A Comparison Between Temperature and Current … · A Comparison Between Temperature and Current...
Transcript of A Comparison Between Temperature and Current … · A Comparison Between Temperature and Current...
A Comparison Between Temperature and CurrentSensing in Photovoltaic Maximum Power Point
Tracking
Daniel BurmesterSchool of Engineering and
Computer Science
Victoria University of
Wellington
PO Box 600 Wellington
New Zealand
Email: [email protected]
Dr Ramesh Rayudu, SMIEEESchool of Engineering and
Computer Science
Victoria University of
Wellington
PO Box 600 Wellington
New Zealand
Email: [email protected]
Prof Winston SeahSchool of Engineering and
Computer Science
Victoria University of
Wellington
PO Box 600 Wellington
New Zealand
Email: [email protected]
Abstract—Maximum power point tracking algorithms, forphotovoltaic modules, commonly use current and voltage sensingas a means of tracking the maximum power point (MPP). Thispaper experimentally compares the use of a temperature sensingin place of the current sensing for MPP tracking. It does soby implementing two commonly used current sensor algorithms(perturb and observe and incremental conductance) and temper-ature sensing algorithm (MPPT-temp [1]) to determine whichis faster and more accurate. The paper shows, the standarddeviation of the two current tracking algorithms is much greaterthan that of a temperature based algorithm. The temperaturesensor also reduces the complexity of circuitry and is a morecost effective solution to maximum power point tracking ofphotovoltaic modules.
I. INTRODUCTION
Since the 1970’s, scientists have grown increasingly aware
of global climate change and its environmental effects. These
effects have manifested as increased average global air and
ocean temperatures, widespread melting of snow and ice and
rising average global sea level [2].
This warming is caused by greenhouse gases in the atmo-
sphere, and although this ”greenhouse effect” occurs naturally,
human activities in recent times have increased this effect [3].
Greenhouse gasses consist of 16% methane (CH4), 6% nitrous
oxide (N2O), 2% F-gases and is dominated by carbon dioxide
(CO2), which accounts for 76% of these greenhouse gases [4].
A break-down of these emissions into the main sectors, reveals
the largest source of CO2 is electricity and heat Generation,
responsible for 41% in 2010 [5].
This is a repercussion of conventional electricity production
where power is generated by burning fossil fuels, producing
CO2 . Despite their negative effects on the environment, fossil
fuels are responsible for more than two-thirds of the total
global electricity production [6].
Encouraged by the Kyoto protocol 1, there has been a global
shift towards renewable energy in an attempt to reduce these
statistics [2].
Renewable energy can reduce greenhouse gasses, utilising
energy sources such as wind, water and sun to produce
electricity. This is a carbon neutral process, meaning no carbon
is produced in the conversion from energy source to electricity.
One such method of generating electricity is photovoltaic (PV)
modules, which convert energy from the sun into electricity.
As of 2012, the world’s cumulative PV capacity surpassed an
impressive 100 gigawatt (GW) installed electrical power mark.
Each year these PV installations save more than 53 million
tons of CO2 [7].
Although PV modules come with the advantage of zero carbon
emissions, and an abundant source (the sun), they also have
their disadvantages.
A PV is a semiconductor device that converts solar radiation to
electrical energy, but does so very inefficiently. The modules
power generation is heavily dependent on environmental con-
ditions such as ambient temperature and sun irradiance [8].
This makes the PV a non-linear, time-variant power source
with a current voltage (I-V) curve similar to that shown in
figure 1 [9].
Figure 1 shows how the PV’s I-V curve varies with changing
solar irradiance and ambient temperature. It reveals, for
a specific temperature and solar irradiance, there is only
one maximum power point (MPP). This maximum power
is located at the ’knee’ of the I-V curve. To maximise the
efficiency of a PV module, this point must be dynamically
tracked using a maximum power point tracker (MPPT).
1An international treaty that sets legal limits and reduction goals ongreenhouse gas emissions978-1-4799-5141-3/14/$31.00 c© 2014 IEEE
Fig. 1. I-V curve of PV module: Left) Under constant temperature. Right)Under constant irradiance
A. Goal and layout of this Paper
This paper will compare current sensing algorithms for
photovoltaic MPPT with a temperature sensing algorithm. It
will look at the speed and accuracy of the MPPT system using
each algorithm.
The paper layout will see section II discuss the functionality
of a maximum power point tracker before the experimental
procedure undertaken for this research is presented in III. The
results of the research are then discussed and concluded in
sections IV and V respectively.
II. MAXIMUM POWER POINT TRACKING
An MPPT consists of a DC/DC converter which forces
the PV to operate at its maximum power point, it does so
by presenting a variable load to the PV [10]. This load is
varied by altering the duty cycle of the pulse width modulation
controlling the switching elements of the DC/DC converter.
This duty cycle is in turn controlled by an algorithm that
uses information gathered by sensors to track the PV modules
MPP. Typically algorithms use current and voltage sensing
to track the PV’s maximum power point. Two examples of
this is the commonly used algorithms, ”perturb and observe”
and ”incremental conductance”. These rely on the ability to
calculate the PV’s output power, a brief description of each
algorithm follows:
A. Perturb and Observe (Hill-climbing)
The output power of the PV is measured (P1), then the
operating point of the PV module is shifted and re-measured
(P2). if P2 > P1 then the algorithm will continue to shift the
operating point in this direction. If not, it will reverse. The
main fault with this algorithm is that it oscillates around the
maximum power point rather than settling directly on it.
B. Incremental Conductance
This algorithm works on the basis that dP/dV = 0 at the
maximum power point. This is achieved by correcting the
operating point, whenever it falls to one side of the maximum
power point. Programming this algorithm is more complex
than Perturb and observe, however it has a faster response
and better accuracy.
The main problem with current sensing arises when you
observe the behavior of the input current to the DC/DC
converter. Figure 2 shows the characteristic current ripple of
the DC/DC converter. To convert this to a usable input for
the algorithm, additional circuitry is required increasing the
complexity and cost of the system.
Fig. 2. DC/DC inductor current
C. MPPT-temp
An alternative to current sensing is proposed in [1] and [9].
It is based on temperature and voltage sensing, eliminating the
need for a current sensor. The algorithm is based on equation
1 which shows the dependence of the maximum power point
on the ambient temperature of the PV module. This equation
takes the datasheet’s voltage at the maximum power point
Vmpp(TRef ) which is measured at a reference temperature
TRef . It then compares the PV’s actual temperature T to
TRef to find how far the temperature has deviated. Multiplying
this by µv , the temperature coefficient (∆V per degrees),
the new Vmpp is found. Figure 3 displays the flow diagram
of the MPPT-temp algorithm, showing how equation 1 is
implemented. It can be seen from figure 3 that the duty
cycle (D(n − 1)) is incremented by δD to give a new duty
cycle D(n). The variable δD takes the difference between the
measured voltage and the voltage at the MPP, then multiplies
this value by k which determine the incremental step size.
Safety checks are made to ensure the algorithm does not
exceed an upper or lower duty cycle limit before incrementing
the SEPIC.
Vmpp = Vmpp(TRef ) + (T − TRef )µv (1)
The disadvantage to this algorithm is the additional
information required from the PV’s datasheet (µv , Vmpp(TRef )
and TRef ), as this information is not always quoted.
D(n)>Dmax
Start
Measure I(t) and V(t)
Vmpp(T) = Vmpp(Tref) + µV(MPP)(T-Tref)
ΔD = [Vmodule-Vmpp(T)]k
D(n) = Dmax
D(n) = Dmin
Return
D(n)<Dmin Yes
D(n) = D(n-1)+ΔD
No
No
Fig. 3. Flow diagram of MPPT-temp
III. EXPERIMENTAL PROCEDURE
So an accurate comparison could be made between current
and temperature sensing, a controlled environment was re-
quired. This would ensure repeatability of conditions for each
system and enable calibration of the PV’s MPP. It was also
important to have multiple conditions so the systems behavior
during transitions could be observed. To achieve this, four
500W lights were used as shown in Figure 4. This gave the
option of either two or four lights being active at any given
time, allowing the simulation of ”full sun” or ”cloud cover”
conditions.
Fig. 4. Test Conditions
Once the controlled environment were implemented, a cal-
ibration of the PV was undertaken. This calibration was to
provide the maximum power output of the PV under each
condition. Figure 5 shows the I-V curve of the test PV and
figure 6 displays the output power as a function of output
voltage. With the MPP of the PV now being a known quantity,
the accuracy of the current and temperature sensing approach
to maximum power point tracking could be assessed.
0 5 10 15 200
0.2
0.4
0.6
0.8
1
1.2
1.4Current vs voltage of photovoltaic module
Voltage (V)C
urr
ent
(A)
Two lights on
Four lights on
Fig. 5. I-V curve of PV module
0 5 10 15 200
5
10
15
20Voltage vs power of photovoltaic module
Voltage (
V)
Power (W)
Two lights on
Four lights on
Fig. 6. Power vs Voltage Curve of PV Module
A. System hardware
A single ended primary inductor converter (SEPIC) was
used as the DC/DC converter of the MPPT. This was due
to its ability to produce and output voltage greater or less
than the input voltage and its non-inverted output. The
control mechanism chosen was an arduino(TM) uno, which
had the required input and output ability and the advantage
of a simple programming environment. The MPPT was then
connected to the 12V, 120W PV and a 10Ω load as shown in
figure 7.
Fig. 7. Block Diagram of Test Rig
1) Current Sensor: A Hall effect current sensor was
employed to sense the panels output current of 0A to 6A. The
output from the sensor was a linear voltage, with a change
of 10 mV/A. Due to the current ripple, a peak detector
circuit was implemented, and this signal was amplified before
analogue to digital conversion.
2) Temperature Sensor: The temperature sensor used was
an LM35 precision temperature sensor. This also gave a
linear output of 10 mV/oC. This output was also amplified to
meet the arduino’s analogue input of 0-5V.
IV. RESULTS
The comparison between the two approaches to maximum
power point tracking, current and temperature, was a measure
of speed and accuracy. The faster the system transitioned
between states, and finds the MPP, the greater the output
power. Likewise, a power increase was gained with increased
MPP tracking accuracy.
This led to the test procedure of switching between “full
sun” and “cloud cover” conditions and observing the systems
output.
Figures 8 and 9 show the plots of the algorithms under varying
conditions. At t = 0 all four lights are switched on, the
algorithms ascend to the maximum power point each reaching
the expected power of ≈ 16W . Two lights are switching off at
t ≈ 19s and the algorithms descend to ≈ 7W , at t ≈ 35s the
additional lights switched back on again and the algorithms
return to their previous state.
While testing, it became apparent that there were three
main regions where the algorithms behaved in a similar
fashion repeatedly. The first of these was the rise from zero
to the direct sunlight MPP.
0 5 10 15 20 25 30 35 40 45 500
10
20Perturb and observe algorithm
Pow
er
(W)
0 5 10 15 20 25 30 35 40 45 500
10
20MPPT−temp algorithm
Pow
er
(W)
0 5 10 15 20 25 30 35 40 45 500
10
20Incremental conductance algorithm
Seconds (s)
Pow
er
(W)
Fig. 8. Individual Responses of MPPT Algorithms
0 5 10 15 20 25 30 35 40 45 500
2
4
6
8
10
12
14
16
18
20Response of MPP algorithms
Seconds (s)
Pow
er
(W)
Perturb and Observe
MPPT−Temp
Incremental Conductance
Fig. 9. Response of MPPT Algorithms
A. Region one: Zero to the direct sunlight MPP
In real world applications region one would only occur
once a day, at sun rise, so is less important than the following
two regions. Another consideration is that early in the
morning, sun irradiance on the PV is low due to the suns
angle [11]. It is also expected that the sun will rise slower
than the algorithms can respond making region one of little
significance.
The incremental conductance algorithm was the slowest of
the three algorithms, deviating on its ascent. When analysis
was performed on the incremental conductance, a fault was
diagnosed. The fault was the algorithms reliance on the ∆I∆V
condition. As the current and voltage only changed a small
amount at the extreme ends of the PV’s I-V curve, ∆I∆V
often
threw “not a number” (NaN) or infinity (Inf). This would
result in the algorithm not shifting the PV’s state, which
caused the algorithm to get stuck at this point. Due to the use
of ∆I∆V
in the algorithm, and its coupling to future iterations
of the loop, amending the code was unachievable.
This displays one disadvantage to current sensing over
temperature sensing, however increasing the resolution of the
current and voltage sensor would some what remedy this
problem.
Despite the P and O being quoted as the slower of the three
algorithms, in this region it consistently had the fastest rise
time with the MPPT-temp a close second.
As one current sensing algorithm was faster, and one
slower than the temperature sensing algorithm, region one was
inconclusive.
B. Region two: On the MPP
Region two is a test of how closely the system remained
on the MPP. Neglecting the rise and fall regions of figure 9,
the standard deviation was taken. Table 1 shows the numerical
results which are plotted with the systems output and mean for
“full sun” and “cloud cover” in figures 10 and 11 respectively.
The results show the temperature sensing algorithm having
the smallest deviation from the MPP with the incremental
conductance a close second. The P and O algorithm had a
largest deviation which was more than twice that of the MPPT-
temp in both four and two light tests.
TABLE ISTANDARD DEVIATION OF ALGORITHMS
Algorithm Two Lights Four LightsP and O 1.48 0.66Incremental Conductance 0.46 0.39MPPT-temp 0.16 0.25
0 1 2 3 4 5 6 7
14
16
18
Pow
er
(W)
Mean and standard deviation of MPPT−temp
0 1 2 3 4 5 6 7
14
16
18
Pow
er
(W)
Mean and standard deviation of incremental conductance
0 1 2 3 4 5 6 7
14
16
18
Pow
er
(W)
Seconds (s)
Mean and standard deviation of P and O
Fig. 10. Algorithm response to “full sun” conditions, where the dotted lineis the standard deviation
A comparison of each algorithm’s mean power output is
plotted in figure 12. It demonstrates which sensing approach
0 1 2 3 4 5 6 7
5
10
Mean and standard deviation of MPPT−temp
Pow
er
(W)
0 1 2 3 4 5 6 7
5
10
Mean and standard deviation of incremental conductance
Pow
er
(W)
0 1 2 3 4 5 6 7
5
10
Mean and standard deviation of P and O
Seconds (s)
Pow
er
(W)
Fig. 11. Algorithm response to “cloud cover” conditions, where the dottedline is the standard deviation
can obtain the highest average power for each given input. This
shows the temperature sensing algorithm’s output power giv-
ing the highest amplitude, with the incremental conductance
second and P and O much lower.
All the algorithms, however, had a higher output power than
the reference tests for the direct sunlight condition. The MPPT-
temp and incremental conductance algorithms also achieved
this for the cloud cover condition. The P and O algorithm’s
mean value for cloud cover was less than the reference MPP,
due to the deviations being large.
The reason the algorithms achieved a higher output power than
the reference, was that discrete resistor values were used to
calibrate the PV module. This meant the MPP was between
the achievable loads, due to these discrete values. A variable
load could be used to test the PV modules output, however to
dissipate high power, the financial cost was the limiting factor.
It is expected that this region will occur regularly, for short
periods of time, on a cloudy day and for longer periods on
a sunny day. This makes it an important region for accuracy
and precision.
In region 2 the temperature sensing algorithm showed
an advantage over the current sensing algorithms with the
lowest standard deviation. The temperature sensor also had
the highest mean power output, making it the best performing
system of the three through this region.
C. Region three: Fall and rise from cloud cover to direct
sunlight
The third region is a better indication, than region one, of
the systems speed. The region will occur more regularly in a
day, as the rise from zero is only expected to occur once a
day. Figure 13 shows the rise and fall of the system between
“full sun” and “cloud cover”. It shows that the current and
temperature sensing algorithms descent is much the same.
However, the temperature sensing algorithm has the fastest
0 1 2 3 4 5 6 7
16
16.5
17Mean value of P and O vs MPPT−temp with four lights
Seconds (s)
Pow
er
(W)
MPPT−temp
Perturb and observe
Incremental conductance
0 1 2 3 4 5 6 74
6
8
10
Mean value of P and O vs MPPT−temp with two lights
Seconds (s)
Pow
er
(W)
MPPT−temp
Perturb and observe
Incremental conductance
Fig. 12. Mean of algorithms
ascent, followed by incremental conductance. The P and O
algorithm dips before it ascends which regularly occurred
when testing the algorithm. The time it took each algorithm
to rise from cloud cover to direct sunlight is listed in table II.
17 18 19 20 21 22
8
10
12
14
16
18
Decreased solar irradiance
Seconds (s)
Pow
er
(W)
32 34 36 38 40 42
8
10
12
14
16
18
Increased solar irradiance
Seconds (s)
Pow
er
(W)
Fig. 13. Transition between states
TABLE IIRISE TIMES FROM CLOUD COVER TO DIRECT SUNLIGHT
MPPT-temp Incremental conductance P and OTime 1.5s 2.75s 5.85s
In region 3 all three algorithms descended at the same
rate. However, the temperature sensing algorithm ascended
at a much faster rate. This meant, through this region, the
temperature sensing algorithm was again the superior of the
three.
V. CONCLUSIONS
This paper compares the use of a current sensor for maxi-
mum power point tracking with a temperature sensor. To do so,
two environmental conditions were simulated in a controlled
environment, “full sun” and “cloud cover”. This enabled the
switching between conditions so the system response could be
observed transitioning to, and remaining on the MPP.
The test was broken into three regions, rise from zero to
direct sunlight, On MPP and fall and rise from cloud cover
to direct sunlight. It was decided that region one was of little
significance and the remaining two regions would decide the
success of the algorithm.
In both cases the temperature sensor out performed the cur-
rent sensor, with faster tracking and better accuracy. It was
also noted the current sensor required additional circuitry so
complexity and cost was higher than that of the temperature
sensor.
VI. FUTURE WORK
The future work would see continued repeatability of the
algorithms so long term power increase could be quantified.
Also external testing to make sure the sensor and algorithm
can withstand dynamic environmental conditions.
VII. ACKNOWLEDGMENTS
The authors would like to acknowledge the NZIRI India
Studies Research Grant for its support. The first author would
also like to thank the Victoria University Doctoral Scholarships
and ECS technical staff member Tim Exley for his help.
REFERENCES
[1] R. F. Coelho, F. M. Concer, and D. C. Martins, “A mppt approach basedon temperature measurements applied in pv systems,” IEEE ICSET, 122010.
[2] Y. de Boer, “Kyoto protocol reference manual on accounting of emis-sions and assigned amount,” United Nations Framework Convention on
Climate Change, 2008.[3] Environmental Protection Authority, “Greenhouse,” Environmental Pro-
tection Authority, 2012.[4] G. Masson, M. Latour, M. Rekinger, I.-T. Theologitis, and M. Papouts,
“The emissions gap report 2012,” United Nations Environment Pro-
gramme (UNEP), 2012.[5] International Energy Agency, “Co2 emissions from fuel combustion
highlights 2012 edition,” IEA STATISTICS, 2012.[6] Worldwide electricity production from renewable energy srouces, “Elec-
tricity prodution in the world: general forecasts,” Fourteenth inventory
2012 edition, 2012.[7] G. Masson, M. Latour, M. Rekinger, I.-T. Theologitis, and M. Papoutsi,
“Global market outlook for photovoltaics 2013-2017,” European Photo-
voltaic Industry Association, 2013.[8] K. C. Kong, M. bin Mamat, and M. Z. Ibrahim, “New approach on math-
ematical modeling of photovoltaic solar panel,” Applied Mathematical
Sciences, Vol. 6, no.8, 2012.[9] R. F. Coelho and D. C. Martins, “An optimized maximum power
point tracking method based on pv surface temperature measurement,”INTECH, Sustainable Energy - Recent Studies, 2012.
[10] D. Burmester, R. Rayudu, and T. Exley, “Single ended primary inductorconverter reliance of efficiency on switching frequency for use in mpptapplication,” APPEEC IEEE, 12 2013.
[11] W. B. Stine and M. Geyer, “Power from the sun,” 2001.