals limited by real values.

The phases, in digital processing, alternate phases of propagating, the constraints and bisecting variable do-mains. Calculations converge towards an external ap-proximation of the solution space, in other words, to-wards a set of value intervals assigned to the variables of the problem being modelled containing the solutions to the design problem.

A. ARBAOUI ET AL.


163

Figure 2. Constraint satisfaction problem solving taking into account 2 constraints and 2 variables.

This approach to solving Constraint Satisfaction Prob-

lems enables us to gradually limit the space containing design solutions, rather than testing different alternative solutions individually and validating them by simulating their functioning. This approach is extremely suitable for preliminary design problems where the aim is to select architectures for studied systems. Also, the solutions domain is explored in its entirety.

Digital processing moves towards a set of domains as-signed to all the variables of the problem defining all the alternatives solutions to the design problem. Where “S” is the set of solutions to the design problem:

ni SSSS ,,,,1 (1)

Each solution is a set of “n” values assigned to “m” variables “Vj” of the problem:

VVV S ni im

ij

ii ,,,,,,1 1 (2)

Values assigned to variables are intervals for variables defined in real domains and integers for variables de-fined in integer domains:

min max

1, , 1, ,

: ,

int :

i i i ij j j j

i i ij j j

i n j m

If V is real V V V

If V is eger V V

(3)

3. Analysis and Structuring of HAWT

Design Problem The wind turbine design problem gives rise to particular

difficulties as wind turbines employ different technolo-gies and concepts. Considering the multiplicity of poten-tial choices, the interaction between the various parame-ters of the problem and the viewpoints to be taken into account, defining a wind turbine appropriate to the site proves quite difficult. In practice, these difficulties rise in anticipating and quantifying the consequences of a given choice. Such difficulties may result in an improper selec-tion of the standard machine and lead to an omission of the potential profits guaranteed by a site specific design [5,6].

Within the general category of horizontal axis wind turbines for grid applications there exists a great variety of possible rotor configurations, power control strategies and braking systems. Inevitably, there are situations in which decisions in one area can affect others. Alongside with these discrete design choices, there are several fun-damental design variables, such as rotor diameter, ma-chine rating and rotational speed, which also have to be established at the start of the design process. Continuous variables such as these lend themselves to mathematical optimization [7]. In this study, the following design variables are chosen to define a horizontal axis wind tur-bine: the nominal power, Pn the hub height, Hhub the rotor diameter, D the rotational speed, N the design speed, Vdes the number of blades, p Control type: the present tool can be applied to

constant-speed “stall” (CSS), constant-speed “pit-

A. ARBAOUI ET AL.


164

ch” (CSP), and variable-speed “pitch” (VSP) sys-tems.

All of these design variables are often given in manu-facturers catalogues.

The need related to the preliminary design of HAWT, consists in being able to characterize these configurations of design the ones compared to the others. We use the quality index, which is the ratio of the electricity pro-duced on the total cost of the wind turbine, to choose the best solutions.

WTC

EQI (4)

To take safety problems into account, the distance between the tip of the blade and the ground should be equal to or more than 15 m:

hubHD

152

(5)

To limit aerodynamic noise from the rotor, blade tip linear speed cannot exceed 80 m/sec:

80120

2

NDVtip

(6)

At this stage of problem definition, generating the constraints related to the cost of wind turbine and those related to the amount of energy produced is sufficient to start the solving phase. 4. HAWT Cost Model The cost model of wind turbine encompasses the aspects related to the design and manufacture of such systems. It is the sum of cost models of the components of the wind turbine. A calibration factor FWT allows using real wind turbine costs [6].

1.1,_ WTi

icomponentWTWT F CFC (7)

Flowcharts were used to identify the models cost of all

components, Figure 3 for example, shows the flowchart used for the rotor.

The choice of level 2 in the flowchart of the rotor is justified to distinguish, firstly, the “pitch” and “stall” concept. In fact, the later encloses in his blade the tip braking mechanism, whereas the first contains the pitch mechanism in his hub. On the other hand, a two-blade rotor must contain the teeter mechanism to compensate his dynamic behavior.

The cost of some components is calculated from weight models developed using engineering estimation rules. These have been applied to the rotor, the transmis-sion system, the nacelle, and the tower. As for the cost of the generator and associated electrical equipment, it is correlated with power rating. All models are calibrated (specific costs) to match the costs market of the compo-nents [8,9]. 5. Annual Electricity Produced Model The amount of calculated electricity depends on the en-ergy available on the site, at the level of the tower, the speed and geometric characteristics of the rotor, the out-put of the power unit, and the start/stop wind speeds of the wind turbine.

The wind in the site is defined as the following Wei- bull distribution:

k

c

Vk

ec

V

V

kVf

)( (8)

Scale parameter c characterises wind average speed, whereas shape parameter k characterises wind distribu-tion which varies with height [10]:

02.003.0)( 0 ZkZk (9)

where k0 is the shape parameter at wind-measurement height Z0.

The vertical gradient of wind speed is considered by introducing the following power law:

“Stall” “pitch” “2 blade”

Figure 3. Flowchart of the rotor.

A. ARBAOUI ET AL.


165

00 Z

Z

c

c (10)

c and c0 are the scale parameters at heights Z and Z0 and is considered constant.

The power recovered by a wind turbine is:

3

2

1VACP e (11)

The efficiency factor depends on both wind speed and system architecture [2]:

2

2

ln2

lnlnexp)(

s

VVCVC des

eme (12)

In this expression, the system is characterised by its maximum efficiency Cem, its optimum operating speed Vdes (design speed), and its operating range s. The nomi-nal power of the wind turbine is given by:

21 9exp ln

2 2n em desP C AV s

(13)

Cem is calculated from the performance of the power conversion unit:

gmpem CC max (14)

The maximum value of Cp is calculated using an ana-lytical relationship [11]:

z

x

P

C

C

p

p

p

p

C

max

2max

2maxmax

67.0

67.0max

max

21

92.1

0025.004.048.1

593.0

(15)

where

desV

ND

60max

(16)

The efficiency of the gearbox is given by [9]:

4/311

P

P n

mm (17)

with 012.089.0 nm P (19)

The efficiency of the generator is given by [9]:

P 6

P

P

P

m

ng

ng

mgg

1511

2

(20)

with 014.087.0 ng P , (21)

and

sgmnng FPP (22)

In this last expression, Fs represents the service factor of the gearbox, which is defined by the following logical constraint:

25,1.

75,1.

2.

s

s

s

FVSPtypeControl

FCSPtypeControl

FCSStypeControl

(23)

Therefore, the annual electricity output in kWh/year of the wind turbine having a rotor with a surface area A, and the start/stop wind speeds (Vi and Vf), is the sum of the energies produced in one year (8,760 hours) which is reduced by the efficiency factor of the system )(VCe :

VVVCVfAEf

i

V

Veap

3)()(

2

760.8 (24)

6. Preliminary Design of a Horizontal Wind

Turbine The decision-making actors need the following data: The Criteria (Cr) are the total cost of the wind

system and the quantity of annual electricity pro-duced. These two criteria allow the calculation of the quality index for a given configuration of wind turbine system.

The design variables represent the parameters serving to define the architecture of the wind sys-tem (Pn, Hhub, D, N, Vdes, Control type and p).

To define the relevance indicators of the solutions, the standard system VESTAS V39-500 is used. It corre-sponds to the ratio of the criteria values obtained by the total model on the values of criteria of the standard sys-tem.

dards

ii Cr

CrRI

tan

(25)

The principle objective of the decision makers is to control the influence of design variables on the criteria. They often seek a machine which has a largest quality index but also which maximizes the annual produced electricity. Furthermore, the search for solutions (satis-faction of all the constraints) is carried out by using the “Constraint Explorer®” solver. According to the need, this tool can modify the fields of the design variables values and the variables which characterize the site.

In this study, a site whose characteristics are given in Table 1 has been chosen. To obtain a good judgement of total field of the solutions, we introduced the variation domain of the design variables gathered in Table 2.

A. ARBAOUI ET AL.


166

Table 1. Characteristics of the investigated site.

Table 2. Design variables and their domain of variation.

Design variable Domain of variation

D (m) [20,80] with a step of 10 m

Pn (kW) [400,2000] with a step of 100 kW

Vdes (m/s) [6,12] with a step of 2 m/s

Hhub(m) [35,70] with a step of 10 m

N (tr/mn) [15,50] with a step of 5 tr/mn

Control type “PVC” or “SVC” or “PVV”

P 2 or 3

Figure 4 shows the Pareto space of solutions obtained.

It also shows some of the best solutions that we have chosen, in addition to standard system which appears as a solution of the problem, too. The best solutions chosen in Pareto front compared with the standard system are exhibited in Table 3.

These results reveal that an increase of the rotor di-ameter causes a diminution of the quality index. Fur-thermore, the increase of this geometrical parameter is associated with an important nominal power output, a weaker rotational speed and a higher tower. These results

are in agreement with those of the reference [7]. We notice that all the given solutions in the table 3 are

two blades with pitch variable speed control. We will return to justify this predominance in the continuation of this paper. Indeed, every solution of this table has a qual-ity index clearly higher than that of the standard system for the studied site. Then, the relevance indicators asso-ciated with the four solutions are respectively: 142.8%, 133.2%, 131.6%, and 118%. This means that the stan-dard system is not adapted to the studied site and hence a redesign in adequacy with the site is necessary.

To improve the performances of the standard machine, we propose to deal with 6 possible redesign scenarios with which we highlighted the influence of the design variables on the performances of the wind system: Scenario 1 (Modification of the rotor): The design

variables concerned with this scenario are the ro-tor diameter D and the design speed Vdes.

Scenario 2 (Modification of the gearbox and the generator): This scenario relates to the nominal power output Pn and the rotational speed N.

Scenario 3 (Modification of the number of blade p)

Scenario 4 (Modification of control type of the rotor): The objective is to compare a stall system with a pitch system.

Scenario 5 (Modification of control type of the generator): The objective is to compare a constant speed system with a variable speed system.

Scenario 6 (Modification of the whole wind sys-tem)

Figure 4. Field of solutions in the Pareto space.

Table 3. Criteria and design variables of standard system and some best solutions in Pareto front.

Criteria Design variable Wind Turbines QI E CWT D Pn Vdes Hhub N Control type p

Standard 3.61 1.37 0.38 39 500 8 40.5 30 CSP 3

Solution 1 5.24 1.19 0.23 30 700 10 35 46.8 VSP 2

Solution 2 4.89 2.21 0.45 40 1200 10 45 33.9 VSP 2

Solution 3 4.83 3.36 0.69 50 1700 10 45 24 VSP 2

Solution 4 4.33 4.59 1.06 60 2000 10 55 20.9 VSP 2

k0 c0 Z0

1.2 8 0.12 30

A. ARBAOUI ET AL.


167

The optimal solutions (with respect to the quality in-dex) obtained for each scenario are given in Table 4.

Thus, the optimal solution of the scenario 1 corre-sponds to the reduction of rotor diameter (18.5%) and an increase in the design speed which reaches 12.5%. As seen in Figure 5, there are also important opportunities to reduce the total cost of the wind system at the level of the nacelle, the tower and the foundation. These last op-portunities must be seized by the decision-makers to be able to improve the quality index of their machine (to reach 4.12 GWh/MEuro instead of 3.67 GWh/MEuro which is equivalent to an indicator of relevance equal to 112.3%). If we just reduce the rotor diameter the ob-tained quality index undergoes a lower reduction than that of the standard system (99%). The gains which must be carried out at the level of the other components will

certainly compensate the shortfall of the produced energy by a smaller rotor (82.5% only).

The retained solution of the scenario 2 provokes an increase of 60% in the nominal power and a weak reduc-tion of 1.6% in the rotational speed. Then, the augmenta-tion in the nominal power allows recovering more energy (126%) and renders the wind system more expensive (an increase of 116%). Figure 6 highlights that the rise of the wind system cost is not only due to the raised costs of the gearbox and the generator but also to the inevitable adaptation of the rotor and the nacelle. Indeed, the aug-mentation in the nominal power is accompanied by an increase in the weight supported by the nacelle and the tower and hence their costs. The rise of the hub and the flanges costs renders the rotor more expensive [9].

Table 4. Criteria and design variables of standard and best solutions for each scenario.

Criteria Design variable Wind Turbines QI E CWT D Pn Vdes Hhub N Control type p

Standard 3.61 1.37 0.38 39 500 8 40.5 30 CSP 3

Scenario 1 4.12 1.13 0.27 31.8 500 9 40.5 30 CSP 3

Scenario 2 4.02 1.73 0.43 39 800 8 40.5 29.5 CSP 3

Scenario 3 3.88 1.36 0.35 39 500 8 40.5 30 CSP 2

Scenario 4 3.7 1.37 0.37 39 500 8 40.5 30 CSS 3

Scenario 5 3.91 1.37 0.35 39 500 8 40.5 30 VSP 3

Scenario 6 5.24 1.19 0.23 30 700 10 35 46.8 VSP 2

Figure 5. Cost reduction in scenario 1.

Figure 6. Cost reduction in scenario 2.

A. ARBAOUI ET AL.


168

Now, let us comment on scenarios 3, 4 and 5 concern-ing the discrete variables design. In fact, for the scenario 3, a two-blade rotor is less heavy and hence less expen-sive although his blade is broader and thicker and that the teeter mechanism is integrated to his hub. The reduc-tion of rotor weight leads to a weak reduction of the cost of the nacelle. The possibilities of reducing the cost of the two components are illustrated in Figure 7. Accord-ing to the Equation (15), we can see that the power coef-ficient of a two-blade rotor is worse than that of three- blade system. This latter recovers more energy than the first rotor system but its production gain does not com-pensate its higher cost; so the quality index obtained is slightly lower than that of a two-blade system.

As for the scenario 4, a stall system seems to be the least expensive. This can be explained by the fact that its tip braking mechanism is less expensive than the pitch mechanism placed in the hub of the pitch system (see Figure 7). The use of a stall control raises the cost of the gearbox and generator. This is essentially due to the gearbox service factor which increases in the case of a stall system. On the other hand, the possible reduction in the rotor cost can not cover the rise in the gearbox and generator costs which gives a slightly higher quality in-dex for stall system [9].

The gearbox and the electric unit play also an impor-tant role in the scenario 5 which relates to the control

type of the generator. By using a variable speed control a decrease in the rotor cost becomes realistic, but the greatest part of this reduction is offered by the gearbox. This fact is due once again to the service factor of the gearbox which decreases in the case of variable speed system [9].

The last scenario has been examined as a combination of the previous ones. Then the wind system is considered as a two-blade system with a variable speed (scenarios 3 and 5). Its nominal power is higher than that of the stan-dard system (scenario 2) whereas its rotor diameter is smaller (scenario 1). In Figure 8, the cost reduction of this case is exposed. In spite of the considerable increase in the electric unit cost, the possible gains on the level of the other components of the system allow to have a less expensive wind turbine. Furthermore, even if energy produced is reduced because of the reduction of the rotor diameter, the quality index is much improved.

Finally, the Table 5 recapitulates the gains in the qual-ity index obtained for all the scenarios. These gains are more important for scenarios 1 and 2 (the design vari-ables concerned are: D, Vdes, Pn and N) with comparison to scenarios 3, 4 and 5 which relate to the discrete design variables (control type and the number of blade). The profit reaches its maximum value for the scenario 6, which represents a combination of the other scenarios.

Figure 7. Cost reduction for scenarios 3, 4 and 5.

Figure 8. Cost reduction for scenario 6.

A. ARBAOUI ET AL.


169

Table 5. Profits in the quality index obtained for all the scenarios.

Redesign Scenarios Gains

Scenario 1: Modification of the rotor 14%

Scenario 2: Modification of the gearbox and the generator 11%

Scenario 3: Modification of the number of blade 7.5%

Scenario 4: Modification of the control type of the rotor 2.5%

Scénario 5: Modification of the control type of the generator 8%

Scenario 6: Modification of the whole design variable 45%

7. Conclusions Decision support systems for the preliminary design of horizontal axis wind turbine is developed by taking into account the wind turbine components and site character-istics.

The present tool is mainly based on the engineering knowledge and it combines a constraint-modelling tech-nique with a solving method derived from artificial intel-ligence (digital CSPs). In this way, it generates solutions and automatically performs the architecture selection and gives the cost of wind turbine components.

The present study highlights the relevance of the site specific design in the decision making process. The im-provements achieved in terms of to the quality index are significant, this criteria is greatly affected by most of the design variables. When applied to redesign of standard wind turbine, our approach proved both its ability to im-plement constraint modelling and its usefulness to the various actors in conducting an appraisal. 8. References [1] J. F. Courtney, “Decision Making and Knowledge Man-

agement in Inquiring Organization: Toward a New Deci-sion-Making Paradigm for DSS,” Decision Support Sys-tems, Vol. 31, No. 1, 2001, pp. 17-38.

[2] C. T. Kiranoudis, N. G. Voros and Z. B. Maroulis, “Short-Cut Design of Wind Farms,” Energy Policy, Vol. 29, No. 7, 2001, pp. 567-578.

[3] D. Scaravetti, J. Pailhès, J.-P. Nadeau and P. Sébastian,

“Aided Decision-Making for an Embodiment Design Problem: Advances in Integrated Design and Manufac-turing in Mechanical Engineering,” Springer, Dordrecht, 2005.

[4] F. Benhamou and W. Older, “Applying Interval Arithme-tic to Real, Integer and Boolean Constraints,” The Jour-nal of Logic Programming, Vol. 32, No. 1, 1997, pp. 1- 24.

[5] P. Fuglsang, C. Bak, J. G. Schepers, T. T. Cockerill, P. Claiden, A. Olesen and R. Van Rossen, “Site-Specific Design Optimization of Wind Turbines,” Wind Energy, Vol. 5, No. 4, 2002, pp. 261-279.

[6] T. Diveux, P. Sebastian, D. Bernard, J. R. Puiggali and J. Y. Grandidier, “Horizontal Axis Wind Turbine Systems: Optimization Using Genetic Algorithms,” Wind Energy, Vol. 4, No. 4, 2002, pp. 151-171.

[7] T. Burton, D. Sharpe, N. Jenkins and E. Bossanyi, “Wind Energy Handbook,” John Wiley & Sons Ltd., London, 2001.

[8] A. Arbaoui, “Aide à La décision pour la définition d’un système éolien, Adéquation au site et à un réseau faible,” PhD Thesis of the Ecole Nationale Supérieure d’Arts et Métiers de Bordeaux, 2006.

[9] R. Harrison and G. Jenkins, “Cost Modelling of Horizon-tal Axis Wind Turbines (Phase 2),” ETSU W/34/00170/ REP, University of Sunderland, 1994.

[10] I. Troen and E. L. Petersen, “European Wind Atlas,” RISO, Commission of the European Communities, Risø National Laboratory, Roskilde, 1989.

[11] A. Spera, “Wind Turbine Technology,” The American Society of Mechanical Engineering, New York, 1998.

A. ARBAOUI ET AL.


170

Notations

A : rotor swept area (m2) c : Weibull distribution scale parameter (m/s) Ccomponent : cost of component Ce : system efficiency factor Cem : maximum system efficiency factor CP : rotor power coefficient Cpmax : maximum power coefficient CX : blade profile drag coefficient CZ : blade profile lift coefficient CWT : total cost of wind turbine (MEuros) D : rotor diameter (m) E : annual electricity produced (GWh/year) Fs : service factor of gearbox FWT : cost calibration factor f : Weibull distribution probability density Hhub : hub height (m) k : Weibull distribution shape parameter N : rotor rotation speed (rev/min) p : blade number Pn : nominal power (kW) Png : generator power rating(kW) QI : quality index RI : indicator of relevance s : operating range V : wind speed (m/s) Vdes : design wind speed (m/s) Vf : network-disconnection speed (m/s) Vi : network-connection speed (m/s) Vtip : blade tip speed (m/s)

Greek symbols

: wind shear factor λmax : maximum tip speed ratio ρ : air density (kg/m3) ηm : gearbox efficiency ηg : generator efficiency πg : generator efficiency factor πm : gearbox efficiency factor



Experimental Investigation of Solar Panel Cooling by a Novel Micro Heat Pipe Array

Xiao Tang, Zhenhua Quan, Yaohua Zhao Architectural and Civil Engineering Institute, Beijing University of Technology, Beijing, China


Abstract A novel micro heat pipe array was used in solar panel cooling. Both of air-cooling and water-cooling condi-tions under nature convection condition were investigated in this paper. Compared with the ordinary solar panel, the maximum difference of the photoelectric conversion efficiency is 2.6%, the temperature reduces maximally by 4.7 , the ou tput power increases maximally by 8.4% for the solar panel with heat pipe using air-cooling, when the daily radiation value is 26.3 MJ. Compared with the solar panel with heat pipe using air-cooling, the maximum difference of the photoelectric conversion efficiency is 3%, the temperature re-duces maximally by 8 , the output power increases maximally by 13.9% for the solar panel with heat pipe using water-cooling, when the daily radiation value is 21.9 MJ. Keywords: Solar Panel Cooling, Photoelectric Conversion Efficiency, Micro Heat Pipe array

1. Introduction Solar cell is the core component of photovoltaic power generation system. The photoelectric conversion effi-ciency of a solar cell is about 6-15% in commercial ap-plication [1]. Most of the radiation has been converted into heat, which results in high temperature of the solar cell and low efficiency even inefficiency. According to Weng et al [2], the temperature increase of 1K corre-sponds to the reduction of the photoelectric conversion efficiency by 0.2%-0.5%. In addition, long-term high temperature of the solar cell will shorten its service life. Therefore, solar cell cooling is of essential importance.

Air-cooling and water-cooling methods are both commonly used in solar cell cooling [2-4]. Heat pipe cooling is deemed to be a promising cooling technology [5-7]. Moreover, there are large thermal contact resistance ex-isting between conventional column heat pipe and flat solar panel, which will results in low heat transfer effi-ciency. The novel micro heat pipe array proposed by Zhao et al. [8,9] has a good contact with the solar panel, as its flat shape. And the heat pipe has higher heat trans-fer efficiency and a uniform temperature distribution that can solve the solar panel cooling issue. 2. Experimental Setup and Scheme The heat pipe consisted of two sections, evaporator sec-

tion and condenser section. The input heat vaporized the liquid inside the evaporator section, transfer to the con-denser section. The condenser section is cooled by air or water. Hence, the heat pipe can transfers the heat from solar panel to air or water, to reduces the temperature of the solar panel and improve the photoelectric conversion efficiency. 2.1. Experimental Setup A silicon solar panel was used in this experiment, with its peak efficiency in the range of 10-15% under standard condition (25, 1000 W/m2 ) , peak power of 10 W and an area of 0.0625 m2.

The cooling setup is schematically shown in Figure 1, which was setting up outdoor. The solar panel should face the south with a tilt angle of 45° [10]. Its total radia-tion area was 0.2049 m2. The air-cooling solar panel is shown in Figure 1(a). The evaporator section of the heat pipe was adhered to the back of the solar panel with its length of 283 mm and width of 300 mm. The condenser section was exposed to the air with its length of 200 mm and width of 300 mm. The schematic of the water-cooling solar panels is shown in Figure 1(b), the evaporator section of the heat pipe was adhered to the back of the solar panel with its length of 283 mm and width of 285 mm. The condenser section was adhered to a water flume with its length of 40 mm and width of 285 mm. The specs of

X. TANG ET AL.


172

heat pipe

solar panel

water tank

water pipe

flume

(a) Air-cooled with heat pipe

(b) Water-cooled with heat pipe

Figure 1. Cooling systems of solar panels.

water flume and water tank are 40 × 25 × 385 mm and 280 × 280 × 280 mm, respectively. There is a distance of 170 mm between them. 2.2. Experimental Scheme In order to discuss the main factors of the solar panel, the following parameters should be measured, such as the output power, the surface temperature and the photoelec-tric conversion efficiency.

1) Power output characteristics: Resistive load two -terminal test method was adopted, due to the small out-put current of the solar panel [11]. The test system is schematically shown in Figure 2. Voltmeter and amme-ter used were of type HC-300C-S-DV in range of 0-50 V and of type HC-300C-S-DA in range of 1-10A, respec-tively. A porcelain-type variable resistor (0-50 Ω, 150 W) was used as the load.

2) Weather parameter: A solarimeter was used to measure the real-time solar radiation intensity (W/m2). A wind speed sensor (YS-CF-X/S) was used to measure the wind speed.

3) Temperature: Temperatures of the solar panel, amb- ience, the water flowing in and out the water flume, and the water in the tank was monitored. Some platinum resi- stances (HT101) and a data acquisition instrument (Agil- ent 34970A) were used to measure and acquire the real- time temperatures.

The photoelectric conversion efficiency is calculated as:

et in t in

P UI

A P A P (1)

where, ηe is the photoelectric conversion efficiency (%), P the output power (W), U the voltage (V), I the current (A), Pin the solar radiation intensity (W/m2), At the total area of solar panel.

RV

A

Solar panel

U

I+

-

Figure 2. Output characteristics test system of solar panels.

In addition, the solar panels have been encapsulated,

so the result calculated by Equation (1) is the photoelec-tric conversion efficiency of solar panel rather than the net efficiency of the solar cell. 3. Results and Discussion Figures 3-5 show the comparing results between the ordinary solar panel without the heat pipe and solar panel with heat pipe using air-cooling one day in May. The maximal air temperature, the radiation intensity, the maximal and average wind speeds are 36, 1001 W/m2, 5.32 m/s and 0.51 m/s, respectively. The daily net radia-tion is 26.3 MJ from 5:00 to 19:30.

Figures 6-8 show the comparing results of the solar panel with heat pipe using air-cooling and wa-ter-cooling one day in May. The maximal air tempera-ture, the radiation intensity, the maximal and average wind speeds are 35 and 858 W/m2, 4.72 m/s and 0.51 m/s, respectively. The daily net radiation is 21.9 MJ from 5:00 to 19:30.

Figures 3-8 show the peak radiation intensity, power, temperature and photoelectric conversion efficiency from 10:00 to 14:00.

Figures 3 and 6 show that the radiation intensity is the primary factor affecting the output power of solar panels under a certain load resistance condition. As the solar radiation intensity first increases and then decreases, the output powers of solar panels also decrease after the first increase. The highest points both appear at the same time range between 12:00 to 13:00. The output power of

Figure 3. Comparison of hourly output power of solar panel cooling by air with heat pipe and without cooling.

X. TANG ET AL.


173

Figure 4. Comparison of hourly temperature of solar panel cooling by air with heat pipe and without cooling.

(a)

(b)

Figure 5. Comparison of hourly efficiency of solar panel cooling by air with heat pipe and without cooling.

the solar panel with heat pipe using air-cooling increases maximumly by 8.4% and averagely by 6.3% compared with ordinary one, as shown in Figure 3. The output power of solar panel with heat pipe using water-cooling increases maximumly by 13.9% and averagely by 9% compared with that using air-cooling, as shown in Fig-ure 6.

Figures 4 and 7 show that the radiation intensity, air temperature and wind speed are the main factors affecting the temperature of the solar panel. As the solar radiation

Figure 6. Comparison of hourly output power of solar panel cooling by air with heat pipe and by water with heat pipe.

Figure 7. Comparison of hourly temperature of solar panel cooling by air with heat pipe and by water with heat pipe.

(a)

(b)

Figure 8. Comparison of hourly efficiency of solar panel cooling by air with heat pipe and by water with heat pipe.

intensity first increases and then decreases, at the mean-while, the temperature of the solar panel also decreases after the first increase with a little lag. The maximum

X. TANG ET AL.


174

temperature point appears at the time between 13:00 to 14:00. The temperature of solar panel with heat pipe us-ing air-cooling reduces maximally by 4.7 and aver-agely by 1.5corresponds to the reduction of the photo-electric conversion efficiency by 0.2%, from 10:00 to 15:00 compared with ordinary one, as shown in Figure 4. The temperature of the solar panel with heat pipe using water-cooling reduces maximally by 8 and averagely by 2.7 compared with that using air-cooling, as shown in Figure 7.

Figures 5 and 8 show that the radiation intensity and the temperature of the solar panel are the main factors that affecting the photoelectric conversion efficiency of solar panel. According to the data, the temperature in-crease of 1 K, the efficiency of solar panel with heat pipe using air-cooling increases maximally by 2.6% and av-eragely by 0.4% compared with ordinary one, as shown in Figure 5. The temperature of solar panel with heat pipe using water-cooling increases maximally by 3% and averagely by 0.5% compared with that using air-cooling. The maximum efficiency of 13.5% can be achieved, as shown in Figure 8. 4. Conclusions A novel micro heat pipe array is used for solar panel cooling. Air-cooling and water-cooling methods used are compared in this study. The results indicate that under cooling condition, the temperature can be reduced to effec-tively increase the photoelectric conversion efficiency of solar panel.

1) Compared with the ordinary solar panel, the tem-perature of that using air-cooling reduces maximally by 4.7, the output power increases maximally by 8.4%, and the efficiency difference is 2.6% (In that day, the maximal air temperature and wind speed are 36 and 5.32 m/s ,the daily global radiation is 26.3 MJ).

2) Compared with the solar panel using air-cooling, the temperature of that using water-cooling reduces maximally by 8, the output power increases maximally by 13.9% and the efficiency difference is 3%. The maxi-mum efficiency of 13.5% can be achieved (In that day,

the maximal air temperature and wind speed are 35 and 4.72 m/s, the daily global radiation is 21.9 MJ). 5. References [1] W. He, T. T. Chow, J. Ji, et al., “Hybrid Photovoltaic and

Thermal Solar-Collector Designed for Natural Circulation of Water,” Applied Energy, Vol. 83, No. 3, 2006, pp. 199-220.

[2] Z. J. Weng and H. H. Yang, “Primary Analysis on Cool-ing Technology of Solar Cells under Concentrated Illu-mination,” Energy Technology, Vol. 29, No. 1, 2008, pp. 16-18.

[3] K Araki, H Uozumi and M Yamaguchi, “A Simple Pas-sive Cooling Structure and its Heat Analysis for 500 × Concentrator PV Module,” 29th IEEE PVSC, New Or-leans, May 2002, pp. 1568-1571.

[4] M. Brogren and B. Karlsson, “Low-Concentrating-Water Cooled PV-Thermal Hybrid Systems for High Latitudes,” 29th IEEE PVSC, New Orleans, May 2002, pp. 1733- 1736.

[5] M. A. Farahat, “Improvement the Thermal Electric Per-formance of a Photovoltaic Cells by Cooling and Con-centration Techniques,” 39th UPEC International, Bristol, Vol. 2, 2004, pp. 623-628.

[6] A. Akbarzadeh and T. Wadowski, “Heat Pipe-Based Cooling Systems for Photovoltaic Cells under Concen-trated solar Radiation,” Applied Thermal Engineering, Vol. 116, No. 1, 1996, pp. 81-87.

[7] W. G. Anderson, P. M. Dussinger, D. B. Sarraf and S. Tamanna, “Heat Pipe Cooling of Concentrating Photo-voltaic Cells,” 33rd IEEE Photovoltaic Specialists Con-ference, San Diego, May 2008, pp. 1-6.

[8] Y. H. Zhao, et al., “A Sort of Micro Heat Pipe Array and Processing Technics,” Chinese Patent: 200810225649.

[9] Y. H. Zhao, et al., “Photovoltaic Cells Radiating Equip-ment,” Chinese Patent: 200810239002.0.

[10] H. Liu, D. C. Wu, Z. G. Yang and Y. H. Zhai, “House-hold Photovoltaic Power System,” Chemical Industry Press, Beijing, 2007.

[11] Z. H. Zhang, L. L. Li, C. P. Ye and P. H. Yang, “Organic Solar Cells and Plastic Solar Cells,” Chemical Industry Press, Beijing, 2007.



A Burning Experiment Study of an Integral Medical Waste Incinerator

Rong Xie1, Jidong Lu1, Jie Li1, Jiaqiang Yin2 1State Key Laboratory of Coal Combustion, Huazhong University of Science and Technology, Wuhan, China

2Huangshi Zhong You Environmental Protection Corporation, Huangshi, China E-mail: [email protected]

Received May 6, 2010; revised June 12, 2010; accepted July 15, 2010

Abstract Mass burning of the medical waste is becoming attractive in China because Chinese government has banned landfilling of medical waste. Many advantages can be found in this method, such as reduction in waste vol-ume, destruction of pathogens and transformation of waste into the form of ash. However, the medical waste with high moisture in China is not suitable to be treated in the present direct mass burning incinerators. In this paper, a novel integral incinerator is developed with combining a feeder, a rotary grate, a primary com-bustion chamber (PCC) and a “coaxial” secondary combustion chamber (SCC) into a unique unit. Its capa-bility is 10 ton/day. As the air excess level in the PCC was only 40% stoichiometric ratio, the PCC acted as a gasifier. The 1.0 excess air ratios in the SCC preserved the purpose of full combustion of flue gas. Tempera-ture and pollutants concentration in the SCC were measured to understand the combustion behavior of vola-tile organics. Emission concentrations of pollutants before stack were also tested and compared with the China National Incineration Emission Standard. Keywords: Medical Waste, Incineration, Mass Burning, Emission Pollutants

1. Introduction With high-rate economic growth and urbanization in China, the amount of medical waste increased continu-ously at the rate of 8.98% since 1980s [1,2]. The citizens and governor have to face the inevitable challenge of the medical waste treatment. As the hazardous waste landfill standard took effect in 2001, the sanitary landfill of medical waste was banned in China [3]. Not only because the medical waste contained a great quantity of bacteria and viruses and threaten the surroundings of the landfill, but also the available area for landfill has become scarce.

Incineration of medical waste and disposal of the resul-tant ash by landfill is now accepted as an environmentally friendly disposal method. Its advantages are the destruc-tion of pathogens, reduction in the volume and transform of waste in the form of ash [4].

The mass burning medical waste incinerators in-cludes three basic types: 1) the modular incinerator, 2) the conventional grate-fired incinerator, 3) the rotary kiln incinerator.

Among them, modular medical waste incinerators are the largest number among mass burn installations [5]. The modular incinerator is a compact furnace in the form

of a cube with multiple internal baffles. Each modular chamber normally has one or two burners to maintain its required operating temperature. The excess air level is well above stoichiometric, typically 150%-200% excess air. This kind of incinerator is not easily adaptable for continuous operation [6].

The grate-fired medical waste incinerator can be di-vided into: 1) fixed grate incinerator, 2) moving grate incinerator. Moving grate incinerators are mainly in use today, as medical waste can be stirred and ash removal can be automated. The moving grate incinerator in-cludes one chamber, where medical wastes go through heating, drying, pyrolysis, ignition and burning on the grate. The total injected air is about 120%-160% of the stoichiometric air requirement. However, experiences and trials in South Africa show that the moving grate- fired incinerators are unsuitable to treat medical for its high emission loads [7].

The rotary kiln system is widely used to treat hazard-ous waste. The raw medical waste can be fed directly to the kiln. All reactions such as organic thermal decompo-sition and char oxidation reactions occur within the ro-tary reactor. The rotary kiln can produce intensive turbu-lence and mix air and solid phases completely [8]. But

R. XIE ET AL.


176

the high combustion efficiency is still depended on high auxiliary fuel consumption. Gas sealing of kiln and fixed parts is also a difficult problem.

The three types of incinerators mentioned above pre-sented good performance in the mass burning of solid waste with high heat value (more than 2000 kcal/kg). However, in China little experience is available of these incinerators to treat medical waste of high moisture.

In this paper, an integral medical waste incinerator is developed from the following consideration: 1) Achieve flue gas emission limits. 2) Reduce heat loss of the incin-erator. 3) Make full use of medical waste’s heat value. The integral medical waste incinerator combines a feeder, a rotary grate, a primary chamber and a “coaxial” secondary combustion chamber into a unique unit. The incineration feeding system continuously feeds medical waste into the incinerator. Heating, drying, pyrolysis of medical waste and char oxidation occur in the primary chamber, while gas phase oxidation reactions occur in the secondary chamber. As there is no similar structure of incinerator to our knowledge, the objective of this study are to introduce its novelty, investigate its performance and measure the combustion pollutants. 2. Incinerator System Description The diagrammatic sketch of incineration system and waste treatment process is shown in Figure 1. The basic com-

ponents in the incineration system are the waste feed system, the combustion system and the air quality control system. The ash disposal and heat recovery are taken as a supplemental process. The details are described as fol-lowing. 2.1. Feeding System The special bin storing medical waste is weighed and loaded by the automatic crane, which charges waste into a 2-meter high chute (C in Figure 2). Two steel clap-boards are installed in the chute to maintain good seal (B and E in Figure 2). So the chute can not only prevent air flowing into the primary chamber, but also prevent fire from the combustion chamber entering the feeding sys-tem. A reciprocating propeller (D in Figure 2) installed in the chute feeds the waste into the primary combustion chamber. 2.2. Primary Combustion Chamber The primary combustion chamber (PCC) consists of a 1.2-meter diameter cylinder and a rotary grate. The grate is installed at the bottom of the PCC. The rotary grate is made up of three discs fixed on the central bearing body and one cone fixed on the top disc. Medical waste, pyro-lysates and bottom ashes are tumbled slowly with the rotation of the grate. Many small holes (5-cm diameter)

Figure 1. Schematic diagram of incineration system.

R. XIE ET AL.


177 A. charge hopper

B. first access slide board

C. vertical bin D. reciprocating propeller

E. secondary access slide board

F. combustion chamber

Figure 2. Schematic diagram of the feeding system.

distribute symmetrically on each disc. So the holes with the space between discs can distribute air homogene-ously.

In the PCC, The raw medical waste is distributed unif- ormly onto the waste bed by the feeding system and forms about 1.5 meters waste bed on the surface of the rotary grate. Continuous feeding and slagging make the waste bed maintained at a fixed height. The raw medical waste undergoes heating and drying process on the the waste bed as soon as the raw medical waste enters. This process removes the moisture in the raw medical waste. When the temperature of medical waste rises up to a level, the pyrolysis and volatilization of the solid medical waste start. After pyrolysis and volatilization process, the remained solid char is further oxidized and forms hot slag.

The air for the solid char combustion is drawn from the bottom entrance of the PCC at the room temperature by a forced draft fan. The “cold” air runs through the hot slag layer on the surface of the rotary grate. So the air is heated up while the hot slag is cooled down. The cooled slag forms a protective layer on the surface of the rotary grate, which isolates the grate from the high temperature combustion region. The grate is free from heat transfigu-ration during the combustion process.

The heated air is then transferred into the combustion area-solid carbon oxidation layer. The char can be oxi-dized completely under full excess air and produce lots of hot flue gases. These hot flue gases pass through the pyrolysis layer and upwards into the heating and drying layer. The heat in the hot gases can supply enough en-ergy for pyrolysis and drying process of the raw medical waste. 2.3. Secondary Combustion Chamber The secondary combustion chamber (SCC) has a distinct cylinder configuration. We call it “coaxial” structure, as

it surrounds the PCC with the same vertical axis. This design significantly decreases the PCC and the SCC outer jacket’s contact area with the environment. No pipeline is needed for connection between the PCC and the SCC, so the heat loss in pipeline occurring in tradi-tional multi-chamber incinerator is eliminated. The baf-fles are installed in the SCC to guide the combustion gases through 180˚ turn in vertical directions. This “U” shape combustion channels can significant increase gas turbu-lence in the high temperature region. In case of the same volume of the combustion chamber, the length of flue gas channel is increased relatively, so the residence time of flue gas at temperatures exceeding 800 is ensured for at least 2 seconds. Secondary air is preheated by a heat ex-changer and supplied at the top of the secondary combus-tion chamber. The secondary air injector is fixed in tan-gential direction to achieve well mixed effect.

Manipulation of combustion in the SCC includes ‘3Ts’ combustion control (temperature, time and turbulence) in order to maximize system efficiency and minimize envi-ronmental adverse effect of the incineration. If the com-bustion temperature is below 800 the ignition equip-ment will start automatically and auxiliary fuel is in-jected to maintain the temperature. In fact the fuel injec-tion is necessary when the mass burning starts. When combustion is stable, the heat release from oxidation reactions can keep temperature exceeding 800. 2.4. Flue Gas Purification System After incineration, the flue gases pass through the pipe and enters the into the flue gas purification system. In order to prevent the erosion by acidic flue gas and flush-ing by the entrained particles, a thin layer of silicon car-bide is coated on the surfaced of the pipe. The flue gas is then induced into semidry scrubbing system. Lime slurry is injected by a spry nozzle in the scrubber. It has high and stable removal efficiency for HCl and SOX. No waste- water treatment equipment is required. After that, flue gas goes through the demister tower. Activated carbon is injected into the flue gas to absorb the dioxins and furans formed in the post combustion process before it enters the baghouse. Finally the purified flue gases are induced into the stack by diversion fun. 3. Medical Waste Incineration Experiments

and Results 3.1. Waste Characteristics The typical medical waste was sampled from the same batch by random sampling method. First typical medical waste sample was sterilization by high temperature steam. Secondly, the large pieces of hard objects removed from

R. XIE ET AL.


178

the sample. Finally sample was grinded for proximate analysis and then dried for ultimate and energy analysis. The ultimate analysis of typical medical waste is show in Table 1. The ultimate analysis of a waste component typically involves the determination of the percent of carbon (C), hydrogen (H), oxygen (O), nitrogen (N) and sulfur(S). Table 1 gives the average chemical composi-tion of combustible component of medical waste. The energy content of the medical waste is determined by a laboratory bomb calorimeter. Typical energy content of medical waste is also estimated and showed in Table 1. 3.2. Air Supply The air supply in the PCC is only about 40% of the stoichiometric requirement, equal to 1/3-1/4 of the recom-mended air levels (140-200%) for traditional mass burn system [9]. In the lower part of the PCC, this air is en-sured to burn out the solid char. However, in the upper zone, the PCC is acting as a gasifier. Pyrolysis is the dominant mechanism in this zone for the production of volatiles from medical waste in the absence of oxygen. The produced volatiles have been identified as CO, CO2, H2O, CH4 and other light hydrocarbons.

The air supply in the SCC is 100% of the stoichiomet-ric requirement for complete waste combustion, so the volatiles from PCC can be decomposed completely. 3.3. Incineration Temperature Armored thermocouples were installed in PCC and SCC outlet as measurement equipment of temperature. The detection accuracy of the thermocouples is 1. Varia-tion of temperature with time in the PCC and SCC is shown in Figure 3. The temperature of PCC is around 640. However the value varied significantly with time. Because materials of medical waste range from food products to pathological waste, there is large variation in the properties of medical wastes. These variations have a dramatic impact on the performance of medical waste incinerator.

Although the combustion and pyrolysis mechanisms in PCC are complex, it is demonstrated that the pyrolysis production of volatiles and char can be regulated by temperature [10]. The decomposition of the organic matter is normally total at these high temperatures. When the py-rolysis temperature is much lower, the production of CO is much decreased. The low temperature may slow down the pyrolysis process and lead to delay the time of raw waste ignition. When the pyrolysis temperature is much higher, the self-combustion process became unstable. Because high temperature present in the PCC may lead to excessive burning of the pyrolysate. It will decrease the quantity of combustive pyrolysate in the SCC. Con-sequently the auxiliary fuel has to be consumed to main-

tain the temperature in the SCC. After several testing before operation, the optimal running temperature for the PCC was set at 660. It can be found in Figure 3 that the actual operating temperature is close to the setting temperature.

The combustion temperature in the SCC is higher than 850, and the temperature varied slightly with time ex-cept the heavy disturbing effect from the PCC. This in-dicates combustion process in SCC is more stable than the process in PCC. The novel coaxial chamber design decreases 40% outer contact area of the SCC with the environment. Heat losses are much reduced. It is easy for SCC to achieve high temperature and remain stable op-eration. 3.4. Pollutants and Oxygen in the SCC In actual operation, whether volatiles are combusted completely in the SCC is an important performance indicator of the incinerator. Gas pollutants at different positions of the SCC were tested by portable infrared flue gas ana-lyzer. Several verification tests of the gas analyzer were carried out before the experiment to avoid measurement errors. As illustrated in Figure 4, five sample points (la-beled as 1-5) were set symmetrically along the axial di-rection. Point “1” is near the PCC exit and point “5” is near the SCC outlet. In the stable operation conditions, each sample point is tested for three times.

Table 1. Medical waste component.

Proximate analysis (wt.%)

Mar Vd Ad LHV

(MJ/kg)

64.1 75.3 23.4 25.63 Ultimate analysis (wt.%)

C H N S O 47.54 7.99 2.02 0.5 18.54

Figure 3. Variation of temperature in the PCC and the SCC.

R. XIE ET AL.


179

Concentration profiles of pollutants (CO, CO2, NOX, and SO2) and oxygen are shown in Figure 5 to Figure 9. It can be seen from Figure 5 that the CO concentration decreases intensely from sample point “1” to “5”. This indicates that CO formed in the PCC is mostly destroyed in the SCC. The destruction efficiency of CO is round 99.95%. It can be found from Figure 5 that CO2 is mainly formed in the SCC. The comparison between Figure 5 and Figure 6 can further explain that CO

Inlet

1

2

3

4

5 Outlet

Figure 4. Sampling points distribution in the SCC.

Figure 5. CO concentration.

Figure 6. CO2 concentration.

mostly translated into CO2 in the SCC. As is illustrated in Figure 7, concentration of oxygen is gradually decreased with the oxidation reaction going-on. Generally speaking, the concentration of oxygen in the SCC outlet is higher than 10%, which meets the demand of Medical Waste Incineration Standard [11]. From Figure 8 and Figure 9, we can find that concentration of NOX and SO2 are lower than 30 ppm and 10 ppm, respectively. These may be due to medical waste containing little N and S elements. However NOX and SO2 mainly formed in the pyrolysis process decrease little in the SCC. Conclusions can be made that combustion in the SCC has no effect on production of NOX and SO2. 3.5. Pollutant Concentrations in the Stack Concentrations of NOX, SOX, HCl and particles in the

Figure 7. O2 concentration.

Figure 8. SO2 concentration.

R. XIE ET AL.


180

Figure 9. NOX concentration.

exhaust were measured respectively according to the Ch- ina National Standards [11]. Concentration of CO, CO2, O2 were also measured by a Combustion Efficiency Mea- surement Instrument.

The maximum air pollutant concentrations in the inlet of the stack are summarized in Table 2. National Incin-eration Emission Standard for each component is also listed in the Table 2. As shown in Table 2, the measured pollutants belong generally to the concentration range set by China National Standards. As Table 2 shows, the experimental value of particulates in the stack is much lower than the limit. In addition to the well operated bag-filter, the low air level in the PCC is an important reason. Compared with the traditional mass burning system, star- ved-air combustion in the PCC produces less solid parti-cles in the gas stream. Researchers have demonstrated that the dioxin formation from carbon particulates is one of the potential mechanisms for PCDD/F formation in the post combustion zone [12,13]. The reduced fly ash entrainment in flue gas is helpful to control the dioxins. The concentration of acid gases such as HCl and SO2 are also lower than the limit. This demonstrates that gas scrubber is in good working condition. 4. Conclusions The integral medical waste incinerator combines a feeder, a rotary grate, a primary chamber and a “coaxial” secon-dary combustion chamber into a unique unit. The temperature of the PCC varied significantly with time because of the intermittent feed and the heterogeneous character-istics of the raw medical waste, however, due to the co-axial SCC design, the combustion temperature in the SCC varied slightly with time. The temperature has great effect during the formation of pyrolysis gas such as CO. The low air level (40%) in the PCC well controlled the

Table 2. Emission concentration of pollutants in the flue gas.

Pollutants Experimental values Standard

CO, mg/m3 34.8 80

NOX, mg/m3 15.1 400

SO2, mg/m3 24.2 260

HCl, mg/m3 48.7 75

O2, mg/m3 16.3 6-11

Particulates, mg/m3 32 80

Blackness 0.8 1

chamber’s temperature. The actual operating temperature in the PCC (640) is close to the setting temperature (660). Consequently, char combustion is also kept stable and complete. Concentrations of pollutants in the SCC were measured on different sample points. In these data, CO level represented the best available estimate of environmentally satisfactory operation for the incineration process. The destruction efficiency of total CO in the SCC is round 99.95%. Emission concentrations of pollutants in the stack were also measured and met the de-mand of the China National Incineration Emission Stan-dard. 5. References [1] R. Xie, W. J. Li, J. Li, et al., “Emissions Investigation for

A Novel Medical Waste Incinerator,” Journal of Haz-ardous Materials, Vol. 166, No. 1, 2009, pp. 365-371.

[2] J. M. Zhu, H. M. Zhu, X. G. Jiang, et al., “Analysis of Volatile Species Kinetics during Typical Medical Waste Pyrolysis Using A Distributed Activation Energy Model,” Journal of Hazardous Materials, Vol. 162, No. 2-3, 2009, pp. 646-651.

[3] State Environmental Protection Administration of China, “Standard for Pollution Control on the Security Landfill for Hazardous Wastes,” National Technical Standard of China (GB 18598-2001).

[4] W. R. Niessen, “Combustion and Incineration Processes,” 3rd Edition, Marcel Dekker Inc, New York, 2002.

[5] C. C. Lee and G. L. Huffman, “Medical Waste Manage-ment Incineration,” Journal of Hazardous Materials, Vol. 48, No. 1-3, 1996, pp. 1-30.

[6] A. F. Shaaban, “Process Engineering Design of Patho-logical Waste Incinerator with an Integrated Combustion Gases Treatment Unit,” Journal of Hazardous Materials, Vol. 145, No. 1-2, 2007, pp. 195-202.

[7] D. E. Rogers and A. C. Brent, “Small-Scale Medical Waste Incinerators Experiences and Trials in South Af-rica,” Waste Management, Vol. 26, No. 11, 2006, pp. 1229-1236.

[8] G. R. Woodle and J. M. Munro, “Particle Motion and Mixing in A Rotary Kiln,” Power Technology, Vol. 76,

R. XIE ET AL.


181

No. 3, 1993, pp. 187-190. [9] A. R. Lawrence, “Energy from Municipal Solid Waste: A

Comparison with Coal Combustion Technology,” Pro-gress in Energy and Combustion Science, Vol. 24, No. 6, 1998, pp. 545-564.

[10] H. M. Zhu, J. H. Yan and X. G. Jiang, “Study on Pyroly-sis of Typical Medical Waste Materials by Using TG- FTIR Analysis,” Journal of Hazardous Materials, Vol. 153, No. 1-2, 2008, pp. 670-676.

[11] State Council of China, “Decree of the Medical Waste

Management,” Chinese Environmental Science Press, Bei-jing, 2003.

[12] L. Stieglitz and H. Vogg, “On Formation Conditions of PCDD/F in Flyash from Municipal Waste Incinerators,” Chemosphere, Vol. 16, No. 8-9, 1987, pp. 1917-1922.

[13] B. K. Gullet, K. R. Bruce, L. O. Beach, et al., “Mechanis-tic Steps in the Production of PCCD and PCDF during Waste Combustion,” Chemosphere, Vol. 25, No. 7-10, 1992, pp. 1387-1392.



Chaotic Optimal Operation of Hydropower Station with Ecology Consideration

Xianfeng Huang1, Guohua Fang1, Yuqin Gao1, Qianjin Dong2 1College of Water Conservancy and Hydropower, Hohai University, Nanjing, China

2College of Water Resources and Hydropower, Wuhan University, Wuhan, China E-mail: [email protected]

Received April 21, 2010; revised June 2, 2010; accepted July 10, 2010

Abstract Traditional optimal operation of hydropower station usually has two problems. One is that the optimal algo-rithm hasn’t high efficiency, and the other is that the optimal operation model pays little attention to ecology. And with the development of electric power market, the generated benefit is concerned instead of generated energy. Based on the analysis of time-varying electricity price policy, an optimal operation model of hydro-power station reservoir with ecology consideration is established. The model takes the maximum annual power generation benefit, the maximum output of the minimal output stage in the year and the minimum shortage of eco-environment demand as the objectives, and reservoir water quantity balance, reservoir storage capacity, reservoir discharge flow and hydropower station output and nonnegative variable as the constraints. To solve the optimal model, a chaotic optimization genetic algorithm which combines the ergodicity of chaos and the inversion property of genetic algorithm is exploited. An example is given, which shows that the pro-posed model and algorithm are scientific and feasible to deal with the optimal operation of hydropower station. Keywords: Hydropower Station Operation, Ecology, Chaotic Genetic Optimization Algorithm,

Time-Varying Electricity Price

1. Introduction The optimal operation of hydropower station can increase the hydropower market competitive ability and realize the optimization of resources. How to manage and utilize the present hydropower station and to obtain the more comprehensive benefits in the case of maintain the extra investment, which have important significance on the development of our national economy and society and solve the problem of the energy shortage in short time. Investigation into the hydropower station optimal opera-tion arises from the 1940s, in 1946, Masse the first in-troduced the concept of optimization into hydropower station optimal operation. American scholar Little [1] propose a random model of reservoir optimal operation which random variable is the runoff. Now, many schol-ars carried out earlier reservoir optimal operation of research and application [2-5]. At present hydropower station reservoir operation have more research on optimization algorithm and reached many great results. Many arith-metic have been applied to hydropower station reservoir optimal operation [6-13], such as dynamic programming,

decomposition-coodination method of large systems, fuzzy mathematics, genetic algorithm, artificial neural network, particle swarm optimization algorithm, ant colony optimization algorithm. With the continuous develop-ment of theory and method for reservoir optimal opera-tion and in-depth development of electricity market re-form, the study of hydropower station reservoir optimal operation is a hotspot research within a period of time in the future. In addition, with the development of people’s knowledge about the ecological environment, some scholars become to concern the ecology for optimal operation of hydropower station [14,15]. But most studies of the existing reservoir are also rarely considered the ecologi-cal water requirements of the river downstream and the reservoir itself. Although the models and algorithms have made considerable progress, but most of the models and the algorithms have bad universality, and is often for the specific reservoir conditions and operating character-istics which develop the specific model and algorithm. In this paper, based on the predecessors, under the condi-tions of researching the electricity market, reasonable consider the ecological water requirements, the chaotic optimal operation model based on time-varying electric-

X. F. HUANG ET AL.


183

ity price is established, chaotic genetic arithmetic(CGA) is exploited to solve the model, the study has a certain reference practice value for optimal operation of hydropower station. 2. Time-Varying Electricity Price In electric power market, competition system is intro-duced, and the policy of time-varying electricity price, such as flood and dry power price, peak and valley pow-er price is carried out. The key of hydropower compa-nies’ business is to achieve the maximum power genera-tion benefit considering the various constraints, through optimal arranging the generation process. Due to the characteristic of hydropower, such as the uncertainty and randomicity of runoff, the regulating ability of reservoir, the hydropower station reservoir optimal operation is much more complex. So, how to forecast the reservoir runoff, how to reasonably arrange generation operation to improve the economy benefit based on integrated uti-lization request and the regulating ability of reservoir and the flood and dry power price, become more and more important.

In electric power market, hydropower companies par-ticipate in market competition through declaring energy price curve. Because the policy of “plant and network separation, electricity price bidding” is carried out, the tra-ditional optimal operation is challenged. The surround-ings up against the hydropower plant have changed pro-foundly, and so the optimal objective of hydropower plant. Along with the independency of property right of hydropower companies, the maximum income and profit is pursued. The rule of maximum power generation en-ergy in the past is substituted by the power generation benefit in the electric power market.

In the condition of electricity price bidding, the opti-mal operation of hydropower station reservoir considers not only the water quantity factor, but also the electricity price factor. At the view of market economy, the hydro-power plants pay more attention to the timeliness of power generation energy and acquire more economic benefit.

In the past few years, especially in the summer, power consumption load has increased rapidly. Many provinces of China suffer the lack of electricity. Besides, people’s life habit in a day makes the electricity load high in the day, and low in the night. The big difference of peak and valley usually causes some bad effects for power system, such as causing the operation difficulty, depressing the economy of the system. The best method to solve the problem is carrying out flood and dry power price and peak and valley power price. That is to say, the power price falls in the flood season, and increases in the dry season. It forms a net power price structure with price difference in different seasons. The peak and valley pow-

er price includes peak load power price, valley load power price and smooth load power price which ascer-tained by the peak load period, valley load period and smooth load period in the daily load curve [16].

The time-varying electricity price can promote uses to avoid peak, and make the best of valley load. It can make the distribution of the power load curve uniform. In addi-tion, at the view of consumer demand and provider bene-fit, the policy of time-varying electricity price can in-crease power energy sales, reduce generation cost and improve the system operation benefit. In the other hand, it can provide preferential electricity price, save the expen- se for uses. So it is favorable for consumer and provider. The one-part rate system price is fixed and unchangeable, and the income of hydropower companies is the product of generated energy and power price. That is to say, the maximum benefit is equal to the maximum of generated energy. But for the time-varying electricity price, the two are different. So the optimal operation based on time-varying electricity price is a new research problem for hy-dropower station reservoir. 3. Optimal Operation Model with Ecology

Consideration 3.1. Guidelines for Optimal Scheduling Optimal scheduling of hydropower station reservoir op-timization operation is generally divided into two groups: the quantity and quality, including which makes the greatest economic benefits of electricity generation and makes the highest quality of electricity and water supply. From the economic point of view the biggest, in the ab-sence of the implementation of power market, for hydropower station, the most common method is to make a maximum generating capacity in operation period. Under the environment of electricity market, the most important is to combine the electricity price, in accordance with peak and valley changes in policy, make reasonable ar-rangements for generation companies which owned the water and thermal power generation capacity at different times, and then the power companies can get the most economic benefits in operation period. 3.2. Objective Functions In order to achieve the most optimal use of water re-sources, in this study, the goal is to meet the premise of water requirements, the criterion is to improve the power generation companies’ earnings, and provide the greatest possible reliability and power to the grid, so, choose the following two objectives:

Objective I: Obtain the maximum annual power gen-eration benefit per every year, by the reservoir regulation,

X. F. HUANG ET AL.


184

so as to increase the profit of hydropower station in the dry season and flat-water period, and then increase rev-enue generation as much as possible in the wet period. The objective can be seen as follows:

1

maxT

t t t tt

F Ap Q H M

(1)

where, F is the effectiveness of the annual hydropower generation(RMB). A is the comprehensive output co-efficient of hydropower. tP is the power price factor;

tQ is the generating flow of hydroelectric power on the

time t (m3/s). tH is the average water head of hydroe-

lectric power on the time t(m). T is the hydropower operation calculation of the total time. tM is the num-

ber of hours in the time t. Objective II: Obtain the maximum output of the mini-

mal output stage in the year. The objective can be seen as follows:

max max min t tNP AQ H (2)

where, NP is the maximization minimum output of hydropower station (MW). Other symbols are the same as (1).

Objective III: Obtain the minimum amount of ecology- ical water shortage for river downstream and the reser-voir itself.

( ) ( )t 1

min ( )N

t tZ RL VL

(3)

where, Z is the total ecological water shortage. ( )tRL is

the ecological water shortage of the river downstream at time t. ( )tVL is the ecological water shortage of the res-

ervoir itself at time t. N is the total time. Ecological water demand computing [17,18]. As the

ecological water requirements of surviving in the de-bate, some scholars have raised the minimum ecologi-cal water and suitable ecological water demand, water demand of ecological environment in this article is based on the minimum water volume of ecological wa-ter demand on the basis of calculation. Therefore, (3) can be written as:

( ) ( ) ( ) ( )t 1

min ( )N

t t t tZ RD VD RS VS

(4)

where, ( )tRD and ( )tVD are the minimum amount of

ecological water demand of the river downstream and reservoir itself respectively. ( )tRS and ( )tVS are the sup-

ply amount of ecological water of the river downstream and reservoir itself respectively.

The traditional optimal operation of hydropower reservoir main considers socio-economic objective, with little regard the requirements of the ecological environment, leading to environmental degradation. The objective of

minimum ecological water shortage of river downstream and reservoir fully reflects the lower reaches of the res-ervoir and river ecology to environmental protection, so that human life and the ecological environment has been the basic water was placed in the same the degree of im-portance for the protection of river health, and promote sustainable utilization of water resources play an impor-tant role. 3.3. Constraints 1) Reservoir water balance constraint.

1 ( )t t t t tV V q Q K (5)

where, 1tV is the reservoir storage capacity in the end

of time t(m3); tV is the reservoir storage capacity in the

beginning of time t(m3); tq is the average inbound flux

on the time t (m3/s); tK is the conversion factor of the

time length. 2) Reservoir storage capacity constraint. At any time,

the reservoir water storage capacity storage capacity should be kept between a minimum and maximum vol-ume of water.

,min ,maxt t tV V V (6)

where, ,mintV is the allowed minimum reservoir capacity

on the time t (m3). ,maxtV is the allowed maximum res-

ervoirs capacity on the time t (m3). 3) Reservoir discharged flow constraint.

,min ,maxt t tQ Q Q (7)

where, ,mintQ is the minimum discharge on the time t

(m3/s); ,maxtQ is the maximum discharge on the time t

(m3/s). 4) Hydropower station output constraint.

,min ,maxt t t tN AQ H N (8)

where, ,mintN is the minimum output which hydro-

power station allowed, it is always the guaranteed output (kW). ,maxtN is the maximum output which hydropower

station allowed, it is always the installed capacity (kW). 5) Nonnegative variable constraint. All of the above decision variables are non-negative

variables(≥ 0).

4. Chaotic Genetic Algorithm 4.1. Thought of Chaotic Genetic Algorithm Chaos is a seemingly rule, similar to the random phenol- menon which emerge in deterministic system, the theo-

X. F. HUANG ET AL.


185

ries marked by the United States meteorologist Lorenz in 1963, which published papers named “Deterministic non-periodic flow” [19], this paper reveals the existence of deterministic chaos in nonlinear equation. Nowadays, scholars generally agreed that chaos with such features, include randomness, regularity, and ergodicity [20], due to chaotic motion can after all states according to their own laws within a certain range of non-repetition. There- fore, use chaotic variables to optimization search, certainly the global optimal solution can be obtained and have high search efficiency. At present, the chaos optimization have been applied in optimal operation of reservoirs [21, 22], on the basis of previous research, this paper coupled the chaos optimization and genetic algorithms, combined with multi-objective decision-making techniques to develop the chaotic genetic algorithm (CGA). The first, CGA use constraint method converted multi-objective problem to single objective problem, using penalty function method dealing with constraints. And then chaotic variables were introduced into the optimization variables, and to enlarge the scope of chaotic motion to the range of opti-mization variables, code chaotic variables which we got, using the search mechanisms of genetic algorithm to obtain the optimal solution. The basic idea of CGA solv-ing hydropower station reservoir optimal scheduling is: the first, scheduling period (usually one year) is divided into a number of time slots T, numbered the sequence numbers of each period, choose every period of the res-ervoir water level value (the value of reservoir storage capacity can also be used) as optimization variables, de-termine the upper and lower limits of the reservoir water level value each time, randomly selected n different ini-tial values which interval is between 0 and 1, through Logistic maps can obtain the n-chaotic trajectories of different sequence, the length of the chaotic sequence is the population size, large it to the range of reservoir wa-ter level at all times, then get n group sequence of the reservoir water level which stand forthe reservoir op-eration control process ( 1 1 1 1

1 2 3, , , , TZ Z Z Z ), ( 2 2 21 2 3, , ,Z Z Z

2, TZ ), …, ( 1 2 3, , , ,n n n nTZ Z Z Z ), and as a mother, accor-

ding to the intended target function evaluate its advan-tages and disadvantages, calculate the fitness value of all chromosomes, carry out selection, crossover and muta-tion operations according to the chromosomes fitness, use the most excellent retention strategy, abandon the low fitness chromosomes, to retain the high fitness chromosomes, and thus get new groups. Add a chaotic small perturbation to the optimization variables, by evolving from generation to generation, then finally converge to one individual which in the most suitable environment, and obtain optimal solution to the problem. 4.2. Multi-Objective Decision The basic springboard of chaotic optimization is the er-

godicity, that chaotic motion can pass all states nonre-curring in a certain range. The characteristic can be ef-fective mechanism of avoiding local optimal solution and the difficulty of the continuity and differentiability of objective functions and constraints. The idea of chaotic optimization is to transfer area coverage from chaotic series to decision variables, use the new chaotic variables for searching and iterative comparison. If the criterion of stop is satisfied, then export the optimal results. Mathe-matically, the general multi-objective constrained opti-mization problem can be stated as follows:

1 2min ( ) ( ), ( ), , ( )

. . ( ) 0 , , ,

( ) 0 , , ,

m

i

j

f f f f

s t g i k

h j l

y x x x x

x

x

(9)

where, x is the decision vector. 1 2( , , , )nx x x x nR X . X is decision space. y is the objective

function vector. 1 2( , , , ) mmy y y R y Y . Y is obj-

ective space. ( )f x is the objective function to be opti-

mized, ( )ig x , ( )jh x are the constraints imposed on

the design, n, m (m > 1) are the dimensions of decision vector and objective functions respectively.

For multi-objective optimization problem, it is diffi-cult to find absolute optimal solution. Mostly, we choose the best equilibrium solution which has precision to a certain extent and practical significance according to the request of problem.

Traditional multi-objective programming method is linear weighted sum method. It converts the problem to single objective problem by weighted coefficient. There are some disadvantages such as the units of different objectives are not the same for comparison and subjec-tivity is obvious. This paper adopts constraint method. Supposing the problem has p objectives. The idea is to ascertain a main objective 1( )f x , and take the p-1 ob-

jectives as secondary targets, choose some threshold values ( 2,3, , )ju j p through the experience of deci-

sion-makers, then change the secondary targets to con-straints. So the problem is to solve the single objective optimal problem as follows.

1min ( )

. . ( ) ,j j

f x

s t x S x S f x u j p

(10)

4.3. Constraints Treatment Chaotic optimization is a direct search method, which requests dealing with the constraints. Penalty function method is effective for the constraints treatment. Its basic idea is to add a penalty item to the objective function as

X. F. HUANG ET AL.


186

(11), converting the initial problem to the new non-restraint optimization problem with the penal function, was- hing out the non-feasibility solution through punishing the solutions dissatisfying the constraints and finally gaining the optimal solution.

( , ) ( ) ( ( ))p f c x x x (11)

where, ( )f x is the objective function of initial problem,

( ( ))c x is the penalty item.

The paper adopts non-differentiable exact penalty function method to convert the constraints to non-restraint optimization problem. It avoids the sequence character in computation, accords the solution of restraint optimization with the minimum points of penalty function. It also avoids the difficulty of the non-differentiability, which is effective for the optimization without grads information. With the work upwards, multi-objective restraint optimization problem is converted to single objective non-re-straint optimization problem.

In the paper, objective II and objective III is trans-ferred into constraints by non-differentiable exact pen-alty function method. The specific action is: firstly, am-plification to guaranteed output, as lower limit output of the minimum time output. The value needs several trials. Secondly, the minimum ecological water shortage objec-tive will be transferred to two constraints, one is to meet the minimum ecological water demand of river down-stream, and the other is to meet the minimum ecological water demand of the reservoir itself.

The objective II can be transferred to guaranteed out-put constraints as follows:

0t tAQ H N (12)

where, 0N is the amplified guaranteed output.

The objective III can be transferred to ecological water demand constraints. Considering the ecological hydropower station reservoir operation, the reservoir and down-stream of the ecological and environmental problems were emphatically solved, the following river ecological water requirements lower bound to determine the process of ecological environment of the reservoir discharged. Therefore, the ecological water demand constraints in-cluding two aspects: ecological storage capacity constraints and the downstream of ecological water demand constraints.

,max( , )t ze t t tVL V V VH (13)

mint tWh Wh (14)

where, ,ze tV is the ecological environment storage ca-

pacity of reservoir; tVL is the dead storage capacity of

reservoir; tVH is the beneficial capacity in non-flood

period, in flood period is the flood control storage capac-ity; tWh is the discharged ecological flow of reservoir

in time t period; minlWh is the minimum ecological wa-

ter demand of downstream in time t period. 4.4. Steps of CGA The idea of multi-objective chaotic genetic optimization algorithm is to decompose the problem into a single un-constrained optimization problem. Chaotic optimization theory and genetic algorithm are coupled to solve the optimization problem. The steps of CGA are as follows:

Step 1. Multi-objective decision. Constraints method is used to deal with the multi-objective problem. The problem is transformed into single problem by (11).

Step 2. Constraints treatment. Non-differentiable exact penalty function method is used to deal with the constraints. We choose a certain penal factor to constitute penalty item. The problem is converted into the non-restraint optimization problem according to (15). Then we gain the optimization problem of continuous object as fol-lows:

1 2min ( , , , ) [ ,n i i if x x x x a b i n (15)

Step 3. Parameters setting. Ascertaining the numbers of variables is n, and the bounds is [ai, bi], the population scale of genetic arithmetic is M, the maximal iteration times of the arithmetic is T, the cross probability of crossover probability is Pc, the mutation probability is Pm.

Step 4. Initialization. Choosing n different initial val-ues and acquiring n chaotic variable serial ,i p through

Logistic mapping. 1,2, ,i n , p is the length of the chaotic variable serial. Logistic mapping is as follows:

, 1 , ,(1 )i j i j i j , 0,1,2, , 1j p (16)

where, is controlled parameter, if 00 1 ,

4 , then the (11) is in chaotic status and has all char-

acteristic of chaotic motion. Step 5. Magnify the ranges of chaotic series into the

confines of optimal variables with (17).

, ,( )i j i i i i jx a b a , 1, 2, ,j p (17)

Step 6. Fitness function values calculation. Choose a proper fitness function to calculate the fitness value. Fit-ness value will be sorted in descending, select the 10 percent group on better fitness directly into the next gen-eration of groups, all populations of the selection, cross-over and mutation were carried out.

Step 7. Calculate the new fitness value and make ad-justments, and sort the group according to the fitness value, then replace the worst of which 10% of fitness, and maintain the files. Calculate the average fitness value and compare with the maximum, if within the allowable error, then end the searching process, output the optimal solution, otherwise continue.

X. F. HUANG ET AL.


187

Step 8. Plus a chaotic disturbance to the current gen-eration of group fitness value less than 90 percent of the corresponding optimization variables, through the carrier means mapped to optimization variables, perform itera-tive calculation, with the increase of iterations number, iteration gradually approach to the optimal solution. Un-til the time, two figures out the difference between the average fitness is less than a pre-given small positive number. Chaotic disturbance can be done as follows: the number of iterations to meet the initial optimal solution ( 1 2, , , nx x x ) is mapped to (0,1) interval, the optimal decision by the initial vector, denoted by ' , the chaotic mapping function iteration times K get chaotic sequence (K is the length of the chaotic sequence), set k in the chaotic sequence of the k value ( 1, 2, ,k K ), write

k composed of the ground n-vector, by Equation (18) Find the Chaos decision vector '

k after disturbance ' '(1 )k k ( 1,2, ,k K ) (18)

The formula for the (0,1), a value range can be adaptive selection, search the initial large, late small, according to (19) to determine :

11 ( )mk

k (19)

The formula m for a positive integer, determined according to the number of objective function, generally greater than or equal to 2; k for the chaotic map iterations.

Step 9. According to the fitness value, resort the groups, calculate the average fitness and compare it with the maximum. If it is in the allowable error, end the process of optimization, and output the optimal solution, otherwise, go to Step 5. 5. Case Study Study a certain hydropower station data as an example.

The curve between reservoir capacity and water level up- stream and the curve between water levels downstream with discharge of the hydropower station reservoir are given. The total reservoir capacity is 896 million m3, regulating capacity is 445 million m3, the normal water level is 977.0 m, dead water level is 948 m, flood-control water level is 966.0 m. The coefficient of output power takes 8.3, guaranteed output is 185 MW. Installed capac-ity is 1080 MW. The largest discharge is 1000 m3/s. Ac-cording to the time-sharing surfing electricity price pol-icy, in dry season period on the basis of the flat water floating upward 50 percent, in flood period, it falls 25 percent on the basis of the flat water. In flat-water period, the benchmark price is 0.247 RMB/kW·h, in dry season the floating price coefficient (December-next April) is 1.50, in flat-water period (May and November) is 1.00, and in flood period (June-October) 0.75. In a day, the power price of normal period is carried out by regulation, spike and peak hours floating upward 22 percent, and valley hours fall 40 percent. According to the annual average runoff data, the fore-mentioned model and CGA is used for optimal operation. The initial population of the model takes 2000. Crossover probability takes 0.9. Mutation probability takes 0.1. Permitted error takes 1.0 × 10-8. The largest iteration number takes 50. Logistic mapping initial value takes the values belong to [0.51, 074]. The software of MATLAB is used for calculation. The program runs 10 times, and we take the best result of them. In each program runs, the results of each genera-tion to be superior than the previous generation, or the results equal to the previous generation, it is the result of using the optimal retention policies. The final results can be seen in Table 1.

The ecological water demand of the river downstream is calculated by Tennant method and minimum monthly runoff method, and takes the bigger. The ecological wa-ter demand of the reservoir itself is less than dead storage reservoirs 451 million m3, so the smallest ecological wa-ter demand of the reservoir itself is met.

Table 1. The result of optimal operation of hydropower station in electricity market environment.

Month Initial level /m Final level /m Generating flow /m3/s Output /104 Kw Generated energy /108 Kw·h Generated benefit /108 RMB

7 976.47 966 945.48 53.03 3.87 1.43

8 966 948 1273.16 55.59 4.06 1.50

9 948 948 1254.29 45.44 3.32 1.23

10 948 976.56 766.30 37.62 2.75 1.02

11 976.56 976.97 535.05 33.03 2.41 0.60

12 976.97 976.79 330.70 20.65 1.51 0.28

1 976.79 976.65 343.55 21.39 1.56 0.29

2 976.65 976.91 382.60 23.80 1.74 0.32

3 976.91 976.58 601.90 37.04 2.70 0.50

4 976.58 979.96 493.82 31.17 2.28 0.42

5 979.96 976.53 322.40 20.50 1.50 0.37

6 976.53 976.47 337.15 20.94 1.53 0.57

Total 29.21 8.53

X. F. HUANG ET AL.


188

From Table 1, the initial and the final water level are meeting the upper and lower limits. The output value of each month is also within the control range. Generating flows of each month are also meeting the ecological wa-ter demand of flow downstream. Therefore, the results meet the requirements.

In order to illustrate the superiority of the algorithm, four kinds of methods including dynamic programming (DP), genetic algorithm (GA) and chaos optimization algorithm (COA) were used in this paper, use the four kinds of methods to calculate the hydroelectric reservoir opti-mal operation model, the results can be seen in Table 2.

From Table 2, CGA has the best results, next are the GA and the COA, and finally is the DP. DP needs the state discrete-time and stores state information in the optimiza-tion. The more discrete points it has, the higher precision it gets. But it increases the optimization run-time very much. GA depends on random variables, and it does not need the storage of eliminated discrete points. Therefore, it shows an obvious superiority in solving these optimization prob-lems. So for the same problems, GA can greatly save the computer’s memory demand. However, because of the probability search features, the results of GA are unstable. COA maps the chaotic sequences which generate by Lo-gistic mapping to the range of optimal variables, and then does iterative searching, fully using the ergodicity charac-teristic of chaotic optimization. The annual generating capacity which is calculated by COA is greater than that of DP, but it needs larger chaotic sequence length, the time consuming of the program is longer. Therefore, the search efficiency of COA needs to be improved.

CGA combines the advantages of GA and COA, using the ergodicity feature of chaotic optimization, mapping the chaotic sequences which generated by Logistic mapping to the ranges of optimization variables, and then using the optimization mechanism of genetic algorithm for selection. After crossover and mutation, a chaotic disturbance is added to the variables corresponding to the degree of op-timization variables in the current generation of groups, and then carrying out the genetic manipulation until the termination of the proceedings meets the conditions. Fi-nally, output the optimal solution. Therefore, CGA has the advantages of high efficiency, good convergence per-formance and it approaches to the global optimal solution better. So it is the best algorithm to solve the optimal op-eration of hydropower station reservoir. Of course, due to the need of larger population size to achieve ergodicity, the calculation time of CGA is longer than GA.

The paper also does the research that not considering the ecology. The objectives take the largest generated benefit and the maximum output of the minimal output stage in the year.

The generated energy of hydropower reservoir optimal operation is 8.62 billion RMB without ecology consid-eration, which is 9 million RMB more than that with ecology consideration. But in January and February, the

generating flow is less than the minimum ecological wa-ter flows. The results can be seen in Table 3.

Can be seen from Table 3, without considering the eco-logical optimal operation of hydropower generating capac-ity of the reservoir and consider the ecological results of optimal operation of hydropower station reservoir compared to an increase of only 0.09 million RMB about 1.06%, the two are close, But do not take into account the environment of the river hydropower reservoir optimal operation of the negative environmental impact is far-reaching. Optimal operation of hydropower and ecological considerations will help protect the river environment and promote sustainable use of water resources; sacrifice in exchange for a small part of the power generation efficiency and harmonious social and economic development and ecological environment is worth it and very necessary. 6. Conclusions In electric power market, as a result of “separate the sta-tion and network, price bidding”, and the independent property of all generating companies, the focus of opti-mal operation for the power plant transforms from single safety production to taking economic benefits as central of all-round integrated development. Power generation benefit will be paid more attention by generation compa-nies. In this paper, a multi-objective chaotic optimal op-eration model based on time-varying electricity price is established in electric power market, considering the ecological water requirements of downstream. To solve the model, a multi-objective chaotic genetic optimization algorithm is exploited, which has the advantages of high search efficiency, good convergence performance, faster pace converge to the global optimal solution. It greatly increases the efficiency and effectiveness of optimal op-eration, and enriches and develops the theory and method of optimal operation for hydropower station reservoir.

Table 2. Results calculated by different algorithms.

AlgorithmsOptimal results

/108RMB

Increase the proportion of annual electricity output /%

Time consuming

/s

DP 8.24 1.10 10.4

GA 8.51 3.86 54.9

COA 8.39 2.74 387.3

CGA 8.53 6.23 265.2

Table 3. Results with and without ecology consideration.

With ecologyConsideration

/108RMB

Without ecology consideration

/108RMB

Generated benefit decrease

/108RMB

Decreaseratio /%

8.53 8.62 0.09 1.06

X. F. HUANG ET AL.


189

As the development of people’s consciousness about the ecological environment and the idea of harmonious coexistence between human and nature, the ecology is necessary in optimal operation of hydropower stations. The paper takes full consideration of the ecology water demand of the river downstream and the reservoirs itself. In the actual calculations, the objective is solved through the constraint method. The example shows that the result without ecology consideration is a little more than that with ecology consideration, but the negative environ-mental impact is far-reaching. 7. Acknowledgements The authors wish to thank Yinqin CHEN of Nanjing Un- iversity of Technology for her careful review and cons- tructive modifications. And the work is supported by the The National Natural Science Fund of China (No. 50909073) and Natural Science Fund of Hohai Univer-sity (No.2009422011) and the Fundamental Research Funds for the Central Universities (2013B1020084). 8. References [1] J. D. C. Little, “The Use of Storage Water in a Hydroe-

lectric System,” Operational Research, Vol. 3, No. 2, March 1955, pp. 187-197.

[2] Y. C. Zhang, F. S. Li, Y. F. Du, Y. F. Huang and S. Y. Xiong, “The Optimization of Hydropower Station Res-ervoir Dispatching,” in Chinese, Journal of Huazhong University of Science and Technology, Vol. 9, No. 6, De-cember 1981, pp. 49-56.

[3] W. Y. Tan, S. X. Huang, J. M. Liu and S. X. Fang, “Ap-plication of Dynamic Programming in Optimizing the Regulation of Reservoirs of Hydroelectric Stations,” in Chinese, Journal of Hydraulic Engineering, Vol. 13, No. 135, July 1982, pp. 1-7.

[4] Q. Huang and Z. Q. Yan, “A Discussion on Methods of Long-Term Optimal Operation of Hydropower Stations Reservoir,” in Chinese, Journal of Hydro Electric Power, Vol. 3, No. 3, March 1987, pp. 1-7.

[5] G. W. Ma, “Study on Stochastic Optimum Problems for Multireservoir Hydroelectric Systems,” in Chinese, Jour- nal of Xi’an University of Technology, Vol. 7, No. 4, De-cember 1988, pp. 65-74.

[6] D. G. Shao, X. Y. Song, J. Xia and Z. Q. Sun, “Research on Real-Time Optimal Flood Dispatching Model for Yanghe Reservoir,” in Chinese, Advances in Water Sci-ence, Vol. 10, No. 2, February 1999, pp. 135-139.

[7] J. Wan and H. Y. Chen, “A Study on Optimal Operation of Hydropower Group by Combined Model of Decompo-sition Coordination and Aggregation-Decomposition of Large System,” in Chinese, Journal of Hydroelectric En-gineering, Vol. 15, No. 2, February 1996, pp. 41-50.

[8] S. Y. Chen and H. C. Zhou, “Fuzzy Optimal Decision Theory and Model of the Multistage and Multiobjective

System,” in Chinese, Water Resources and Power, Vol. 9, No. 1, January 1991, pp. 9-17.

[9] W. Robin and S. Mohd, “Evaluation of Genetic Algo-rithms for Optimal Reservior Resources,” Journal of Wa-ter Resource Planning and Management, Vol. 125, No. 1, January 1999, pp. 25-33.

[10] G. W. Ma and L. Wang, “Application of a Genetic Algo-rithm to Optimal Operation of Hydropower Station,” in Chinese, Advances in Water Science, Vol. 8, No. 3, March 1997, pp. 275-280.

[11] T. S. Hu, Y. H. Wan and S. Y. Feng, “Research on the Artificial Neural Network Methodology for Multi-Res-ervoir Operating Rules,” in Chinese, Advances in Water Science, Vol. 6, No. 1, January 1991, pp. 9-17.

[12] S. H. Zhang, Q. Huang, H. S. Wu and J. X. Yang, “A Modified Particle Swarm Optimizer for Optimal Opera-tion of Hydropower Station,” in Chinese, Journal of Hy-droelectric Engineering, Vol. 26, No. 1, January 1996, pp. 41-50.

[13] G. Xu and G. W. Ma, “Optimal Operation of Cascade Hydropower Stations Based on Ant Colony Algorithm,” in Chinese, Journal of Hydroelectric Engineering, Vol. 24, No. 5, May 1991, pp. 9-17.

[14] L. S. Suo, “Water Conservancy’s ‘Special Features’— The New Ideas about the Water Conservancy Construc-tion,” in Chinese, China Water Resources, Vol. 38, No. 1, January 2003, pp. 25-26.

[15] Z. R. Dong, D. Y. Sun and J. Y. Zhao, “Multi-Objective Ecological Operation of Reservoirs,” in Chinese, Water Resources and Hydropower Engineering, Vol. 38, No. 1, 2007, pp. 28-32.

[16] G. W. Ma, L. Wang and X. M. Guo, “A Optimal Opera-tion Model for Multiple Reservoir Power with Time- Varying Electricity Price,” in Chinese, Advances in Water Science, Vol. 13, No. 5, May 2002, pp. 583-587.

[17] H. P. Hu, D. F. Liu and F. Q. Tian, “A Method of Eco-logical Reservoir Reoperation Based-on Ecological Flow Regime,” in Chinese, Advances in Water Science, Vol. 19, No. 3, 2008, pp. 325-332.

[18] W. G. Yu, Z. Q. Xia and G. R. Yu, “Analysis of Eco-logical Ideas and Measures in Reservoir Scheduling,” in Chinese, Journal of Shangqiu Teachers College, Vol. 22, No. 5, 2006, pp. 148-151.

[19] E. N. Lorenz, “Deterministic Non-Periodic Flow,” Jour-nal of Atomsphere Science, Vol. 20, No. 2, June 1963, pp. 130-141.

[20] B. Li and W. S. Jiang, “Chaos Optimization Method and its Application,” in Chinese, Control Theory and Appli-cations, Vol. 14, No. 4, April 1997, pp. 613-615.

[21] L. Qiu, J. H. Tian, C. Q. Duan, X. N. Chen and Q. Huang, “The Application of Chaos Optimization Algorithm in Res-ervoir Optimal Operation,” in Chinese, China Rural Water and Hydropower, Vol. 30, No. 7, July 2005, pp. 17-19.

[22] W. Liang, S. L. Chen, C. Y. He, D. Y. Liu and J. Rui, “Op-timal Operation of Cascaded Hydropower Stations Based on Chaos Optimal Algorithm,” in Chinese, Water Resources and Power, Vol. 26, No. 1, January 2008, pp. 63-66.



Simulation on SO2 and NOX Emission from Coal–Fired Power Plants in North-Eastern North America

Shuangchen Ma School of Environmental Science and Engineering, North China Electric Power University, Baoding, China


Abstract MM5-SMOKE-CMAQ regional air quality modeling system was used to simulate pollutants emission from coal–fired power plants in North-Eastern North America. The effects of SO2 and NOX on air quality produc-ing from coal-fired power plants in the summer of 2001 were analyzed. Simulations show the contributions of SO2 and NOX emission from coal-fired power plants using different scenarios, coal-fired power plants from US and Canada contribute 67.2% and 32.8% for total SO2 concentration, 17.6% and 6.0% for total NOX concentration in researched domain. Some control measures for coal-fired power plants were discussed. Further controls for the emissions of SO2 and NOX from coal-fired power plants are necessary to reduce the adverse environmental effects. Keywords: Electric Generating units, CMAQ, Air Quality, Pollutants Control

1. Introduction There is a long history of public concern about the emis-sion of SO2 and NOX, these emissions contribute to acid deposition, ultimately leading to a wide range of envi-ronmental impacts, including damage to forests and soils, fish and other living things, materials, and human health. as well as the formation of fine particles and gases that can impair visibility. SO2 and NOX all are the precursors of acid rain [1]. Coal-fired power plants are regarded globally as the major contributor to local air quality deg-radation and to global environmental impacts such as acid rain and greenhouse phenomenon [2]. In the United States, over 70 percent of this electricity was produced through the combustion of fossil fuels, primarily coal and natural gas, while 26 percent of this generation was pro-duced from fossil fuels in Canada.

Legislation enacted for control the emission of these two pollutants during the past two decades. In the Clean Air Act Amendments (CAAA) of 1990, US Environ-mental Protection Agency (EPA) has promulgated several regulations and proposed rules to decrease these ad-verse effects by reducing the emissions of SO2 and NOX from coal-fired power plants. Since Congress passed the Clean Air Act (CAA), power plants have cut emissions of SO2 and NOX dramatically [3,4]. Many of these re-ductions have been achieved by conversion to lower- sulfur fossil fuels, primarily natural gas, with attendant

increases in costs. An alternative to conversion is to use mitigation technologies to reduce the emissions. The most common such technologies are reducing fuel sulfur levels and filtering emissions using flue gas desulfuriza-tion (FGD) and denitrification systems [5,6].

Analyses of the environmental effects arising from power plants using a variety of models [7-10] suggest that air quality effects depend on a wide variety of local atmospheric parameters as well as on the combustion technology. In some cases [11], regional transport domi-nates local sources during pollution episodes, while in other cases [12], transport is strongly affects by local topology. In view of this variable sensitivity, the effects of thermal power generation on air quality should be assessed on a case by case basis for well defined geo-graphical locations and time periods. Plumes from power plants in certain locations have been implicated in sig-nificantly reduced air quality [13,14].

In this study, we use regional atmospheric modeling to explore the air quality implications of coal-fired power plants in North-Eastern North America. We design dif-ferent scenarios to study the contributions of coal-fired power plants from United States and Canada. The con-cerned pollutants are SO2 and NOX, which all are the leading precursors of acid rain; they have large environ-mental influences on the acid sensitive regions of North- Eastern North America. Research results can be made references for pursuing further emission reductions in US and Canada.

S. C. MA


191

2. Modeling Description EPA’s Models-3/Community Multi-scale Air Quality (C- MAQ) [15] modeling system is a numerical transport and chemical model which is developed by US EPA and is implemented in a full modular approach. It means that the user can select different chemical schemes (CBM-IV, V, SAPRAC-99, RADM, etc.) and different numerical schemes. It has different capabilities such as: Process analysis, IRR analysis, etc.). It can handle interactions at different dynamic scales and among multi-pollutants with one modeling system. In this research, CMAQ (v4.5) is employed to simulate SO2 and NOX for the summer season of 2001 (May 1 through September 1). The model uses the Carbon Bond 4 (CB4) gas phase chemical mechanism with 2005 updates to extend the inorganic reac-tions [16].

2001 national emissions inventories was used and processed by the Sparse Matrix Operator Kernel Emissions Modeling System (SMOKE, v2.2) [17]. SMOKE system provides an efficient tool for converting emissions in-ventory data into the formatted emissions files: gridded, temporalized and speciated, which are required by CM- AQ. Area sources, non-road sources, mobile emissions, biogenic emissions and point sources are treated sepa-rately and merged.

Meteorological input data for the modeling runs are processed using the National Center for Atmospheric Re- search (NCAR) 5th generation Mesoscale Model (MM5, v3.0) [18,19].Important MM5 parameterizations and physics options apply to each summer include mixed phase microphysics, planetary boundary layer (PBL), and the land surface module. Meteorology-Chemistry Interface Processor (MCIP, v2.0) is used to process the MM5 output fields and generate the meteorological pa-rameter fields required by SMOKE and CMAQ as well as the dry deposition velocity fields of chemical species required by CMAQ.

The model, as applied here, uses a horizontal resolute- ion of 36 × 36 km2, with 23 layers in the vertical. The model domain covers most Eastern of America in order to minimize the effects of boundary conditions on model results. Additional model simulation was performed over an embedded domain covering the North-Eastern North America including the Great Lakes, with a grid spacing of 12 km. Modeling coarse domain and 2-way nested 12km domain are shown in Figure 1.

We explore the effects of the three emission scenarios described below:

1) Scenario 1 (Base case): All emissions in the 2001 criteria inventories were used.

2) Scenario 2: Coal emissions from power plants of US and Mexico removed from point source inventory, but kept the emission inventory of Canada same as Sce-nario 1.

3) Scenario 3: Coal emissions from power plants in

Canada were removed from point source inventory, but kept the emission inventories of US and Mexico same as Scenario 1. 3. Modeling Evaluation To validate the basic calculation, we compared the val-ues of SO2 and NOX produced by the model for the time period from May 1 to September 1, 2001 with the corre-sponding measurements from AQS air quality monitor-ing stations located in Domain 2.

The simulation for SO2 and NOX aren’t very good, the model has an under-prediction for SO2 and NOX average concentration; Normalized Mean Bias (NMB) and Normalized Mean Error (NME) for SO2 are –51.75% and 71.28%; for NOX are –9.77% and 55.38% respec-tively. The reason that modeled deviates from observa-tions is that measurements are made near the sources where the model emission schemes will never be able to reproduce small-scale fluctuations observed. Sub-grid scale variability in emissions will have a major impact on the comparison between the model and observation. More efforts should be done to improve the simulation for SO2 and NOX. Moreover, the study’s conclusions are obtained mainly by subtracting the scenarios from base case, thereby reducing the effects of errors in re-search results. 4. Modeling Results and Analysis 4.1. SO2 Emission and Changes Resulting from

Different Scenarios SO2 are formed from fuel containing sulfur (mainly coal and oil) is burned at power plants and during metal smelt-

Domain 1

Domain 2

110 w 100 w 90 w 80 w 70 w

50 N

40 N

80 N

50 N

40 N

80 N

100 w 90 w 80 w

Figure 1. The set of air quality modeling domain.

S. C. MA


192

ing and other industrial processes. SO2 emissions from power plants react with other chemicals in the atmos-phere to form sulfate particles, an important contributor to the fine particle mix that circulates with the air we breathe.

Figure 2 and Figure 3 show the average SO2 emission flux and SO2 concentration in North-Eastern North America in the summer season of 2001. The average SO2 emission flux was 0.021mol/s in whole domain in the summer. Several regions such as Chicago, southern Ohio, northern West Virginia and western Pennsylvania etc. were responsible for a significant proportion of SO2 emissions. Many large coal-fired power plants encompasses in Ohio River Valley, which is a well-known emission source of precursors of acid rain in the North America. Same as emission, these areas had higher SO2 average concentration in this summer.

Figure 4 shows the average difference of SO2 concen-tration from base case and Scenario 2, which produce by

Average SO2 emission flux in the summer of 2001

a = emission.maindata.d2 96

1 1

1.000

0.900

0.800

0.700

0.600

0.500

0.400

0.300

0.200

0.100

0.000 moles/s 102

Figure 2. Average SO2 emission flux in the summer season of 2001 in North-Eastern North America.

Average SO2 concentration in the summer of 2001

a = CCTM_2001 sulfurcon.d2 96

1 1

0.010

0.009

0.008

0.007

0.006

0.005

0.004

0.003

0.002

0.001

0.000 ppmV 102

Figure 3. Spatially resolved hourly Average SO2 concentra-tion in the summer season of 2001 in North-Eastern North America.

the average of SO2 concentration simulating from base case minus those of Scenario 2. The regions of high SO2 concentration are the areas affected by the SO2 emission of coal-fired power plants apparently. The highest dif-ference of SO2 concentration exists at the region of west Pennsylvania about 0.012 ppmv, it is easy to see that the Ohio River Valley is biggest source of SO2 from coal- fired power plants in researched domain.

Figure 5 shows the average SO2 concentration simu-lating from three scenarios in the summer of 2001. The average SO2 concentration of base case in domain 2 is 0.001 ppm in this summer. Average SO2 concentration getting from Scenarios 2 and 3 are 0.00033 and 0.000672 ppm, these mean coal-fired power plants from US and Canada contribute 67.2% and 32.8% for total SO2 con-centration in researched domain.

According to EPA report, power plants account for 69% of total SO2 emission in the US in 2001 [20]. So, this is consistent with our research results for the most part.

Average SO2 concentration between base case and scenario 2a = CCTM_2001basecase.d2, b = CCTM_2001scenario2.d2

96

11

0.012

0.011

0.010

0.008

0.007

0.006

0.005

0.004

0.002

0.001

0.000ppmV 102

Figure 4. Spatially resolved hourly average difference of SO2 concentration from base case and Scenario 2 in the summer season of 2001 in North-Eastern North America.

0

0.0002

0.0004

0.0006

0.0008

0.001

0.0012

Scenario 1 Scenario 2 Scenario 3

SO

2

co

nc

en

tra

tio

n,

pp

mV

Figure 5. Average SO2 concentration simulating from dif-ferent scenarios in the summer of 2001.

S. C. MA


193

4.2. SO2 Emission Control from Coal-Fired Power Plants

To help reduce acid rain, EPA is implementing a pro-gram to reduce releases of SO2 and other pollutants from coal-fired power plants. The first phase began in 1995 for SO2 and targets the largest and highest emitting power plants. The second phase (started in 2000) sets tighter restrictions on smaller coal-, gas-, and oil-fired plants. This program will reduce annual SO2 emissions by 10 million tons (almost half the 1980 level) between 1980 and 2010. The control of SO2 emissions is currently achieved using an innovative strategy called cap and trade, established by the CAAA of 1990 and adminis-tered by EPA, which is the world’s first large-scale cap- and-trade program for air pollution. The program is de-signed to reduce electric power sector emissions of SO2 through a national, market-based cap-and-trade system for SO2 emissions with the goal of reducing the adverse effects of acid rain. The total SO2 release allowed is set at a maximum of 8.95 million tons by the year 2010—ap-proximately half of 1980 emissions [21].

Four main technology strategies for SO2 emissions control have been used by the electricity industry [22]:

1) Tall gas stacks that disperse emissions away from immediate areas;

2) Intermittent controls which involve routine opera-tional adjustments to reduce power plant SO2 emissions in response to atmospheric conditions;

3) Pre-combustion reduction of sulfur from fuels; and 4) Removal of SO2 from the post-combustion gas

stream, the main method is FGD. There are a wide vari-ety of FGD techniques, of which the most common are wet scrubbers. Wet scrubbers work by using a slurry or solution to absorb SO2, producing an initially wet by- product. Frequently, limestone is used as the absorbent, generating gypsum as a by-product. Typically, FGD can achieve SO2 removal efficiency more than 90%. 4.3. NOX Emission and Changes Resulting from

Different Scenarios NOX is the term used to describe the sum of nitric oxide (NO), nitrogen dioxide (NO2), and other oxides of nitro-gen. NOX plays a major role in the formation of acid rain, secondary PM, ground level ozone, and smog in the atmosphere through a complex series of reactions with vola-tile organic compounds (VOCs), sunlight and water va-pors. The Relationship between VOC and NOX and O3 has extensive studies [23-26]. NOX also contributes to visibility impairment. The major sources of human-produced NOX emissions are high-temperature combustion processes such as those that occur in electrical power plants and automobiles. Anthropogenic NOx emissions in the US are estimated to total 22.2 Tg (as NO2) per year,

with 53% from transportation, coal-fired plants are an-other major sources of NOX because of the fuel they burn [27].

Figure 6 and Figure 7 show the average NOX emission flux and concentration in North-Eastern North Am- erica in summer season of 2001. The average NOX emis-sion flux was 0.20 mol/s in whole domain in the summer. Big cities such as Chicago, Detroit, and Toronto etc. have much more emission of NOX than other areas, which arising from local motor vehicle emissions primarily. The biggest average NOx concentration is 0.056 ppmv in Toronto city.

Figure 8 shows the average NOX concentration simu-lating from three scenarios in the summer of 2001. The average NOX concentration of base case in domain 2 is 0.00199 ppm in this summer. Average NOX concentra-tion getting from Scenario 2 in Toronto city change to 0.029 ppmv, this mean the emissions of coal-fired power plants from US can transport long distance through the

Average NOX emission flux in the summer of 2001

a = emission.maindata.d2 96

11

8.000

7.200

6.400

5.600

4.800

4.000

3.200

2.400

1.600

0.800

0.000moles/s 102

Figure 6. Average NOX emission flux in summer season of 2001 in North-Eastern North America.

Average NOX concentration in the summer of 2001

a = CCTM_2001NOX.d2 96

11

0.056

0.045

0.034

0.022

0.011

0.000ppmV 102

Figure 7. Average NOX concentration in summer season of 2001 in North-Eastern North America.

S. C. MA


194

0

0.0005

0.001

0.0015

0.002

0.0025

Scenario 1 Scenario 2 Scenario 3

NO

Xc

on

ce

ntr

ati

on

, p

pm

V

Figure 8. Average NOX concentration simulating from dif-ferent scenarios in the summer of 2001.

atmosphere to have obvious effect on Toronto. Average NOX concentration getting from Scenario 2 and Scenario 3 is 0.00164 ppmv and 0.00187 ppmv. So, we can get coal-fired power plants from US and Canada contributing 17.6% and 6.0% for NOX concentration in researched domain. This is lower with same EPA research results [20], coal-fired power plants contributed about 22% of all U.S. NOX emissions in 2002, and this may be the cause of under-prediction of model for NOX and the dif-ferent object domain. 4.4. NOX Emission Control from Coal-Fired

Power Plants In the past few years, EPA has promulgated several rules to reduce NOX emissions, new rules for stationary sources (the Clean Air Interstate Rule, the Clean Air Mercury Rule, and the Clean Air Visibility Rule), as well as State plans to attain the National Ambient Air Quality Stan- dard (NAAQS) of fine particle and ozone, will also significantly reduce future NOX emissions. Beginning with the CAAA of 1990, existing generators also have faced increasingly stringent regulation of their nitrogen oxide (NOX) emissions. Restrictions on summer emissions of NOX from electricity generators in a majority of eastern states are expected to become even tighter during the next decade with the implementation of the call for amend-ments to state implementation plans (SIPs) from the US EPA, known as the NOX SIP Call. This new regulation is designed to address the long-range transport of NOX as a contributing factor to summer air pollution in cities on the East Coast. Recent lawsuits filed by EPA and New York State also have raised the possibility that many existing generating sources were negligent in not bring-ing their facilities into compliance with New Source Performance Standard (NSPS) when they made substan-tial investments enabling greater electricity generation at these facilities [28].

Reduction of NOX emissions from industrial combus-

tion sources, especially coal-fired power plants is an im-portant consideration in efforts undertaken to address the NOX environmental concerns. New regulations announ- ced by the US EPA facilitate utilities to develop new, efficient, and robust post-combustion NOX control tech-nologies [29]. The popular primary control technologies in use in the United States are low-NOX burners (LNB) and over-fire air. The average NOX reductions for spe-cific primary controls have ranged from 35% to 63% from 1995 emissions levels. The secondary NOX control technologies applied on U.S. coal-fired utility boilers include re-burning, selective catalytic reduction (SCR) and selective non-catalytic reduction (SNCR) [30]. 5. Conclusions In this study, we use a regional atmospheric modeling system-CMAQ to compute the air quality effects of coal- fired power plants in North-Eastern North America. For the time period from May 1 to September 1, 2001, we consider three different scenarios to study the effects of SO2 and NOX emission from coal-fired power plants. the contributions of the coal plants for SO2 and NOX are found to be big, coal-fired power plants from US and Canada contribute 67.2% and 32.8% for SO2 concentra-tion, 17.6% and 6.0% for NOX concentration in re-searched domain. The use of existing emission reduction rules and technologies can diminish the contributions of coal-fired power plants for total emission of SO2 and NOX. Further control measures for the emissions of SO2 and NOX from coal-fired power plants are necessary to reduce the negative environmental effects. 6. Acknowledgements The authors would like to acknowledge necessary condi-tions provided by Waterloo Centre for Atmospheric Sci-ences, University of Waterloo. The work is sponsored by SRF for ROCS, SEM. 7. References [1] L. Yang, I. Stulen, L. J. De Kok and Y. Zheng, “SO2,

NOX and Acid Deposition Problems in China Impact on Agriculture,” Phyton-Annales Rei Botanicae, Vol. 42, 2002, pp. 255-264.

[2] M. V. Toro, L. V. Cremades and J. Calbo, “Relationship between VOC and NOX Emissions and Chemical Produc-tion of Tropospheric Ozone in the Aburra Valley (Co-lombia),” Chemosphere. Vol. 65, No. 5, 2006, pp. 881- 888.

[3] F. B. Chaaban, T. Mezher and M. Ouwayjan, “Options for Emissions Reduction from Power Plants: An Eco-nomic Evaluation,” International Journal of Electrical Power & Energy Systems, Vol. 26, No. 1, 2004, pp. 57-

S. C. MA


195

63.

[4] M. Victor, “Recent EPA Regulatory Actions and Effects on NOX Controls,” Proceedings of the 2006 Environ-mental Controls Conference, U.S. Department of Energy, National Energy Technology Laboratory, Pittsburgh, 2006, pp. 1-2.

[5] J. R. Swinton, “Phase I Completed: An Empirical As-sessment of the 1990 CAAA,” Environmental & Re-source Economics, Vol. 27, No. 3, 2004, pp. 227-246.

[6] J. H. Goo, M. F. Irfan, S. D. Kim and S. C. Hong, “Ef-fects of NO2 and SO2 on Selective Catalytic Reduction of Nitrogen Oxides by Ammonia,” Chemosphere, Vol. 67, No. 4, 2007, pp. 718-723.

[7] D. Stevenson and R. Dale, “Limestone FGD System Ret-rofit to San Juan Generating Station: Start-up Problems and Performance Following Start-up,” Combined Power Plant Air Pollutant Control Mega Symposium, 2004, pp. 823-836.

[8] J. M. Hao, L. T. Wang, M. J. Shen, L. Li and J. N. Hu, “Air Quality Impacts of Power Plant Emissions in Bei-jing,” Environmental Pollution, Vol. 147, No. 2, 2007, pp. 401-408.

[9] F. Mehdizadeh and H. S. Rifai, “Modeling Point Source Plumes at High Altitudes Using a Modified Gaussian Model,” Atmospheric Environment, Vol. 38, No. 6, 2004, pp. 821-831.

[10] R. P. Hermann, et al., “Predicting Premature Mortality from New Power Plant Development in Virginia,” Ar-chives of Environmental Health, Vol. 59, No. 10, 2004, pp. 529-535.

[11] M. J. Martin, et al., “High Performance Air Pollution Modeling for a Power Plant Environment,” Parallel Computing, Vol. 29, No. 11-12, 2003, pp. 1763-1790.

[12] J. C. H. Fung, et al., “Observational and Modeling Analy-sis of a Severe Air Pollution Episode in Western Hong Kong,” Journal of Geophysical Research, Vol. 110, No. D9, 2005, p. 9105.

[13] J. Chen, R. Bornstein and C. G. Lindsey, “Transport of a Power Plant Tracer Plume over Grand Canyon National Park,” Journal of Applied Meteorology, Vol. 38, No. 8, 1999, pp. 1049-1068.

[14] S. R. Springston, et al., “Chemical Evolution of an Iso-lated Power Plant Plume during the TexAQS 2000 Stu- dy,” Atmospheric Environment, Vol. 39, No. 19, 2005, pp. 3431-3443.

[15] J. C. St John and W. L. Chameides, “Possible Role of Power Plant Plume Emissions in Fostering O3 Exceedence Events in Atlanta, Georgia,” Journal of Geophysical Re-search, Vol. 105, No. D7, 2000, pp. 9203-9211.

[16] D. W. Byun and J. K. S. Ching, “Science Algorithms of the EPA Models-3 Community Multiscale Air Quality (CMAQ) Modeling System,” U.S. Environmental Protec-tion Agency, NERL, Research Triangle Park, 1999.

[17] M. W. Gery, G. Z. Whitten, J. P. Killus and M. C. Dodge, “A Photochemical Kinetics Mechanism for Urban and Re-gional-Scale Computer Modeling,” Journal of Geophysical

Research, Vol. 94, No. D10, 1989, pp. 12925-12956.

[18] M. Houyoux, J. Vukovich and J. Brandmeyer, “Sparse Matrix Kernel Emissions Modeling System: SMOKE User Manual,” MCNC-North Carolina Supercomputing Center, 2000.

[19] G. A. Grell, J. Dudhia and D. R. Stauffer, “A Description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5),” NCAR Technical Note, NCAR, Boulder, CO, 1995.

[20] Y. R. Guo and S. Chen, “Terrain and Land Use for the Fifth-Generation Penn State/NCAR Mesoscale Modeling System (MM5): Program TERRAIN,” NCAR/TN-397+ IA NCAR Technical Note, National Center for Atmos-pheric Research, Boulder, 1994.

[21] U.S.EPA, “Acid Rain Program 2002 Progress Report,” EPA 430/R-03-011, Washington DC, 2003.

[22] S. Napolitano, J. Schreifels, G. Stevens, M. Witt, M. La-Count, R. Forte and K. Smith, “The U.S. Acid Rain Pro-gram: Key Insights from the Design, Operation, and As-sessment of a Cap-and-Trade Program,” The Electricity Journal, Vol. 20, No. 7, 2007, pp. 47-58.

[23] NAPAP, “NAPAP Report to Congress: An Integrated Assessment National Acid,” Precipitation Assessment Program Office of the Director, Washington, D.C., 2005.

[24] M. R. Taylor, E. S. Rubin and D. A. Hounshell, “Control of SO2 Emissions from Power Plants: A Case of Induced Technological Innovation in the US,” Technological Fo-recasting and Social Change, Vol. 72, No. 6, pp. 697- 718, 2005.

[25] Y. P. Peng, K. S. Chen, C. H. Lai, P. J. Lu and J. H. Kao, “Concentrations of H2O2 and HNO3 and O3-VOC-NOX sensitivity in ambient air in southern Taiwan,” Atmospheric Environment, Vol. 40, No. 35, 2006, pp. 6741-6751.

[26] I. Filella and J. Penuelas, “Daily, Weekly and Seasonal Relationships among VOCs, NOX and O3 in a Semi-Ur-ban Area near Barcelona,” Journal of Atmospheric Che- mistry, Vol. 54, No. 2, 2006, pp. 189-201.

[27] A. F. Stein, E. Mantilla and M. M. Millan, “Using Meas-ured and Modeled Indicators to Assess Ozone-NOX- VOC Sensitivity in a Western Mediterranean Coastal En-vironment,” Atmospheric Environment. Vol. 39, No. 37, 2005, pp. 7167-7180.

[28] U.S. EPA, “National Air Quality and Emissions Trends Report, 1900-1998,” EPA 454/R-00-003, U.S. Environ-mental Protection Agency, Washington, DC, 2000.

[29] D. Burtraw, K. Palmer, R. Bharvirkar and A. Paul, “Re-structuring and Cost of Reducing NOX Emissions in Electricity Generation,” Discussion Paper 01-10REV, Resources for the Future, Washington, DC, 2001.

[30] Y. Fu and M. D. Urmila, “Cost Effective Environmental Control Technology for Utilities,” Advances in Environ-mental Research, Vol. 8, No. 2, 2004, pp. 173-196.

[31] R. K. Srivastava, R. E Hall and S. Khan, “Nitrogen Ox-ides Emission Control Options for Coal-Fired Electric Utility Boilers,” Journal of the Air & Waste Management Association, Vol. 55, No. 9, 2005, pp. 1367-1388.



Multi-Bias Model for Power Diode Using a Very High Description Language

Hassene Mnif, Montassar Najari, Hekmet Samet, Nouri Masmoudi National School of Engineering of Sfax, LETI Laboratory, Sfax, Tunisia

E-mail: [email protected] Received May 3, 2010; revised June 12, 2010; accepted July 16, 2010

Abstract This study focused on the determination and analysis of an accurate analytical model for PIN diode under different bias conditions. This approach employs analytically derived expressions including the variation of the depletion regions in the device to make the used model available over a wide range of testing conditions without remake the parameters extraction procedure. The validity of the proposed extraction procedure has been verified by the very good agreement between simulated and measured current and voltage waveforms reverse recovery at different range of the operation conditions. The model is developed and simulated with the VHDL-AMS language under Ansoft Simplorer® Environment. Keywords: Power PIN Diode, VHDL-AMS, Modeling, High Injection, Parameter Extraction

1. Introduction Design of integrated power systems requires prototype- less approaches. Accurate simulations are necessary for analysis and verification purposes. Simulation relies on component models and associated parameters.

Most of the power PIN diode models provided by the literature are not valid for a mutli-bias simulation be-cause of their limitation to accurately describe the de-vice’s physical rules.

One of these physical limitations can be due to sup-posing a fixed depletion region width which is the case of several studies [1]. In [2], the power PIN diode model includes three empirical and non physical equations to escape the extraction procedure when using the model for multi-bias conditions.

In this study, the developed power PIN diode model supposes that the width of the intrinsic bulk region is a function of reverse applied voltage as it’s reported by Neudeck and Streetman [3].

Our study is structured as follows. In Section 2, a de-tailed presentation of a new moving boundary diffusion model for the power PIN diodes is made. Section 3 treats the test bench circuit and demonstrates a first comparison with experimental data. Section 4 investigates on the influence of the parasitic inductance in simulation result and the improvement of simulated test circuit by the cor-rection of this parasitic inductance value. Finally, Section 6 gives some conclusions and future study.

All the operations (implementation, extraction, optimization …) are made under the free mixed signal simula-tor “Simplorer-sv 7.0” by Ansoft®. 2. PIN Diode Model Description As it is reported in [2], the model is formulated for a P+-

IN+ type diode based on the Lauritzen power diode model [1]. Lauritzen and Al.’s model is based on a sys-tematic technique for analytical modeling; the lumped- charge modeling technique (LC) developed by Linvill [4].

Figure 1 shows the charge distribution in a conducting pin diode under forward conduction.

The charge distribution in the i-region is assigned to four charge storage nodes following the lumped model approach developed by Linvill [4] with:

Figure 1. Charge distribution profile in diode forward conduction [1].

H. MNIF ET AL.


197

In the left node: o q1 is the total charge o p1 is the average hole concentration o δ is the respective width

In the adjacent node: o q2 is the total charge o p2 is the average hole concentration o d is the respective width

Table 1 summarizes the complete set of “Lauritzen” and Al.’s model equations [1], collected together in a simple and consistent form.

More information about the original model can be found on [1].

By seeing this previous set of equations, one can clearly remark that there is no variation of the special dimen-sions of depletion regions in the device.

So, by considering the fact that the width of the intrin-sic bulk region reduces significantly during the reverse biased condition, more accurate simulation results for reverse recovery current and voltage can be found.

An improvement was made in [2], which consists of introducing some empirical equations. In this study the developed PIN diode model considers the variation of depletion regions width in the p+-i and i-n+ junctions as depicted in Figure 2.

Table 1. Lauritzen model’s equations [1].

Equation form Description Observations Number

E MM

M

q qi t

T

iM(t): Diode current with time dependency. ----- (1)

EE S

T

vq I exp 1

V

qE: Injected charge level at both the p+ I junction

and n+i junction qE = 2q0 with q0 represents the variable remain-ing after δ 0

(2)

EE SE

T

2vi I exp 1

V

iE: End region recombination current Occurs at very high current level (3)

E MM M

M

q qdq q0

dt 2T

qM: Total charge in the i-region

qM = 2qE ; analysis made to only half of the structure

(4)

T M M0M

M M0 T M

V T R iv

q R V T

vM: Voltage across half of the i-region ----- (5)

E M Sv 2v 2v R i V: Total diode voltage Kirchoff voltage law (6)

j

E M

dqi i i

dt I: Total diode current

Kirchoff current law with qj = ∫Cjd(2vE) is the total charge in the capacitor

(7)

Iregion N- regionP- region

q1

q2 q3

q4

dδ2WI

X(μm)

(a)

Iregion N- regionP- region

WJ

q2 q3

q 01

dδ =

0

2WI

X(μm)

Wt

Depletedregion

q 04

(b)

Figure 2. Improved charge distribution profile in diode forward conduction: (a) static depletion region width; (b) dynamic depletion region width.

H. MNIF ET AL.


198 3. A Moving Boundary Diffusion Model In fact, by applying a reverse potential and due to the as- ymmetrically doped junction, the junction region exte- nds primarily into the less doped intrinsic side. The half width of the dynamic intrinsic bulk region Wt can be calculated as:

Wt = WI – WJ (8)

in which WI is the half physical intrinsic region. Unlike the low injection phenomenon, the junction-

depletion width WJ variation effects under the high in-jection phenomenon can be written as [5]:

12

S bi AJ

diodeI

sat

2 V VW

iq N

qAv

(9)

where: εs = The silicon permittivity; NI = The doping level of the bulk region; q = The electron charge; Vbi, VA = Respectively the bult-in potential and the

applied voltage to the junction; idiode = The total current under the device; A = The device active area; vsat = The carrier saturation velocity. By adding this new concept to Najjari’s and Al. Model

[2], many parameters, especially the carrier transit time, will be affected and it’s crucial to investigate on this.

Transit time TM: As it’s reported in Table 1, to cal-culate the diode current due to carriers injected to the i-region by the two junctions, i.e. p+/i and i/n+ junctions, the carrier transit time must be defined. So, by varying the depletion region width in the power PIN diode model with respect to the two (8) and (9), the carrier transit time TM will be varied and the following equation is crucial for establishing this relation:

2t

Ma

(W )T

4D (10)

with Da is the ambipolar diffusion constant defined by:

n pa T

n p

D 2 U

(11)

So, the set of equations in Table 1 will be affected by this new concept and this variation of the depletion re-gion with respect to the applied bias on the device will be considered in our new formulation of the power di-ode model.

In addition to the nine classical Lauritzen and Al.’s model parameters (τ, TM, ISE, IS, Cj0, RS, m, ФB and RM0), three new parameters describing the moving boundary effect are added, i.e. (WI, NI and τrr). The new improved

model has the following VHDL-AMS structure, Figure 3. 4. New Model’s Parameters Discussion In this section, an investigation about the new model parameters (WI, NI and τrr) influence will be done.

After writing the VHDL-AMS code for this power PIN diode model, the device is simulated under the cir-cuit simulator Simplorer from Ansoft.

The same switching cell used in [2] is considered like it’s shown in Figure 4.

The switching cell main operating conditions are the forward current IF and the reverse voltage VR. These conditions are imposed by a current and a voltage source respectively. The “IRF740” transistor is a fast NMOS type. The parasitic inductance Lm has to be estimated carefully since it controls the current rate to some extent. In Section 4, we will discuss how to fix the value of this parasitic capacitance in the right way.

In order to avoid damaging the device, generally, due to the voltage overshoot, a RC-snubber was placed in parallel with the diode. Because of the small values of the snubber resistance and capacitance, the varia-tion in component values with respect to temperature change could be neglected in so far as these variations might alter the diode switching waveforms [6].

Since the diode turn-off behavior is of more concern to engineers than the turn-on because of the important energy losses, only the experimental results for diode turn-off are performed.

VR

WI

IF

NI

tau

Turn-off characteristics

PIN VHDL-AMS model

Figure 3. New VHDL-AMS power PIN diode structure.

C

R D

Vdiode

VR RG

Lm

VG

Idiode

IF

IRF 740

Figure 4. Principal circuit for the measurements.

H. MNIF ET AL.


199

4.1. Influence of the Half Physical Intrinsic Region Width WI

Under this turn-off state, we perform a simulation of the device with different values of the physical intrinsic re-gion width WI and we capture the curves of the forward current, IF and the reverse voltage, VR.

Figure 5 shows that with increasing WI, we give per-mission to the device to accumulate more charge under the on-state. When switched to the off-state, the device gets more time to evacuate these accumulated charges, so we can see a higher maximal reverse voltage and current. 4.2. Influence of the Doping Level NI Figure 6 shows that with increasing the doping profile NI in the device, the device gives more important values of reverse voltage and current. This can be explained by the fact that the depletion region width WJ becomes higher and so, injected charges rise more and more.

4.3. Influence of the Carrier Life Time τ With respect to the following equation of the reverse recovery time τRR [2]:

MRR

M

T

T

(12)

The plot of Figure 7 is established with supposing a constant transit time set to 22ns. This plotted function is growing-up with increasing the carrier life time τ, i.e., the reverse recovery time τRR is growing-up.

In fact, carriers injected to the bulk region recombine less and they have more chances to reach the side con-tacts of the device. So in the turn-off state, the device gives higher values of reverse voltage and current which is mentioned in Figure 8. 4.4. Comparison with Experimental Data The model parameters for the power diode are estimated

Time (n sec)

5.00

2.50

0

–2.50

–5.00

–7.50

–10.00

–12.50

–15.00

Dio

de c

urre

nt (

A)

25.0 50.0 66.7 83.3 100.0 125.0

W1 = 34.10-4 cm

W1 = 34, 5.10-4 cm

W1 = 35.10-4 cm

Figure 5. Influence of the intrinsic region width WI on the diode current turn off characteristic.

Time (n sec)

5.00

0

–5.00

–10.00

–15.00

–20.00

–25.00

–30.00

Dio

de c

urre

nt (

A)

25.0 40.0 60.0 80.0 100.0 120.0 150.0

N1 = 3.1014 cm-3

N1 = 3, 5.1014 cm-3

N1 = 4.1014 cm-3

Figure 6. Influence of the intrinsic region doping NI on the diode current turn off characteristic.

τRR (n sec)

–2.6 × 10-8

–2.8 × 10-8

-3 × 10-8

–3.2 × 10-8

–3.4 × 10-8

–3.6 × 10-8

–3.8 × 10-8

τ RR (

n se

c)

5 × 10-8 10 × 10-8 1.5 × 10-7 2 × 10-7

Figure 7. Variation of the reverse recovery time τRR in re-spect to the carrier life time τ.

Time (n sec)

5.00

2.50

0

–2.50

–5.00

–7.50

–10.00

–12.50

–15.00

Dio

de c

urre

nt (

A)

40.0 50.0 62.5 75.0 87.5 100.0 120.0

τ = 100 n sec

τ = 200 n sec

τ = 400 n sec

Figure 8. Influence of the carrier life time τ on the diode current turn off characteristic.

H. MNIF ET AL.


200 using the same parameter extraction procedure described in [2] at reverse recovery condition. The Table 2 gives a summary of these model parameters values for the BYT12P600 power diode [7] which are valid for all op-eration conditions.

Figure 9 illustrates a comparison of the measured and the simulated voltage and current waveforms for reverse recovery on three different operating conditions: a) VR = 60V and IF = 10A, b) VR = 70V and IF = 2.05A, c) VR = 120V and IF = 10A. The two simulated curves are re-spectively the simulation result of model [2] and the im-proved model developed in this study.

These previous Figures 9(a), (b) and (c) show a good agreement between simulation and experimental results except at the end of turn-off where oscillations appear. These oscillations at the end of the reverse recovery voltage waveforms are due to residual phenomena and are very difficult to simulate. However, the effect of these oscillations is not very interesting in design appli-cations [8].

In this figure, the agreement is good enough about the current (IRM and RR) but not so good about the voltage.

On the other hand, the estimation of the stray induc-tance in the switching loop of the circuit is critical for modeling the switching process.

One cause of error is due to the current and voltage probes. In fact, as stated in [8], probes interact with the device under test, create delays due to propagation in the cable and worst of all degrade the signal due to distortion in the probes and the cables. The overall accuracy of the extraction procedure requires the probe effects to be taken into account in simulation.

Simulated forward current curve for the model cited in [2] for IF = 2A and VR = 120A presents an error on the slope compared with the measurement forward current curves. In fact, textbooks [9,10] detail that the diode cur-rent slope is approximated by di/dt = –VR/Lm at the begin-ning of the reverse recovery, inside a switching cell circuit where a unique wiring parasitic inductance is considered.

For this reason, an improvement on the parasitic test- bench inductance was made in this work. The next sec-tion will treat this part. 4.5. Parasitic Test Bench Inductance Investigation For estimating the behavior of the device in circuit simulation, a complete model of the wiring influence is required which demands the development of a parasitic induc-tive matrix [8]. For more simplicity and short simulation time, in this study, a simple model for this wiring para-sitic is formulated. This simulated test bench inductance improvement is based on the extracted values of Lm (see test bench circuit, Figure 4) from the measured turn-off characteristics according to the following relation:

R

m

Vdi

dt L (13)

Figure 10 presents the variation of this test bench parasitic inductance extracted from the measured turn-off characteristics for 3 different test conditions. To capture this inductance influence on the simulated switching cell, a 2nd order polynomial fitting equation is introduced to predict its right value:

R Rm 2 1 0

F F

V VL a a a

I I

(14)

With a2, a1 et a0 are fitting parameters set to: a2 = 0.0268.109 [HA/V] a1 = 1.3582.109 [HA/V] a0 = 103.19.109 [HA/V] This equation is added to simulated test bench circuit via

a VHDL-AMS entity in substitution of the fixed inductance. Figure 9(c) shows the correction of the slope on the

simulated current curves compared to measurement. 5. Conclusions A physical improvement is made for a power diode lumped-charge model which consists of varying the width of the intrinsic bulk region with respect to the reverse ap-plied voltage and current variations. A second improve-ment is made to the simulated test-bench circuit to esti-mate the right value of the parasitic inductance value. The new model is described in VHDL-AMS language and has been implemented in the SIMPLORER simulator. A good agreement is observed between the new model simulated data and measured data. The results of this improved model are verified with a wide range of oper-ating conditions.

This model can be extended by including the selfhea- ting effect which can greatly affect the turn off charac-teristics and the device performance.

Table 2. New model’s parameters list.

Parametersymbol

description Parameter

value Unity

WI Bulk region width 37.6 µm

NI Doping level of the I-region

3.12 × 1014 cm3

carriers lifetime 103.5 n secTM carriers transit time Equation 10 n secIS Saturation current 2.78 × 10 24 A

RM0 Initial resistance 500 Ω ISE Recombination current 45.7386 × 109 A Cj0 Junction capacitance 2.2281 nF RS Serial resistance 15.7632 mΩ m Gradient coefficient 430.9960 × 103 -- ФB Bult-in potential 5.9294 mV

H. MNIF ET AL.


201

Time (n sec)

15.00

10.00

5.00

0

-5.00

-10.00

-15.00

-20.00

-25.00

Dio

de c

urre

nt (

A)

10.0 25.0 50.0 75.0 100.0 125.0 150.0 185.0

Model (2) Measurement

This study

Time (n sec)

10.00

-25.0

-50.0

-75.0

-100.0

-125.0

-150.0

-180.0

Dio

de V

olta

ge (

V)

50.0 80.0 100.0 120.0 140.0 160.0 185.0

Model (2) MeasurementThis study

(a)

Time (n sec)

15.00

2.50

0

-2.50

-5.00

-7.50

-11.00

Dio

de c

urre

nt (

A)

30.0 50.0 62.5 75.0 87.5 100.0 112.5 125.0


Time (n sec)

20.00

0

-25.0

-50.0

-75.0

-100.0

-125.0

-150.0

Dio

de V

olta

ge (

V)

30.0 37.5 50.0 62.5 75.0 87.5 100.0 115.0


(b)

Time (n sec)

15.00

10.00

5.00

0

-5.00

-10.00

-15.00

-20.00

-25.00

Dio

de c

urre

nt (

A)

10.0 40.0 60.0 80.0 100.0 120.0 150.0


Time (n sec)

10.00

-50.0

-100.0

-150.0

-200.0

-250.0

-320.0

Dio

de V

olta

ge (

V)

24.0 40.0 60.0 82.0 100.0 120.0 140.0 160.0


(c)

Figure 9. Comparison of measured and simulated current and voltage waveforms reverse recovery for model [2] and this study’s model (a) VR = 60 V and IF = 10 A, (b) VR = 70 and IF = 2.05A, (c) VR = 120 V and IF = 10 A.

VR/IF(V/A)

140

120

100

80

60

40

20

0

Lm

(nH

)

0 20 40 60

Figure 10. Extracted parasitic inductance test-bench (cir-cles) for three measurement test conditions with 2nd order polynomial fitting function (black line).

This research is being continued at our laboratory to have a variety of basic library of elementary semicondu- ctors in VHDL-AMS. 6. Acknowledgements The authors thank H. Morel from “INSA de Lyon-CE-GELY laboratory” for providing us with the reverse re-covery measurement data and helpful discussions. 7. References [1] C. L. Ma and P. O. Lauritzen, “Simple Power Diode Mo-

H. MNIF ET AL.


202

del with Forward and Reverse Recovery,” IEEE Transac-tions on Power Electronics, Vol. 8, No. 4, 1993, pp. 342- 346.

[2] M. Najjari, H. Mnif, H. Samet and N. Masmoudi, “New Modeling of the Power Diode Using the VHDL-AMS Language,” The European Physical Journal Applied Phy- sics, Vol. 4, No. 1, 2008, pp. 1-11.

[3] H. Zhang and J. A. Pappas, “A Moving Boundary Diffu-sion Model for PIN Diodes,” IEEE Transactions on Mag- net, Vol. 37, No. 1, 2001, pp. 406-410.

[4] P. Antognetti and G. Massobrio, “Semiconductor Device Modeling with SPICE,” MCGraw-Hill Inc, New York, 1988.

[5] C. L. Ma and P. O. Lauritzen, “Modeling of Diodes with the Lumped-charge Modeling Technique,” IEEE Trans-actions on Power Electronics, Vol. 12, No. 3, 1997, pp. 398-405.

[6] K. X. A. Caiafa, E. Santi, J. L. Hudgins and P. R. Palmer,

“Parameter Extraction for a Power Diode Circuit Simu-lator Model Including Temperature Dependent Effects,” 17th Annual IEEE Applied Power Electronics Conference and Exposition, Dallas, March 2002, pp. 452-458.

[7] BYT12 600 Datasheet, “Datasheet Archive,” 1998. http:// www.datasheetArchive.com

[8] H. Garrab, B. Allard, H. Morel, K. Ammous, S. Ghedira, et al., “On the Extraction of PiN Diode Design Parame-ters for Validation of Integrated Power Converter,” IEEE Transactions on Power Electronics, Monastir, Vol. 20, No. 3, May 2005, pp. 660-670.

[9] N. Mohan, T. M. Undeland and W. R. Robbins, “Power Electronics: Converters, Applications and Design,” 2nd Edition, John Wiley & Sons, New York, 1995.

[10] R.W. Erickson and D. Maksimovic, “Fundamentals of Power Electronics,” 2nd Edition, Kluwer Academic Pulishers, Berlin, 2001.



Insulation State On-Line Monitoring and Running Management of Large Generator

Qiudong Sun, Zhengxin Zhou, Weiqin Guo School of electronic and electrical Engineering, Shanghai Second Polytechnic University, Shanghai, China

E-mail: [email protected], zxzhou, [email protected] Received March 23, 2010; revised May 18, 2010; accepted July 4, 2010

Abstract This study presented an insulation state monitoring method for large generator based on radio frequency (RF) technique. As an on-line condition monitor and the precondition of condition-based maintenance (CBM), the RF monitor used the high frequency current mutual inductor to detect the partial discharge signal from neutral wire of stator windings. According to the magnitude of indicative value of RF monitor, a five phase model was also proposed to manage the generator’s running better. The practices show that the proposed method is effective. Keywords: Large Generator, Partial Discharge, Radio Frequency Technique, On-Line Monitoring, Running

Management

1. Introduction Today, more and more power utilities are switching to mo- ney-saving and effective condition-based maintenance (CBM) programs for scheduling of machine maintenance and testing. Such a system can overcome the disadvantage of excess maintenance brought by the preventive maintenance [1]. It will determine the equipment’s health, and act only when maintenance is actually necessary. Development in recent years have allowed extensive instrumen- tation of equipment such as condition monitoring to observing the state of the system, and together with better diagnosis tools for analyzing con-dition data, the maintenance personnel of today are more than ever able to decide what is the right time to perform maintenance on some piece of equipment. Ideally CBM will allow the maintenance personnel to do only the right things, minimizing spare parts cost, system downtime and time spent on maintenance.

A large generator is a complicated machine system. Its breakdown is paroxysmal. If the accident once happens, the imperilment will be great, and the maintenance cost will also be great. A majority reason of its breakdown is the short circuit caused by its insulation being destroyed [1,2]. Due to the manufacturing and long time running of generator, the partial discharge (PD) of its stators is un-avoidable [1]. This state can lead to aging of its main insulation, and eventually lead it to breakdown. There-

fore, it is necessary for generator to be equipped an on- line condition monitoring system to observe its insulation state. Meanwhile, the on-line condition monitoring is the precondition of CBM. For a good CBM, it is far from enough if there is only an on-line condition monitoring system without being supported by a partial discharge analyzing technique. So, it is also necessary to study the relationship between the value of partial discharge and the insulation state of generator.

At present, the main methods for measuring the stat- or’s partial discharge of large generator are the neutral- point coupling detecting, coupling capacitor detecting, radio frequency detecting, detecting by partial discharge analyzer (PDA) and detecting by stator slot coupler (SSC) [3-6]. Generally, the partial discharge detecting systems are classified into two types by their frequency band- widths. One works in narrow bandwidth, and the other in wide bandwidth, such as SSC. Generally, the former has a stronger anti-jamming capability and is sensitive for serious partial discharge, but it can not distinguish the discharge signal occurred inside or outside of generator [2,3,5]. Although the latter can collect plenty signals of partial discharge for analyzing, it is unacceptable because it needs imbedding a coupling sensor under the stator slot wedge and changing the insulation structure of stator windings [2,3].

In this study, we propose an insulation state monitor-ing method for generators based on radio frequency monitoring technique. We choose the narrow bandwidth

Q. D. SUN ET AL.


204

technique to design the partial discharge detecting sys-tem, which can monitor the insulation state of generator. We also give some advices for generator’s running management according to the RF signal level produced by partial discharge and CBM experience.

This study is organized as follows. In Section 2, we give the radio frequency monitoring technique about the partial discharge, insulation deterioration and RF measurement method. We also give the frame of RF Monitor and the on-line monitoring architecture in this section. Section 3 presents an on-line evaluation method of insulation state. Section 4 gave some application exam-ples, and shows that the proposed scheme yields more effective through practices. Section 5 gives the conclu-sion of this study. 2. Radio Frequency Monitoring 2.1. Partial Discharge Partial discharge can be described as an electrical pulse or discharge in a gas-filled void or on a dielectric surface of a solid or liquid insulation system. This pulse or discharge only partially bridges the gap between phase insulation to ground, or phase-to-phase insulation. A full discharge would be a complete fault between line potential and ground [2].

These discharges might occur in any void between the copper conductor and ground. The voids may be located between the copper conductor and insulation wall, or internal to the insulation itself, or between the outer insulation wall and the grounded frame [2].

These discharges also might occur at the terminal of winding. Its surface contamination and moisture creates the surface discharge and lightning. 2.2. Insulation Deterioration and Complete Failure Insulation degradation is frequently linked to partial discharges. The partial discharges are effectively small sparks occurring within the insulation system, therefore deteriorating the insulation, and can eventually result in complete insulation failure [2].

At many times, the winding insulation deterioration can be represented by the developing and forming process of its strand breaking. Those wires of windings, which are located at the upper winding-bar with the same slot and the same phase endured maximal magnetic force, upmost temperature and supreme electric strength, and the wires near by slot wedge, especially the wires located at seamed edge, are easily deteriorated. The deterioration process of stator windings can be described as Figure 1. In this figure, those phenomena labeled by symbol “*” can be detected by RF monitor or super- heater.

2.3. Partial Discharge Measurement Method Partial discharges are high frequency pulses originating at various sections within an insulation system [1,2]. These pulses generate a voltage and current signal into the insulation, returning through a ground path. The partial discharge measurement can be implemented by radio frequency monitoring techniques (RFMT).

We use the HF current transducer (CT), which is nip- ped tightly at the proper unshielded position of neutral wire of stator winding, to detect the RF current. The RF measurement connection and technique are as shown in Figure 2. In this figure, the RF monitor is a high sensitive measurement meter with a magnitude of μV quasi- peek value.

Partial

discharge* Work temperature

Additional temperature rise caused by

electromagnetic vibration of wires near by slot wedge

Damage agglutinant of enamel covered wire,

especially nearthe slot wedge

Wire lost integrity

Wire vibrated and worn

Wire milled and strand broken

Main insulation attenuated

Wire short-circuited

HF spark discharge*

Main insulationsuperheated

Gas particle* Main insulation dielectric strength

lowered

Main insulation failure

Generator shut down

Insulation resistance lowered

Figure 1. Deterioration process of stator windings.

On-Line Generator

RF Monitor

HF CurrentTransducer

Neural Earthing Transformer

Figure 2. The measurement connection method.

Q. D. SU ET AL.


205

2.4. RF Monitor As mentioned above, the RFMT can be classified into two types: narrow bandwidth and wide bandwidth. The narrow bandwidth technique is applied in our monitoring system because it has strong anti-jamming capability.

Our RF monitor (or Insulation Monitor) is composed of receiver (includes crystal filter, RF logarithm amplify- ier, quasi-peek value detector), logical judging process- or and output alarming circuit as shown in Figure 3. Its centre frequency is 1 MHz, -3 dB bandwidth is 5 kHz, –60 dB bandwidth is 20 kHz, the dynamic input range is 10 μV ~ 10000 μV, input resistance is 50 Ω. It also has a greater than 60 dB suppressing circuit to attenuate the false signals and to improve its capacity to defense ab- rupt disturbing. 2.5. On-Line Monitoring Architecture The on-line monitoring system is composed of RF monitor, HF current transducer (CT) and telecontrol board mainly as shown in Figure 4. As a probe, the HF current transducer is used to detect the RF current. The RF Monitor is used for on-line monitoring the state of interior faults of generator and unwonted alarming. The telecontrol board equipment is placed far from the generator to re-mote monitoring its insulation state. It has the same con-trol functions, indication buttons and the voice and light alarming set with the main monitor. It also has a graph recorder for plotting and saving the output signal from RF Monitor. Finally, the telecontrol board is connected by USB interface into a distributed control system (DCS), which can store the RF signal timely, process and ana-lyze the RF signal, and be accessed remotely by the users (diagnosis personnel ) far from internet.

3. On-Line Evaluation of Insulation State 3.1. Running Experience According to the experience from more than 360 set RF monitoring practices, we have obtained the following knowledge:

When the RF signal level is less than 300 μV or fluctuating about this level and it is independent of generator’s load, there is only a small discharge in the generator system. The fact shows that the insulation state of generator system is good or regular.

The RF signal level of greater than 1000 μV indicates that there is a bigger discharge in the generator system. Here, we should compare this moment data with that of forepassed and consider the factors as follows to evaluate the insulation state:

1) The relationship between RF signal level and generator’s load.

2) The amplitude of RF signal is relative invariable or changing randomly.

3) The RF signal rises tardily or abruptly. 4) We should also integrate the RF signal with the

traditional testing and the work condition change testing to judge the insulation fault happened inside or outside the generator.

Receiver

Crystal filter

Log AMP

Quasi-peek DET

Logical judging

processor

Alarmingcircuit

Figure 3. Frame of insulation monitor.

Figure 4. The on-line monitoring architecture of RF Monitor.

Q. D. SUN ET AL.


206

We have to admit that the generator need not always shut down when the RF signal rises to a set level, and 1000 μV RF signal is not an absolute threshold to judge the discharge happened inside the stator of generator.

In the condition-based maintenance system, it is im- possible to judge what kind fault of generator happened or whether the generator need shut down or not, just by one apparatus or one signal. Therefore, we need other testing equipment and some experienced field experts to support the condition-based maintenance for generator. 3.2. Relationship between Insulation State and

RF Signal According to the running experience and analysis in Sub- section 3.1, we propose a five advice model for generator’s running management corresponding to the five phases of insulation state as shown in Figure 5. This model is also a result of practice statistics. It is an eligible model for most circumstances.

RF signal level detected by SJY

10 100 300 1000 2000 3000 5000 10000V

0 33.3 50 66.7 75 80 90 100% |---------------------Good---------------------|--Aging--|--- Notice---|-Alarm-|-Danger|

Insulation state

Figure 5. Relationship between insulation state and RF signal.

3.3. Five Advice Model for Running Manage-ment

In Figure 5, each phase has a corresponding advice for generator’s running management as shown in Table 1. 4. Applications Up to now, more than 360 sets of our RF monitor (SJY) have been launched into various power generating sets, which include thermal power generators, water power generators, nuclear power generators and gas turbine generators. Their equipped capacitors are from 100 MW to 1000 MW.

Our RF monitors have predicted various insulation failures successfully in practices. For example, a 300 MW hydrogen-cooled generator of a Power Plant has been run 7 years. Some day the equipped RF monitor output an abnormal signal as shown in Figure 6 and alarmed. The magnitude of abnormal signal was about hundreds of μV. By analyzing the RF signal, we judged that the terminal of stator winding has damaged to result in the partial discharge like this. And we also gave an advice to user to reconstruct the stator winding. After its stator winding reconstructed, the generator got back in order. And the RF monitor went back to indicate a nor-mal signal as shown in Figure 7. The magnitude of RF signal had dropped to a normal level of tens of μV.

Table 1. Insulation states of generator evaluated by RF signal and its running management advices.

Insulation state RF signal level (μV) Running management of generator from field experts

Good ≤ 300 The generator is allowed long-term running.

Aging 300~1000 The insulation state of generator is in the interim from good to bad. It is allowed running unceasingly. But we should notice to observe the trend of insulation degradation.

Notice 1000~3000 We should notice to observe the running state of generator or arrange the maintenance scheme.

Alarm 3000~5000 We should pay attention to the trend of the running state change and the work condition change testing. The generator is needed to shut down at appropriate time.

Danger ≥ 5000 We should integrate the indicative value of RF signal with other symptoms to judge whether the generator is necessary to shut down or not.

Figure 6. An abnormal signal of RF Monitor.

Q. D. SU ET AL.


207

Figure 7. A normal signal of RF Monitor.

As mentioned in Section 1, it is necessary to install a

RF monitor into the complicated generator set for on- line monitoring its insulation state. If the RF monitor has been installed, but we can not unscramble its signal, then the RF monitor like this is just an ornament and it is useless. We should better to integrate the RF signal with other diagnosing system to monitor the generator’s insulation state and give a diagnosis conclusion and a running management advice.

Also we should indicate that the RF monitor is not enough apparatus or RF signal is not enough symptoms to judge what kind fault of generator happened or whether the generator need shut down or not. Therefore, we need some other testing equipment and some experienced field experts to support the condition-based maintenance for generator. 5. Conclusions This study presented an insulation state monitoring approach for CBM of generators. In our approach, the RF monitor with strong anti-jamming capability and the on- line monitoring system were designed to detect the partial discharge signal caused by the insulation degradation of generators. In order to run generator better, we proposed a five phase running management model according to the magnitude of indicative value of RF monitor (SJY). The practices demonstrated that the proposed method is effective.

6. Acknowledgements This research project was supported by the Key Disci- plines of Shanghai Municipal Education Commission under Grant No. J51801. 7. References [1] W. Q. Guo, “On-Line Insulation Monitoring and

Condition Maintenance to Generators,” Journal of Shanghai Second Polytechnic University, No. 1, 2002, pp. 20-26. [2] G. Paoletti and A. Golubev, “Partial Discharge Theory and Technologies Related to Traditional Testing Methods of Large Rotating Apparatus,” 34th IAS Annual Meeting, Phoenix, Vol. 2, October 1999, pp. 967-981.

[3] C. J. Huang, W. Y. Yu, P. Gabe and W. Wei, “Partial Discharge On-Line Monitoring and its Application to the Large Generators,” Large Electric Machine and Hydraulic Turbine, No. 6, 2000, pp. 33-38.

[4] X. L. Chen, X. P. Cao, Y. H. Lu, B. Yue, Y. H. Cheng, and H. K. Xie, “Field Detection of Ultra-Wide Band Partial Discharge for Generator Stator Insulation,” Electric Power, Vol. 35, No. 3, 2002, pp. 31-34.

[5] G. Stone, V. Warren and M. Fenger, “Diagnostic Info- rmation Obtained from Examining a Large Stator Winding PD Result Database,” Proceedings of the International Symposium on Electrical Insulating Materials, Vol. 33, 2001, pp. 635-640.

[6] G. Stone, “Advancements during the Past Quarter Cen-tury in On-Line Monitoring of Motor and Generator Winding Insulation,” IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 9, No. 5, 2002, pp. 746- 751.



The Design of New Sensorless BLDCM Control System for Electric Vehicle

Zhibin Ren, Xiping Liu School of Mechanical and Electrical Engineering, Jiangxi University of Science and Technology, Ganzhou, China

E-mail: [email protected] Received March 11, 2010; revised April 29, 2010; accepted June 3, 2010

Abstract In order to meet the requirement of reliably running which is by Electric Vehicle for motor controller, the paper is focused on a sensorless brushless DC motor controller design and a commutation point method. By utilizing the saturation effect of stator iron core, six short voltage pulses are employed to estimate the initial rotor position. After that a series of voltage pulses are used to accelerate the motor. When the motor reaches a certain speed at which the back-electromotive force (EMF) method can be applied, the running state of the motor is smoothly switched at the moment determined by the relationship between the terminal voltage waveform and the commutation phases. “Lagging 90˚-α commutation” is bring forward to overcome the shortages existing in the traditional method. The experimental results verify the feasibility and validity of the proposed method. Keywords: Brushless DC Motor, Electric Vehicle, Ensorless Control, Commutation Point

1. Introduction Due to the advantages of high power density, robust structure, and ease of control, the brushless DC motor (BL- DCM) has played an important role in many applications. It is necessary for high-performance applications, such as servos in machine tools and robotics, to use position sensors for successful starting and operation. The issues on reducing cost, low-performance applications, space-res- tricted applications, and reliability of the position sensors have motivated research on sensorless control [1-6]. In addition, the fast and continuing improvements of pow-erful and economical microprocessors and digital signal processors (DSPs) have speeded up the development of sensorless control technology. In fact, [7] enumerates many applications of the BLDCM, for example, air-con- ditioning compressor, engine cooling fan, fuel/water pump, electric vehicle.

The sensorless control technology of the brushless DC motor (BLDCM) based on the back-electromotive force (EMF) detection method has been widely used in the industrial and commercial fields. As we know, the mag-nitude of the back-EMF is proportional to the motor speed, so the back-EMF detection method cannot be ap-plied properly when the motor is at standstill. In order to solve this problem, many methods have been developed.

One of them, often referred to as a 3-step startup method, is used to align the rotor first in a predetermined direc-tion, and then accelerate the motor in an open-loop scheme before the back-EMF method is applied. This startup method is easy to implement but it tends to be affected by the load and may temporarily cause reverse rotation which is not allowed in some applications.

In this paper, the short pulse sensing method, which is based on the saturation effect of the stator iron and will not cause any reverse rotation or vibration during the startup process. The key hardware implementation is the current sensor detected by Ri and the resistance network used as the voltage divider. The terminal voltage which reflects the back-EMF information is sampled by the A/D converter integrated in the micro-controller. 2. Initial Rotor Position Estimation The phase inductance of the stator is determined by

L =Φ / I (1)

where I is the phase current and Φ is the flux due to magnet rotor and stator coils and core. Figure 1 shows the inductance of stator windings with nonlinear mag-netization characteristics of the stator core, depending upon the position of rotor. As the pole of magnet rotor is

Z. B. REN ET AL.


209

close to the stator winding, the ratio of the change of the current in the stator winding flowing in the magnetizing direction is larger than that in the opposite direction because of the magnetic saturation of the stator core. So the value of the current would be different according to the rotor posi-tion if a constant voltage vector from inverter is applied to the stator winding of the motor for a constant time period. The estimation of the rotor position is based on strongly magnetized stator field. Three situations of a magnetic pole of permanent magnet of the rotor close to the stator core are considered as shown in Figure 2. A smaller phase induc-tance L(sat) is defined when the stat- or field is in phase with rotor field, shown in Figure 2(a). Similarly, a larger phase inductance L(linear) is defined when the stator field is out of phase with rotor field, shown in Figure 2(b). Figure 2(c) depicts the case of middle value L(mid). 3. Initial Rotor Position Detection Based on the operation principle mentioned above, six voltage pulses are injected into the phase windings and the peaks of the response current are compared with each other to determine the rotor position.

As shown in Figure 3(a), the high-side power device VT1 and the low-side power device VT6 are activated first, which can be denoted as A+ B–. The resultant mag-netic field is represented by arc line. Then the high-side power device VT3 and the low-side powerdeviceVT2 are activated, and are denoted as B+ C–, and arc line Figure 3(b) represents the resultant magnetic field. If the north pole of the rotor is in the same direction as that of the resultant magnetic field arc line and in the opposite direction from that of arc line, the peak of the response cur-rent is greater when the north pole of the rotor is in the

i

F

L1

L2L3

Figure 1. The inductance of stator windings with nonlinear magnetization characteristics of the stator core.

s

V

i

s

N

N

V

i

s

N

s

V

i

s

N

N

(a) (b) (c)

Figure 2. Magnetic fields (a) saturated magnetic field, (b) linear (non-saturated) magnetic field and (c) middle case.

A

BC

A

BC

A

BC

(a) (b) (c)

Figure 3. The schematic diagram of initial rotor position detetion. (a) Energized stator winding and rotor position within180°; (b) Rotor position within 60°. same direction as that of the resultant magnetic field arc line and in the opposite direction from that of arc line, the peak of the response current is greater when the north pole of the rotor is in the same direction as that of the resultant magnetic field arc line. Thus the north pole of the rotor can be narrowed down to 60°, as shown in Fig-ure 3(c).

When the initial rotor position is identified, the motor is accelerated to a certain speed. Generally, a self-con-trolled BLDCM with trapezoidal BEMF waveforms is driven by a three-phase inverter with six-step commutation. Each conducting phase is called one step of two phase conducting. The conducting interval for each phase is 120° by electrical angle. Therefore, only two phases conduct current at any time, leaving the third phase floating. In order to produce maximum torque, the inverter should be commutated every 60°, and the com-mutations occur at 30° delay from the corresponding zero-crossing points (ZCP) of the BEMF waveforms. 4. Lagging 90˚-α Commutation Method According to intensive analysis of the zero-crossing detec-tion of Back EMF, a detecting method with band-bass filter is proposed. By analyzing the method in the detect-ing Back EMF, zero point of Back EMF is lagged, see Figure 4(a), so the corresponding correction method and the new commutation approach are presented to improve the performance of the BLDCM, see Figure 4(b). When detecting the zero-crossing of Back EMF, the commuta-tion approach is lagged for 90˚-α, see Table 1. The posi-tion detection can be achieved over a wide speed range. 5. Experimental Results The specifications of the test BLDCM are: 8 poles, 400 W. According to Figure 3 for the initial position detec-tion and based on Table 1, Figure 5 display the response of velocity waveform. To verify the proposed method, we conducted some experiments. In experiment, the cur- rents are sampled to verify the rotor position, the key hard ware of the current sensor and powerful micropro-

Z. B. REN ET AL.


210

A phase

B phase

C phase

(a)

A phase

B phase

C phase

(b)

Figure 4. Lagging 90˚-α commutation method.

Table 1. The commutation approach is lagged for 90˚-α.

zero-crossing point

Delay angle

Commute point

Commute switch

P1 90˚-α n1 V1 to V3

P2 90˚-α n2 V2 to V4

P3 90˚-α n3 V3 to V5

P4 90˚-α n4 V4 to V6

P5 90˚-α n5 V5 to V1

P6 90˚-α n6 V6 to V2

cessors of digital signal processors (DSPs) should be used Traditional method of 3-step startup is shown in Figures 5(a, b) and new method is shown in Figure 5(c). In the experiments, the results show that the use of the methods makes the drive better, with better follow performance. 6. Conclusions In this paper, new startup and smooth switching method of a sensorless brushless DC motor is presented. By us-

Z. B. REN ET AL.


211

Tek View

(a)

Tek View

(b)

Tek View

(c)

Figure 5. Speed response curve. ing this method, the rotor position at standstill can be estimated with a resolution of 60° and the motor is acceler-ated to a certain speed at which the back-EMF detection method can be applied. The method will not cause any reverse rotation or vibration during the startup process. The hardware implementation of the driving circuit is simple. It is very suitable to use in the low-cost applica-tions. 7. References [1] R. Krishnan and R. Ghosh, “Staring Algorithm and Per-

formance of A PMDC Brushless Motor Drive System with No Position Sensor,” 20th Annual IEEE Power Electronics Specialists Conference, Milwaukee, Vol. 2, June 1989, pp. 815-821.

[2] J. P. Johnson, M. Ehsani and Y. Guzelgunler, “Review of Sensorless Methods for Brushless DC,” IEEE-IAS/PCA Cement Industry Technical Conference, Roanoke, Vol. 1, April 1999, pp. 143-150.

[3] T. Kim and M. Ehsani, “Sensorless Control of the BLDC Motors from Near-Zero to High Speed,” IEEE Transac-tions Power Electronics, Vol. 19, No. 6, 2004, pp. 1635- 1645.

[4] J. Holtz, “State of the Art of Controlled AC Drive with-out Speed Sensor,” IEEE Proceedings of the 1995 Inter-national Conference on Power Electronics and Drive Systems, Vol. 1, February 1995, pp. 1-6.

[5] T. H. Kim, H. W. Lee, and M. Ehsani, “State of the Art and Future Trends in Position Sensorless Brushless DC Motor/Generator Drives,” 31st Annual Conference of IEEE Industrial Electronics Society, Michigan, Novem-ber 2005, pp. 1718-1725.

[6] P. P. Acarnley and J. F. Watson, “Review of Position- sensorless Operation of Brushless Permanent-Magnet Ma-chines,” IEEE Transactions on Industrial Electronics, Vol. 53, No. 2, April 2006, pp. 352-362.

[7] J. Shao, “An Improved Microcontroller-Based Sensorless Brushless DC (BLDC) Motor Drive for Automotive Ap-plications,” IEEE Transactions on Industry Applications, Vol. 42, No. 5, 2006, pp. 1216-1231.

Energy and Power Engineering - scirp.orgProf. Qingquan Chen Prof. Yusheng Xue Prof. Ryuichi Yokoyama...

Documents

Transcript of Energy and Power Engineering - scirp.orgProf. Qingquan Chen Prof. Yusheng Xue Prof. Ryuichi Yokoyama...