Distributed Control of HVDC Transmission Grids

DAVOOD BABAZADEH

Doctoral Thesis

Stockholm, Sweden 2017

TRITA-EE 2017:018

ISSN 1653-5146

ISBN 978-91-7729-310-1

Electric Power and Energy Systems

KTH, Royal Institute of Technology

Stockholm, Sweden

Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy

© Davood Babazadeh, March 2017. Copyrighted articles are reprinted with kind permission

from IET, IEEE and Elsevier.

Printed by Universitetsservice US AB

Abstract

Priority access of renewable resources such as offshore wind recommended by European

energy directives, new market models and trading the electric energy among countries lead

to new requirements on the operation and expansion of transmission grids. Since AC grid

expansions are limited by legislative issues and long distance transmission capacity, there

is a considerable attention drawn to application of HVDC transmission grids on top of, or

in complement to, existing AC power systems. Potential benefits of HVDC transmission

grids includes the possibility to access remote energy sources thereby increasing renewable

penetration, improving grid security and decreasing congestion in the system. However, the

secure operation of HVDC grids requires a hierarchical control system to manage different

functions such as voltage or power flow control. In HVDC grids, the primary control action

to deal with power or DC voltage deviations is communication-free and local which can be

carried out by different control schemes such as DC voltage droop control. In addition to

primary local actions, the higher supervisory control actions are needed to guarantee the

optimal operation of HVDC grid.

This thesis presents distributed control of an HVDC grid. To this end, three functions

are investigated to be deployed in HVDC supervisory system; coordination of power injec-

tion set-points in the presence of large wind farms, DC slack bus selection and two-stage

network topology identification. However, the implementation of supervisory control func-

tions is linked to the arrangement of system operators; i.e. an individual HVDC operator

(central structure) or sharing tasks among AC system operators (distributed structure). In

this thesis, all three functions are first investigated for the central structure. As main contri-

bution, this thesis presents the distributed solutions for the determined supervisory control

applications. Furthermore, to study all aspects of proposed algorithms, a co-simulation

platform is introduced.

In this thesis, two different distributed algorithms based on Alternating Direction Method

of Multipliers (ADMM) and Auxiliary Problem Principle (APP) are used to solve coordi-

nation of power injection. However, for distributed implementation of DC slack bus, the

choice of parameters for quantitative ranking of converters is important. These parame-

ters should be calculated based on local measurements if distributed decision making is

desired. To this end, the short circuit capacity of connected AC grid and power margin

of converters are considered for the evaluation of converters to work as slack bus. To es-

timate the short circuit capacity as one of the required parameters for selection of DC

slack bus, the result of this thesis shows that the recursive least square algorithm can be

very efficiently used. Besides, it is possible to intelligently use a naturally occurring droop

response in HVDC grids as a local measurement for this estimation algorithm. Regarding

the network topology, a two-stage distributed algorithm is introduced to use the abstract

information about the neighbouring substation topology to determine the grid connectivity.

Key words: co-simulation, cyber-physical system, DC slack bus, distributed control,

HVDC grids, power injection, topology processor, wind farms

ii

Acknowledgements

First and most, Im very grateful to my main supervisor Lars Nordstrom who has given

me very helpful and valuable support and guidance throughout the PhD project. Similarly,

my best thanks go to my co-supervisors Pontus Johnson and Karl Henrik Johansson that

helped me develop my skills in this journey.

Special thanks go to my friends and colleagues Annica Johannesson, Kaveh Paridari,

Mikel Armendariz and Hassan Fidai who have been helpful in the proofreading of my thesis

material and a source of motivation to wrap up the thesis. Furthermore, I would like to

thank the administration group of the department Elvan Helander, Annica Johannesson,

Eleni Nylen and JudyWesterlund who provided a calm and fruitful environment to research.

In general, I am very thankful to the great atmosphere at KTH both in former ICS

department and new EPE department, where discussions and collaborations with my col-

leagues and friends made these five years interesting and brought a great opportunity to

learn new skills for personal and work life. Here, special regards goes to Mikel Armen-

dariz, Kaveh Paridari, Moustafa Chenine, Mohammad Nazari, Hassan Fidai, Matus Kor-

man, Margus Vlja, Daniel Brodn, Fabian Hohn, Harold Chamorro Vera, Wei Li, Mehrdad

Kazem Tabrizi, Arshad Saleem, Claes Sandels, Liv Gingnell, Khurram Shahzad, Robert

Lagerstrom, Mathias Ekstedt and the PSMIX research group.

I would like to acknowledge the colleagues in my PhD technical reference group from

industry who have contributed significantly to my research development, from ABB: Tomas

Larsson, Pinaki Mitra, Hans Bjorklund, Lidong Zhang, Inger Segerqvist, Magnus Callavik,

and from Svenska Kraftnt, Magnus Johansson and Gran Ericsson. I would like to thank the

sponsors of the PhD project - ABB, Svenska Krafnt, SweGRIDS and the Swedish Energy

Agency.

Here, I also would like to acknowledge all co-authors in my publications for the enjoyable

and constructive collaboration, especially Alberto Borghetti, Carlo Alberto Nucci, Dirk Van

Hertem, Arvind Muthukrishnan and Kun Zhu.

Very special thanks to my family who have been very supporting and loving in my life

- Marita, Farideh and Darab. Also an acknowledgement goes to my friends Kaveh, Mikel,

Eneko, Piotr, Amir, Reza, Ingrid.

Thank you all!

Stockholm, February 2017

Davood Babazadeh

iii

Papers

List of included papers

Paper 1: D. Babazadeh, D. Van Hertem, L. Nordstrom, et al., ”Study of Centralized

and Distributed Coordination of Power Injection in Multi-TSO HVDC Grid with Large

Off-shore Wind Integration”, Journal of Electric Power Systems Research, Volume 136,

July 2016, pp. 281-288, ISSN 0378-7796.

Paper 2: D. Babazadeh, A. Muthukrishnan, P. Mitra, T. Larsson, L. Nordstrom, ”Se-

lection of DC Voltage Controlling Station in an HVDC Grid”, Journal of Electric Power

Systems Research, 2016, pp. 1-8.

Paper 3: D. Babazadeh, A. Muthukrishnan, P. Mitra, T. Larsson, L. Nordstrom, ”Real-

Time Estimation of Grid Short Circuit Capacity for HVDC Control Application”, IET

Generation, Transmission and Distribution, 2016, pp. 1-8

Paper 4: D. Babazadeh, and L. Nordstrom, ”Distributed Security-Constrained Sec-

ondary Control of HVDC grids in the Presence of Wind Uncertainty”, submitted to Sus-

tainable Energy, Grid and Network, 2016

Paper 5: D. Babazadeh, and L. Nordstrom, ”Analysis of Centralized versus Distributed

Two-stage Network Topology Processor for HVDC Grid”, submitted to Journal of Engi-

neering Applications of Artificial Intelligence, 2016.

Paper 6: D.Babazadeh, M. Nazari, M. H. Fidai, M. Chenine, M. Ghandhari and L.

Nordstrom, ”Implementation of agent-based power flow coordination in AC/DC grids us-

ing co-simulation platform”, IEEE International Conference on Smart Grid Communication

(SmartGridComm), Venice, 2014, pp. 188-193.

Author contributionsIn Paper 1, the research problem formulation and solution implementation were carried

out by Babazadeh. Van Hertem and Nordstrom supported and reviewed the research. The

paper was fully authored by Babazadeh.

In Paper 2, the general research concept was initiated and developed by Babazadeh and

Mitra, whereas the article was authored by Babazadeh. The modeling and programming

was done by Muthukrishnan. Larsson and Nordstrom supported and reviewed the research.

In Paper 3, the general research concept was initiated and developed by Babazadeh

and Mitra, whereas the article was authored by Babazadeh. The modeling was assisted by

Muthukrishnan. Larsson and Nordstrom supported and reviewed the research.

In Paper 4, the problem formulation and the solution algorithm were done by Babazadeh.

The programming of the algorithm was assisted by Muthukrishnan. The paper was fully

authored by Babazadeh. Nordstrom supported and reviewed the research.

iv

In Paper 5, the general research concept was due to Babazadeh and the authoring was

fully done by Babazadeh. Nordstrom supported and reviewed the research.

In Paper 6, the research concept and design of the test-bed platform were carried out

by Babazadeh. The modeling has been assisted by Fidai and Nazari. Chenine provided

valuable related research on test-bed. Ghandhari and Nordstrom supported and reviewed

the research. The paper was fully authored by Babazadeh.

v

Publications not included in the thesis

Paper 7: D. Babazadeh, A. Muthukrishnan, P. Mitra, T. Larsson and L. Nordstrom,

”Short circuit capacity estimation for HVDC control application,” 2016 Power Systems

Computation Conference (PSCC), Genoa, Italy, 2016, pp. 1-7.

Paper 8: M. Armendariz, D. Babazadeh, D. Broden and L. Nordstrom, ”Strategies

to Improve the Voltage Quality in Active Low-Voltage Distribution Networks Using DSOs

Assets,” IET Generation, Transmission and Distribution, 2016.

Paper 9: D.Babazadeh, A.Tonti, M.Armendariz, A. Borghetti, C.A. Nucci, L. Nord-

strom ”Two-stage Network Processor for an Independent HVDC Grid Supervisory Con-

trol,” in IEEE PES GM, 2016.

Paper 10: W. Yiming, D.Babazadeh and L. Nordstrom, ”Modeling of Communica-

tion Infrastructure Compatible to Nordic 32 Power System,” in IEEE PES GM, 2016.

Paper 11: M. Armendariz, D.Babazadeh, L. Nordstrom and M. Barchiesi, ”A method

to place meters in active low voltage distribution networks using BPSO algorithm,” 2016

Power Systems Computation Conference (PSCC), Genoa, Italy, 2016, pp. 1-7.

Paper 12: M. T. Khan, K. Larsson, M. Zakaria, M. H. Fidai, D.Babazadeh and L.

Nordstrom ”Distributed Secondary Frequency Control Considering Rapid Start Units Us-

ing Alternating Direction Method of Multipliers”, in MSCPES 2016, Vienna, Austria, April

2016.

Paper 13: L. Nordstrom and D.Babazadeh, ”Cyber physical Approach to HVDC grid

control,” in Cyber Physical Systems Approach to Smart Electric Power Grid, Springer-

Verlag New York, 2015, pp. 75-101.

Paper 14: M.H. Fidai, D.Babazadeh, J. Hanning, T.Larsson and L. Nordstrom, ”Real-

time Implementation of Optimal Power Flow Calculator for HVDC Grids,” in 2015 Cigre

Symposium, Lund, Sweden, May 2015.

Paper 15: D.Babazadeh, D. V. Hertem, M. Rabbat and L. Nordstrom, ”Coordina-

tion of Power Injection in HVDC Grids with Multi-TSOs and Large Wind Penetration,” in

ACDC 2015, Birmingham, UK, 2015.

Paper 16: D.Babazadeh, and L. Nordstrom, ”Multi-Stage Network Processor for an

Independent and Integrated HVDC grid Supervisory Control Architecture,” in The 13th

International Workshop on Electric Power Control Centers, Bled, Slovenia, May 2015.

Paper 17: D.Babazadeh, F. Freizadeh and L. Nordstrom, ”Distributed Ancillary Service

Support for Independent AC-Areas through HVDC Grid,” in IEEE PowerTech Eindhoven,

Netherlands, June 2015.

vi

Paper 18: M.H. Fidai, D.Babazadeh, J. Hanning, T.Larsson and L. Nordstrom, ”Real-

time Implementation of Optimal Power Flow Calculator for HVDC Grids,” in Cigre Sym-

posium, Lund, Sweden, May 2015.

Paper 19: L. Nordstrom and D. Babazadeh, ”Requirements on Communication and

Control Systems for HVDC Grids,” in IEEE PES Conference on Innovative Smart Grid

Technologies (ISGT); Washington, DC, US, February 2015

Paper 20: M.H. Fidai, D.Babazadeh, J. Hanning, T.Larsson and L. Nordstrom, ”Super-

visory Control for VSC-HVDC Grid Interconnecting AC Systems,” in IEEE PES General

Meeting, Denver, July 2015.

Paper 21: D.Babazadeh, and L. Nordstrom, ”Agent-based Control of VSC-HVDC

Transmission Grid : A Cyber Physical System Perspective,” in MSCPES 2014, Vienna,

Austria, March 2014.

Paper 22:, Y. Wu, D. Babazadeh and L. Nordstrom ”Stateful Data Delivery Service

for Wide Area Monitoring and Control Applications,” in Proceedings of the International

Conference on Dependable Systems and Networks, 2014, pp. 768-773.

Paper 23: D.Babazadeh, K. Zhu, M. Chenine, A. Al-Hammouri and L. Nordstrom,

”A Platform for Wide Area Monitoring and Control System ICT Analysis and Develop-

ment,” in IEEE Grenoble Conference PowerTech, France, 2013.

Paper 24: R. Bottura, D.Babazadeh, A. Borghetti, C.A. Nucci and L. Nordstrom,

”SITL and HLA Co-simulation Platforms : Tools for Analysis of the Integrated ICT and

Electric Power System,” in 2013 IEEE EUROCON, Zagreb, Croatia, 2013.

Paper 25: D.Babazadeh, M. Chenine and L. Nordstrom, ”Survey on the Factors Re-

quired in Design of Communication Architecture for Future DC grids,” in IFAC Proceedings

Volumes 2013, pp. 58-63.

Paper 26: K. Zhu, S. Deo, A. Al-Hammouri, N. Honneth, M. Chenine, D.Babazadeh

and L. Nordstrom, ”Test Platform for Synchrophasor based Wide-Wrea Monitoring and

Control Applications,” in IEEE Power and Energy Society General Meeting (PES), Van-

couver, Canada, 2013.

Paper 27: D.Babazadeh, K. Zhu, M. Chenine, A. Al-Hammouri and L. Nordstrom,

”Real-Time Smart Grid Application Testing using OPNET SITL,” in OPNETWORK2012,

US, 2012.

Paper 28: K. Zhu, D.Babazadeh, Y. Wu and L. Nordstrom, ”Design of Robust Wide-

Area Damping Controller Based on Synchronized Phasor,” in APSCOM, 2012.

vii

Contents

1 Introduction 1

1.1 Background and motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Research objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.3 Main contribution related to research objective . . . . . . . . . . . . . . . . 6

2 Research Context 8

2.1 HVDC grids operation and control . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 Short circuit capacity estimation for HVDC application . . . . . . . . . . . 14

2.3 Tools for simulation of cyber-physical systems . . . . . . . . . . . . . . . . . 17

3 Related Works 21

3.1 Coordination of power injection . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.2 DC slack bus selection and related practical implementation . . . . . . . . . 23

3.3 Network topology processor . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4 Research Results and Discussions 27

4.1 Investigation of advanced control applications for HVDC grids operation . . 27

4.2 Distributed solution for control applications in an HVDC grid . . . . . . . . 32

4.3 Benefit of real-time co-simulation platform . . . . . . . . . . . . . . . . . . . 38

5 Conclusion and Future Work 41

Bibliography 43

viii

Chapter 1

Introduction

This doctoral thesis is divided into two parts. In part I, a brief introduction to the research

topic, the research context and a summary of methods and findings are presented by sum-

marizing some of author’s research articles. Then, these research articles are presented in

part II of this thesis. This chapter in part I includes motivation to the research topic and

the research objectives, followed by a summary of the research contribution.

1.1 Background and motivation

In the recent power system expansion driven by growing energy demand, more attention

is being put on integration of Renewable Energy Sources (RES). This is tangible by the

intention of the European Union to give priority access to RES or to produce 20 % of its

electric power demands through RES by 2020 [30]. The benefits of integrating renewable

resources such as wind or solar have been justified through many researches and to some

extent in real-world projects [28, 39]. One of the challenges in increasing the share of

renewable production in the power system is the remote location of the resources and

consequently the problem in transmitting bulk electric energy to load centers. Current AC

power transmission grids operate close to their limits. However, the expansion of the AC

grid involves problematic legislative rights-of-way efforts limiting the speed of expansion.

The expansion is also limited by long distance transmission capacity of AC grids. Taking

these challenges into account, recently there is a significant attention drawn to application

of High Voltage Direct Current (HVDC) transmission grids on top of, or in complement

to, existing AC transmission grids. Potential benefits of an HVDC grid as an alternative

solution include but are not limited to increasing access to remote energy sources, improving

power system security and decreasing congestion in the system [31, 52, 71, 84, 108].

Topologies

Several works have been carried out to investigate the proper solution for building HVDC

systems or grids [20, 60, 82]. To this end, there are two HVDC technologies, based on

either the Line Commutated Converter (LCC) or the Voltage-Source Converter (VSC).

Based on certain parameters such as the topology of the HVDC systems, expected transfer

capacity, strength of the connecting AC system or expected control functionality, a single

1

technology or a mix of both technologies can be used for the solution. Some examples of

suggested topologies are parallel or serial point-to-point links, star and ring topologies [47].

In the literature, the term ”HVDC Grid” is used for the meshed network architecture

to differentiate it from the rest of topologies while ”Multi-terminal HVDC” is a general

term for all configurations that connect more than two converters. Regarding HVDC grid

development, the VSC technology due to its power flow flexibility is the most suitable

solution to build meshed topology grids [3,47]. On the other hand, in order to transfer higher

power capacity in a series link topology LCC technology is more suitable [8,78]. This thesis

focuses on meshed HVDC grids based on VSC technology. As examples of proposals and

projects on HVDC grids, Supergrid (see Fig.1.1) has been suggested to integrate Europe’s

abundant offshore wind in the North Sea and Desertec similarly to harness sustainable

power from the sun-rich regions [32, 96]. Along with the main objective of the bulk power

transmission from generation nodes to load centers, HVDC grids are able to offer a reliable

infrastructure for connected AC areas to exchange active power or control local reactive

power, thereby providing ancillary services to the AC areas. These services can be offered,

for example in the context of frequency control, voltage support or damping of electro-

mechanical oscillations [7, 13, 52, 60].

Figure 1.1: The European Supergrid structure [23].

Protection

To make HVDC grids come to reality, there are still some challenges to be studied and

solved. The configuration of the grid, vendor interoperability and standardization [4], DC

breaker and protection system [52, 92, 103], ancillary services provided by the HVDC grid

and finally supporting automation and communication systems dedicated for HVDC grids

2

are some of those challenges [11,31,52]. One time-sensitive constraint in the secure operation

of HVDC grid is the protection system. In contrast to the conventional HVDC which do not

experience a large overcurrent during the faults due to its large DC smoothing reactance,

the discharge of the DC link capacitor in VSC-HVDC can lead to high levels of overcurrent.

Therefore the protection system should be designed in such a way that it detects and isolates

a faulty part of the system within the range of few milliseconds. Most protection schemes

proposed for DC faults e.g. [93, 103] include algorithms that do not use DC breakers and

instead trips AC breakers. As an alternative, a full-bridge modular multi-level converter can

be used to interrupt the fault current and then mechanical switches isolate the fault [92].

In all those schemes the whole HVDC system needs to be interrupted and de-energized. To

avoid HVDC grid interruption and increase the availability of the whole system, schemes

with DC circuit breakers have been proposed to detect and isolate the faulted segment in

the DC side [36, 59]. This development enables to decouple the interruption of DC and

AC side and therefore brings better opportunity to operate HVDC grids and AC system

independently.

Control

In addition to the protection challenges, secure operation of HVDC grids also requires a

hierarchical control system to manage different functions such as voltage or power flow

control. In a VSC-based HVDC grid, the variation in demand and/or generation (i.e.

power injection and extraction at converters) introduces DC voltage and power deviations

on the DC side [57, 83]. The primary response to any disturbance within the HVDC grid

is normally carried out by a communication-free DC voltage droop strategy where certain

converters change their power set-points proportional to the deviation in DC voltage to

compensate the imbalances, which is similar to primary frequency control concept in AC

system [1, 20, 58, 100]. However, other different strategies for DC voltage control of HVDC

converters have also been proposed and studied in literature. DC voltage droop with or

without dead-band, pilot voltage droop control and adaptive droop control are some of

these strategies [1,24,76,97,101]. While some schemes such as droop with or without dead-

band as well as adaptive droop proposed in [22, 76, 97] are communication-free and just

use the local DC voltage information to compensate the power mismatch, the pilot voltage

droop method in [24] needs to assign a global DC voltage and then communicate the value

with certain converters. Similarly, Apart from the choice of conventional or advanced droop

control, the optimal design of droop constants has been studied intensively as well [2,37,45].

In addition to primary response, similar to AC grid, the secondary control action is

needed to optimally tune the DC power and voltage set-points, and prepare the system

for new disturbances considering future demand, uncertainty in generation and also market

signals [28, 46, 89, 106]. The design of the secondary control for HVDC grids has been

approached from different perspectives such as market concerns [28,41], grid security [26,91],

system uncertainties [102] and optimization problem formulation [9, 17, 42, 74].

In most works, the formulation of secondary correction of set-points has been studied in

the context of Optimal Power Flow (OPF) with various objective functions, system bound-

aries, constraints and optimization techniques [25, 26, 43, 50, 64]. The combined AC/DC

optimal power flow has been studied for both the non-linear [9, 19, 100], and the linearized

power flow equations [64, 91, 102]. The objective of secondary control in literature varies

3

from minimizing the losses to trying to follow the AC pre-defined schedule [25, 43]. Opti-

mization techniques such as second order cone programming [19] and convex relaxation [17]

have been used to handle the non-convexity of OPF problem and the interior point method

has been commonly used to solve the problem. However heuristic algorithms such as genetic

algorithm in [85], particle swarm optimization in [88] and differential evolution in [74] have

also been considered. Apart from the overall AC/DC OPF formulation approach to directly

correct the set-points in the HVDC grid, some other works such as [55,77] propose an archi-

tecture based on integral control action similar to automatic generation control (AGC) of

AC grids to carry out the secondary control within HVDC grid. In [5], distributed secondary

control has been studied for HVDC grids by proposing controllers which are designed on

top of voltage droop controllers. The proposed control schemes bring the voltages closed to

their nominal values while minimizing a quadratic cost function of injected DC currents.

AC / DC grid operation

· Combined AC/DC OPF

· Economic consideration

ms~100sms

sec~min

15min

HVDC grid

Current Control

Outer Control· DC voltage

· Power

AC

area A~

=

HVDC Supervisory Control

Control Applications : OPF

Network

Processor

Control Mode

Selection

State

Estimation

AC area C

~=

AC

area B~

=

Current Control

Outer Control· DC voltage

· Power

Figure 1.2: HVDC grid control architecture.

Considering different level of control actions, a hierarchical architecture has been as-

sumed for the HVDC grid control system [45, 56]. The primary control actions in HVDC

grids takes place at converter level and secondary control actions happen at the higher

level that is referred as supervisory controller or sometime master controller [43, 75]. The

supervisory control level manages disturbances in an HVDC grid and bridges the gap be-

tween slower AC Energy Management Systems (EMS) and fast converter controls (see Fig.

1.2). To carry out the secondary actions, similar to AC EMS, the supervisory controller

requires certain applications. For example, state estimator (with bad data detection) is

applied to estimate HVDC grids’ state in the presence of errors in measurements, and a

network processor is used to recognize the topology of the HVDC grid and detecting is-

landing scenarios [15, 34]. However, taking into account the different control strategies in

HVDC grids compared to AC grids, the new EMS applications have to be considered when

designing HVDC supervisory control system. Ancillary service coordination and DC volt-

4

age droop gain optimizer are examples of such applications that have been proposed and

studied in [6, 31, 37, 52].

Given the HVDC grid supervisory control level to coordinate between the AC energy

management system and local converter controls, the implementation of supervisory control

functions can be coupled tightly to the arrangement of system operators. As shown in Fig.

1.3, there are two possible operational strategies: an individual HVDC operator (central

structure) or sharing the tasks among AC Transmission System Operators (TSO) i.e. dis-

tributed structure [16, 81]. Looking at possible operational strategies of HVDC grids, two

points must be taken into account when designing the HVDC control schemes. First, the

control needs to function properly when the coordination of converters is shared between

connecting TSOs. Second, if one TSO coordinates the grid centrally, a back-up distributed

coordination plan can be considered to take over the responsibility in case of central failure.

TSO1TSO2

Wind Farm

Operators

TSO3

(b)(a)

VSC 1

VSC 2

VSC 4

VSC 3

VSC5

~=

~=

~=

~=

~=

Wind Farm

AC1

AC1

AC2

AC3

VSC 1

VSC 2

VSC 4

VSC 3

VSC5

~=

~=

~=

~=

~=

Wind Farm

AC1

AC1

AC2

AC3

VSC

~=AC2

VSC

~=AC3

TSO2

TSO3

TSO1

Wind Farm

OperatorsVSC5~

=

Wind Farm

VSC 4

~=AC2

VSC 1

VSC 2

~=

~=

AC1

AC1

VSC 3

~=AC3

C 1

SC 2

~=

~=

AC1

AC1

V

V

4

3

VSC5~=

Wind Farm

DC grid Operator

Figure 1.3: HVDC grid operation concepts: (a) central structure (an independent HVDC

grid operator), (b) distributed structure.

1.2 Research objective

Considering the aforementioned issues, the following research question was formulated at

the initial step of the thesis:

• Will distributed control of HVDC grids offer the same levels of reliability as centralized

control?

Based on the research question, the following objectives were determined for this thesis:

• OBJ1: Identify the need of advanced control applications for an HVDC grid supervi-

sory controller and define their requirements (in terms of computational time as well

as required measurements or information).

• OBJ2: Develop distributed solutions for the determined control applications and

investigate if they fulfill the requirements

• OBJ3: Design a tool that enables modeling and analysis all aspects of proposed

algorithms from control, communication and power system perspectives.

5

1.3 Main contribution related to research objective

The structure of objectives and corresponding contributions are shown in Fig.1.4.

Regarding OBJ1, different types of control functions for the HVDC supervisory con-

troller have been already studied such as secondary control of DC voltage, AC/DC combined

power flow and state estimation [21,34,45]. However, taking into account the literature and

also possible need of supervisory actions inside HVDC grids, specially when they are con-

nected to large wind farms, coordination of power injection set-points of HVDC converters

has been suggested as a control function in Paper 1. The requirement on how fast and

frequent the application should be run, have been also addressed in Paper 1. However, the

formulation of this control function or other control functions proposed in literature require

two other types of functions: i) network topology processor to provide the Y matrix and

grid topology and ii) a function to select the control mode of converters. Paper 2 and 5

focus on these two supplementary control functions. Paper 2 presents an online evaluation

framework to select DC voltage controlling station and has investigated which parameters

are needed for such a selection. Paper 2 suggested power margin of converter and short

circuit capacity of connected AC grid as possible evaluation metrics. By addressing OBJ1,

the type of functions, their relations, required calculation time and finally required mea-

surements and information were determined.

Paper 1: coordination of power injection in

HVDC grids with large-scale wind farms and

its required updating rate

Paper 2: why and how to

select DC voltage controlling

station (slack bus)

OBJ1: Identify supervisory control functions

and their relations and requirements

Grid and

substation

topology

Paper 5: two-stage network

topology processor and its

requirements

Grid topology

and islanding

Control mode

Paper 1: APP-based distributed optimal

coordination for power injection in HVDC

grids with multiple TSOs.

Paper 4: ADMM-based distributed power

injection coordination with N-1 security

consideration and wind uncertainty

Paper 5: distributed identification of grid

topology in a two-stage network topology

processor using successive-over-relaxation

OBJ2: Develop distributed solution

Paper 3: estimation of grid parameters based

on local information to be used to evaluate

the credibility of HVDC converters to work

as DC slack bus

OBJ2: Develop distributed solution

OBJ3: Design and build a co-simulation platform

Paper 6: introduce the real-time co-simulation called

PSMIX that can be used for cyber-physical system

modelling and simulation

Figure 1.4: An overview of thesis contributions in relation to the research objectives.

6

OBJ2 addressed the distributed solution and corresponding challenges for the three

above-mentioned control functions as follows:

• In Paper 1 distributed solution for the coordination of power injection was demon-

strated by formulating the problem as a non-linear constrained optimization problem.

Then it has been decomposed to sub-problems using the Auxiliary Problem Principle

(APP) method for the distributed structure. This distributed optimization problem

is solved by exchanging the required information between the AC TSOs.

• In Paper 4 the performance of distributed solution presented in Paper 1 has been

improved by formulating the sub-problems as a convex optimization problem and

adopting a modified version of Alternating Direction Method of Multipliers (ADMM)

to solve the problem. Furthermore, it has been shown that the distributed solution

can consider and handle other important aspects such as the N-1 secure criterion and

wind uncertainty in the formulation of the power injection coordination in HVDC

grids.

• Paper 3 studied the practical challenges in implementing an algorithm to estimate

the grid parameters that can be used for selection of slack bus. This practical as-

pects included non-matrix and non-complex modelling for hardware implementation,

studying type of measuring points and sensitivity of algorithm to design parameters in

HVDC applications. As a major contribution, it suggested to intelligently use a nat-

urally occurring droop response in HVDC grids as a measuring point for estimating

the grid parameters and consequently short circuit capacity.

• Paper 5 proposed a distributed algorithm that uses the neighboring information to de-

termine the grid connectivity. Regarding distributed islanding detection, the connec-

tivity problem has been formulated as a set of linear equations and solved iteratively

using successive-over-relaxation method.

To address OBJ3, Paper 6 has presented a real-time co-simulation test-bed which en-

ables the studies regarding the design, testing and implementation of real-time control and

operation applications in power system. This real-time platform reflects the characteris-

tics of the supporting Information and Communication Technology (ICT) and the physical

process, as well as the interfacing devices or systems as close as possible to the real life sce-

narios. This platform includes real-time power system simulator, real-time communication

network simulator, software-based and real interfacing devices/measurement, HVDC con-

verter controllers and supervisory applications.The performance of the platform has been

tested by a simple distributed algorithm for power sharing in Paper 6. However the plat-

form has also been used in Paper 2 and 3 when the practical implementation of algorithm

were the issue.

7

Chapter 2

Research Context

This chapter is to a large extent based on the author’s contribution in the book

chapter of ”L. Nordstrom and D. Babazadeh, Cyber physical approach to HVDC

grid control, In Cyber Physical Systems Approach to Smart Electric Power Grid,

pages 75101, Springer Berlin Heidelberg, Berlin, Heidelberg, 2015”. This partial

extract is included to provide research background and context.

2.1 HVDC grids operation and control

In order to convert high voltage AC power to DC power, two technologies are available,

classical Line Commutated Converter (LCC) and the Voltage Source Converter. LCC

technology is designed based on a semiconductor-based switch named thyristor. Unlike

diodes, thyristors need to be turned on, or fired, to start conducting current. These switches

can withstand the AC voltage in either polarity. But current can only flow in one direction

and can be limited by adjusting the time the thyristors are turned on. This time, or angle in

a sinusoid, at which the thyristors are turned on is called the firing angle, or valve ignition

delay angle, and is used to control power flow between the HVDC stations [53].

Voltage Source Converter technology is developed based on Insular Gate Bipolar Tran-

sistors (IGBT). The IGBT semiconductor can be controlled both with regards to being

turned on or off. In VSC technology, the DC current can flow in both directions. That is a

benefit over the LCC technology in which the current can flow in one direction. Considering

the bi-directional capability of the DC current flow in VSC, there is no need to change the

DC voltage polarity of the converters to change the power flow direction between converters.

Compared to LCC technology, it is possible for VSC to be connected to weak grids which

has low short-circuit level [107]. In contrast to LCC, VSC has two degrees of control which

enables control of the active and reactive power separately. This control freedom comes

from controlling the VSC using the Pulse Width Modulation (PWM) technology [38, 44].

However, one challenge with IGBT based VSC is that they have less overload capability

compared to LCC [66]. Taking into account the flexible power flow control of VSC tech-

nology, it is a recommended technology for the meshed topology HVDC grids. Therefore,

8

this section studies the VSC-based HVDC grid.

2.1.1 HVDC grids control system

In this section a possible control system for future VSC-based HVDC grid is described. The

HVDC grid control system can be presented by two general levels: converter station control

level and system control level. As shown in Fig.2.1, the converter station control consists of

inner and outer control layers. The phenomena within inner or outer control layers takes

place in the range of respectively milliseconds and several of milliseconds to seconds. Con-

trol functions at the station level control layer such as voltage or AC frequency stability

considerations or even local power flow calculation can be implemented with delays up to

a few seconds. The comprehensive description of converter control level is presented in the

next section.

AC Area

SCADA (C)

T2

VSC 2

DC – DC Converter

T5

VSC4

Grid

GridGrid

T5

VSC5

Grid

T3

VSC 3

Grid

DC Supervisory Control

OPFState Estimation

Topology ProcessorData GateWay

PWM

Converter Station Control

dq/abc

dcP

dcV

Ui iqá

&i iP Q

VSC 1

_PI

_+

*

dcV

PI

_+ acU

*

acU

PI

Inner Current

Control

PI +

Q

*

dcP

_+

*Q

dcPdcdc

Vdc

_PI

__++

*

dcVdc

PI

__++ acUU

*

acacU

PI

PI ++++

Q

*

dcPPdc

__++++

*Q

Outer Control

Communication Network

AC Area

SCADA (B)

Communication Network

DC Substation

Inner Current

Control

Inner Control

CSC

Inn Cont

Out Cont

==

AC Area

SCADA (A)

measurement and

breaker status

Figure 2.1: HVDC grid control system [78].

The system control level can be designed in different ways based on the market, technical

and political consideration which brings different data exchange limitations. For example,

9

the control of HVDC grid can be merged with the connected AC grids and a super hy-

brid AC/DC operator manages the entire system [25, 89]. However, as an alternative, an

individual HVDC supervisory management system can be introduced to control the phe-

nomena in an HVDC grid and bridges the gap between slower AC EMS and fast converter

controls [78,81,91]. In comparison to converter level, when it comes to supervisory control,

based on communication infrastructure, the requirements for control functions can vary

from tens of milliseconds for wide area protection system to minutes or longer for tertiary

power flow control. A simple schematic of HVDC grid control architecture involving a

separate HVDC supervisory control is presented in Fig. 2.1.

Given a separate HVDC supervisory controller, it consists of state estimation application

that estimates the state of the power system in the presence of errors in measurements. State

estimation also includes bad data detection to detect and identify the measurements with

error. Furthermore, the supervisory control requires a network processor to recognize the

topology of the HVDC grid and also detect islanding scenarios. The output of the network

processor can be used in state estimation or control application to form the admittance

matrix. Supervisory control is also responsible for real-time balancing in HVDC grids.

This can be carried out through power flow calculation after a disturbance, after a change

in HVDC grid topology or periodically.

2.1.1.1 Converter control level

In the literature, two different approaches have been introduced to control the VSC, i.e.

direct control and vector control. Direct control is based on controlling the voltage in the

VSC. This means by controlling phase angle and amplitude of the voltage transmitted active

power and reactive power is controlled. Vector control on the other hand sets the converter

to work as a controllable current source. The vector control method has some advantages

compared to direct control. This includes better power quality since it is less influenced

by grid harmonics and disturbances. Besides, in vector control decoupled control of active

and reactive power is possible. Finally it also provides the capability of inherent protection

during over-current events [38, 78].

In vector control approach, converter is set to work as a controllable current source.

In this approach, the injected current vector is set to follow a reference current vector.

Therefore, each VSC needs to have an internal current controller. In this scheme, dq

reference frame is used in order to project current vector into d and q axes (i.e. id and iq)

and respectively, decouple the control of active and reactive power [51]. Assume a typical

VSC station shown in Fig.2.2. R and L are the resistance and inductance, respectively, on

the AC side of the converter. These resistance and inductance consists of the transformer

and phase reactor parameters. Considering this model, the equation of the AC side in abc

coordinates can be written as:

vabc − uabc = Ldiabc

dt+R.iabc (2.1)

Where uabc is the voltage at point of common coupling, vabc is the converter’s AC voltage

and iabc is the current flow. To follow the vector control approach, the representation in

abc coordinates can be transformed to dq coordinate based equation 2.2:

10

Converter

Vdc

v u

LR i

Figure 2.2: Schematic of a typical VSC station [78].

[

vdvq

]

−

[

ud

uq

]

=

[

R −ωL

ωL R

] [

id

iq

]

+

[

L 0

0 L

]

d

dt

[

id

iq

]

(2.2)

In these coordinates, both id and iq currents can be controlled separately by inner

controller which leads to the decoupled control of active and reactive power in the converter.

The complete block diagram of an inner current controller is presented in Fig.2.3. The inner

controller follows the reference set-points that are ordered by outer controller (i.e. i∗d and

i∗q).

*

di

_

_

1

R sL+

*

qv

du

*

dv

*

qi

_di

qi

PI

1

R sL+PI

Lw

Lw

qu

Figure 2.3: Block diagram of the inner current control [78].

The outer controller can control reactive power (or AC voltage) on the AC side and active

11

power (or DC voltage) on the DC side. The outer controller provides the inner controller

with the reference current values in dq coordinate (i.e. i∗d and i∗q). This controller is slower

than the inner controller. Usually the PI controller is used in industrial applications due

to its simplicity. However, to control the power flow in the HVDC grid, different strategies

can be used to set the DC voltage and power of the outer control loop on the DC side.

These strategies basically define the primary reaction of converters to any disturbance in

the HVDC grid. In the next section, some of those strategies are presented.

2.1.2 Primary control strategies for DC voltage and power

In HVDC grids, real-time mismatch of power injection can be compensated by the DC

voltage controlling converter(s). Similar to frequency in AC grid, the DC voltage deviation

is a local indication of a power mismatch in an HVDC grid. The control of DC voltage

at the grid level and the corresponding power flow control can be influenced by different

voltage control schemes at the converter level. There are several control schemes suggested

in literature to maintain power balance within the HVDC grid. In this section, the two

most popular and most referenced control schemes are presented, namely the voltage-margin

method (VMM) and voltage droop control.

2.1.2.1 Voltage margin method

In the voltage margin method, one converter is set to control the DC voltage in the HVDC

grid (i.e. like a DC slack-bus) and the remaining converters are set to active power control

[54]. The Vdc − P characteristics of this converter is presented in Fig. 2.4. When the

converter operates on the B-C-D line, it is on constant DC voltage mode, i.e. Vdc = Vrefdc .

In A-B mode the converter acts as an inverter and is in constant active power level, i.e.

Pac = Pminac . In D-E mode the converter acts a rectifier and Pac = Pmax

ac .

12

Inverter Rectifier

P

Vdc

Vdcref,1

PP max,1min,1

A

B C D

E

0

Vdcref,2

Pmax,2Pmin,2

F

G H I

J

Voltage Margin

Figure 2.4: Voltage-power characteristics of two converters in voltage-margin method

In the HVDC grid with voltage-margin method, once the converter hits the limits and

goes from constant DC voltage mode to constant power mode (i.e. B-C-D line to A-B

or D-E line), another converter needs to work in DC voltage control mode to maintain

the voltage levels in the entire grid. This can be achieved by having the other converters

operating at different reference values (see Fig.2.4), thus establishing a voltage margin

between the converters. As shown in Fig.2.4, when converter 1 reaches the minimum active

power level, V min,1dc , the DC voltage increases to V

ref,2dc , and converter 2 is now responsible

to maintain the DC voltage level.

2.1.2.2 DC voltage droop method

In this method, some converters are assigned to change the injected active power based

on the local deviation of DC voltage in order to provide the active power balance in the

HVDC grid [20]. Different types of voltage-power characteristics have been considered in

literature. As an example, in DC voltage droop without dead-band the operating point of

the power injection changes as a linear function of the local voltage change (see Fig. 2.5.a).

To avoid changes in power injection set-point due to small changes in the DC voltage, droop

with dead-band is used (see Fig. 2.5.b). When it comes to DC voltage droop concept, it

is communication free during the operation. The droop setting and/or new set-points can

be calculated by higher level control functions such as optimal power flow at EMS/SCADA

every few hundred seconds or minutes and then be sent to converters.

13

0

Vref.

Vdc

P

PrefPmin Pmax 0

.

Vdc

P

PrefPmin Pmax

.Vlow

Vhigh

(a) (b)

Figure 2.5: Voltage-power characteristic of converters in DC voltage droop method,

(b)without dead-band, (b) with dead-band.

2.2 Short circuit capacity estimation for HVDC appli-

cation

The estimation of the AC grid parameters (i.e. SCC) connected to an HVDC system can

be used to adjust the converter control parameters or to select the converter’s operational

control mode. Fast and accurate estimation of the grid parameters can also lead to more

autonomy in terms of control adjustments. There are different passive and active methods

available to estimate the SCC. Extended Kalman Filter (EKF) as a passive method provides

various benefits such as cancellation of different types of noise, no forced grid disturbance

and also proven experience in other applications. However, it requires high computational

power due to the calculation of the Jacobian matrix. Furthermore, the convergence depends

on the choice of the initial values [62]. When the predict and update functions of a state

space model are non-linear in nature, EKF algorithm performs poorly. The modified version

of EKF, i.e. Unscented Kalman Filter (UKF) is developed to deal with the non-linearity of

the models. UKF also shares the same advantages and disadvantages of EKF in terms of

implementation [67].

Considering certain requirements for the practical implementation of the algorithm such

as computation, operational complication and accuracy, Recursive Least Square (RLS) can

be an alternative choice for the SCC estimation. The RLS algorithm forms a regression

problem using algebraic complex equations with an objective of minimizing the error be-

tween the estimated parameters and the calculated parameters based on the measurements.

This regression problem is solved recursively in a discrete domain. This algorithm benefits

from the low computational efforts and the possibility of non-complex and non-matrix trans-

formation for easy hardware implementation. However, it requires two operating points to

converge. The brief description of the algorithm is provided here.

14

2.2.1 Recursive least square algorithm

The recursive least square algorithm has been mainly used to estimate short circuit capacity

in distribution networks with high penetration of renewable generations [29]. This section

on RLS algorithm has been provided from the author’s previous conference paper that is

not included as the contribution in the thesis [14].

V

ZI

E

Figure 2.6: Equivalent circuit [14].

Fig. 2.6 shows an AC equivalent circuit for the connected AC grid that can be seen

from the point of common coupling (PCC) at each converter. This equivalent model is

considered for the RLS algorithm. In steady state conditions,

V = IZ + E (2.3)

where V and I are the voltage and current at PCC, Z is the equivalent grid impedance

and E is the equivalent grid voltage. All the parameters are expressed as complex numbers

in the dq reference frame. Note that V and I are measured parameters, and E and Z are

estimated parameters. Considering n different operating points,

V1 = I1Z1 + E1

V2 = I2Z2 + E2

. . .

Vn = InZn + En.

Assume that the grid is stationary during the measurements, then Z1 = Z2 = . . . = Zn

and E1 = E2 = . . . = En. Hence, two constant parameters Z and E0 can be assumed

for all such operating points. Considering the stationary situation, equation 2.4 can be

expressed in matrix form as

Y = A.X (2.4)

where,

Y =

V1

V2

...

Vn

; A =

I1 1

I2 1...

...

In 1

; X =

(

Z

E0

)

; (2.5)

15

To estimate the parameters Z and E0 from equation 2.4, a minimum of two operating points

in a stationary grid is required. Equation 2.4 is a linear regression problem expressed in

the complex plane. Consider e as the error between the estimated voltage and the actual

measurement:

e = A.X − Y (2.6)

where X is the estimated grid parameter. The best-fit for the estimated parameter vector

X can be found by minimising a positive function of error magnitude (J):

J = |e|2= e

T ∗ e = (AX − Y )T ∗ (AX − Y ) (2.7)

∂J

∂t= 0 =⇒ X = (AT ∗A)−1(AT ∗ Y ) (2.8)

Equation 2.8 provides an optimal off-line estimation after all measurements are available.

In order to estimate the parameters, the problem is needed to be transformed to a recursive

problem using the following equations:

Pk = ([AT ]k[A]k)

−1; (2.9)

B =

(

I∗k+1

1

)

(2.10)

Wk+1 = P

kB(I +B

T ∗ P kB)−1 = P

k+1B (2.11)

where, P k is cross correlation matrix and I is the identity matrix. B and W k+1 are

the intermediate variables and I∗k+1is the measured currents at iteration k + 1. As shown

in [29], considering intermediate variables, the problem can be iteratively solved by:

(

Zk+1

E0

k+1

)

=

(

Zk

E0

k

)

+Wk+1.ek+1 (2.12)

where, Z is the estimated grid equivalent impedance, E0 is the estimated grid equivalent

voltage and ek+1 is the error between the measured PCC voltage and the calculated PCC

voltage based on estimates. Equation 2.12 provides a straight forward method to recursively

estimate the equivalent impedance and voltage. Fig. 2.7 shows the final flowchart for this

recursive algorithm. The RLS algorithm can thus estimate the equivalent grid voltage

and equivalent grid impedance at the PCC. In addition, thanks to the small dimension of

the regression problem, the matrix inversion during computation of W k+1 reduces to a

complex number inversion which implies low computational requirement of the algorithm.

As presented in Fig. 2.7, the initial values P0 and X0 are required and can be found from

an off-line identification. Furthermore, an evaluation can be incorporated to find whether

the grid parameters have changed by time. This evaluation is based on equation 2.13 where

n is the moving window of the terms.

τk =1

2k

k+n∑

i=k−n

‖Vi − IiZi − Ei‖2

|Ii|(2.13)

The evaluation parameter at time instant k is given as τk. The iteration variable i com-

putes the summation in a loop throughout the window n. The threshold for the evaluated

16

B = (Ik+1 1)

Wk+1 = PkB (I + BT*PkB)-1

k++Off-line

Identification

Pk+1 = (I –Wk+1BT*) Pk

Convergence?

End

No

(Zk+1 E0k+1)T=(Zk E0

k)T+ Wk+1.ek+1^ ^ ^^

Figure 2.7: Final Estimation Algorithm

parameter τk is τγ . If τk is greater than the threshold (i.e τk > τγ), the grid parameters

have changed to such extend that a new SCC estimation must be carried out. Therefore,

the estimated grid equivalent voltage and impedance are no longer correct and Ck must be

reset to its initial value.

2.3 Tools for simulation of cyber-physical systems

To study the distributed control applications needed for future HVDC grids, simulation

that considers all aspects of the involving systems is necessary. This section describes the

modelling of distributed control for HVDC grids from a Cyber-Physical System (CPS) per-

spective. The term cyber-physical system is recently being used to refer to systems in

which the computational entities including control/communication units (cyber) as well as

physical processes are strongly coupled [68]. The integration of ICT system with traditional

power system is more noticeable nowadays by implementation of new control and monitor-

ing applications based on new technologies like Phasor Measurements Unit (PMU). The

detailed study of cyber-physical systems to e.g. understand the interdependency of these

sub-components and their impact on the overall quality of the system requires a multi-

domain simulation tool or a co-simulation platform. Present simulators are limited by the

fact that none of them can be used to carry out the detailed modelling of all the domain.

In addition, developing such a simulator requires a huge investment. On the other hand,

co-simulation is a solution that runs each part of the model in its relevant simulator and

then coordinates all these simulators in terms of data exchange and time synchronization.

Several studies have been carried out to implement the co-simulation through different

approaches. The Electric Power and Communication Synchronizing Simulator (EPOCHS)

has been designed based on federated simulation with multiple discrete-event and continu-

ous time simulators [63]. In this approach, a mediator software is used as an interface which

17

enables the simulators to exchange data periodically and synchronize them. It integrates

the PSLF as an electromechanical transient simulator, the PSCAD/EMTDC as an electro-

magnetic transient simulator, and the NS2 as a communication network simulators in such

a way that their internal time clocks progress simultaneously. In this approach, events are

stored in an event queue and are executed only at the next synchronization point thus accu-

mulating errors. On the other hand, Global Event-driven Co-Simulation (GECO) consists

of a global event queue to store the interleaved simulation events from power system and

communication simulators and process them in chronological order [72]. Events generated

can be executed with minimal delay.

In comparison to this method, The Discrete Event System Specification (DEVS) method

does not federate, instead, it requires continuous time simulations such as power system

should be transferred into discrete events by using different state event detection mecha-

nisms such as zero crossing [79]. This is a tedious work and one of the DEVS drawbacks.

MOSAIK as an open-source discrete event simulator have been proposed in which the ex-

ecution of simulators is performed in an event-based manner [86]. The APIs of simulators

are language agnostic and this reduces the learning overhead with new tools. All the men-

tioned approaches do not aim for hardware-in-the-loop tests. On the other hand, the Virtual

Grid Integration Laboratory (VirGIL) is proposed very recently as a modular co-simulation

platform which can be extended to have a Hardware-in-the-loop implementation [87]. It

is based on industrial grade standard, Functional Mockup Interface (FMI), that brings the

modularity feature. Any FMI compliant simulator can be integrated to work with VirGIL.

A master algorithm coordinates data exchange. The drawback is that only static analysis

for power systems can be carried out as developing FMI interfaces for dynamic analysis is

complicated.

2.3.1 PSMIX real-time co-simulation platform for HVDC grid stud-

ies

The Power System Management and Information eXchange (PSMIX) is a real-time co-

simulation platform that make it possible to study the design, testing and implementation

of real-time applications for control and operation of a power system. PSMIX real-time

platform mirrors the characteristics of physical process, ICT system and the interfacing

devices or systems as close as possible to the real life situation. PSMIX is a general real-

time architecture that can be re-configured for different studies from wide-area monitoring

and control of power transmission system to distribution grid control scenarios [10, 12].

This platform comprises of real-time power system simulator, real-time communication

network simulator, supervisory control applications, and software-based or real interfacing

control/measurement component. The key factors that can impact the overall performance

of any such real-time platform is accuracy of the software-based replications of the real

devices and the implementation of industrial automation protocols. For HVDC studies, the

PSMIX platform is set-up to simulate the HVDC grid and its supporting control and com-

munication system (see Fig. 2.8). The detailed information of the components is described

as follows.

18

SoftPMU

OPNETDC control Unit

Supervisory

ControllerLocal Controller

AC Grid

SoftPMU

DC control Unit

Data

Concentrator

State

Estimater

Voltage ...Control

Mode

SelectionOptimal

Power

Injection

Power Simulator

T1

VSC 1

T2

VSC 2

VSC 5

VSC4

Grid

Grid

VSC 6

VSC7

T3

VSC 3

Grid

VSC 1

VSC 2

VSC 5

VSC4

VSC 6

VSC7

VSC 3

T1

TT22

Griddd

Gridd

TTT333

Gridd

Communication Simulator

DC grid

Analog I/Os

DMU

Figure 2.8: Real-time co-simulation platform [78].

2.3.1.1 Real time power simulator

The eMEGAsim is a commercial real time simulator which combines electrical circuit

solvers, SimPowerSystem, distributed processing software and hardware for high speed real-

time simulations of power system for both steady state and transient analysis [94]. This

simulator can be customized to meet I/O requirements enabling the Hardware-in-the-Loop

(HIL) simulations. The simulator used in PSMIX is able to simulate the Cigre DC grid

test system with the time-step of 50 µs using the average model of converter (not switching

model).

2.3.1.2 Measurement units

The HIL feature of the real time power simulator enables the simulated power system

to interact to outside word via different means such as analog I/Os. Since the HVDC

controller is able to communicate with specific analog I/Os, a special DC measurement unit

(DMU) can be developed inside the OPAL-RT simulator to send/receive the DC voltage

and active power measurement with specific accuracy to/from analog I/Os. The I/Os have

16-bits resolution. The I/Os use EtherCAT protocol to communicate [42]. For the AC

side, a Software-based Phasor Measurement Unit (SoftPMU) or a real PMU device can be

considered (detailed information has been presented in [12]).

2.3.1.3 HVDC industrial controller

In this type of real-time platform, an industrial HVDC controller can be used. For the

purpose of this thesis, the controller that runs Windows embedded integrated with INtime

reliable real time operating systems (RTOS) has been considered. This controller communi-

cates with the analog device via EtherCAT protocol. The data exchange among the HVDC

control substations are carried out via Ethernet or UDP protocol.

19

2.3.1.4 OPNET communication network simulator

In the real-time platform, any communication network software-based simulator with the

ability of connection to real-world and real-time capability can be considered to model

and study the communication aspect of the understudied system. However, this can be

extended to real-time commercial network emulators as well. In this thesis, OPNET has

been considered as a communication system modeler to provide comprehensive develop-

ment environment for modeling and studying communication networks. OPNET solution

provides the System-in-the-loop (SITL) module enabling the connection of the simulation

model with live network hardware [12].

2.3.1.5 Supervisory applications

The application component of the PSMIX platform consists of openPDC as the phasor data

concentrator and the KTH PowerIT as the application hosting platform that connects to

openPDC to receive the synchronized measurements [27]. Besides, it is able to receive the

HVDC grid information i.e. DC voltage and active power from the converters using indus-

trial defined Ethernet protocol. Several applications have been implemented in PowerIT

platform, such as average frequency visualization and electro mechanical mode estimation

for AC grid, and monitoring and control application for HVDC grid. Note that for dis-

tributed schemes, there is no need of a centralized application to be run on the PowerIT

platform.

20

Chapter 3

Related Works

This chapter presents the related works that are directly relevant to studied control func-

tions. Therefore, the related work is divided into three categories; first the works that cover

optimal power flow in HVDC or hybrid AC/DC grids. Second, the work related to HVDC

converter control mode selection algorithm including its parameter’s estimation algorithm.

Third, the works related to network topology processing architecture and design.

3.1 Coordination of power injection

Normally the primary control action to small disturbances in the HVDC grid takes place

by communication-free schemes such as DC voltage droop control [20, 57]. However, the

secondary control is needed to optimally tune the DC power and voltage set-points, and

prepare the system for new disturbances considering future demand, wind uncertainty and

also market signals. In most works, the formulation of secondary control of set-points

has been studied in the context of Optimal Power Flow (OPF) with various objective

functions, boundary of the system, constraints and optimization techniques [25, 26, 50, 64].

Several researches have presented the combined AC/DC optimal power flow solutions for

single AC grid with embedded MTDC networks or an HVDC grid connecting different

AC areas. In [9], a comprehensive tool for solving OPF in hybrid AC/DC systems for

grid integration of large offshore or onshore wind power plants has been presented. The

power flow of AC and DC side have been linked through an equation on the power balance

in the converter (PAC = PDC − Plossconverter). Similarly, the work in [25] tackled the

combined AC/DC problem but with extra constraint on DC current and voltage. These

constraints come from considering two different VSC control strategies, i.e. constant DC

voltage control (master-slave control) and DC voltage droop control in the formulation.

Since in the practical implementation of HVDC grid, there might be various types of non-

linear DC voltage droops, [100] proposed a generalized OPF algorithm that considers those

control modes also. In these works, the objective function usually covers the transmission

and converter losses in both AC and DC systems as well as generations costs. Since most of

the works on combined AC/DC OPF (with similar formulation as [9,25]) do not guarantee

to obtain the global optimal solution, [17, 19] have aimed to determine the global optimal

21

solution by using convex relaxation techniques and transform the optimization problem to

a semidefinite program or second-order cone programming.

Apart from the type of formulation, some works have separated the combined AC/DC

calculations to different parts to optimize the computational efficiency. For example in [18],

the system is divided into two sub-systems: 1) the HVDC grid and the AC area connected

to slack converter of HVDC grid 2) the rest of AC area. The problems in these two

subsystems are solved in an iterative way. The decomposition of AC/DC power flow has

also been analyzed in [73] based on network decomposition concept. In this approach, an

AC grid OPF calculator solves the problem for the AC area, the DC grid OPF calculator

solves for the HVDC grid and a master coordinator handles the data exchange between

different calculators.

Given the separation of calculation in some works, they still focused on the centralized

combined AC/DC power flow with one single entity or operator. However, a valuable study

has been carried out in [64] to introduce a distributed power flow formulation for multi-

area AC systems with embedded HVDC systems. In this work, three operational schemes

are considered: 1) whole power system is operated by one single operator, 2) in addition

to the AC operators, the HVDC grid is operated by a separate operator 3) there is no

HVDC grid operator and areas are separated based on their geographical borders and each

area includes the corresponding AC and part of HVDC system. The proposed distributed

method for scheme 2 and 3 is compared with the central approach in scheme 1 where there

is a single operator to dispatch the set-points. The converter is modeled as a generating unit

with positive or negative value based on the direction of power flow. The shared variable

between areas are power exchanges on both HVDC links and AC lines that connect two

areas. Therefore the objective function includes three parts; 1) the cost of AC generating

units, 2) power losses and 3) a penalty for the shared variable. In the problem formulation,

multi-area power systems are synchronous, that means the areas are interconnected by

AC and DC links or grids. Besides, a simplified linear model is used for AC power flows

based on deviation of angles and for DC power flow based on deviations of DC voltages.

This work does not include security-constrained optimal power flow. The idea of having

individual operator for HVDC grid has also been presented in [50] and a simple OPF has

been introduced for an HVDC grid with the objective of minimization the DC losses.

All previous mentioned related works formulate the OPF in a way that the generation

(e.g. wind power production both in AC or DC grid) and the demand from the AC grid

are assumed to be known. This means they emphasize on a deterministic set-up and do not

take into account any uncertainty in their formulation. However in [102], uncertainty in

the generation and security constraints have been considered in a linear AC/DC power flow

formulation where the AC voltage angles and the DC voltage magnitudes are eliminated.

As can be seen, these related works have approached the power flow problem in HVDC

grids or hybrid AC/DC grids from several angles such as the boundary of the system,

efficiency of algorithm and type of formulation (see table 3.1). In relation to these related

works, paper 1 has first proposed a separate power injection coordination for HVDC grids

with large wind offshore. This coordination is a bridge between slower overall AC/DC

power flow calculator and fast primary control actions at the converter level. It aims

at finding an optimal operation point within the HVDC grid especially during large wind

disturbances while it tries to follow the converters’ schedules set by the connecting AC-TSOs

22

Table 3.1: The references and their OPF problem areas

operational

strategies

system

configurationtype of

formulation

other

consideration

SingleAC/HVDCoperator

multipleAC/HVDCoperators

separateHVDC and

ACoperators

embeddedAC/HVDC

asynchronousAC/HVDC central distributed uncertainty

N-1criterion

Baradar [19] * – – * – * – – –

Aragues [9] * – – * * * – – –

Baradar [18] * – * * * * – – –

Liu [73] * – – * – – * – –

Gonzales [50] – – * – * * – – –

Wang [100] * – – * – * – –

Wiget [102] * – – * – * – * *

Iggland [65] * * * * – * * – –

every 15 − minutes. Paper 1 also has aimed to show through a sensitivity analysis that

HVDC grids benefits from this type of coordination when they are subjected to variation of

wind production. Furthermore, the requirement for the updating interval of power injection

coordination have been studied which can be an interesting information for all related works

to check the efficiency and speed of their algorithm. Based on this time requirement, as the

main contribution paper 1 has proposed a fully distributed algorithm based on the auxiliary

problem principle to divide this coordination among the connecting AC TSOs. Paper 4 has

completed the idea in paper 1 by considering the wind uncertainty and N-1 criterion in the

formulation of distributed coordination. However, paper 4 has improved the efficiency of

the distributed coordination by creating convex subproblems and using the ADMM method

to solve them.

3.2 DC slack bus selection and related practical imple-

mentation

The concept of slack bus in AC power system problems is a mathematical requirement to

absorb all the resulting uncertainty of the solution but it has no physical interpretation

to any generator bus. Research in [33] addressed the mathematical challenges introduced

by the slack bus in AC load flow solutions with uncertain nodal powers. However, in the

context of HVDC grid, DC slack bus is a physical HVDC converter controller needed to

be set in term of its controlling modes. A comprehensive study on classification of DC

voltage nodes and possible HVDC grid control strategies in [98] suggests that the future

HVDC grid comprises of different complex voltage controllers including a DC slack bus. The

effects of DC voltage control strategy on the dynamic behavior of bus voltages have been

studied through simulations for a multi-terminal HVDC following a system disturbance

in [49] and after a converter loss in [48]. Two different DC voltage control methods are

simulated in these papers: voltage margin method and DC voltage droop method. However

in [35], the effect of droop setting on the post-outage steady-state have been mathematically

assessed and some suggestions have been made on assigning the DC voltage droop values.

Furthermore, as a general note, [35] has expressed that sensitive voltage controllers (i.e.

the one connected to weak grid) may have disadvantages for system stability and should be

investigated.

23

Given a slack converter in an HVDC grid control architecture, requirements for select-

ing the specific converter to take the role of a DC slack bus may be considered in static

planning problem. The off-line selection of DC slack bus in the planning phase is a matter

of optimization which takes time and requires careful consideration. However, the online

selection of slack bus is needed and can be significant after loss of the slack converter in

an HVDC grid. To this point, it seems to be a gap in the literature on how to deter-

mine the validity of converters to act as the DC slack bus in the HVDC grid and what

parameters to be considered for this online validity check. Therefore, paper 2 presented a

real-time evaluation framework to choose the appropriate DC slack bus in an HVDC grid.

This online quantitative evaluation examines the post-contingency capacity margin of the

converter and the strength of all AC grids connected to converters (characterized by grid’s

short circuit capacity). Paper 3 studied different aspects of implementing recursive least

square algorithm as a potential candidate for the estimation of SCC for the DC slack bus

evaluation.

There are passive and active methods to estimate grid parameters to calculate the

short circuit capacity. Passive methods use disturbances present in a grid to estimate the

grid parameters, whereas in active methods, the grid is disturbed along with its normal

operation. There are also quasi-passive methods which usually requires the change in the

operating point of the power converter [29, 95]. The comparison of the Kalman filter and

the Recursive Least Squares (RLS) algorithms as two examples of passive and quasi-passive

methods for the estimation of the grid impedance has been studied in [90]. The estimation

of the equivalent grid impedance seen from a power converter connected to the real electric

distribution network using Extended Kalman Filter (EKF) is addressed in [61]. The study

has focused on the typical low voltage distributed power generation networks where the grid

impedance is (in most cases) inductive-resistive. Other related work on EKF in [62] have

presented various benefits of the algorithm such as cancellation of different types of noises,

no forced grid disturbance and also a proven experience in other applications. However, this

paper as well has shown the challenges regarding EKF such as need of high computational

power due to the calculation of the Jacobian matrix and also the dependency of convergence

on the choice of the initial values. In addition, when the predict and update functions of

a state space model are non-linear in nature, EKF algorithm performs poorly. In [67],

a modified version of EKF, i.e. Unscented Kalman Filter (UKF) has been developed to

deal with the non-linearity of the models. However, UKF algorithm also shared the same

advantages and disadvantages of EKF in terms of implementation.

Given either of these methods to estimate the grid parameters, accuracy and sensitivity

of the estimation algorithm to the system operational changes and computational require-

ments are also of importance, especially when it comes to real-life implementation of the

algorithms. Therefore, Paper 3 tries to fill the gap in the literature on studying the op-

erational and practical challenges in the implementation of the SCC estimation algorithm

focused on the RLS method. Since RLS algorithm needs at least two operating points for

estimation, as the major contribution, paper 3 proposes to intelligently utilize a naturally

occurring disturbance in HVDC grids to use as the second operating point to obtain a

fairly correct estimate of SCC. This paper also has studied the practical aspect of the es-

timation algorithm regarding the selection of operating points. Furthermore, the original

algorithm has been also reformulated and simplified to make it non-complex without the

24

use of matrices, in order to be able to implement on an industrial real-time controller.

3.3 Network topology processor

In the context of AC grids, different approaches such as the application of the intelligent

methods or the integration of Phasor Measurement Units (PMU) for the purpose of topology

processing have been investigated [104, 105]. A centralized topology processor for a PMU-

only state estimator is presented in [40]. An alternative method for topology processing

using an expert system for detection of device malfunction at the distribution substations

based on local state estimation is proposed and evaluated in [80]. A method for the iden-

tification of breaker statuses in power system state estimation is presented in [69]. In this

method, additional pseudo measurements for each circuit breaker are also introduced to

facilitate the identification process. This identification happens at a central level. Sim-

ilarly, [70] presents a model for breaker status identification and power system topology

estimation based on auto-associative neural networks.

Most of the previous works have emphasized processing the topology at central level

(one-stage processor) which limits applicability of the topology processor for carrying out

local and distributed control functions. Considering more flexibility in power system au-

tomation and communication due to the recent evolutions in computerized substation au-

tomation system and moving toward more distributed solutions for power system operation,

the study of different methods than conventional one-stage solution for processing the grid

topology is necessary. Therefore, in paper [104] a two-level architecture has been proposed

for a linear state estimator. This paper has shown that the two-level processing removes

the bad data and topology errors, which are major problems today, at the substation level

and therefore, resulting in a more accurate two-level state estimator. This work has de-

scribed a new substation level calculations, called the substation level state estimator or

zero impedance state estimator, that pre-filters all the real-time data at that substation.

Furthermore, the paper focused on the mathematical algorithms for each level of state es-

timation. However, it just assumed that the substation level network topology is available

for the estimation application without any further comment on how to calculate or process

it.

Looking at possible operational strategies in HVDC grids, two requirements must be

taken into consideration for design of the topology processor. First, the topology processor

needs to function when coordination of converters is shared between connecting TSOs

(i.e. distributed strategy). Second, if one TSO coordinates the grid centrally, a back-

up distributed coordination plan can be considered to take over the responsibility in case of

central failure. This back-up distributed processor must be able to fulfill time requirements

of grid coordination functions such as power injection without any additional reformation

of data exchange or processing. Therefore, paper 5 studied a two-stage network topology

processor that uses either centralized or distributed method to identify grid connectivity.

The proposed two-stage topology processor examines the DC substation topology in the first

stage based on an automated graph-based algorithm. Then, the second stage determines

the grid level connectivity including islands detection. Furthermore, as main contribution

to related works, it formulated and presented a distributed methods for grid connectivity

25

identification in the second stage. The performance of the distributed grid identification

has been tested to check if it can be used for distributed control application.

26

Chapter 4

Research Results and

Discussions

This chapter presents the outcome and contribution of the research by summarizing the

results of related papers for each individual objective.

4.1 Investigation of advanced control applications for

HVDC grids operation

The first aim of the thesis is to investigate control applications that are needed for HVDC

grids supervisory management. Furthermore, the requirements that should be considered

to design of such control applications have been presented. In literature, different types of

control functions for the HVDC supervisory controller have already been studied such as

secondary control of DC voltage, AC/DC combined power flow and state estimation [21,

34,45]. Taking into account the literature and the possible needs for control and operation

of HVDC grids, three types of control functions have been determined in this thesis to be

studied namely, coordination of power injection, selection of DC voltage controlling station

(DC slack bus) and finally network topology processor. Among these three functions,

selection of DC voltage controlling station is a new function that is missing in the literature.

Therefore has been considered in this thesis. The idea of having coordination of power

injection and network topology processor has been presented in the literature and then

studied from different perspective. However in this thesis these functions are presented and

formulated based on different considerations which are provided in following sections.

4.1.1 Coordination of power injection

The first control application is ”coordination of power injection” in HVDC grid. The

difference between the proposed control function and the combined AC/DC optimal power

flow in literature such as [25,26,41] is in their objective functions, the boundary of the system

and the rate of calculation (how often the control applications needs to be carried out). In

paper 1, the coordination of power injection in HVDC grids has been first formulated

27

as a non-linear constrained optimization problem for the centralized architecture. This

optimization problem aims to tune the power injection set-point of converters to follow the

schedules set by the connecting AC TSOs while minimizing the operational costs of HVDC

grid. For this, the divergence of the converter power injection from the connected TSOs’

desired power schedule has been considered as a penalty to be minimized. This means

working as close as possible to the desired value, P desi . The desired value is the set-point

normally assigned each 15-minutes by the AC TSOs. The same penalty has been considered

for the DC voltage. This term helps converters to avoid large deviations in DC voltage and

approach the limit after each disturbance. In addition, the power loss of the DC lines and

converter loss have been also considered in this formulation (see equation 4.1).

f =

Nconv∑

i=1

Cp,i(Pdci − P des

i )2 + Cu,i(Udci − Udes

i )2

+

Nconv∑

i=1

Ploss,i +

Nline∑

i=1

Pi

(4.1)

where,

• P desi is the desired DC power injection at the converter i defined by the corresponding

TSO in each balancing period (e.g. 15 minutes),

• P dci is the optimal power injection set-point to be calculated for converter i,

• Cp,i is the cost for the deviation from each TSO’s power injection plan at converter i,

• Cu,i the penalty for the DC voltage deviation at converter i,

• Ploss,i is the loss at converter i

• Nconv is the number of converters,

• Nline is the number of lines,

• Pi is power flow on the line i.

This formulation also includes equality constraints for DC power at each injection or

consumption node (converter node) as well as inequality constraints on power and DC

voltage limitations of converters and DC lines. Since the boundary of the problem involves

just the HVDC grid and its controllable converters, this control application is capable to be

run much faster that overall AC/DC combined power flow which considers the entire power

system. Therefore this control function enables the converters to update the set-points

more frequently in a system with fluctuating power and voltage profiles such as HVDC

grids with large wind farms. Paper 1 studied the benefit of faster updating frequency i.e.

within the 15-minutes intervals using this HVDC grid control application. For this purpose,

the optimal power injection set-points have been calculated in different intervals from every

1-min interval to every 15-min interval and all the variation of power in the HVDC grid

between the updating of set-points has been assumed to be taken by DC voltage droop

control. As shown in Fig.4.1, for the specific system and wind data, the 1-minute-based

power coordination can decrease the total cost function (equation 4.1) by around 16%

compared to having the DC voltage control taking over the wind power mismatches.

28

0 3 6 9 12 15

2.7

2.8

2.9

3

3.1x 10

-3

Updating period (minute)

Av

erag

e co

st (

p.u

.)

Coordination method

Droop method

Figure 4.1: Average of cost function versus updating frequency for 100 simulations

4.1.2 DC slack bus selection

Selection of DC slack bus is always a matter of optimization which takes time and requires

careful consideration. This selection can be significant after loss of the slack converter in

an HVDC grid or in an islanding scenarios. Considering this requirement, it is important

to find algorithms that examine the validity of the stations to take the role of DC slack bus.

Paper 2 proposed that this control application uses a real-time quantitative evaluation of

HVDC converters’ in an HVDC grid to rank and select the suitable DC slack converter.

Various parameters could be considered for quantitative evaluation of a converter. Since the

final aim of the thesis is to focus on the distributed solution, the parameters that required

local information and measurements are of interest. On-line power margin of the converter

and the strength of connected AC grid (i.e. short circuit capacity) are two examples of

such parameters. Paper 2 highlighted the importance of these two parameters by a set

of scenarios for the CIGRE B4 DC grid model available in [99]. For power margin, it is

assumed that the current slack converter does not have enough margin to compensate the

mismatch. This phenomena drops the DC voltage profile in the entire HVDC grid. The

decrease in DC voltage corresponds to an increase in the current to maintain the same

amount of power exchange. Higher current profile leads to increase in transmission losses.

For short circuit capacity, different SCC have been assumed for the AC grid connected to

the DC slack bus. The result shows that the lower SCC value of the AC grid connected to

the slack converter, the higher is the disturbance in the AC voltage.

Given the importance of these parameters to evaluate the converter’s capability, a sys-

tematic ranking function based of these parameters have been proposed in paper 2 and

examined through a scenario. In this scenario, converter Cb-A1 has high SCC and rela-

tively high ranking value and therefore works as slack bus. This is the case until t = 20s

when an AC line trips. This trip reduces the SCC of the AC grid connected to Cb-A1.

After the change in SCC of one AC grid, the evaluation of converters based on new SCC

and power margin is carried out at t = 25s that is detailed in Table 4.1. The ranking table

shows that Cb-B1 is the best choice to act as a DC slack bus in the new situation while

the reduction in SCC have caused the ranking of Cb-A1 to go to 3rd place. Two cases have

been considered: Case 1) Cb-A1 still continues to work as DC slack bus, although it is low

29

Figure 4.2: CIGRE DC grid test model.

ranked and CASE 2) Cb-B1 is selected as DC slack bus based on ranking table. To check

these two cases, at t = 30s, the active power set-point of Cb-B2 is increased. In Case 1

where Cb-A1 still continues to work as slack bus, the DC voltage profile in entire HVDC

grid is reduced since the power margin in Cb-A1 is low and it hits the limit. On the other

hand, as shown in Fig. 4.3, the DC voltage profile is improved in the second case when

Cb-B1 is chosen as the slack converter based on the ranking table.

Table 4.1: Slack Converter Ranking at t = 25 s

HVDC

Station

Estimated SCC

(GVA)

Power Margin

(MW)

Evaluation

Value

Slack converter

Ranking

Cb-A1 13.15 316.32 0.0693 3

Cb-B1 14.89 878.16 0.2182 1

Cb-B2 9.06 699.36 0.1057 2

Regarding the effect on the AC voltage, since SCC is low in Cb-A1, if it continues to

work as a slack bus (i.e. case 1), the change in the operating point causes the larger distur-

bance in the connected AC grid compared to the case 2. Consequently, as shown in 4.4.a

the AC voltage response of Cb-A1 is better in case 1 compared to case 2. The same applies

for AC voltage response at Cb-B1 as shown in Fig. 4.4.b. The results in paper 2 show that

the usage of this control application (i.e. selection of the slack station) improve AC system

response and DC voltage drops during disturbances.

30

time (s)29.5 30 30.5 31 31.5 32

DC

Vol

tage

(pu

)

0.97

0.975

0.98

0.985

0.99

0.995

1

1.005

Cb-B1 (Case1)Cb-B2 (Case1)Cb-B1 (Case2)Cb-B2 (Case2)

time (s)29.5 30 30.5 31 31.5 32

DC

Vol

tage

(pu

)

1

1,003

1,006

1,009

1,012

1,015

1,018

Cb-A1 (Case1)Cb-C2 (Case1)Cb-D1 (Case1)Cb-A1 (Case 2)Cb-C2 (Case 2)Cb-D1 (Case 2)

Figure 4.3: Comparison between Cb-A1 and Cb-B1 as choice of DC slack converter

time (s)29 30 31 32 33 34

Vol

tage

(pu

)

0.997

0.998

0.999

1

1.001 Cb-A1 (Case 1)Cb-A1 (Case 2)

time (s)29 30 31 32 33 34

Vol

tage

(pu

)

0.9985

0.999

0.9995

1

1.0005

1.001

1.0015

Cb-B1 (Case 2)Cb-B1 (Case 1)

(a) (b)

Figure 4.4: Effect of slack converter selection on (a) AC voltage at Cb-A1, red line when it

works as slack and blue line when it works on constant power and Cb-B1 is slack, (b) AC

voltage at Cb-B1, red line shows when Cb-B1 works as slack and blue line shows when it

works on constant power and Cb-A1 is the slack bus.

4.1.2.1 Network topology processor

The network processor uses the real-time circuit breaker status within the substation to

determine the system level topology. The role of the topology processor becomes more

critical in case of islanding where the converters in each island should be recognized as soon

as possible and in some cases, be tuned in terms of their control modes (e.g. selecting DC

slack bus). Most of the previous works carried out for AC grid (e.g. [40, 104, 105]), have

emphasized processing the topology at central level (one-stage processor) which limits ap-

plicability of the topology processor for carrying out local and distributed control functions.

The proposed two-stage topology processor in paper 5 examines the DC substation topology

in the first stage based on an automated graph-based algorithm. Then, the second stage

determines the grid level connectivity including islanding detection. The results show that

the two-stage topology processor benefits from: 1) a faster process compared to one-stage

method since each calculation at the substation is simpler and it is distributed in parallel,

2) the second stage (central level) is less complicated and needs smaller storage capacity as

it has to manage data with reduced dimensions, and finally 3) the two-stage architecture

makes it possible to use the information processed at the first stage for both distributed

and central coordination. Therefore, the distributed coordination can simply be used as a

31

VSC 1

VSC 2

VSC 4

VSC 3

VSC5

~=

~=

~=

~=

~=

Wind Farm

AC1

AC1

AC2

AC3

VSC 4

~=AC2

VSC 1

VSC 2

~=

~=

AC1

AC1

Border Bus

{Pij , Uij }TSO B

TSO A

Figure 4.5: APP for multi-terminal HVDC

back-up plan for the central coordination without any further modification.

4.2 Distributed solution for control applications in an

HVDC grid

The previous section emphasized the need of advanced control applications as part of an

HVDC grid supervisory management system. Those control functions have been first de-

signed and tested for the centralized architecture. However, the main contribution of the

thesis is to investigate the feasibility of distributed solutions for the determined control ap-

plications, so they can be decomposed and implemented for distributed operational struc-

ture presented in Fig.1.3.b. In addition, this solution could be also used in the central

structure (Fig.1.3.a) as a complementary back-up if the grid central coordinator fails.

4.2.1 Distributed coordination of power injection

In paper 1, the first control application (i.e. coordination of power injection) has been

decomposed for the distributed implementation using Auxiliary Problem Principle (APP).

To do so, as shown in Fig.4.5, a fictional border bus has been defined in the middle of the

DC line that connects the converter i in area A to converter j in area B. This border bus

added state variables of DC power (P dcij ) and DC voltage (Udc

ij ) to the optimization problem.

These two variables are shared variables between areas A and B. Given the sub-problem in

area A, it consists of the local variables of DC voltage (Udci ) and DC power injection (P dc

i )

for the local converter and the shared variables of P dcij,A and Udc

ij,A for border bus in the

area. The new objective function and the related constraints have been formulated based

on these variables in each subproblem and an iterative method has been used to solve the

optimization problem. The detailed formulation is presented in paper 1.

32

The result of paper 1 shows that the convergence rate of distributed solution based

on APP for coordination power injection in a 5-terminal HVDC grid (shown in Fig.4.5) is

around 240 iterations. Taking into account the average delay of few hundred milliseconds

for the data exchange in each iteration, update frequencies of once per minute that has

been suggested in previous section might give problems, but slower update rate can be

managed using this distributed algorithm. In paper 4, the boundary of the problem has been

extended by considering the N-1 security constraints and uncertainty of wind production

in the optimization problem. Besides, a linear approximation has been used to deal with

non-linearity of constraints in each subproblem. In addition, to improve the performance of

distributed coordination, a modified version of ADMM has been proposed instead of APP

to solve the problem.

To evaluate the performance of proposed distributed coordination in paper 4, three

different values have been set for the convergence tolerance. Then, the result of these three

distributed cases have been compared with the centralized solution. The converters’ set-

points and the corresponding objective function calculated in the centralized method have

been presented in table 4.2. For distributed implementation, at first, a relaxed tolerance of

ǫ = 10−5 has been set. The result shows that the algorithm converges in 14 iterations. But

the deviation from the optimal solution (objective function) is about 2.9%. By tightening

the tolerance to ǫ = 10−8, as expected, the deviation from optimal objective function

reduces to 0.06% which is a very acceptable error level. However, this accuracy comes with

the expense of increase in iterations. Given the maximum communication delay of 50 ms

between HVDC stations [109], it takes around 5 seconds for the algorithm to exchange

data to converge. However, the computation time of solving subproblems in each iteration

should also be added to the total time. This computation time can vary based on the power

of local solvers. In this simulation, on average, it takes around 0.23 second to solve the

local problem. Therefore, for the third case with the tight tolerance of ǫ = 10−8, the whole

distributed coordination process take less that 30 seconds which satisfies the assumption of

1-minute requirement for updating the set-points.

Table 4.2: Comparison of ADMM-based distributed approach and centralized solution for

security constrained optimal power injection

node central distributed

ǫ = 10−5ǫ = 10−6

ǫ = 10−8

Pinj Pinj error Pinj error Pinj error

1 -0.4805 -0.4812 0.0008 -0.4807 0.0002 -0.4805 4.1E-07

2 -0.5633 -0.5647 0.0014 -0.5636 0.0003 -0.5633 8.7E-06

3 0.7170 0.7091 0.0079 0.7161 0.0010 0.7170 2.0E-05

4 -0.3994 -0.3959 0.0036 -0.3988 0.0006 -0.3994 -3.3E-05

5 0.7560 0.7597 0.0037 0.7565 0.0005 0.7560 -1.0E-05

iter. – 14 34 83

cost 1.2878 1.3254 1.2895 1.2886

33

4.2.2 Distributed parameter estimation for DC slack bus selection

As mentioned earlier, a quantitative ranking framework has been proposed to evaluate

the credibility of converters to take over the DC slack bus responsibility. To be able to

select the slack bus in a distributed manner, each converter should be able to calculate

its quantitative evaluation value based on local information. Then, with a light consensus

protocols, they can agree on the selected converter. To obtain the quantitative ranking

value, each converter needs to calculate two parameters (i.e. power margin and short circuit

capacity). Given easy access to the first parameter i.e. power margin, the performance of the

distributed implementation of slack bus selection in terms of accuracy, time and complexity

would be very dependent on the estimation of SCC. Therefore, in paper 3, it is shown

that the recursive least square algorithm can be very efficiently used for online estimation

of SCC of AC grids from the PCC. As mentioned, to estimate two different parameters

of the grid i.e. equivalent voltage and impedance for short circuit capacity calculation,

RLS algorithm needs minimum two operating points to form the algebraic equations. Both

active and passive methods can be used to introduce the second operating point to the

estimation algorithm. Natural load changes in AC grid and change in DC side due to droop

characteristics of the controller are two examples of passive methods that are presented in

paper 3. A scenario has been considered to show the performance of algorithm in case of

introducing the second operating point with passive methods. To do so, a point-to-point

HVDC system shown in Fig.4.6 is considered.

VSC 2VSC 1~

=~

=AC

System 2

AC

System 1

AC System 1

ABC

Line 1 : R= 5 ohm, L=150 mH – closed always

Line 2 : R= 5 ohm, L=150 mH – opened @ 9s

R= 1 ohm

L=15 mH

Scc= 11.09 GVA

Figure 4.6: AC and DC system model.

The SCC of the AC grid connected to Cb-A1 is changed from 12.82 GVA to 10.04 GVA

by tripping a transmission line at t = 20s. This change in SCC at t = 20s second is detected

by the estimation algorithm and it resets the algorithm. The change in the operating point

of Cb-A1 due to the change in active power schedule at Cb-B1 provides the second operating

point for the algorithm at t = 30s. Fig. 4.7 shows that the estimation algorithm in station

34

Cb-A1 converges with adequate accuracy, immediately after the introduction of the second

operating point by passive droop response. In this case, the latency in parameter estimation

is dependent on when this natural response (either load change or droop response) occurs.

10 15 20 25 30 35 4023456789

10111213141516

time (s)

SC

C (

GV

A)

Cb-A1

Figure 4.7: Estimated short circuit capacity using the droop response as the second oper-

ating point.

However, these passive methods might not be available in all circumstances. Therefore,

to have faster estimation, three active approaches have been introduced in paper 2 and 3

to assure the availability of the second operating point for the estimation algorithm. In

active approach, the active or reactive power set-points of the HVDC station or both of

them can be changed manually by a small percent. Fig. 4.8 compares the estimated SCC

for different types of changes in the set-points. The comparison is limited to a 3% change

in the set-points. Note that paper 3 elaborates the choice of 3%. Fig. 4.8 shows that all

three cases have similar ranges of accuracy. In all cases, the algorithm converges to an

acceptable value in a few seconds. Taking into account that minimal disturbances to the

grid are desired, the change in reactive power set-point can be recommended.

In addition to the choice of operating points, the performance of estimation algorithm

is dependent on other parameters such as execution time, filter time constant, noise in

measurements, X/R ratio of the AC grid and etc. In paper 3, a comprehensive sensitivity

analysis has also been carried out to assess the impact of these parameters and operational

conditions on the performance of estimation algorithm. However, two parameters of exe-

cution time as well as filter time constants have been shown to have critical impact on the

performance and convergence of RLS algorithm.

To show the effect of algorithm’s execution time, it is assumes that the power system

model is running with a sample time of 7.407µs, whereas the controllers are operated with a

sample time of 74.07µs and the RLS estimation algorithm can be run with the same sample

time as the controllers (1x) or slower. The results of the estimated short circuit capacity for

the different execution times are shown in Fig.4.9. The speed of ”2x” that corresponds to

a sample time of 148.14µs (2 ∗ 74.07µs) has the best accuracy. The result also suggests that

the sample time of 222.21µs (”3x”) has enough precision for the practical implementation

35

4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 91.83

1.84

1.85

1.86

1.87

1.88

1.89

time (s)

SC

C (

GV

A)

Active power change (P)

Reactive power change (Q)

Active and reactive change (PQ)

Figure 4.8: SCC estimation for 3% change in P, Q and PQ.

which also does not enforce a high speed requirement.

10 11 12 13 14 15 16 17 18 19 200.99

1

1.01

1.02

1.03

1.04

1.05

1.06

time (s)

SC

C (

GV

A)

1x

2x

3x

5x

10x

Figure 4.9: Comparison of the different execution times.

Regarding the time constants of measurements’ filter, result in Fig.4.10 shows that the

algorithm diverges for both small and large time constants. The algorithm is accurate for

the range of 20ms to 50ms. There is no major difference in the results of estimated SCC

within this range of time constants.

To summarize, comprehensive studies in paper 2 and 3 show that the RLS algorithm

can provide a fair estimation of SCC in range of a few seconds, given the algorithm is tuned

properly in terms of its operational parameters and conditions.

4.2.3 Distributed identification of grid topology

One of the prerequisites for distributed coordination of converters is the knowledge of neigh-

boring stations’ connectivity. One motivation to choose the two-stage network processor

36

10 11 12 13 14 15 16 17 18 19 200.96

0.97

0.98

0.99

1

1.01

1.02

1.03

1.04

1.05

1.06

1.07

time (s)

SC

C (

GV

A)

5ms

30ms

50ms

100ms

200ms

Figure 4.10: SCC estimation for the different filter time constants.

is the possibility to extend the approach for distributed implementation. Since the first

level of substation connectivity processing takes place locally, the neighboring substations

have the opportunity to ask for the processed information (e.g. in the form of Sm ma-

trix presented in paper 5) from each substation for further grid level topology processing.

Then, the local adjacency matrix A can be updated using Sm matrices received from all

neighbors. This connectivity to other neighboring stations is the information needed for the

distributed power injection coordination during normal operation presented in paper 1 and

4. However, for islanding events, additional information must be provided to the control ap-

plication regarding which converters are within the island. A distributed algorithm to detect

the connectivity during islanding scenarios has been developed in paper 5. This algorithm

checks the connectivity of each node to a specific node called source node by forming linear

equations. Then, it uses successive-over-relaxation method to solve the linear problem.

The performance of the distributed two-stage architecture is compared with the conven-

tional centralized architecture and central two-stage architecture in terms of computational

efforts. For conventional centralized architecture, it is assumed that all breaker status are

sent to a SCADA system and there the topology processor is detected. For central two-

stage architecture, it is assumed that the first-stage is carried the similar to the proposed

method, but the second-stage is performed centrally in one entity instead of our distributed

proposal. This comparison has been carried out for an islanding scenario and a Core-i5 ma-

chine with 2.30 GHz CPU and 8.00 GB memory has been used. To have a fair estimation of

the process time, this scenario has been run 100 times and the results have been averaged.

To compare different architectures, the worst case (i.e. longest substation processing

time or slowest substation) has been chosen for the substation topology detection. In the

two-stage architecture, for the 5-terminal grid model in paper 4, the longest local processing

is done by substation 4 with 9 breaker. This design time of 42.26 s and operation time

of 0.2845 s are presented in this table. For the distributed coordination, the islanding

detection happens at the substation level. As shown in Table 4.3, each islanding detection

algorithm takes 0.00057 s at the local level. As mentioned before, to realize the connectivity

to other nodes, a parallel detection algorithm should be run based on number of nodes in the

37

grid. Given the 5-terminal HVDC grid model, each node needs to run 5 sets of algorithm

in parallel, 4 sets for checking the connectivity to other nodes and one set of algorithm to

act as source node to others. Therefore, the total time of the islanding detection in the

distributed coordination is calculated by multiplying the number of iterations for each set of

algorithm to converge, number of parallel algorithm and execution time of each algorithm

(i.e. total time = 5 × 5 × 0.00057 s ). This results in the execution time of 0.01425 s for

only islanding detection. However, in the central two-stage processor, the grid topology is

processed in 0.0063 s and in the conventional centralized architecture, the grid topology

detection takes 151.57 s and 0.4055 s, respectively. There is no substation processing time

since all the breaker statuses are sent to the central processor.

Table 4.3: Comparison of different topology processing architectures.

Two-stage(central)

Conventionalcentralized

Two-stage(distributed)

Substationtopology

Design 42.26 s – 42.26 s

Operation 0.2845 s – 0.2845 s

Islanding – – 0.00057 s

Gridtopology

Design – 151.57 s –

Operation 0.0063 s 0.4055 s 0.01425 s

Totalwithout delay 0.290 s 0.405 s 0.2987 s

with delay 0.340 s 0.4555 s 0.5485 s

The last two rows of Table 4.3 present the total processing time in all three architectures

without considering communication delay and also with an identical scenario but consid-

ering communication delay. The total grid topology processing time is in the same range

for the two-stage processor with either central or distributed coordination. Considering a

communication delay of 50 ms, the distributed coordination requires longer processing time

than the two other methods.

4.3 Benefit of real-time co-simulation platform

The distributed control of HVDC grid involves the interaction of physical power systems

and distributed decision makers, which are basically computational/communication units.

Therefore, such an interaction can be studied from a cyber-physical system perspective.

To do so, as mentioned before, it requires development of a multi-domain co-simulation

platform. In this thesis, the key considerations to develop the co-simulation platform have

been: easy configuration and operating system compatibility, ability to interconnect with

external third-party program/system, hardware-in-the-loop (HIL) capability, time resolu-

tion of simulation steps, time synchronization, level of detail of modelling, interoperability

to import models and scalability.

The real-time co-simulation platform developed in this thesis is a general real-time

architecture that can be rearranged for different studies from distribution grids control

scenarios [10] to wide area control of transmission grid [12]. This platform includes real-

38

SoftPMU

OPNETDC control Unit

Supervisory

ControllerLocal Controller

AC Grid

SoftPMU

DC control Unit

Data

Concentrator

State

Estimater

Voltage ...Control

Mode

SelectionOptimal

Power

Injection

Power Simulator

T1

VSC 1

T2

VSC 2

VSC 5

VSC4

Grid

Grid

VSC 6

VSC7

T3

VSC 3

Grid

VSC 1

VSC 2

VSC 5

VSC4

VSC 6

VSC7

VSC 3

T1

TT22

Griddd

Gridd

TTT333

Gridd

Communication Simulator

DC grid

Analog I/Os

DMU

Figure 4.11: Real-time co-simulation platform configured for HVDC grid control studies.

time power system simulator, real-time communication network simulator, applications,

and software-based or real interfacing devices/measurement. The main key factor in the

development of such real-time platform is accuracy and performance of the software-based

models of the real devices and the implementation of industrial automation protocols such

as synchronized phasor measurement units. For the scope of this thesis, the platform is

configured to support the modelling of HVDC grid and its corresponding HVDC converter

controllers and communication system (see Fig.4.11). The detailed information of the com-

ponents has been described in Paper 6.

time (s)4 6 8 10 12 14 16 18 20

SCC

(G

VA

)

0.5

1

1.5

2

2.5

3

Real-time Non-real-time

X: 14.01Y: 1.012

X: 6.75Y: 1.872

X: 6.75Y: 1.815

Figure 4.12: Performance of short circuit capacity estimation algorithm in non real-time

and real-time platforms.

The results in Paper 3 and 6 show that design, testing and implementation of distributed

control applications for HVDC grids would significantly benefit from developing a proper co-

simulation platform. As an example, in paper 6 the comparison of pure simulation versus

real-time co-simulation implementations of a distributed power flow has been presented.

The result shows that some dynamic of the real controller and errors in A/D conversion

cannot be captured in the pure simulation. Similarly in paper 3, the performance of the

39

SCC estimation algorithm (to be used for DC slack selection) has been evaluated using a

non-real-time environment (pure simulation) as well as the real-time co-simulation platform.

The implementation in real-time co-simulation leads the estimation algorithm inside real

industrial controller hardware faces the noises of measurement devices as well as the small

delay in communication. The comparison of short circuit capacity calculated in simulation

and co-simulation platforms shows that the estimation algorithm is able to provide accurate

results (error less than 2%) even in presence of noise and harmonics (see Fig.4.12).

40

Chapter 5

Conclusion and Future Work

In this thesis, three functions are investigated to be deployed in HVDC supervisory system:

coordination of power injection set-points in the presence of large wind farms, DC slack bus

selection and two-stage network topology identification. Considering the coordination of

power injection in HVDC supervisory brings the possibility of updating the operating point

of HVDC converters more frequently (every 1-min) when the system is subjected to variation

in wind production. Since the new operating points calculated by this function are forced

to work as close as possible to the set-points set by overall AC/DC management system,

the generation reserves in each AC area can be improved. To coordinate the power flow

inside the HVDC grids, the control mode of converters especially the DC voltage controlling

converter (DC slack bus) is of importance. The choice of DC slack converter can impact the

voltage respond of both AC and DC system after a disturbance in the system. Furthermore,

this selection procedure can be significant after loss of the slack converter in an HVDC

grid. In this thesis, a close-to-real-time evaluation framework is developed to choose the

appropriate DC slack bus in an HVDC grid. This online quantitative evaluation examines

the strength of all AC grids connected to the converters (characterized by grid’s short

circuit capacity) and post-contingency capacity margins of the converter. Both parameters

are local information.

When it comes to implementation of supervisory control functions, this was coupled

tightly to the arrangement of system operators i.e. if there is an individual HVDC operator

(centralized) or tasks are shared among different entities e.g. AC operators (distributed).

This arrangement of operators brings two considerations for the design of HVDC supervisory

functions. First, the control functions need to work properly when the coordination of

converters is shared between connecting TSOs (i.e. distributed structure). Second, if

one operator coordinates the grid centrally, a back-up distributed coordination plan is

determined to take over the responsibility in case of central failure.

Based on type of control functions, different algorithms can be considered to decompose

and distribute the function. For coordination of power injection, the updating rate of 1-

minute is considered as a requirement to evaluate the distributed solution. In this thesis,

two different distributed algorithms based on alternating direction method of multipliers

and auxiliary problem principle are used to solve coordination of power injection which

satisfy the required updating rates. However, in DC slack bus selection, for distributed

41

implementation, the choice of parameters for online quantitative ranking is important.

These parameters should be calculated based on local measurements if distributed decision

making is desired. To estimate the short circuit capacity as one of the required parameters

for selection of DC slack bus, the result of this thesis determines that the recursive least

square algorithm can be very efficiently used. Besides, it is possible to intelligently use a

naturally occurring droop response in HVDC grids as a local measurement/information for

this estimation algorithm instead of active methods to introduce measurement points.

Regarding the identification of grid topology for control purposes, considering a two-

stage network topology processor provides a faster process compared to one-stage method

since each calculation at the substation is simpler and it is distributed in parallel. Besides,

the second stage (central level) is less complicated and needs smaller storage capacity as

it has to manage data with reduced dimensions. As main advantage, the two-stage ar-

chitecture enables to use the information processed at the first stage for both distributed

and central coordination. Therefore, the distributed coordination can simply be used as a

back-up plan for the central coordination without any further modification.

Future work

There is more room to expand this research topic from different perspectives. Looking

at OBJ 1 (i.e. to investigate advanced control functions), although there have been some

studies on different control functions for HVDC supervisory control level, but still compre-

hensive studies on other functions can be carried out such as HVDC grid state estimation,

ancillary support and droop setting design.

Regarding the implementation of control functions, this thesis focused on the solution

for distributed operational structure. However, in future it might be the case that one of

the connected AC areas’ operators take over the responsibility of HVDC grid. This scenario

is also valid if an embedded HVDC grid in one AC area is of interest. This can impact the

boundary and objective function of the power injection coordination problem.

One area that might need further studies is the selection of DC slack bus. First, this

selection can be generalized for any type of converter that contribute in power sharing e.g.

DC voltage droop controllers. In that case the problem will be the adaptive tuning of control

parameters such as droop coefficient. In addition, other parameters can be considered for

the evaluation of slack bus such as generation reserve of AC areas. On the SCC estimation,

the equivalent grid estimation can be more challenging when the AC grid includes large

amount of converter-based generating units such as PV and wind plants.

42

Bibliography

[1] A. S. Abdel-Khalik, A. E. B. Abu-Elanien, A. A. Elserougi, S. Ahmed, and A. M.

Massoud. A droop control design for multiterminal hvdc of offshore wind farms

with three-wire bipolar transmission lines. IEEE Transactions on Power Systems,

31(2):1546–1556, March 2016.

[2] A. A. Abdel-khalik, Ayman S, Massoud, Ahmed M, Elserougi and S. Ahmed. Opti-

mum Power Transmission-Based Droop Control Design for Multi-Terminal HVDC of

Offshore Wind Farms. IEEE Transactions on Power Systems, 28(3):3401–3409, 2013.

[3] N. Ahmed, S. Norrga, H. P. Nee, A. Haider, D. V. Hertem, L. Zhang, and L. Harnefors.

Hvdc supergrids with modular multilevel converters; the power transmission backbone

of the future. In Systems, Signals and Devices (SSD), 2012 9th International Multi-

Conference on, pages 1–7, March 2012.

[4] V. Akhmatov, M. Callavik, C. M. Franck, S. E. Rye, T. Ahndorf, M. K. Bucher,

H. Mller, F. Schettler, and R. Wiget. Technical guidelines and prestandardization

work for first hvdc grids. IEEE Transactions on Power Delivery, 29(1):327–335, Feb

2014.

[5] M. Andreasson, D. V. Dimarogonas, H. Sandberg, and K. H. Johansson. Distributed

controllers for multi-terminal hvdc transmission systems. IEEE Transactions on Con-

trol of Network Systems, PP(99):1–1, 2016.

[6] M. Andreasson, R. Wiget, D. V. Dimarogonas, K. H. Johansson, and G. Andersson.

Distributed primary frequency control through multi-terminal hvdc transmission sys-

tems. In 2015 American Control Conference (ACC), pages 5029–5034, July 2015.

[7] M. Andreasson, R. Wiget, D. V. Dimarogonas, K. H. Johansson, and G. Andersson.

Distributed secondary frequency control through mtdc transmission systems. In 2015

54th IEEE Conference on Decision and Control (CDC), pages 2627–2634, Dec 2015.

[8] B. Andrew, L. Joachim, and N. Anna. Asia Pacific Super Grid Solar electricity

generation, storage and distribution. Green, 2(4):189–202, Jan. 2012.

[9] M. Aragues-Penalba, A. Egea Alvarez, S. Galceran Arellano, and O. Gomis-Bellmunt.

Optimal power flow tool for mixed high-voltage alternating current and high-voltage

direct current systems for grid integration of large wind power plants. IET Renewable

Power Generation, 9:876–881(5), November 2015.

43

[10] M. Armendariz, M. Chenine, L. Nordstrm, and A. Al-Hammouri. A co-simulation

platform for medium/low voltage monitoring and control applications. In Innovative

Smart Grid Technologies Conference (ISGT), 2014 IEEE PES, pages 1–5, Feb 2014.

[11] D. Babazadeh, M. Chenine, and L. Nordstrom. Survey on the factors required in

design of communication architecture for future dc grids. volume 2, pages 58–63,

2013.

[12] D. Babazadeh, M. Chenine, K. Zhu, L. Nordstrm, and A. Al-Hammouri. A plat-

form for wide area monitoring and control system ict analysis and development. In

PowerTech (POWERTECH), 2013 IEEE Grenoble, pages 1–7, June 2013.

[13] D. Babazadeh, F. Freizadeh, and L. Nordstrm. Distributed ancillary service support

for independent ac-areas through hvdc grid. In PowerTech, 2015 IEEE Eindhoven,

pages 1–6, June 2015.

[14] D. Babazadeh, A. Muthukrishnan, L. Nordstrom, P. Mitra, and T. Larsson. Short

circuit capacity estimation for hvdc control application. In 2016 Power Systems

Computation Conference (PSCC), pages 1–7, June 2016.

[15] D. Babazadeh, A. Tonti, M. Armendariz, A. Borghetti, L. Nordstrom, and C. A.

Nucci. Two-stage network processor for an independent hvdc grid supervisory control.

In Power Energy Society General Meeting, 2016 IEEE, pages 1–5, 2016.

[16] D. Babazadeh, D. Van Hertem, M. Rabbat, and L. Nordstrom. Coordination of power

injection in hvdc grids with multi-tsos and large wind penetration. In AC and DC

Power Transmission, 11th IET International Conference on, pages 1–7, Feb 2015.

[17] S. Bahrami, F. Therrien, V. W. S. Wong, and J. Jatskevich. Semidefinite relaxation of

optimal power flow for ac dc grids. IEEE Transactions on Power Systems, PP(99):1–

16, 2016.

[18] M. Baradar and M. Ghandhari. A multi-option unified power flow approach for

hybrid ac/dc grids incorporating multi-terminal vsc-hvdc. IEEE Transactions on

Power Systems, 28(3):2376–2383, Aug 2013.

[19] M. Baradar, M. Hesamzadeh, and M. Ghandhari. Second-order cone programming

for optimal power flow in vsc-type ac-dc grids. Power Systems, IEEE Transactions

on, 28(4):4282–4291, Nov 2013.

[20] J. Beerten and R. Belmans. Analysis of power sharing and voltage deviations in

droop-controlled dc grids. Power Systems, IEEE Transactions on, 28(4):4588–4597,

Nov 2013.

[21] J. Beerten and D. V. Hertem. Analysis of power redispatch schemes for hvdc grid

secondary voltage control. In 2015 IEEE Power Energy Society General Meeting,

pages 1–5, July 2015.

[22] J. Beerten, D. V. Hertem, and R. Belmans. VSC MTDC Systems with a Distributed

DC Voltage Control - A Power Flow Approach. In IEEE Trondheim PowerTech,

pages 1–6, 2011.

44

[23] C. Belknap. The american clean energy initiative -2050 - design earth synergy.

http://www.christopherbelknap.com/?page_id=73. Accessed: 2016-10-08.

[24] B. Berggren, K. Linden, and R. Majumder. Dc grid control through the pilot voltage

droop concept- methodology for establishing droop constants. Power Systems, IEEE

Transactions on, 30(5):2312–2320, Sept 2015.

[25] J. Cao, W. Du, H. Wang, and S. Bu. Minimization of transmission loss in

meshed ac/dc grids with vsc-mtdc networks. Power Systems, IEEE Transactions

on, 28(3):3047–3055, Aug 2013.

[26] J. Cao, W. Du, and H. F. Wang. An improved corrective security constrained opf

for meshed ac/dc grids with multi-terminal vsc-hvdc. IEEE Transactions on Power

Systems, 31(1):485–495, Jan 2016.

[27] M. Chenine, L. Vanfretti, S. Bengtsson, and L. Nordstrom. Implementation of an

experimental wide-area monitoring platform for development of synchronized phasor

measurement applications. In 2011 IEEE Power and Energy Society General Meeting,

pages 1–8, July 2011.

[28] S. Chondrogiannis and M. P. Blanco. Market integration scheme of a multi-terminal

hvdc grid in the north seas. IEEE Transactions on Power Systems, 31(3):2415–2422,

May 2016.

[29] S. Cobreces, E. Bueno, D. Pizarro, F. Rodriguez, and F. Huerta. Grid impedance

monitoring system for distributed power generation electronic interfaces. Instrumen-

tation and Measurement, IEEE Transactions on, 58(9):3112–3121, Sept 2009.

[30] E. Commission. The 2020 climate and energy package.

http://ec.europa.eu/clima/policies/strategies/2020/index_en.htm. Ac-

cessed: 2016-08-17.

[31] J. Dai, Y. Phulpin, a. Sarlette, and D. Ernst. Coordinated primary frequency control

among non-synchronous systems connected by a multi-terminal high-voltage direct

current grid. IET Generation, Transmission & Distribution, 6(2):99, 2012.

[32] M. O. Dessouky. The environmental impact of large scale solar energy projects on the

mena deserts: Best practices for the desertec initiative. In EUROCON, 2013 IEEE,

pages 784–788, July 2013.

[33] A. Dimitrovski and K. Tomsovic. Slack bus treatment in load flow solutions with un-

certain nodal powers. International Journal of Electrical Power and Energy Systems,

27(910):614 – 619, 2005. Special Issue on Probabilistic Methods Applied to Power

Systems 8th International Conference on Probabilistic Methods Applied to Power

Systems.

[34] V. Donde, X. Feng, I. Segerqvist, and M. Callavik. Distributed state estimation of

hybrid ac/hvdc grids by network decomposition. IEEE Transactions on Smart Grid,

7(2):974–981, March 2016.

45

[35] J. Dragon, L. Werner, and J. Hanson. Effects of dc voltage droop characteristics on

contingency behaviour of ac/dc systems. In Power Engineering Conference (UPEC),

2014 49th International Universities, pages 1–6, Sept 2014.

[36] F. B. Effah, A. J. Watson, C. Ji, E. Amankwah, C. M. Johnson, C. Davidson, and

J. Clare. Hybrid hvdc circuit breaker with self-powered gate drives. IET Power

Electronics, 9(2):228–236, 2016.

[37] A. Egea-Alvarez, F. Bianchi, A. Junyent-Ferre, G. Gross, and O. Gomis-Bellmunt.

Voltage control of multiterminal VSC-HVDC transmission systems for offshore wind

power plants: Design and implementation in a scaled platform. IEEE Transactions

on Industrial Electronics, 60(6):2381–2391, 2013.

[38] A. Egea-alvarez, S. Fekriasl, F. Hassan, and O. Gomis-Bellmunt. Advanced Vector

Control for Voltage Source Converters Connected to Weak Grids. IEEE Transactions

on Power Systems, 30(6):3072–3081, 2015.

[39] P. Fairley. Germany jump-starts the supergrid. Spectrum, IEEE, 50(5):36–41, May

2013.

[40] M. Farrokhabadi and L. Vanfretti. An efficient automated topology processor for

state estimation of power transmission networks. Electric Power Systems Research,

106:188 – 202, 2014.

[41] W. Feng, L. A. Tuan, L. B. Tjernberg, A. Mannikoff, and A. Bergman. A New Ap-

proach for Benefit Evaluation of Multiterminal VSC HVDC Using A Proposed Mixed

AC/DC Optimal Power Flow. IEEE Transactions on Power Delivery, 29(1):432–443,

2014.

[42] M. H. Fidai, D. Babazadeh, J. Hanning, T. Larsson, and L. Nordstrom. Real-time

implementation of optimal power flow calculator for hvdc grids. In 2015 Cigre Sym-

posium in Lund Sweden, 2015.

[43] M. H. Fidai, D. Babazadeh, A. Muthukrishnan, and L. Nordstrm. Supervisory control

for vsc-hvdc grid interconnecting ac systems. In 2015 IEEE Power Energy Society

General Meeting, pages 1–5, July 2015.

[44] N. Flourentzou, V. G. Agelidis, and G. D. Demetriades. Vsc-based hvdc power trans-

mission systems: An overview. IEEE Transactions on Power Electronics, 24(3):592–

602, March 2009.

[45] C. Gavriluta, I. Candela, A. Luna, A. Gomez-Exposito, and P. Rodriguez. Hierarchi-

cal control of hv-mtdc systems with droop-based primary and opf-based secondary.

IEEE Transactions on Smart Grid, 6(3):1502–1510, May 2015.

[46] C. Gavriluta, I. Candela, A. Luna, A. Gomez-Exposito, and P. Rodriguez. Hierarchi-

cal control of hv-mtdc systems with droop-based primary and opf-based secondary.

Smart Grid, IEEE Transactions on, 6(3):1502–1510, May 2015.

46

[47] O. Gomis-Bellmunt, J. Liang, J. Ekanayake, R. King, and N. Jenkins. Topologies of

multiterminal HVDC-VSC transmission for large offshore wind farms. Electric Power

Systems Research, 81(2):271–281, Feb. 2011.

[48] F. Gonzalez-Longatt and J. Roldan. Effects of dc voltage control strategy on voltage

response on multi-terminal hvdc following loss of a converter station. In 2013 IEEE

Power Energy Society General Meeting, pages 1–5, July 2013.

[49] F. Gonzalez-Longatt and J. M. Roldan. Effects of dc voltage control strategies of

voltage response on multi-terminal hvdc following a disturbance. In 2012 47th Inter-

national Universities Power Engineering Conference (UPEC), pages 1–6, Sept 2012.

[50] F. M. Gonzalez-Longatt. Optimal power flow in multi-terminal hvdc networks for

dc-system operator: Constant current operation. In Power Engineering Conference

(UPEC), 2015 50th International Universities, pages 1–6, Sept 2015.

[51] C. Guo, W. Liu, C. Zhao, and R. Iravani. A frequency-based synchronization approach

for the vsc-hvdc station connected to a weak ac grid. IEEE Transactions on Power

Delivery, PP(99):1–1, 2016.

[52] R. H. Renner and D. Van Hertem. Ancillary services in electric power systems with

hvdc grids. IET Generation, Transmission & Distribution, 9:1179–1185(6), August

2015.

[53] C. Hahn, A. Geu, and M. Luther. Modeling and control design of hybrid - lcc and vsc

based - hvdc systems. In 2016 IEEE/PES Transmission and Distribution Conference

and Exposition (T D), pages 1–6, May 2016.

[54] T. Haileselassie, M. Molinas, and T. Undeland. Multi-terminal vsc-hvdc system for

integration of offshore wind farms and green electrification of platforms in north sea.

In Nordic Workshop on Power and Industrial Electronics, pages 1–5, June 2008.

[55] T. M. Haileselassie, A. G. Endegnanew, and K. Uhlen. Secondary control in multi-

terminal vsc-hvdc transmission system. In 2015 IEEE Power Energy Society General

Meeting, pages 1–5, July 2015.

[56] T. M. Haileselassie, A. G. Endegnanew, and K. Uhlen. Secondary control in multi-

terminal vsc-hvdc transmission system. In 2015 IEEE Power Energy Society General

Meeting, pages 1–5, July 2015.

[57] T. M. Haileselassie and K. Uhlen. Impact of DC Line Voltage Drops on Power Flow

of MTDC Using Droop Control. IEEE Transactions on Power Systems, 27(3):1441–

1449, Aug. 2012.

[58] T. M. Haileselassie and K. Uhlen. Impact of DC Line Voltage Drops on Power Flow

of MTDC Using Droop Control. IEEE Transactions on Power Systems, 27(3):1441–

1449, Aug. 2012.

[59] A. Hassanpoor, J. Hfner, and B. Jacobson. Technical assessment of load commutation

switch in hybrid hvdc breaker. IEEE Transactions on Power Electronics, 30(10):5393–

5400, Oct 2015.

47

[60] D. v. Hertem, O. Gomis-Bellmunt, and J. Liang. HVDC Grids : For Offshore and

Supergrid Future. IEEE Press Series on Power Engineering. Wiley-IEEE Press, 2016.

[61] N. Hoffmann and F. W. Fuchs. Estimation in Inductive Resistive Power Networks

Using Extended Kalman Filter. IEEE Transactions on Power Electronics, 29(2):631–

641, 2014.

[62] N. Hoffmann, F. W. Fuchs, and C.-a.-u. Kiel. Online Grid Impedance Estimation for

the Control of Grid Connected Converters in Inductive-Resistive Distributed Power-

Networks Using Extended Kalman-Filter. In IEEE Energy Conversion Congress and

Exposition (ECCE), pages 922–929, 2012.

[63] K. Hopkinson, X. Wang, R. Giovanini, J. Thorp, K. Birman, and D. Coury.

Epochs: a platform for agent-based electric power and communication simulation

built from commercial off-the-shelf components. IEEE Transactions on Power Sys-

tems, 21(2):548–558, May 2006.

[64] E. Iggland, R. Wiget, S. Chatzivasileiadis, and G. Anderson. Multi-area dc-opf for

hvac and hvdc grids. IEEE Transactions on Power Systems, 30(5):2450–2459, Sept

2015.

[65] E. Iggland, R. Wiget, S. Chatzivasileiadis, and G. Anderson. Multi-area dc-opf for

hvac and hvdc grids. IEEE Transactions on Power Systems, 30(5):2450–2459, Sept

2015.

[66] D. Jovcic and K. Ahmed. VSC HVDC Applications and Topologies, Performance and

Cost Comparison with LCC HVDC, pages 121–140. John Wiley & Sons, Ltd, 2015.

[67] S. J. Julier and J. K. Uhlmann. New extension of the kalman filter to nonlinear sys-

tems. In AeroSense’97, pages 182–193. International Society for Optics and Photonics,

1997.

[68] S. K. Khaitan and J. D. McCalley. Design techniques and applications of cyberphysical

systems: A survey. IEEE Systems Journal, 9(2):350–365, June 2015.

[69] G. Korres and P. Katsikas. Identification of circuit breaker statuses in wls state

estimator. Power Systems, IEEE Transactions on, 17(3):818–825, Aug 2002.

[70] J. Krstulovic, V. Miranda, A. Simoes Costa, and J. Pereira. Towards an auto-

associative topology state estimator. Power Systems, IEEE Transactions on,

28(3):3311–3318, Aug 2013.

[71] J. Liang, T. Jing, O. Gomis-Bellmunt, J. Ekanayake, and N. Jenkins. Operation and

control of multiterminal hvdc transmission for offshore wind farms. IEEE Transac-

tions on Power Delivery, 26(4):2596–2604, 2011.

[72] H. Lin, S. S. Veda, S. S. Shukla, L. Mili, and J. Thorp. Geco: Global event-driven co-

simulation framework for interconnected power system and communication network.

IEEE Transactions on Smart Grid, 3(3):1444–1456, Sept 2012.

48

[73] J. Liu, X. Feng, B. Berggren, I. Segerqvist, and M. Callavik. Distributed power flow

analysis for hybrid ac/hvdc grids by network decomposition. In 2014 IEEE PES

General Meeting — Conference Exposition, pages 1–5, July 2014.

[74] A. K. Marten, F. Sass, and D. Westermann. Mixed ac/dc opf using differential

evolution for global minima identification. In PowerTech, 2015 IEEE Eindhoven,

pages 1–6, June 2015.

[75] P. Mitra, J. Hanning, J. Khlstrm, T. Larsson, and J. Danielsson. Dc grid system be-

havior: A real-time case study. In 2015 IEEE Power Energy Society General Meeting,

pages 1–5, July 2015.

[76] S. Nanou, O. Tzortzopoulos, and S. Papathanassiou. Evaluation of dc voltage con-

trol strategies for multi-terminal hvdc grids comprising island systems with high res

penetration. In AC and DC Power Transmission, 11th IET International Conference

on, pages 1–7, Feb 2015.

[77] M. Nazari and M. Ghandhari. Application of multi-agent control to multi-terminal

hvdc systems. In Electrical Power Energy Conference (EPEC), 2013 IEEE, pages

1–6, Aug 2013.

[78] L. Nordstrom and D. Babazadeh. Cyber physical approach to hvdc grid control.

In Cyber Physical Systems Approach to Smart Electric Power Grid, pages 75–101.

Springer Berlin Heidelberg, Berlin, Heidelberg, 2015.

[79] J. Nutaro, P. T. Kuruganti, L. Miller, S. Mullen, and M. Shankar. Integrated hybrid-

simulation of electric power and communications systems. In Power Engineering

Society General Meeting, 2007. IEEE, pages 1–8, June 2007.

[80] S. Park, E. Lee, W. Yu, H. Lee, and J. Shin. State estimation for supervisory moni-

toring of substations. Smart Grid, IEEE Transactions on, 4(1):406–410, March 2013.

[81] Y. Phulpin and D. Ernst. Ancillary services and operation of multi-terminal hvdc

systems. In International Workshop on Transmission Networks for Offshore Wind

Power Plants, pages 1–6, October 2011.

[82] E. Prieto-araujo, F. D. Bianchi, and A. Junyent-ferre. Methodology for Droop Con-

trol Dynamic Analysis of Multiterminal VSC-HVDC Grids for Offshore Wind Farms.

IEEE Transactions on Power Delivery, 26(4):2476–2485, 2011.

[83] E. Prieto-Araujo, A. Egea-Alvarez, S. Fekriasl, and O. Gomis-Bellmunt. Dc volt-

age droop control design for multiterminal hvdc systems considering ac and dc grid

dynamics. IEEE Transactions on Power Delivery, 31(2):575–585, April 2016.

[84] A. Reidy and R. Watson. Comparison of vsc based hvdc and hvac interconnections

to a large offshore wind farm. In Power Engineering Society General Meeting, 2005.

IEEE, pages 1–8 Vol. 1, June 2005.

49

[85] S. Rodrigues, R. Pinto, P. Bauer, E. Wiggelinkhuizen, and J. Pierik. Optimal power

flow of vsc-based multi-terminal dc networks using genetic algorithm optimization. In

Energy Conversion Congress and Exposition (ECCE), 2012 IEEE, pages 1453–1460,

Sept 2012.

[86] S. Rohjans, S. Lehnhoff, S. Schtte, S. Scherfke, and S. Hussain. mosaik - a modular

platform for the evaluation of agent-based smart grid control. In IEEE PES ISGT

Europe 2013, pages 1–5, Oct 2013.

[87] S. Rotger-Griful, S. Chatzivasileiadis, R. H. Jacobsen, E. M. Stewart, J. M. Domingo,

and M. Wetter. Hardware-in-the-loop co-simulation of distribution grid for demand

response. In 2016 Power Systems Computation Conference (PSCC), pages 1–7, June

2016.

[88] K. Rouzbehi, A. Miranian, A. Luna, and P. Rodriguez. Optimized control of multi-

terminal dc grids using particle swarm optimization. In Power Electronics and Ap-

plications (EPE), 2013 15th European Conference on, pages 1–9, Sept 2013.

[89] K. Rouzbehi, A. Miranian, A. Luna, and P. Rodriguez. Dc voltage control and

power sharing in multiterminal dc grids based on optimal dc power flow and voltage-

droop strategy. Emerging and Selected Topics in Power Electronics, IEEE Journal

of, 2(4):1171–1180, Dec 2014.

[90] S. Sanchez and M. Molinas. Large Signal Stability Analysis at the Common Coupling

Point of a DC Microgrid : A Grid Impedance Estimation Approach Based on a

Recursive Method. IEEE Transactions on Energy Conversion, 30(1):122–131, 2015.

[91] V. Saplamidis, R. Wiget, and G. Andersson. Security constrained optimal power flow

for mixed ac and multi-terminal hvdc grids. In PowerTech, 2015 IEEE Eindhoven,

pages 1–6, June 2015.

[92] J. Sneath and A. D. Rajapakse. Fault detection and interruption in an earthed hvdc

grid using rocov and hybrid dc breakers. IEEE Transactions on Power Delivery,

31(3):973–981, June 2016.

[93] L. Tang and B. T. Ooi. Locating and isolating dc faults in multi-terminal dc systems.

IEEE Transactions on Power Delivery, 22(3):1877–1884, July 2007.

[94] O.-R. Technologies. emegasim. http://www.opal-rt.com/system-emegasim/. Ac-

cessed: 2016-12-11.

[95] S. S. Tsai and K. Wong. On-line Estimation of Thevenin Equivalent with Varying

System States. In IEEE Power and Energy Society General Meeting, pages 1 – 7,

2008.

[96] D. Van Hertem and M. Ghandhari. Multi-terminal VSC HVDC for the European

supergrid: Obstacles. Renewable and Sustainable Energy Reviews, 14(9):3156–3163,

dec 2010.

50

[97] J. C. Vasquez, J. M. Guerrero, A. Luna, P. Rodrıguez, and R. Teodorescu. Adaptive

droop control applied to voltage-source inverters operating in grid-connected and is-

landed modes. IEEE Transactions on Industrial Electronics, 56(10):4088–4096, 2009.

[98] T. K. Vrana, J. Beerten, R. Belmans, and O. B. Fosso. A classification of DC node

voltage control methods for HVDC grids. Electric Power Systems Research, 103:137–

144, 2013.

[99] T. K. Vrana, Y. Yang, D. Jovcic, J. Jardini, and H. Saad. The cigre b4 dc grid test

system. System, 500:D1.

[100] W. Wang and M. Barnes. Power flow algorithms for multi-terminal vsc-hvdc with

droop control. Power Systems, IEEE Transactions on, 29(4):1721–1730, July 2014.

[101] S. Wenig, Y. Rink, and T. Leibfried. Multi-terminal hvdc control strategies applied

to the cigre b4 dc grid test system. In Power Engineering Conference (UPEC), 2014

49th International Universities, pages 1–6, Sept 2014.

[102] R. Wiget, M. Vrakopoulou, and G. Andersson. Probabilistic security constrained

optimal power flow for a mixed hvac and hvdc grid with stochastic infeed. In Power

Systems Computation Conference (PSCC), 2014, pages 1–7, Aug 2014.

[103] J. Yang, J. E. Fletcher, and J. O’Reilly. Multi-terminal dc wind farm collection and

transmission system internal fault analysis. In 2010 IEEE International Symposium

on Industrial Electronics, pages 2437–2442, July 2010.

[104] T. Yang, H. Sun, and A. Bose. Transition to a two-level linear state estimator-part i:

Architecture. Power Systems, IEEE Transactions on, 26(1):46–53, Feb 2011.

[105] Z. Yongli, T. Sidhu, M. Yang, and L. Huo. An AI-based automatic power network

topology processor. Electric Power Systems Research, 61(1):57–65, Feb. 2002.

[106] K. Yunus and T. Thiringer. Primary and secondary control strategy for meshed hvdc

transmission grids at steady state. In Power and Energy Engineering Conference

(APPEEC), 2015 IEEE PES Asia-Pacific, pages 1–7, Nov 2015.

[107] L. Zhang. Modeling and Control of VSC-HVDC Links Connected to Weak AC Sys-

tems. PhD thesis, 2010.

[108] L. Zhang, L. Harnefors, and H. P. Nee. Interconnection of Two VeryWeak AC Systems

by VSC-HVDC Links Using. IEEE Transactions on Power Systems, 26(1):344–355,

2011.

[109] K. Zhu, M. Chenine, L. Nordstrom, S. Holmstrom, and G. Ericsson. Design require-

ments of wide-area damping systems using empirical data from a utility ip network.

IEEE Transactions on Smart Grid, 5(2):829–838, March 2014.

51

Part IIPapers 1 to 6

Paper 1Study of Centralized and Distributed Coordination of Power

Injection in Multi-TSO HVDC Grid with Large Off-shore WindIntegration

Paper 2Selection of DC Voltage Controlling Station in an HVDC Grid

Paper 3Real-Time Estimation of Grid Short Circuit Capacity for HVDC

Control Application

Paper 4Distributed Security-Constrained Secondary Control of HVDC

grids in the Presence of Wind Uncertainty

Paper 5Analysis of Centralized versus Distributed Two-stage Network

Topology Processor for HVDC Grid

Paper 6Implementation of agent-based power flow coordination in AC/DC

grids using co-simulation platform