Contribucion A La Seguridad de Los SEP

Diss. ETH No.
16492
Contributions to Security of Electric Power Systems
A dissertation submitted to the SWISS FEDERAL INSTITUTE OF TECHNOLOGY ZURICH for the degree of Doctor of Sciences presented by MAREK ZIMA MSc. Royal Institute of Technology (KTH), Stockholm born February 11th , 1977 in Cadca, Slovakia accepted on the recommendation of Prof. Dr. Gran Andersson, examiner o Prof. Dr. Ian Dobson, co-examiner Prof. Dr. Manfred Morari, co-examiner 2006
Acknowledgments
I would like to express my deep thanks to Prof. Gran Andersson. He o introduced me into research community and provided a lot of support, many valuable advices and opportunities for development of various professional skills. My PhD studies have been done externally, during my engagement in industry, explicitly ABB Switzerland Ltd, which was represented in the advisory committee of my PhD studies by Dr. Joachim Bertsch and Dr. Christian Rehtanz. I am very grateful to Christian, his active constructive attitude and his couching have been very valuable inspiration, from which I tried to learn as much as I could. I acknowledge the continuous support of Joachim during the whole PhD studies, who provided me opportunities to interact both with research community (e.g. via time allocation for traveling to conferences) as well as industrial partners (e.g. engineers from companies operating transmission systems). At several occasions in the recent years I had a luck to experience very inspiring discussions with Prof. Dr. Ian Dobson. Therefore I am very happy to have him as co-referee of my thesis. As a part of my thesis deals with Model Predictive Control, I am honored to have Prof. Dr. Manfred Morari as my second co-referee. His course on Model Predictive Control triggered my interest in this topic and in control as such. During my PhD studies, I have interacted with many researchers in various extend. Very enriching were discussions and correspondence with Prof. Dr. Ian Hiskens, from University of Wisconsin in Madison. 3
Acknowledgments
As a dierent type of inspiration, I would like to mention Prof. Dr. Imrich Sol my uncle. Although starting from unfavorable conditions, k, thanks to his talent and hard work, he achieved a maximum in circumstances and time given to him, both in the industry and academia. Let this thesis also be a small memory of him. Most of all, I want to thank my family, especially my mom. She has been always very supportive and encouraging in taking new professional and personal challenges. Marek Zima
Abstract
Electricity supply is a fundamental need of a modern society. A crucial property of electricity supply - reliability, is achieved by keeping a power system in a secure state most of the time. A secure state means, that the risk of a disturbance to spread further and endanger the system integrity, resulting into supply interruptions, is minimal. Keeping a power system secure is a very challenging interdisciplinary multidimensional problem, which is aected by many factors. Besides the ones, which can not be inuenced - such as extreme weather conditions or natural disasters, there is a set of factors given by the design of the system and its operation, which can be denoted security concept. Recent trends have introduced conditions, in many power systems, in which a traditional security concept experiences shortcomings. This thesis proposes modications of the principles of the existing security concept. In case of several modications, this thesis goes further into details. Explicitly, this thesis describes an index expressing the system security in an alternative manner to well known N-1 criterion, proposes a control method handling a larger dimensional problems , which are typical for power systems, and suggests a framework for a better control impact evaluation. All detailed proposals are supported by simulation and computation examples.
Kurzfassung
Ein grundlegendes Bedrfnis jeder modernen Gesellschaft ist die sichere u und ausreichende Versorgung mit Elektrizitt. Die Versorgung erfolgt a dabei im Hinblick auf das Kriterium der Zuverlssigkeit, d.h. es wird a angestrebt, das Energieversorgungssystem in einem sicheren Zustand zu betreiben. In einem sicheren Zustand ist das Risiko, dass sich eine lokale Strung ausbreitet und die Systemintegritt gefhrdet, minimal. o a a Die stndige Aufrechterhaltung der Systemzuverlssigkeit ist ein intera a disziplinres multidimensionales Problem, das von verschiedensten Faka toren beeinusst wird. Whrend ein Teil dieser Faktoren, wie extreme a Wetterverhltnisse oder Naturkatastrophen, als exogen gegeben angea sehen werden muss, existieren andere Faktoren, die durch das System selbst und seinen Betrieb gegeben sind. Letztere knnen unter dem o Begri Sicherheitskonzept zusammengefasst werden. Aktuelle Entwicklungen haben die Rahmenbedingungen in denen Energieversorgungssysteme heutzutage betrieben werden nachhaltig verndert. a In diesem Zusammenhang oenbaren traditionelle Sicherheitskonzepte zunehmend Mngel. a Vorliegende Arbeit schlgt Modikationen bereits existierender Sicherheitskonzepte auf grundlegender Ebene vor. In diesem Zusammenhang wird ein Konzept - alternativ zum N-1 Kriterium - entwickelt, mit dessen Hilfe sich die Zuverlssigkeit eines Systems in angemessenerer a Art und Weise determinieren lsst. Weiterhin beschreibt vorliegende a Arbeit einen Regelalgorithmus fr typische, hher-dimensionale Probu leme in Energiesystemen. Der Algorithmus basiert auf Methoden der Model Predictive Control und Trajectory Sensitivities. In einem letzten Schritt wird ein Konzept zur besseren Evaluierung von Regelstrategien abgeleitet. 7
Kurzfassung
Die vorgeschlagenen Modikationen und Konzepte werden ausfhrlich u mit Hilfe von Beispielen und Simulationen dargestellt.
Contents
1 Introduction 1.1 1.2 1.3 1.4 Positioning and Goal of the Work . . . . . . . . . . . . . Dissertation Outline . . . . . . . . . . . . . . . . . . . . Contributions of the Work . . . . . . . . . . . . . . . . . List of Publications . . . . . . . . . . . . . . . . . . . . . 13 13 15 15 16 19 19 21 23 27 34 36 39 40 40 41 42 42 45 45
2 Power Systems Security 2.1 2.2 Threats of Power Systems Security . . . . . . . . . . . . Present Power Systems Control . . . . . . . . . . . . . . 2.2.1 2.2.2 2.3 2.4 Normal and Preventive Control . . . . . . . . . . Emergency Control . . . . . . . . . . . . . . . . .
Last Decade Trends . . . . . . . . . . . . . . . . . . . . Shortcomings of Present Security . . . . . . . . . . . . .
3 Improvements in Power Systems Security 3.1 A New Security Index . . . . . . . . . . . . . . . . . . . 3.1.1 3.1.2 3.1.3 3.1.4 3.1.5 3.1.6 Introduction . . . . . . . . . . . . . . . . . . . . Cascading Failures . . . . . . . . . . . . . . . . . Operation Criteria . . . . . . . . . . . . . . . . . N-1 Criterion . . . . . . . . . . . . . . . . . . . . N-2 Criterion . . . . . . . . . . . . . . . . . . . . Relevant Topics in Literature . . . . . . . . . . . 9
10 3.1.7 3.1.8 3.1.9 3.2 3.3
Contents Proposed Security Index . . . . . . . . . . . . . . Example . . . . . . . . . . . . . . . . . . . . . . . Remarks . . . . . . . . . . . . . . . . . . . . . . . 45 48 48 49 51 55 55 57 57 58 60 62 62 67 73 73 76 77 79 83 83 84 84 85 86 86 88 92
Proposed Control Philosophy Essentials . . . . . . . . . Proposed Control Structure . . . . . . . . . . . . . . . .
4 MPC Employing Trajectory Sensitivities 4.1 4.2 Introduction - Problem Description . . . . . . . . . . . . Model Predictive Control Fundamentals . . . . . . . . . 4.2.1 4.2.2 4.3 4.4 General (Nonlinear) Model Predictive Control . . Approaches to Reduction of Complexity of General MPC Formulation . . . . . . . . . . . . . . .
Proposed MPC Formulation . . . . . . . . . . . . . . . . Voltage Control Example . . . . . . . . . . . . . . . . . 4.4.1 4.4.2 Illustrative Example . . . . . . . . . . . . . . . . Application to Realistic System . . . . . . . . . .
5 Benets of Better Power Systems Control 5.1 5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . Economical Benets of Control . . . . . . . . . . . . . . 5.2.1 5.2.2 5.2.3 5.3 5.3.1 5.3.2 5.3.3 5.4 5.4.1 5.4.2 5.4.3 Methodology and Modeling . . . . . . . . . . . . Simulation Results . . . . . . . . . . . . . . . . . Remarks . . . . . . . . . . . . . . . . . . . . . . . Methodology and Modeling . . . . . . . . . . . . Simulation Results . . . . . . . . . . . . . . . . . Remarks . . . . . . . . . . . . . . . . . . . . . . . Methodology and Modeling . . . . . . . . . . . . Simulation Results . . . . . . . . . . . . . . . . . Remarks . . . . . . . . . . . . . . . . . . . . . . .
Security Benets of Optimal Control . . . . . . . . . . .
Considerations . . . . . . . . . . . . . . . . . . . . . . .
Contents 6 Conclusions and Future Prospects 6.1 6.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . Directions for Future Work . . . . . . . . . . . . . . . .
11 95 95 96 99 . . . . . . . . 99 99
A Test Systems Data A.1 The IEEE Reliability Test System - 1996 A.2 A Simple Radial Test System . . . . . . . . . . . . . . .
A.3 The Nordic 26 Bus Test System . . . . . . . . . . . . . . 101 B Computation of Trajectory Sensitivities B.2 Numerical Computations 107
B.1 Trajectory Sensitivities . . . . . . . . . . . . . . . . . . . 107 . . . . . . . . . . . . . . . . . 109 . . . . . . . . . . . . . . 110 B.2.1 Nominal Trajectory . . . . . . . . . . . . . . . . 109 B.2.2 Trajectory Sensitivities B.2.3 Remarks . . . . . . . . . . . . . . . . . . . . . . . 111 C Benets of Better Control - Computations 113
C.1 Traditional Generation Dispatch . . . . . . . . . . . . . 113 C.2 Traditional Generation Dispatch Including a FACTS Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 C.3 Proposed Control Including a FACTS Device . . . . . . 114 C.3.1 Generation Dispatch . . . . . . . . . . . . . . . . 114 C.3.2 Emergency Conditions . . . . . . . . . . . . . . . 115 Bibliography 115
Chapter 1
Introduction
This chapter briey explains the objective of the research work described in this dissertation and positions it in the context of power systems research. Contributions and publication activities related to the carried out research work are listed too.
1.1
Positioning and Goal of the Work
The purpose of power systems is to supply electric energy to consumers, i.e. industry, commercial and residential subjects and entities. Main criteria for assessing capabilities of power systems to do so, are: Reliability Economy Quality Environmental impact Environmental impact on the nature and humans shall be kept minimal, e.g. minimal pollution, noise, radiation and taken space. Quality, or more frequently used term power quality, expresses how the supplied energy in form of current and voltage waveform diers from the regular sinusoidal shape. Power quality problems have usually a local nature, 13
14
Chapter 1. Introduction
i.e. signicantly inuencing only certain limited area. Typical examples of power quality problems are distortions of the voltage shape by power electronics equipment, signicant short-term drops of the voltage etc. Economy has an obvious meaning of minimal cost related to assuring power supply under the three other criteria (environmental friendliness, power quality and reliability). According to the ocial denition published in [49], reliability is probability of the power system satisfactory operation over the long run. It denotes the ability to supply adequate electric service on a nearly continuous basis, with few interruptions over an extended time period. According to the same source, another important term is security: Security of a power system refers to the degree of risk in its ability to survive imminent disturbances (contingencies) without interruption of customer service. It relates to robustness of the system to imminent disturbances and, hence, depends on the system operating condition as well as the contingent probability of disturbances. Relations between reliability and security are also provided in the same publication as: Security and stability are time-varying attributes which can be judged by studying the performance of the power system under a particular set of conditions. Reliability, on the other hand, is a function of the time-average performance of the power system; it can only be judged by consideration of the systems behavior over an appreciable period of time. Note that the term stability is discussed later in the next chapter. In the context of this chapter it is not relevant. Reliability is the overall objective in power system design and operation. To be reliable, the power system must be secure most of the time.
1.2. Dissertation Outline
15
When the security is jeopardized and a power system is subjected to a disturbance, which is not eliminated, it can lead to catastrophic scenarios [65], [46], [69], [35] and [3], having a very serious impact on the society. If this happens, a complicated and complex restoration procedure must take place [30]. This thesis focuses on the aspect of security, however, the other aspects have to be considered too. Finding an appropriate balance and compromise between above listed criteria, which are often in contradiction (for example, a cheaper power plant may cause more severe pollution etc.), is generally very challenging in most of power systems design problems.
1.2
Dissertation Outline
The thesis starts with chapter 2 identifying shortcomings of the security of present power systems originating in inadequacy of traditional control and operation approaches in the context of recent trends in power systems. Chapter 3 suggests improvements of power systems security as a compact concept of operation philosophy addressing power systems security, possessing following main features: using a dierent index, instead of well known N-1, for expressing power systems security (section 3.1); modifying existing power systems control philosophy (section 3.2) and the structure in which it is implemented (section 3.3). Chapter 4 goes deeper into the principle and an application example of a particular control method, which could be employed within the control concept mentioned in the previous sentence. Chapter 5 provides a reasoning and justication of advantages, which would be brought by such a new control method.
1.3
Contributions of the Work
Major contributions of the PhD studies research work summarized in this dissertation are as follows: Outline of a compact concept of operation philosophy addressing power systems security in several ways (chapter 3). In particular, by using an index expressing the system security better than existing one (e.g. N-1), by applying a new control philosophy implemented in a modied control structure.
16
Chapter 1. Introduction Proposal of a new security index (as an alternative to N-1), allowing to measure the extent in which the system is robust (i.e. how much is the system secure) against large disturbances (section 3.1). Furthermore, the proposed security index provides a basis for a generation dispatch providing highest probable security for a given amount of load. Proposal and derivation of a Model Predictive Control (MPC) scheme employing trajectory sensitivities to reduce complexity while acceptably allowing capturing of nonlinear system behavior of power systems (chapter 4). This introduces much higher control performance, allowing modication of power systems control philosophy, i.e. moving towards control decisions formulated and adjusted very closely to the actual system state and lumping together preventive (i.e. control, to be executed in normal and/or alert conditions, for more details, see section 2.2) and emergency control. Proposal of a framework and criteria to be considered when evaluating a control method (chapter 5). The proposed framework suggests how to measure both economy and security impacts and possible benets if a new control philosophy and/or controls, in this particular case a FACTS device, are under consideration.
1.4
List of Publications
The work reported in this dissertation has been partially covered by the following publications: 1. M. Zima and G. Andersson, Model Predictive Control Employing Trajectory Sensitivities for Power Systems Applications, 44th IEEE Conference on Decision and Control, 12 - 15 December, 2005, Seville, Spain. 2. M. Zima and G. Andersson, Model Predictive Control of Electric Power Systems under Emergency Conditions, chapter in the book Real Time Stability in Power Systems - Techniques for Early Detection of the Risk of Blackouts, Kluwer Academic Publishers, 2005.
1.4. List of Publications
17
3. M. Zima and D. Ernst, On Multi-Area Control in electric Power Systems, 15th Power Systems Computation Conference, 22 - 26 August, 2005, Lige, Belgium. e 4. M. Zima and G. Andersson, On Security Criteria in Power Systems Operation, IEEE PES General Meeting, 12 - 16 June, 2005, San Francisco, USA. 5. M. Zima and G. Andersson, Wide Area Monitoring and Control as a Tool for Mitigation of Cascading Failures, Probability Methods Applied to Power Systems (PMAPS), 13 - 16 September, 2004, Ames, USA. 6. M. Zima, C. Rehtanz and G. Andersson, Aspects of Wide Area Control of FACTS Devices, Bulk Power System Dynamics and Control VI (IREP), 22 - 27 August, 2004, Cortina D. Ampezzo, Italy. 7. M. Zima and G. Andersson, Emergency Voltage Control Using Trajectory Sensitivities, IX SEPOPE, 23 - 27 May, 2004, Rio de Janeiro, Brazil. 8. M. Zima and G. Andersson, Stability Assessment and Emergency Control Method Using Trajectory Sensitivities, 2003 IEEE Bologna PowerTech, 23 - 26 June, 2003, Bologna, Italy. 9. M. Zima, P. Korba and G. Andersson, Power Systems Voltage Emergency Control Approach Using Trajectory Sensitivities, IEEE Conference on Control Application, 23 - 25 June, 2003, Istanbul, Turkey. Besides the above list, I participated in several other publications which deal with topics relevant to the dissertation, but the material covered by them is out of the scope of this dissertation: 1. M. Weibel, K. Imhof, W. Sattinger, U. Steinegger, M. Zima and G. Biedenbach, Overhead Line Temperature Monitoring Pilot Project, CIGRE Session 2006, B2-311, Paris, to appear. 2. M. Zima, M. Larsson, P. Korba, C. Rehtanz and G. Andersson, Design Aspects for Wide-Area Monitoring and Control Systems, in Proceedings of the IEEE, Vol. 93, No. 5, May 2005.
18
Chapter 1. Introduction 3. M. Zima, Wide Area Monitoring Systems: For which Purposes?, 8th International Workshop on Electric Power Control Centers, 5 - 8 June, 2005, Les Diablerets, Switzerland. 4. M. Zima, Wide Area Monitoring - a Move towards Future Supervision of Power Systems, ETG Tagung: Forschung und neue Technologien im Energiebereich, 8 March, 2005, Zurich, Switzerland. 5. C. Carnal, J. Bertsch and M. Zima, Guarding the Grid, ABB Review Special Report, September, 2004. 6. M. Zima, T. Krause and G. Andersson, Evaluation of System Protection Schemes, Wide Area Monitoring and Control Systems, Advances in Power System Control, Operation and Management (APSCOM), 11 - 14 November, 2003, Hong Kong, China. 7. M. Zima, Blackout a Deregulacia Trhu s Elektinou, SME, 21 August, 2003, in Slovak. 8. J. Bertsch, M. Zima et al., Experiences with and Perspectives of the System for Wide Area Monitoring of Power Systems, Quality and security of Electric Power Delivery Systems, CIGRE/IEEE PES International Symposium, 8 - 10 October, 2003, Montreal, Canada. 9. C. Rehtanz, M. Larsson, M. Zima et al., System for Wide Area Protection, Control and Optimization based on Phasor Measurements, Power Systems and Communication Systems Infrastructures for the Future (CRIS), 23 - 27 September, 2002, Beijing, China.
In addition, some attitudes expressed in this thesis have been formed also taking into account experience collected during 5 years of engagement in power industry, which has resulted into 7 pending or already registered patents in the area of power systems stability monitoring and control.
Chapter 2
Power Systems Security

This chapter provides an overview of the present situation in security of power systems and main factors inuencing it. Weaker points in the concept of power systems control, which may not be adequate anymore in the present and future conditions and thus should be addressed, are identied.
2.1
Threats of Power Systems Security
Power systems have originally arisen as individual self-sucient units, where the power production matched the consumption. In the case of a severe failure, a system collapse was unavoidable and meant a total blackout and interruption of the supply for all customers. But the restoration of the whole system and synchronization of its generators were relatively easy due to the small size of the system. Power systems size and complexity have grown to satisfy a larger and larger power demand. Phenomena, having a system-wide, global nature, endangering the normal operation of power systems have appeared, explicitly: Frequency Instability - is inability of a power system to maintain steady frequency within the operating limits. Keeping frequency within the nominal operating range (ideally at nominal constant value) is essential for a proper operation of a power system. A 19
20
Chapter 2. Power Systems Security maximal acceptable frequency deviation (usually 2 Hz) is dictated by an optimal setting of control circuits of thermal power plants. When this boundary is reached, unit protection disconnects the power plant. This makes the situation even worse - frequency further decreases and it may nally lead to the total collapse of the whole system. Frequency instability is in its nature rather a tracking than truly a stability control problem. Voltage Instability - is the inability of a power system to maintain steady acceptable voltages at all buses in the system under normal operating conditions and after being subjected to a disturbance. A system enters a state of voltage instability when a disturbance, increase in load demand, or change in system conditions causes a progressive and uncontrollable drop in voltage. Transient Angular Instability (also called Generators Out-of-step) - is the inability of the power system to maintain synchronism when subjected to a severe transient disturbance. The resulting system response involves large excursions of generator angles and is inuenced by the nonlinear power-angle relationship [47]. In case of transient angle instability, a severe disturbance is a disturbance, which does not allow a generator to deliver its output electrical power into the network (typically a tripping of a line connecting the generator with the rest of the network in order to clear a short circuit). This power is then absorbed by the rotor of the generator, increases its kinetic energy, which results in the sudden acceleration of the rotor above the acceptable revolutions and eventually damage of the generator. Small-signal Angular Instability (also mentioned as Generators Swinging or Power Oscillations) - is the inability of the power system to maintain synchronism under small disturbances. Such disturbances occur continually on the system because of small variations in loads and generation. The disturbances are considered suciently small for linearization of system equations to be permissible for purposes of analysis. Local modes or machine-system modes are associated with the swinging of units at a generating station with respect to the rest of the power system. The term local is used because the oscillations are localized at one station or small part of the power system [47]. Inter-area modes are associated with the swinging of many machines in one part of the
2.2. Present Power Systems Control
21
system against machines in other parts. They are caused by two or more groups of closely coupled machines being interconnected by weak ties [47]. Note, that transient phenomena, having purely local nature, such as short-circuits, usually do not have a direct system-wide impact (Although they may be an initiating trigger of instabilities, e.g. a slow short-circuit fault clearing may cause transient instability of a closely located generator etc.), therefore they are excluded from further considerations. With the rising importance of the electricity for industry (and the entire society), the reliability of supply, and thus power systems security, has become a serious issue. Interconnection of the separated/individual power systems have oered a number of benets [33] [26], such as sharing the reserves both for a normal operation and emergency conditions, dividing of the responsibility for the frequency regulation among all generators and a possibility to generate the power in the economically most attractive areas, thus providing a good basis for the power trade. Although this has reduced some negative features mentioned above, on the other hand it has also introduced new problems - such as a potential for spreading of disturbances over large distances and thus paralyzing vast geographical areas etc.
2.2
Present Power Systems Control
Control methods mitigating the listed dangerous phenomena (frequency, voltage, transient and small-signal instability) and keeping the power system in a secure state are mainly based on the classication of power system states. Explicitly, according to [47], these states are: Normal All system variables are within the normal range and no equipment is being overloaded. The system operates in a secure manner and is able to withstand a contingency without violating any of constraints. Alert All system variables are still within the acceptable range and all constraints are satised. However, the system has been weakened to a level where a contingency may cause an overloading of equipment that places the system in an emergency state.
22
Chapter 2. Power Systems Security
Normal
Restorative
Alert
In Extremis
Emergency
Figure 2.1: Power system operating states. Arrows express possible transitions among them. Emergency Some system variables are outside of acceptable range (e.g. voltages too low, lines overloaded). If no control changes are introduced, system progresses into In Extremis. In Extremis Cascading spreading of system components outages resulting in partial or system-wide blackout (loss of supplied load). Restoration Energizing of the system or its parts and reconnection and resynchronization of system parts. Mutual relations and possible transitions among the operation states are shown in gure 2.1. With respect to the above categorization of operation states, control approaches for keeping power systems secure are usually applied in two stages: Normal and preventive control This control is applied in the normal and alert state. Its objective is either to stay in or to return into normal state.
23
Emergency control This control is applied in emergency or in extremis state to stop the further progress of the failure and to bring the system into normal or alert state.
2.2.1
Normal and Preventive Control
Typical representatives of normal and preventive control are: Hierarchical automatic control: Frequency control Voltage control Centralized manual control based on: Contingency screening Operator judgment Control measures usually include: Change of active power generation set-points, i.e. redispatch. Change of reference points of ow controlling (FACTS) devices. Start-up of generation units. Change of voltage set-points of generators and Static VAR Compensators (SVC). Switching of shunts elements (reactors, capacitors). Change of substation conguration (e.g. splitting of busbars). Common Practice The hierarchical frequency control concept has been established in power systems for decades. Its description can be found in any textbook dealing with power systems control or operation, e.g. [47]. On the other hand, hierarchical voltage control has been implemented only in a few countries (e.g. France, Italy). Two additional higher levels - Secondary Voltage Regulation (SVR) and Tertiary Voltage Regulation
24
(TVR), enrich primary voltage regulation. National TVR shall coordinate SVRs that control the areas voltage proles. The objective here is to create a system-wide voltage prole minimizing the transmission of reactive power over longer distances and maximizing the reactive power generation reserves. Larger reactive power generation reserves mean that they can be activated in case of a disturbance, i.e. the system is made more robust. More details about implemented hierarchical voltage control schemes are reported in [70]. Hierarchical frequency and voltage control concepts are usually fully automated, i.e. the control loop is closed and does not involve any human intervention. Possible negative interactions among layers are minimized by appropriate selection of their time constants, e.g. primary layer dynamics is in order of seconds, secondary layer in tens of seconds to minutes and tertiary layer minutes to hour. As mentioned above, in a large majority of power systems, hierarchical frequency control is the only fully automated closed loop control in Normal and Alert state. In all other cases, human factor is involved in form of power system operator. Then a control scheme/loop basically consists of: Data acquisition and monitoring This stage is usually handled by Supervisory Control and Data Acquisition (SCADA) system. Operators decision Operator based on the knowledge about the present state of the system, coming from SCADA system, decides, whether to intervene by modifying actual values of controls. Operator can base his intervention decision either on his experience and judgment or on decision support tools. Control execution This stage materializes either via link SCADA Substation Automation, or communication with the personnel executing the control manually in substations or/and power plants. In this control scheme, the main objective is to keep the power system in a secure state, expressed by the compliance with N-1 criterion. That means that a possible outage of any single component shall not create an unacceptable stress of other component(s) or instability problem. In most power systems, a procedure called Security Assessment is employed for this purpose. Security Assessment is usually implemented as a program belonging to the Energy Management System (EMS) processing
25
present state information given by SCADA system. Security Assessment is then done in a continuous cycle, typically every 5 or 15 minutes. The consequences of possible component outages, which are then examined, are components operated outside their limits (line overloads and undervoltage or overvoltage), voltage instability, transient instability). Usually, each type of consequences is analyzed by a separate software package, independently of others. Security Assessment is often executed in two steps: 1. Complete set of possible contingencies is processed using fast (sometimes only approximate) static analysis, neglecting system dynamics. This procedure is often referred to as Contingency Screening. 2. Reduced set of contingencies, identied as possibly most severe ones in the Contingency Screening, are analyzed in detail in form of time domain simulation considering all relevant dynamics aspects. For checking of unacceptable components stresses, employment of purely static methods (i.e. Contingency Screening) is sucient. Essentially, for each assumed contingency a power ow computation is done, followed by a simple comparison of computed post-contingency state and components operational limits. The relation between the active power consumed in the monitored area and the corresponding voltages is expressed by so called PV-curves (often referred as nose curves). The increased values of loading are accompanied by a decrease of voltage. When the loading is further increased, the maximum loadability point is reached, from which no additional power can be transmitted to the load under those conditions. In case of constant power loads, i.e. loads whose power consumption is independent of the voltage magnitude, the voltages in the nodes become uncontrollable and rapidly decrease, resulting in voltage collapse. However, the voltage level close to this point is sometimes very low, what is not acceptable under normal operating conditions, although it is still within the stable region. But in emergency cases, some utilities accept it for a short period. There are also other alternative graphical representations, e.g. QV-curves (amount of needed reactive power to keep a certain voltage at a given active power loading). PV-curves (or QVcurves) starting from actual system state are computed for considered set of contingencies. If any of the resulting PV-curves indicates that
26
the system would be unstable (or would have an unacceptably low voltage prole) after a contingency, the operator takes preventive measures, such as switching shunt capacitors, generation redispatch etc. Proposed Improvements Most of the research in power systems security area follows the framework of the existing control structure described above. The major part of the research activities in voltage instability area has focused on the steady state aspects, i.e. nding the maximum loadability point of the PV-curve. The solution of the Newton - Raphson power ow calculations becomes unfeasible close to this critical point due to the singularity of Jacobian matrix. This provides a basis for a number of indices, expressing the proximity to the voltage collapse, which has been derived, e.g. [21], [14], [72] and [8]. Continuation Power Flow (CPF) can overcome the numerical problems indicated above. In principle, it is a slightly reformulated conventional power ow. The equations are augmented by the term quantifying the load increase and containing a new variable - the load parameter. A new equation is introduced, which basically forces a continuation parameter chosen in the predictor step to hold its value in the iterative correction process. This continuation parameter is optimally loading in the beginning of the PV -curve and when approaching the nose, voltage. Various techniques have been developed for predictor step in order to speed up the computations and increase the accuracy. A very good explanatory example of tangent method is in [1]. Secant predictor (in fact linear approximation estimate) can be found in [13]. An application on inter area power transfer security evaluation is demonstrated in [28]. CPF has probably become the most widely accepted tool for the voltage instability assessment/evaluation and a huge number of papers have been written about it. A step forward is the QSS (Quasi Steady-State) approximation proposed in [17]. This method consists of voltage stability evaluation based on the time domain simulation with a simplied description of the power system dynamics, such as load behavior etc. An application of traditional measure of transient angle instability equal area criterion (expressing a balance between the accelerating and decelerating energy), on emergency control has been presented in [22],
27
which describes the method called SIME (single machine equivalent). The angles of the generators in the system are predicted approximately 200 ms ahead. According to this, the machines are ranked and grouped into two categories. For the generators from the critical category, OMIB (one machine, innite bus) equivalent is modeled and extended equal area criterion is applied to assess their stability. Pre-assigned corrective action is executed if an unstable generator is identied. A similar procedure/algorithm is used in the commercially available program TSAT intended for both o-line and on-line use [48], developed by Powertech Labs. Here the Dynamic Extended Equal Area Criterion is employed for screening of the most severe contingencies that are then analyzed in detail in form of time domain simulation.
2.2.2
Emergency Control
Typical representatives of emergency control in today power systems are: Protection based systems: Under frequency load shedding (UFLS) schemes Under voltage load shedding (UVLS) schemes System Protection Schemes (SPS) Damping control Emergency control measures may include: Tripping of generators Fast generation reduction through fast-valving or water diversion Fast HVDC power transfer control Load shedding Controlled opening of interconnection to neighboring systems to prevent spreading of frequency problems Controlled islanding of local system into separate areas with matching generation and load Blocking of tap changer of transformers Insertion of a braking resistor
28 Common Practice
The main challenge in emergency control is the urgency, in which it has to be applied. Since historically very high demands for high performance communication system and control decision logic could not be met, emergency control strategy relies on devices reacting to their local measurements based on their setting determined o-line by simulations of assumed dangerous scenarios. Under frequency load shedding (UFLS) schemes Local devices used for UFLS schemes are UFLS relays. UFLS schemes and relays might be sorted in various categories [20], but their functionality is essentially the same. They are usually triggered when frequency drops to a predened level and/or with a predened rate of change. Their action is disconnection of the load in several steps (5 - 20 % each) from the feeders they supervise. However, their eectiveness is strongly dependent on their careful tuning based on prestudies, since there is no on-line coordination between them. Under voltage load shedding (UVLS) schemes Under voltage load shedding relays are a conventional local solution to prevent voltage instability [64]. The criterion triggering the load shedding action is a predened voltage level in the supervised node (For example 88 % and 86 % of the nominal voltage in one particular isolated network.). System Protection Schemes (SPS) The abbreviation SPS is often also referred to as Special Protection Schemes. However, throughout this thesis the System Protection Scheme interpretation is assumed due to more appropriate descriptive value (expression special may be misleading due to large room left for its interpretation). SPS dier from UFLS and UVLS schemes and relays essentially in two aspects: 1. SPS use in addition to (or instead of) measurements also a particular topology change (i.e. contingency) information to detect a dangerous system state. 2. SPS consist of several relays, which often use an information from a remote location (e.g. measurement taken by one relay is sent to other relay, which processes it and executes a control action).
2.2. Present Power Systems Control According to [2], SPS are dened as: a protection scheme that is designed to detect a particular system condition that is known to cause unusual stress to the power system and to take some type of predetermined action to counteract the observed condition in a controlled manner. In some cases, SPSs are designed to detect a system condition that is known to cause instability, overload, or voltage collapse. The action prescribed may require the opening of one or more lines, tripping of generators, ramping of HVDC power transfers, intentional shedding of load, or other measures that will alleviate the problem of concern. Common types of line or apparatus protection are not included in the scope of interest here.
29
In 1992, CIGRE and IEEE performed a survey about the installed SPS among utilities. The detailed statistical results of the answers reporting 111 installed SPSs can be found in [2]. Very important and interesting observations and information can be extracted from it, although they are not explicitly stated there: The trend is quite obvious; the most SPSs have been commissioned in the nineties. The degree of complexity is rapidly increasing and the solutions are more and more sophisticated. All installed SPSs are dedicated solutions for particular power systems. There is no scheme that could be applied to another power system with minor modications. SPS should be armed (i.e. ready for operation) all the time, not only in the periods when the power system is heavily stressed. An explicit quotation: ... the costs of the false trips is generally much lower than the cost of failure of the SPS to operate when required ... This implies, that even with the risk of malfunction, SPS installations have generally proven to bear a relatively low economic risk (at least the cases discussed in the reference).
30
Damping Control Some power systems lack a natural damping of oscillations, which may occur, and they would be unstable when subjected to any minor disturbance and sometimes even under normal operation conditions [68] if no measures increasing the damping were introduced. As stated in [61], extension of the transmission capacity by adding a new line does not necessarily improve the damping signicantly and solve the problem (as the authors show on the eigen-sensitivity analysis applied on Korean network). A traditional way of introducing an additional damping in the system is using of Power System Stabilizer (PSS), which modulates the output voltage of the generator. Besides generators, PSS can be installed and used for modulation of FACTS devices control. The coordinated tuning of PSSs is a complex task, since they should be robust - work in the wide range of operation conditions and provide the best possible performance. This process is done o-line. Data needed for tuning of PSSs are usually obtained by modal analysis of power system model linearized around its operating point. Proposed Improvements Under frequency load shedding (UFLS) schemes Using of Neural Network to estimate the dynamic response of the power system to the under-frequency load shedding is proposed in [55]. This information is then used to calculate an optimal amount of load to be disconnected. Under voltage load shedding (UVLS) schemes An alternative local approach for detection and evaluation of voltage instability has been presented in [71] and [31]. The grid is in the supervised node represented/replaced by a Thevenin equivalent and the load is modeled by an impedance. The point of equal impedances (rule known from the basic circuit theory) is then representing a boundary between stable and unstable conditions. However, there is an open issue of coordination of several such local devices. The exclusion of the system dynamics might bring a risk of missed information about the inertia of the system, with which it responds to disturbances. An attempt to include it and optimize the load shedding may involve a genetic algorithm [56]. However, there is a danger, that an
31
important scenario may be omitted from the training/tuning procedure of the algorithm and failure of SPS in case of occurrence of such situation in reality. Alternative solutions proposed by the same research group for implementation in Hydro-Qubec network are more or less rule based relying on the o-line studies and setting of local relays [18]. The roots of these considerations and probably the most complete coverage can be found in [4] and in a bit more condensed form in [5]. The author explores the dynamics of loads, especially the timing aspect of load shedding and its location. A centralized approach based on PV-curves has been implemented in practice and proven to work properly [44] when the automatic load shedding system VSHED, installed in Entergy (utility serving Arkansas, Mississippi, Louisiana and a portion of Texas), successfully operated in case of major disturbance on September 22nd, 1998. System Protection Schemes (SPS) The mentioned weakness of UFLS relays (uncoordination) can be overcome by centralized load shedding schemes. Some of them are already in the operation. One part of the SPS included in a new Hydro-Qubecs defence plan [67], commissioned in 2000, is Remote Load Shedding System (RLSS). RLSS is triggered by Extreme Contingencies Detection Systems (ECDS) monitoring and supervising the highest 735 kV transmission network [15]. RLSS calculates the power to be shed depending on the severity of the event. RLSS sends the commands to the communication processors. Each of them communicates with 24 Programmable Load Shedding Systems (PLSS). PLSSs are in operation in Hydro-Quebec for more than 20 years, currently installed in 140 distribution substations. PLSS is a device that executes the received load shedding commands and in extreme cases it works as a back up using the same criteria as any conventional local UFLS relay. It is also capable of taking measurements, performing self-diagnostics etc. Simple rules derived from the classical relations of OMIB equivalent are proposed in [54] and compared with the conventional out-of-step relay. The simulation results of Tokyo Interconnected Network show that the prediction of the loss of synchronism is 0.7 second faster, thus providing longer time for countermeasures. Simulation results show how the blackout of Taiwans power system on July 29th, 1999 could have been avoided employing a protection scheme against transient instability described in [74]. The proposed algorithm
32
aims at the protection of EHV-tie line connecting the generation area in the south with the load area in the north of the country. Therefore the line power ow limit was used to modify the Equal Area Criterion for stability of the generators in the south. Reference [40] suggests an algorithm, which does not require knowledge of the system and uses only the on-line measurements of generators rotor angles and power mismatches to predict the transient angular stability of the generator. It implies that there would not be any need for tuning/adaptation procedure when applying this method to another power system. The valves of the steam turbines aecting the mechanical input of a generator are controlled in order to stabilize the generator. Simulation tests on the WSCC system have been carried out, showing the successful operation of the proposed method. The experience with the transient stability control systems (TSC Systems) is reported in [43]. CEPCO (Chubu Electric Power Co.) in Japan has commissioned two systems, one in June 1995 and the other one in May 1996. In the paper, mainly the statistics are listed. The principle behind, as mentioned, is an on-line pre-calculation cycle including all possible operating scenarios but no details are provided. In case of a dangerous situation, the result of pre-calculations is recalled and appropriate generator is disconnected. The calculation time of the cycle is always less then 5 minutes (usually 3) for 30 cases and power system model consisting of 300 nodes, 400 branches and 30 generators. This impressive number is achieved by employing several arithmetic units performing parallel calculations. The TSC system has the features: automatic recognition of change in operating conditions of generators and transmission lines which information determines the generators to be controlled for stabilization CEPCO system is stabilized with the minimum amount of generation shedding, since the controlled generators are chosen on-line coping with an extension (new built generators or lines) of the CEPCO system is relatively easy by update of network data database Damping Control Various techniques aiming at the identication of oscillation modes from measurements of various quantities have been
33
reported. This should compensate for model inaccuracies aecting the result of modal analysis. For estimation of NORDEL inter-area oscillation modes, the frequency measurements from the common distribution network have been utilized [32]. However, the distribution network is probably not the best choice since the measurements contain quite a lot of noise (distortion from higher harmonics) and extraction of information about the two typical NORDEL oscillation modes was dicult. The monitoring of frequency on the transmission level, triggered by disturbances, shows more promising results [24], although another important factor, which has played a certain role, is the size of the system and thus the frequency of the oscillations. UCTE/CENTREL system is a magnitude larger than NORDEL system, so the recorded oscillations (measurements have been taken directly on the transmission level) have much lower frequency and the measurement noise is more easily ltered out. The authors also make an important statement about the increase of meaning/importance of inter-area oscillations monitoring using WAMS (Wide Area Measuring System) due to the growing size of the power systems. The oscillations along the north-eastern Australian coast (Queensland) have been investigated in [51]. The voltage angle at the ends of two long lines have been measured and analyzed. The authors state that the angle signals have greater potential for modal identication than power. Promising simulation results with voltage angles measured with PMUs and fed into the PSS designed for it and placed at two generators in Norwegian network are demonstrated in [68]. Employment of Kalman ltering techniques is reported in [45]. A Remote Feedback Controller (RFC) design methodology using PMU measurements is presented in [62]. The simulation results show robustness and good performance of the RFC applied to the damping of low frequency inter-area oscillations. A research group in Hydro-Qubec under lead of Kamwa has done signicant work in the eld of damping of inter-area oscillations. In [38] two-loop PSSs are proposed. The speed sensitive local loop operating the usual way is extended with a global loop using wide-area measurements from two suitably selected areas, in this case frequency measurements. Five control sites comprising 4 generators and one synchronous condenser have been chosen for implementation of the proposed method. The simulations (without considering a time delay caused by
34
communication synchronization of values, processing and execution of a command) have proved a signicant improvement in the damping of inter-area oscillations, which have been excited by a contingency (trip of one of the major lines). The device that is assumed to be used for measuring in the practice is a Phasor Measurement Unit (PMU). The locations for their placement are shortly discussed. The author suggests placement of 65 PMUs for multifunctional SPS in Hydro-Quebec system. The described work has been extended in [39].
2.3
Last Decade Trends
Majority of control concepts described in the previous section, which represent an absolutely dominant part of present power systems control, have been developed for the traditional structure of power systems, which prevailed several decades ago, based on the limitation and capabilities of infrastructure existing in the time of their development. However, power systems operation environment has been changing quite signicantly as well as technological implementation possibilities (such as computers, communication infrastructure etc.). Most signicant factors inuencing power systems operation in the last decade are related to: Electricity Market Liberalization Distributed Generation Environmental Restrictions Electricity Market Liberalization There are many publication describing in detail process of market liberalization, the new organizational framework and all related aspects. But in the context of this thesis it is important to stress only some consequences of them. Unbundling of utilities, part of liberalization, caused separation of transmission, distribution and generation parts of power systems. Responsibility for the security of power systems lies then on the System Operator (SO) operating transmission system. There is no uniform electricity market design. In dierent power markets the entity bearing responsibility for the system operation have dierent names, such as Transmission System Operator (TSO), Independent System Operator (ISO),
2.3. Last Decade Trends
35
Transmission System Coordinator (TSC), related mainly to the fact if the possess, or only operate transmission assets. To avoid confusion and restriction to only specic cases, the abbreviation SO is used throughout this thesis hence, since ownership of transmission assets does not play any role in the context of this thesis. The main complication, brought by the separation of ownership of generation and transmission, is lack of coordination in the long-term system expansion planning. This results in the very reduced predictability of utilization of transmission assets and correct allocation of controls. Involvement of many parties in generation and distribution of energy creates a more complex economical environment - many more economical links and large volatility of economical factors. This in turn introduces a strong variation of power ow patterns. Interconnecting links (frequently called tie-lines) , that were originally built for power exchange under emergency conditions, are now used for regular electricity trading over long distances. Now tie-lines are often the most stressed/loaded elements in transmission systems. Distributed Generation Penetration of distribution (Hence the expression distributed generation. Additionally, this expression also refers to the fact of spatial sparse geographical spreading of many units with low power rating.) systems with small generation units has been rapidly increasing because of technological advances in this area and a very strong legislation support in some countries. As an example, utilities in Germany are obliged to connect new distributed generation and buy energy from distributed generation any time it produces it. Similar rules apply in many other countries. A large portion of distributed generation is wind and solar generation, whose production is weather dependent and thus very volatile, less predictable and its controllability is very limited. Many generators of low ratings are asynchronous machines, which need to be energized from the network side (This implies that they can not be used for so-called black start in case of system restoration.) and do not oer a voltage support and control. Flows in distribution networks are strongly aected from distribution generation, which is injecting its power directly there. Environmental Restrictions Last, but not least, increasing pressure from public side, due to environmental concerns, has made it very
36
dicult to build any larger primary equipment assets both on the generation (power plants) and transmission side (lines, substations). Thus only existing primary assets can be used to serve continuously increasing power demand. (Primary equipment is the expression referring to the equipment in power systems, which is subjected to high voltage and/or high currents. Measurement, monitoring and control systems are referred to as secondary equipment.)
2.4
Shortcomings of Present Security
Main assumptions, which were used in the design of control concepts in the traditional power system, are: Predictability - such as typical loading pattern etc. as a consequence of the fact that all aspects were managed by a single entity. Substantial equipment loading margin. Good controllability of generators. Distribution systems having purely consumption character. These assumptions led to the fundamental features of the control concept described in the section 2.2 as follows: Control schemes dedicated separately to each dangerous phenomenon, i.e. separate scheme for control of frequency, voltage etc. Strict distinction between preventive and emergency control. Control policies (especially for emergency control) formulated oline for a reduced set of typical system conditions. N-1 criterion used as a quantitative measure of the system security. However, present conditions in power systems dier signicantly from the above listed assumptions, as shown in the previous section. This discrepancy may lead to many dangerous situations, two examples can illustrate this:
2.4. Shortcomings of Present Security
37
When avoiding one instability problem, system can be pushed into another one. For example, when executing Under Voltage Load Shedding, unconsumed generated power can result in a sharp frequency rise yielding frequency stability problems. Under Frequency Load Shedding acting by disconnection of a distribution feeder may actually disconnect a signicant portion of distributed generation and make the situation even worse. Power systems security is a very complex multidimensional problem. It is suggested that improvements of power systems security can be achieved by complementing and/or modifying both operational philosophy and control, which naturally includes technology and infrastructure. However, suggestions for such improvements should be supported by demonstration of incentives. These statements also form basis for the work summarized in this dissertation.
Chapter 3
Improvements in Power Systems Security

Power systems security can be inuenced in many ways. This chapter attempts to put the security of power systems into a wider perspective for possible future power systems operation and control. Essential for this proposal is a use of dierent criteria expressing system security; shift to real-time formulation of both preventive and emergency control policies and a modication of the control structure. There are many factors inuencing power systems security. Overview of their today form and weaknesses have been provided in sections 2.2 and 2.4, respectively. In this chapter a proposal of a compact concept of operation philosophy addressing power systems security is presented. The following features represent the essence of this concept: Use of new index expressing security The objective of the new security index, replacing the standard N-1 criterion extensively used today, is to drive a power system on a trajectory more robust against spreading of large system disturbances. This topic will be further discussed in the section 3.1. Adopting a new control philosophy Central for the proposed philosophy is that normal, preventive and emergency control should be combined (That means there should not be a strict distinction 39
40
Chapter 3. Improvements in Power Systems Security between these two types of control, which are today implemented even on two dierent completely decoupled hardware platforms.) and their control policies should be formulated based on the actual system conditions, that means not based on o-line studies of typical situations as is common practice today. More detailed discussion follows in the section 3.2.
Modication of the existing control structure This may seem to be necessary in order to accommodate a new control philosophy. Section 3.3 elaborates on this topic further.
3.1
A New Security Index
In this section a new way of minimizing the risk of cascading events is presented. A method to dispatch the system in such a way that the risk for subsequent line trips after the rst initial event is minimized is presented. This method is based on the assumption that the probability of a line trip is a continuous increasing function of the line overloading. A slight overload implies a smaller probability of a line trip than a larger overload does. The new method is compared with the N-1 criterion and it is shown that the proposed method makes a system more robust against multi contingencies than the N-1 criterion.
3.1.1
Introduction
A secure operation of power systems requires that the system has appropriate security margins so it can withstand various disturbances. Ideally, the security margin should be calculated and monitored online, but due to the size and complexity of modern power systems this is either done using very simplied models of the system or not at all. Instead dierent simplied security criteria are used to ensure that the security margins are sucient. One such criterion is the N-1 criterion, which was introduced after the 1965 Northeast USA blackout. This criterion, in its simplest form, says that the system should be able to withstand the loss of any component, e.g. line, transformer, generator, etc, without jeopardizing system operation. This criterion, with slight modications, is widely used in power system operation today all over the world. In some systems where a higher security is required, the N-2
3.1. A New Security Index
41
criterion is applied, and one can also envisage in general N-K criterion, with obvious meaning. The recent large blackouts have initiated a discussion regarding the relevance of the N-1 criterion. It can be argued that the situation today is signicantly dierent from the one thirty years ago when this criterion was generally adopted. The power systems today are generally more stressed due to higher loadings than earlier, and the power transfers due to trading, which are hard to predict, have increased dramatically. The N-1 criterion deals with one single event, but recent blackouts have often been triggered as a consequence of several independent or coupled events. Particularly, so called cascading sequences of events have played crucial roles in several blackouts. Another important issue is the temporal aspects of the N-1 criterion. A relevant question in this context concerns the time delay to bring the system back to N-1 security after a contingency. Also, the discrete nature of N-1 criterion can be questioned, since from the N-1 denition, the system is in a higher risk, if its component is loaded 100% of its rating 1% of the time, than 99% of its rating 100% of the time, what is perceived as secure according to N-1 criterion. In the last years, several blackouts have been experienced worldwide. Their origins were various, but as a general consequence the traditional operation criteria started to be questioned, e.g. [60]. One main suggestion, which appear in the literature, is that moving from deterministic criteria towards probabilistic ones could be the right step, e.g. [75]. Perhaps the most comprehensive discussion with a particular proposal can be found in [41]. A mixed approach will be proposed here. The probability of occurrence of the initial event is not modeled, instead, we focus on the minimization of the risk that any initiating event will proceed to a cascading failure. An assumption in this approach is that it is possible to assign approximate probability characteristics of the component trip as a function of its loading.
3.1.2
Cascading Failures
A cascading failure is a process, in which an initiating event/failure increases the stress of other system components, resulting in possible overloading above the limits for which they were designed. The outage
42
Chapter 3. Improvements in Power Systems Security
of overloaded components can progress either slowly (steady-state progression), or quickly (transient progression) [52]. The elements involved in the cascading failures are primarily branches (lines and transformers). However, we would like to emphasize that cascading process may generally involve any type of component failures (even a wrong operators control action). The initiating event is usually a component trip due to a factor, which cannot be inuenced (e.g. such as lightning strike into the transmission line tower and subsequent line tripping by the short-circuit relay). The transient progression usually involves dangerous phenomena like voltage instability, frequency instability and small-signal instability (power oscillations), and its time scale is between one second and several tens of seconds, depending on the type of instability. A typical scenario is for example an outage of large power plant(s) as initiating event(s). The sudden lack of the active power generation results in a signicant drop of the frequency below the operating range of other thermal power plants, which then are tripped. The line loading pattern changes dramatically and overloaded lines are tripped yielding in worst case a blackout. A similar sequence of events occurred in the Swedish blackout on September 23rd 2003, but there thermal power plants were tripped due to a double busbar protection. The slow progression involves line tripping between fairly large time intervals, in order of minutes. In this case the line tripping occurs either after exceeding a short-term emergency line loading limit by the line protection, or the line sags and short-circuit between the line and ground takes place, which also implies a line trip. An example of the slow progressing cascading is the initial phase of the Northeastern US blackout on August 14th, 2003.
3.1.3 3.1.4
Operation Criteria N-1 Criterion
As mentioned in the Introduction, the N-1 criterion says that no other element should be overloaded after an outage of any single element. Some utilities continuously monitor on-line their compliance with N1 criterion according to the actual power system state, whereas some other utilities apply N-1 check only in their day ahead dispatch plan. Most of the time the N-1 criterion serves well. However, there are some weak points we want to discuss in this paragraph. The denition of
43
the N-1 criterion does not say anything about the degree of the yielded overloads. Also, the depth of possible cascading is not treated at all, although it is a very crucial question which portion (in terms of size) of the power system that will be aected by negative consequences, such as unserved load , i.e. a blackout. The N-1 criterion does not contain any reference to the power system size, thus actually the relative security margin is shrinking as the system grows. The relative system loading is increasing with its size when adopting the N-1 criterion, furthermore the probability of multiple outages increases as well, and consequently the system becomes more vulnerable. One more reason has historical roots. Originally, interconnections of independent power systems have been built for exchange of power under emergency conditions. Meantime, they have often become more loaded in the network due to heavy transfers caused by power trading. We introduce several assumptions in order to simplify the analysis and to keep the focus on the main principle we want to discuss. To reduce the dimension of the problem, the DC power ow model can be employed to describe the power system behavior, thus neglecting dynamics and voltage/reactive power aspects. Then mathematically, a dispatch yielding maximal possible loading (while fullling N-1 criterion) in a particular power system can be formulated as follows:
Nl
max
Pg n=1
Pln
(3.1)
Ng
Nl
Pgm =
m=1 n=1
Pln
(3.2)
|Fi | Fmax
(3.3)
Pg 0
(3.4)
Pl 0
(3.5)
44

x 10
4
60
System Load [pu]
40
Number of Optimization Constraints 0 1 K 2
50
30
20
10
1 K
Figure 3.1: Properties of N K dispatch. System load represents the sum of possibly consumed power while being compliant with the security criterion. Number of constraints refers to the equation 3.3.
Pg RNg Fi , Fmax RNb i = 1, 2, . . . , Nc
(3.6)
where Pg and Pl are the injected generation and load respectively. Fi denotes post-contingency ows in the remaining branches (i.e. lines and transformers) after the outage of the branch i. Nc is the number of considered contingencies (here single element outages, thus number of branches), Ng number of generators and Nl number of loads. We assume that a node can be only either generation or consumption type in the analysis presented here. This assumption can be relaxed and it is believed to have no signicant inuence on the conclusions drawn here.
45
3.1.5
N-2 Criterion
A suggestion, which sometimes appears, is that a N-2 security criterion could be adopted. However, its practical implementation has several drawbacks. First, the utilization of the network would be very low, second, the computation burden when doing the security analysis would increase very signicantly as shown in gure 3.1. As seen the maximum system load decreases with about 35 % for the N-2 dispatch as compared with the N-1 one, while the computational burden increases with an order of magnitude. Here the comparison has been made on data from [23].
3.1.6
Relevant Topics in Literature
Besides aspects directly dealing with the operation of power systems, there are several topics, which we consider to be closely related. Above all, the mechanism of a blackout (i.e. cascading failure) evolution and factors having a signicant impact on it. We believe that there is a certain potential for cascading failures related to the power system topology structure [76]. Actually, several papers on hidden failures in power systems have conrmed that there might be several paths, or system regions, which are much more vulnerable than others in terms of the further spreading of a contingency, see [42] and [66]. Here we would like to point out that it is not clear that all contingencies shall be treated equally from the system wide impact point of view.
3.1.7
Proposed Security Index
Assumptions The assumptions we list in this subsection form a basis for the proposed security index. In the traditional N-1 operation criterion, overloads are aimed to be strictly avoided for all single element contingencies. This means that it is assumed that the probability for line trip is zero when line ow |F | < Fmax , and 1 if |F | Fmax . However, it can be believed that the probability will essentially rather follow a curve shown in gure 3.2. The authors in [12] use such a curve to describe a probability of a relay malfunction (i.e. trip of a healthy line), but other factors contributing to the line trip may have a similar characteristics. For example, the
46
Probability of the trip 0 Fmax k.Fmax
Figure 3.2: Probability of the line trip as a function of its loading. temperature of an overloaded line starts to increase, thus resulting in a sag of the line, which may nally be so severe that a ashover towards the ground or trees may occur and cause the line trip. It should once again be stressed that we do not intend to determine the probability of an initial event occurrence, but the focus is on the possible subsequent evolution. As mentioned in the previous section, severity of the contingency on the integrity of entire system may be determined by various types of studies or methods, and only these types of contingencies should be considered. Thus, we suggest that for example a single element contingency threatening only a single load, not a larger portion of a system, may be tolerated and thus removed from the security dispatch procedure. Proposed Method Similarly to N 1, we also focus on avoiding cascading by targeting its rst triggering stages. The main idea is to minimize the overall system cascading exposure expressed as a sum of highest probabilities for the possible post con-
47
tingency trips of subsequent lines. In this minimization a probability distribution of line trips as shown in gure 3.2 is assumed. This minimization can be mathematically formulated in the framework of Mixed Integer Linear Programming. However, post-contingency overloads are very rarely exceeding value k.Fmax (Which would typically represent 140 % line loading.), thus the computationally more ecient Linear Programming formulation can be used as follows:
Nc
min
Pg i=1
max{Wi .si }
(3.7)
|Fi | Fmax + si |Fnom | Fmax

Ng Nl
(3.8) (3.9)
Pgm =
m=1 n=1
Pln
(3.10)
0 si 0 Pg Pgmax Pg RNg W RNb Nb si , Fi , Fnom , Fmax RNb i = 1, 2, . . . , Nc
(3.11) (3.12)
(3.13)
where Nb is the number of branches and diagonal elements of W are weight factors corresponding to the slope of the characteristics shown in gure 3.2, which can be individually dened for each branch. Nondiagonal elements of W are zero1 . An element of vector si is a slack variable expressing the overload severity (when a post-contingency ows
1 This implies that the outage of a component is only a function of its own loading. However, this assumption should be further studied, since an overload of one component may impose a higher outage probability also on other components.
48
are lower than the line ratings, its value is 0) of the corresponding branch as a consequence of contingency i. In other words, the result of proposed optimization is a generation dispatch under a contingency list (which may include multiple contingencies).
3.1.8
Example
One zone of the Reliability Test System from [23], which is also described in the Appendix A.1, is chosen. A DC load ow modeling framework similar to the one described in [66] is used. First, we compute a generation dispatch for a maximum possible load, when the system complies with the N-1 criterion. When doing so, the power injection is restricted in such a way, that in each node, there can be either generation or load, not both of them. After obtaining the load values, we apply the proposed new dispatch method. The set of dangerous contingencies considered in the procedure has been chosen out of all possible double contingencies. As shown in gure 3.3, only small modications have been introduced. Then a Monte Carlo simulation to estimate the average blackout size after applying random double contingencies is performed. The probability distribution of the trip of other lines is used as displayed in gure 3.2, where we chose k = 1.4 as in [12]. When a load or generation is lost in the cascading process, remaining generators are scaled to match the load demand. As it can be observed in gure 3.4, convergence of the estimated average blackout is quite good. Its value is 3.8388 pu for the N-1 criterion and 3.6389 pu for the proposed new method. From gure 3.5 it is seen that the frequency of outages is smaller for the new dispatch method for all sizes of the blackouts. For the largest ones, i.e. lost power greater than 25 pu, the dierence is insignicant. For these large blackouts the initiating events have been so severe that a completely dierent approach had to be taken to save the system
3.1.9
Remarks
The value of the cost function given by equation (3.7) may serve as an index expressing overall system vulnerability. Theoretically, the best value would be 0, and the worst case Nc . This index could then also
3.2. Proposed Control Philosophy Essentials

15 Proposed dispatch N 1 dispatch 10
49
Injected Power [pu]
10 0
10 15 Node Number
20
25
Figure 3.3: Bus power injections. The negative ones correspond to the loads, positive ones to the generators. be used for transmission pricing or for congestion management. For example by evaluating how each transaction between a generator and a load contributes to the system vulnerability by increasing or decreasing this index. Relaxing some of the listed assumptions represents a future challenge in the direction towards the practical implementation of the proposed method. Main issues of interest are validation and clarication of the shape of the probability characteristics in gure 3.2, consideration of the voltage stability problems etc.
3.2
Proposed Control Philosophy Essentials
As previously outlined, control policies should be adjusted as much as possible to the actual system conditions and actual available controls and their cost. In addition, all relevant (That means for example excluding economical interactions not having a direct impact on the short-term system
50

10 9 Estimated Average Blackout Size [pu] 8 7 N 1 dispatch 6 5 4 3 2 1 0 0 0.5 1 1.5 Iterations 2 2.5 x 10 3
4
Proposed dispatch
Figure 3.4: The convergence of Monte Carlo simulation estimating average blackout size.
0.25
Cummulative Distribution Function
0.2
0.15
N 1 dispatch
0.1 Proposed dispatch 0.05
0 0
10
15 20 25 Lost Power [pu]
30
35
40
Figure 3.5: A distribution of the blackout sizes after convergence of Monte Carlo simulation.
3.3. Proposed Control Structure
51
security.) types of interactions in power systems should be considered. The next chapter proposes a particular control method following this control philosophy and the next section describes how such control philosophy could be implemented.
3.3
Proposed Control Structure
Features of the control philosophy presented in the previous section put very high demands on the control performance and thus on the control structure, which should accommodate them. Note that interconnected power systems are complex large-scale systems featuring: continuous dynamic states; discrete states; dynamics of various time scales (from milliseconds to years); nonlinearities; interactions between above four phenomena; interactions between various parts of the system. To capture all types of interactions, a fully centralized control structure, supervising a complete interconnected system, would be a natural choice. However, there are many other obstacles to implement a fully centralized power systems control, in particular: 1. legislation - interconnected power system may cover countries with dierent laws and operation rules 2. competition and conict of commercial interests between the involved parties 3. reliability of all involved components 4. robustness - vulnerability of the centralized scheme to the outage of one of its components 5. willingness to share the information
52
Chapter 3. Improvements in Power Systems Security 6. cost 7. technology performance limitations: computation power communication infrastructure dierent communication and computation standards
Except in some isolated systems, there is today no such fully centralized control scheme supervising interconnected power system. In particular, it is very dicult to imagine a single agent supervising for example the entire European interconnection (including part of North Africa and Eastern Europe). Thus, we propose the control structure consisting basically from the same layers as today hierarchical control: 1. Primary Control 2. Secondary Control 3. Tertiary Control However, we suggest to redene the roles of layers, their mutual relations as well as level of fragmentation (This refers to the secondary layer.). Note, that the complete discussion, which follows, refers to a complete interconnected power system. That means, for example in European conditions, UCTE system consisting of many national power systems interconnected via tie-lines. Similar ideas have naturally already appeared in a literature, e.g. [25]. Primary Control shall keep its present form of tracking local quantities according to the reference values provided by the secondary layer. Secondary Control shall be performed from area controllers. The size of an area should correspond to a region. In geographical sense a region within a country. That means closer to the voltage control secondary layer in contrast to present frequency secondary control, where the supervised area usually corresponds to a country within the interconnected system. Note that this discussion refers rather to the European context. An area controller shall be capable of addressing both normal and emergency conditions. An area controller would fulll the tasks of secondary
3.3. Proposed Control Structure
53
frequency control, voltage control and avoidance of cascading spreading of components overloads. Mutual coordination of neighboring areas would be used to redistribute the tertiary voltage control functionality and spreading of cascading. In a case of very severe disturbance, a possible control action would be fragmentation of the secondary layer into individual areas, which would be later resynchronized. The role of Tertiary Control would be signicantly reduced to handle exclusively tasks, which are very dicult to redistribute. Explicitly: Coordination of tuning of damping controls Tertiary frequency control Electricity market functionalities It is important to stress, that even in the case of addressing issues related to coordination (especially of the secondary layer) in a very ecient manner, it will probably still be very dicult to completely avoid the specic dynamics introduced by mutual interaction of secondary layer controllers. In the economy context of transmission pricing, this control structure may actually support the concept of zonal pricing, where zones could be associated with secondary control areas, i.e. pricing based on controllability, which would be especially convenient for pricing of auxiliary services, e.g. reactive power support. The value of the power to be exchanged between two zones could be determined by an iterative procedure similar to the decomposition methodology of Optimal Power Flow, described for example in [57] or [73].
Chapter 4
MPC Employing Trajectory Sensitivities

This chapter starts with the description of a classical Model Predictive Control formulation and outlining its complexity when applied to large nonlinear systems. Reduction of complexity by use of trajectory sensitivities and their use in Model Predictive Control are then described.
4.1
Introduction - Problem Description
This chapter deals with the topic of a single area controller as a part of the concept discussed in the section 3.3. Requirements on such an area controller are quite strict, as outlined in the section 3.2. Both normal as well as emergency conditions should be automatically detected and addressed. Nonlinear behavior should be appropriately captured. A suciently high performance (i.e. fast control decision making process) shall be provided to allow acting under emergency conditions. Traditional control schemes treat the dierent types of the problem separately, e.g. frequency control, voltage control etc. The area controller shall treat them simultaneously. Optimal control framework seems to t the above requirements best, in particular Model Predictive Control. 55
56
Chapter 4. MPC Employing Trajectory Sensitivities
Model Predictive Control (MPC) is a method, which determines a control law by solving an optimization problem, whose formulation explicitly incorporates control quality criteria and constraints imposed on system inputs, outputs and states, which is a very attractive feature. The eect of controlled inputs on the system behavior (i.e. states and outputs) is predicted by using a model of the system. However, drawbacks of MPC are its full dependence on an accurate system model and computational burden associated with the solution of the optimization problem. The rst obstacle can be overcome by introducing an implicit feedback in form of repetitive computation of control laws in a receding horizon manner, what has been proven for innite horizon control. The second obstacle has restricted the application of MPC mostly to control of slower processes, with the dynamics in order of minutes (e.g. chemical industry), or to smaller systems. This is also probably the factor limiting a wider spreading of MPC in power systems up to now. Power systems are large systems, comprising many components (thus number of variables is very high), which interact with each other in various ways and in dierent time scales, i.e. dynamics speed ranging from milliseconds to years. Dynamics present in power systems may be very nonlinear and have a hybrid nature, i.e. both continuous and discrete state variables are involved. All these aspects complicate the employment of MPC in power system. Several authors have addressed the diculties to apply MPC in power systems. Reference [50] analyzes Euler nonlinear and linear approximations with numerical computation of sensitivities to controlled inputs for prediction of the power system behavior. Reference [29] has proposed mixed Logic Dynamics (MLD) (A comprehensive explanation can be found in [7].) framework for dealing with the hybrid dynamics in MPC of power systems. A signicant acceleration of MPC computations has been reported in [6]. Coordinated secondary voltage control applied in the western part of France by EdF and reported in several publications (e.g. [70]) may probably be considered as the rst industrial implementation of MPC in a real power system. Although it is not referred to as MPC in the literature, it possesses its features, i.e. continuous repetitive employment of optimization for computing voltage set points of generators in the supervised region.
4.2. Model Predictive Control Fundamentals
57
Another work, possessing features of MPC based voltage control has been described in [16]. Our goal is to describe how an employment of trajectory sensitivities for the prediction of the system behavior and an eect of controlled inputs can help to signicantly reduce the computational burden, while keeping a good reproduction of the system nonlinearities. By this, a further use of MPC for control applications in power systems would be possible. The chapter starts with a short overview of MPC and computation burden reduction techniques. Then we introduce a formulation of MPC based on trajectory sensitivities, which is described further in the Appendix. After that simulation results for voltage control of a test power system are presented.
4.2
4.2.1
Model Predictive Control Fundamentals

General (Nonlinear) Model Predictive Control
In a general system (no restrictions concerning the type of dynamics etc.), MPC computes a sequence of control inputs, which yield an optimum (in most cases a minimum) of the objective function, while satisfying constraints imposed on the system inputs, outputs and states. The objective function usually comprises several parts and represents a trade o between the quality of the control process and the eort needed to enforce it. The quality of the control process is usually expressed as a dierence between the desired and actual value of the output (when a tracking problem is considered). The control eort is the deviation of control inputs from their original values and/or their optimal values. Similarly to [27] or [58], this can be formulated in the continuous time domain for a general case, applicable also to nonlinear systems, as follows: min (x(t), y(t), u(t)) (4.1)
u(t)
subject to equality constraints (usually describing system dynamics): x(t) = f (x(t), y(t), u(t)) (4.2)
58
Chapter 4. MPC Employing Trajectory Sensitivities 0 = g(x(t), y(t), u(t)) (4.3)
and inequality constraints: 0 h(x(t), y(t), u(t)) (4.4)
where x(t) are dynamic states, y(t) are algebraic states and u(t) are manipulated inputs within the control time interval (horizon) of the length Th , thus t [t0 , t0 + Th ]. Furthermore, all types of variables may be nonlinearly bounded by spaces X , Y and U, respectively: x(t) X y(t) Y u(t) U (4.5)
The solution of the above problem at the time t0 is an optimal input sequence u (t). Since the above formulation fully relies on the system model and the system state at the time t0 , an actual response of the system to the computed control sequence can dier due to the model imperfections and disturbances present in the system. Therefore applying the initial part of control inputs and then solving MPC formulation again introduces a feedback. This procedure will be repeated at regular time intervals.
4.2.2
Approaches to Reduction of Complexity of General MPC Formulation
The complexity of solving (4.1) - (4.5) plays a crucial role in the feasibility of a control scheme based on MPC. There are several factors inuencing it: Complexity of the cost function (4.1) Complexity of the boundaries (4.4) and (4.5) Size of the controlled system (i.e. number of variables in the system model) Complexity of the model behavior, e.g. internal structure, type of interactions within the model etc., expressed by constraints (4.2) and (4.3) Length of the prediction horizon Th
4.2. Model Predictive Control Fundamentals
59
The length of the prediction horizon Th should ideally be innite, but it is practically impossible, so it is one of the design parameters of the MPC to be considered in the tradeo between the accuracy and the computation time of the controller. Practically Th is chosen to be suciently long to capture the slowest system dynamics to be controlled. The cost function itself has usually a simple form of either a linear or a quadratic function; boundaries are usually constant. However, both cost function and boundaries are coupled to the system behavior description, i.e. (4.2) and (4.3). Thus, the most decisive factor is the system dynamics. Many ecient, powerful and robust computation techniques have been developed for linear systems, so in such a case, even a larger system size does not represent a major problem. But often nonlinear and/or hybrid systems have to be controlled. In that case, solving (4.1) - (4.5) may represent a challenge. Therefore signicant eort has been invested into the research of techniques allowing reproduction of nonlinear system behavior with a minimal computational eort. The rst option is the use of a full nonlinear model and computations, which is of course most accurate but also the most time consuming approach. Therefore this is normally applicable only either for small sized systems or slow processes, where the slow dynamics implies that long computation times can be tolerated. The second frequently used method is to linearize the system equations around the present operating point (equilibrium) and apply the linear MPC. Discrepancy between the linearized model and the actual system behavior is then compensated in the next controller step. However, this involves a risk that the system may undergo large excursions from the optimal trajectory and even violate imposed constraints. The third approach is based on the approximation of the expected trajectory (Euler prediction) of the system if control inputs would remain unchanged and the numerical (A numerical computation of sensitivity to a change of a parameter is obtained by dividing the dierence of trajectories with and without considering an incremental change of the parameter by the incremental change of the parameter.) computation of sensitivities of control inputs impact on this trajectory. A linear Euler prediction has similar properties as the ones described in the previous paragraph and is based on the strong assumption of monotonous time evolution of the system trajectory. Nonlinear Euler prediction is much more accurate. However, a numerical computation of sensitivities then
60
means repetition of the system trajectory prediction as many times as the number of control inputs.
4.3
Proposed MPC Formulation
The proposed approach is closest to the last one described in the previous section. Let us assume a discrete time control. The controller would then be employed (i.e., MPC optimization problem would be solved) in the regular time intervals of Tcs with the outlook of the prediction horizon (i.e., the length of the sliding window) of Th . Assume that the control should be executed at the time t0 . Then the control method consists of three stages: 1. A nominal trajectory of the system is predicted for the time interval [t0 , t0 + Th ] by employing the discrete time form of (B.1) (B.6) with the integration step of Tps . This may be perceived as a time domain simulation. As a byproduct, trajectory sensitivities are computed with a little additional eort according to [36] and as explained in the Appendix. 2. Controls are computed according to the equations (4.6) - (4.13) and the description in the next paragraphs. 3. The rst set of computed controls is applied at the time t0 (neglecting the time delay). Note, that although we use a discrete time framework, we do not explicitly mention a particular number of samples, but rather time instants. This is due to possible discrete events. To capture the discrete events properly, two samples at the same time instant are necessary. Moreover, it may happen between two regular samples, thus the time of the discrete event occurrence has to be computed as well and the integration step Tps is divided accordingly. In most cases, the integration step has a relatively small size to capture dynamics properly. Therefore Tps Tcs Th . Using a quadratic cost function, the control objective can be expressed by (4.6). The rst term represents a cost of the deviation yk of the system output from the reference value, so the corresponding algebraic variable is dened in the model description in the equality constraints
4.3. Proposed MPC Formulation
61
(4.7). The third term penalizes the control employment, whereas the second term the deviation of the control from its initial value at the time t0 . Note that all control deviations u refer to the initial point and not to the previous sample. However, a penalty on the deviation of the control from its previous sample value can easily be included by an appropriate choice of the non-diagonal elements of the matrix Rdu . 1 min U 2
Nps 1 T {yk Qyk + uT Rdu uk + uT Ruk } k k k=0
(4.6)
yk uk
nom = yk +
yk uk uk = unom + uk k
max yk umax k umax k 1
(4.7) (4.8) (4.9) (4.10) (4.11) (4.12) (4.13) (4.14)
min yk yk umin uk k umin uk k N
ps U = {uk }k=0
uk RNu yk RNy
uk RNu k K
yk RNy Nu uk K = {0, , Nps 1}
Nu is the number of available controls (manipulated inputs), Ny number of controlled variables (i.e., tracked outputs) and Nps number of samples in the prediction horizon (as discussed earlier, it is not necessarily constant). Unless employed controls change signicantly, trajectory sensitivities reproduce the system behavior quite accurately, even considering nonlinear dynamics. However, it is very dicult to bound a region, in which the changes can be considered reasonably small. Another possible source of errors may be certain type of discrete events. A discrepancy between the model and the actual system response can be corrected in the receding horizon manner. The above dened controller uses a quadratic cost function and hard constraints. But this is not a limitation of the method. Soft constraints may be introduced as well as slack variables and the whole problem can be restated in a form of linear programming.
62
4.4
Voltage Control Example
We will now present two examples. The rst one explains the principle of MPC based on trajectory sensitivities in a very simple system. The second example shows a particular application for which the proposed control would suit well.
4.4.1
Illustrative Example
Consider the system described in the Appendix A.2. The role of the controller is to keep the voltage in node 3 in the operation range (i.e. above 0.85 pu) by applying smallest possible load shedding (in node 3). Load shedding is expressed by the factor k, which expresses which percentage (0 means no load shedding at all, 1 represents disconnection of entire load) of the initial value of the load is disconnected. Note that transformers tap changer is controlled by its corresponding independent controller, which is included in the system model, but not under supervision of the MPC controller discussed here. Let us assume that the controller is activated at the time t0 as shown in Figure 4.1. First the voltage magnitude in node 3 is predicted if no control change from its present value (i.e. in this case load shedding, present value of the load shedding is zero) is applied. Simultaneously, trajectory sensitivities are computed using parts of the Jacobian evaluated during the prediction of the voltage evolution. Since the voltage drops under the minimal allowed value, MPC following formulation (4.6) - (4.13) computes the necessary load shedding to avoid it, see Figure 4.3. The new trajectory predicted by the MPC controller would be as shown in Figure 4.1. The controller predicts the nominal trajectory (i.e. the one without any control input changes) and computes the corresponding trajectory sensitivities for the time horizon of Th equal to 28 seconds, which is divided into two intervals Tcs (equal to 14 seconds). Within each of them control inputs remain constant. The integration step Tps (prediction sampling), which is used for computations, is 0.5 second. Figure 4.2 shows the relationship between all described time intervals. Use of time constants instead of the number of samples (as it is a well established practice in discrete time control) can be also explained with the help of Figure 4.2. The transformer tap changes its position (according to the
4.4. Voltage Control Example

1.02
63
1 actual output 0.98
0.96
0.94 Voltage [pu] predicted output with control 0.92
0.9
0.88
0.86 minimum voltage 0.84 predicted output without control 0.82 0 5 10 t0 15 20 Time [s] 25 30 35 40 45
Figure 4.1: Controller prediction at the time t0 .
prediction) in the beginning of the second control interval Tcs . It is a discrete event and it requires two samples at one time instant to capture this behavior appropriately. Therefore the second control interval has one more sample than the rst one. Although two control intervals have been used, only the control inputs values of the rst one are applied at the time t0 (neglecting the time delay). At the time t0 + Tcs the entire described procedure with the prediction and computation of necessary control inputs changes is repeated as shown in Figure 4.4 and Figure 4.5. As Figure 4.6 shows, even when no external disturbance takes place, the system response diers slightly from the predicted one. Therefore also the value of the optimal control inputs to be applied at the time t0 + Tcs may dier from the one envisioned for this time instant at the time t0 , see Figure 4.5. The same procedure is then repeated at time instants corresponding to multiples of the control intervals Tcs . Thus an implicit feedback control is introduced. The nal voltage magnitude trajectory as well as the load shedding factor after the decay of the transient voltage drop is plotted in Figure 4.7.
64
0.92 Tps 0.9
Voltage [pu]
0.88
0.86
0.84 Tcs 0.82 Th Tcs
15
20
25 Time [s]
30
35
40
Figure 4.2: Overview of time constants of the controller.

0.25
0.2
computed control
Load Shedding Factor []
0.15
0.1
0.05 actual control
10 t0
15
20 Time [s]
25
30
35
40
45
Figure 4.3: Necessary load shedding as computed by the controller at the time t0 .

1.02
65
1 actual output 0.98
0.96
0.94 Voltage [pu]
predicted output with control adjustment
0.92 predicted output without control adjustment
0.9
0.88
0.86
0.84 t0 + Tcs 0.82 0 10 20 30 Time [s] 40 50 60
Figure 4.4: Predicted trajectories at the time t0 + Tcs .

0.25 computed control at t0
0.2
Load Shedding Factor []
0.15
actual control 0.1 computed control at t0 + Tcs
0.05
t0 + Tcs 0 0 10 t0 20 30 Time [s] 40 50 60
Figure 4.5: Update of the value of the optimal load shedding at the time t0 + Tcs .
66
0.92
0.91
0.9 Voltage [pu] actual output
0.89 predicted output with control
0.88
0.87
0.86 t0 15 20 25 Time [s] 30 35 40
Figure 4.6: Deviation of the actual trajectory from its predicted value even under the condition of no disturbance and with using an ideal model.
1.05
1 Voltage [pu]
0.95
0.9
0.85
20
40
60
80 Time [s]
100
120
140
160
0.2 Load Shedding Factor []
0.15
0.1
0.05
20
40
60
80 Time [s]
100
120
140
160
Figure 4.7: Final voltage trajectory and applied control (i.e. load shedding).

Generated Reactive Power [pu] 45 40 35 30 25 20 15 0 5 10 Time [s] 15 20 25
67
1.3 Bus Voltages [pu] 1.2 1.1 1 0.9 0.8
voltage in bus 14 maximum voltage
voltages in buses 15 and 16 minimum voltage 0 2.4 5 line trip 10 Time [s] 15 20 25
Figure 4.8: Controlled system outputs. The test system is subjected to a severe disturbance. The upper plot represents the sum of absolute values of reactive powers produced by all generators. The lower plot shows buses voltage magnitudes.
4.4.2
Application to Realistic System
Voltage control (sometimes also referred to as a reactive power management) in power systems has several goals. They can be divided into two categories: economy and security oriented. From the long-term economical operation point of view, it is most desirable to avoid frequent under- or over- voltages, which increase aging of the equipment, and to keep the losses generated in the components low. This can be achieved by decreasing owing current values (most of the losses are resistive: R.I 2 ) by keeping the voltage prole high. Other type of economic objective of voltage control is to maximize power transfers constrained by voltage collapse risk. From the security point of view, the main concern is a danger of encountering a voltage instability and resulting voltage collapse. Here the defense strategy is usually dependent on the system conditions. When
68
Generators Voltage References [pu]

1.15
1.1
1.05
1 generator 7 0.95 0 5 10 Time [s] 15
generator 6
20
25
0.01 Load Shedding Factor [] 0.008 load in bus 3 0.006 0.004 0.002 0 load in bus 1
10 Time [s]
15
20
25
Figure 4.9: Controls computed and executed by the controller. The test system is subjected to a severe disturbance. The upper plot shows reference voltages for generators AVRs. The lower plot shows load shedding factors of all loads, closer explained in the text. the system is in emergency state and voltages signicantly drop, undervoltage load shedding relays trigger a disconnection of some loads. In normal operation conditions, the focus is on keeping system robust and to strengthen its ability to withstand a disturbance (e.g. a line or a generator trip), which could possibly cause a voltage collapse. So it is desirable to control the system in a way keeping large reactive power reserves of generators (in other words, keeping a low reactive power production of generators). Consequently there should be a sufcient amount of reactive power available to avoid voltage problems in the case of a disturbance (consult [47] or [19] for explanation of coupling between the reactive power and voltage). Consider the 26 bus test system described in the Appendix A.3. Tracked variables/outputs are voltages in all buses and reactive power production of all generators. Possible controls are reference voltages of all AVRs and load shedding of all loads. The load shedding factor

Generated Reactive Power [pu] 20 19 18 17 16 15
69
10 Time [s]
15
20
25
1.1
Bus Voltages [pu]
1.05
0.95
10 Time [s]
15
20
25
Figure 4.10: Controlled system outputs. The test system is under normal conditions. The upper plot represents the sum of absolute values of reactive powers produced by all generators. The lower plot shows buses voltage magnitudes. of each load can have a value between 0 (no load shedding at all) and 1 (complete load is shed) and determines to which extend the load is disconnected. The voltage boundaries are 0.85 pu (per unit, referred to the nominal voltage equal to 1 pu) and 1.15 pu, respectively. Limits imposed on the reference voltages (i.e. controls, manipulated inputs) are 0.9 pu and 1.1 pu, respectively. Note that voltage boundaries both of the outputs and inputs are usually narrower in reality. The goal of the controller is to continuously supervise the power system and minimize the absolute value of reactive power production of all generators within its supervised area. Under normal conditions, this should be achieved only by selecting the most suitable reference voltages for AVRs. Only when the system is seriously endangered, load shedding can be employed to recover the system into the acceptable operation region. The controller parameters have been chosen accordingly to serve this purpose. The nominal prediction sampling (i.e., integration step) Tps is 0.5 second, prediction horizon Th is 10 seconds and controls sam-
70
Generators Voltage References [pu]

1.15 generator 2 1.1
1.05
1 generator 7 0.95 0 5 10 Time [s] 15 20 25
0.01 Load Shedding Factor [] 0.008 0.006 0.004 0.002 0
10 Time [s]
15
20
25
Figure 4.11: Controls computed and executed by the controller. The test system is under normal conditions. The upper plot shows reference voltages for generators AVRs. The lower plot shows load shedding factors of all loads, closer explained in the text. pling Tcs is 5 seconds. The controller follows the form dened by (4.6) - (4.13). To show the controllers behavior, we have chosen two dierent scenarios. First the system is subjected to a severe disturbance, which is disconnection of the line between the buses 3 and 16. This could be the consequence of an earth fault, caused by lightning hitting the line. As it is shown in gure 4.8, the line outage triggers an oscillation driving voltages outside of the secure operation region. Therefore, the controller orders load shedding at buses 1 and 3 in addition to the adjustment of generator voltage reference set points. The latter control action would be insucient and too slow to save the system from more severe consequences, see gure 4.9. In the second case, the system starts from the same initial conditions, but there is no disturbance. Therefore the employed controls shown in gure 4.11 aim at the decrease of the reactive power produced by
71
generators as shown in gure 4.10. Explicitly, only generators controls are activated (the top part of gure 4.11) while load shedding is not employed at all (the bottom part of gure 4.11 has thus all waveforms equal to zero).
Chapter 5
Benets of Better Power Systems Control

Introduction of a real-time optimal supervisory control in power systems may minimize severe consequences of disturbances and thus increase security. Controlling FACTS devices in a coordinated manner ,within the real-time optimal control concept, implies an increase of economical eciency by better distribution of line ows (i.e. decrease of dierences in Locational Marginal Prices). However, these benets come with a price representing cost of such a control system (both acquisition and running cost, i.e. installation and maintenance) and keeping a certain level of available reserves and their activation for the control. Even when a massive amount of reserves are available, a complete avoidance of load loss (in form of load shedding) is not possible. Therefore a compromise in the amount of reserves and controllability given by them has to be looked for. This chapter elaborates on the issues outlined above.
5.1
Introduction
Several authors have provided reasoning of benets of optimal control applied in power systems, e.g. with respect to secondary voltage control [70] and [63]. These studies have dominantly focused on a single typical 73
74
Chapter 5. Benets of Better Power Systems Control

3000
2500
System Load [MW]
2000
1500
1000
500
1000
2000
3000
4000
5000
6000
7000
8000
9000
Time [hour]
Figure 5.1: Example of the system load (sum of all active power consumptions) variation during one year, based on the data from reference [23]. situation in which the proposed optimal control concept has clearly showed better performance. However, power systems are subjected to permanently varying conditions. As discussed in chapter 2, major causes for these are: Load and generation variation in the magnitude, location and character. This is today an especially relevant issue, as section 2.3 describes. Topology changes both planned ones (i.e. scheduled maintenance) and unplanned ones (i.e. contingencies). Figure 5.1 shows the variation of the system load (i.e. sum of all load consumed active powers) of the RTS system described in Appendix A.1 during one year. This gure has been constructed based on the data available in the reference [23]. A histogram showing frequency of occurrence of certain load levels is constructed and shown in gure 5.2. We used 20 system load levels,
5.1. Introduction
900
75
800
700
600
500 [hours]
400
300
200
100
0 500
1000
1500
2000
2500
3000
Load Levels [MW]
Figure 5.2: Histogram grouping the system load values from the gure 5.1 into 20 categories. which means that the load values are redistributed in 20 levels within the range between minimum and maximum load values that occur during one year. Majority of studies focus only on one scenario, as mentioned, and often that scenario corresponds to the peak system loading. As the histogram shows, in this particular case it represents only a very small fraction of the time. Therefore to accurately capture and assess the impact of a control, ideally all conditions should be considered during the entire expected lifetime of the control system. But this is not possible because of: Dimensionality of the problem is enormous, given by the combination of the number of power system describing variables (i.e. states) and parameters, their possible values and the time frame. Uncertainties related to both the variety of the system states and their frequency of occurrence. Again, recent trends play a signicant role here, see section 2.3. Thus, to achieve a feasible assessment framework providing a fair qualitative estimate, several justiable simplifying assumptions have to be
76
introduced: Control system observes and controls the entire power system. This implies that each dynamics and dangerous state of the system can be identied and addressed. Unavailability of the control system is neglected. The control system is always at the disposal to determine and execute necessary control actions, in other words, the control system is 100% available. Probability of the control system missoperation is neglected. This means that the control system behaves exactly as intended. Fast transients are neglected. This assumption reduces the dimensionality of the analysis by focusing on the phenomena having slower dynamics, such as cascading spreading of network branches (i.e. lines and transformers) overloads and static voltage collapse. Power system structural changes are excluded. This means that no system extensions are assumed during the assessed operation period. However, this assumption could be relatively easily relaxed and the known system modications could be included in the assessment. Only the magnitude of the load variation is considered. Locational variation of the load is assumed to have a negligible impact as well as the changes of the load character (e.g. impedance load to the constant power load etc.).
5.2
Economical Benets of Control
Economical benets of the system wide application of the proposed control concept can be derived from increase of power ow from areas with the cheaper electricity production to the areas with the high electricity demand, while keeping acceptable security standards. In the framework of deregulated electricity markets this implies reduction of the transmission restrictions (i.e. congestion) and increase of market eciency by decreasing dierences in locational marginal prices.
5.2. Economical Benets of Control
77
A powerful way to attain the above objective of an increase of controllability in a transmission system is to utilize FACTS (Flexible AC Transmission System) devices. Some FACTS devices aim at the introduction of damping in a power system but most of them focus today on the control of power ow and reactive power support under steady state conditions. However, both concepts are based on the use of local measurements to follow a predened reference value (voltage magnitude, line power ow etc.). Several publications (e.g. [59]) have recently appeared, proposing control of FACTS based on remote measurements - wide area control. This concept oers several advantages, such as coordination of the FACTS device with other types of controls, coordination of several FACTS devices, considering entire system conditions etc. This section uses a particular power system example to show economical benets of such a wide area control of a FACTS device.
5.2.1
Methodology and Modeling
The qualitative assessment of possible economical benets can be done by studying a situation when the power system consists of two or more zones with dierent generation price levels. The three dierent situations can then be considered: Power system is operated without FACTS devices. Power system is operated with only one FACTS device controlled using local measurements Power system is equipped with one FACTS device and a wide area controller. We assume that the generation pattern is initially constant. The further analysis and calculations, based on the power ow and sensitivity computations, are then performed for each load level within the spectrum between the minimal and maximal system load (for example 20 load values in the case shown in gure 5.2) and importance of the results can be weighted with the distribution coecient (i.e. hours). We have considered the following ow control FACTS devices in this study: Controllable Series Capacitor (CSC) and Phase Shifter (PS).
78
These are the most common devices of the FACTS family, besides Static Var Compensators (SVC). The practical implementations of these devices could be with traditional switches, Thyristor Valves, Voltage Source Converters or combinations of them. The basic functionality varies with the speed and resolution of the controllable steps. It is assumed that generation in zone A is much cheaper than generation in zone B. Therefore the most economical way to operate the system is to cover the load demand in the entire system from zone A as much as possible. However, this policy can be applied only to a certain extent since there are restrictions: Dierence between the generation in zone A and system load, i.e. if there is not enough generation to cover the whole system load, the generators in zone B have to be employed as well. System security concerns. These are in most cases expressed in terms of N-1 criteria, i.e. no system element shall be overloaded after an outage of any other system component. This rule is however interpreted in many ways; we assume that it refers to the line short-term emergency limits. These limits can be approached only for 15 minutes, and then the level of loading has to be alleviated. In reality, system security concerns are dominating in restriction of power transfer capacity from one zone to another zone via a transfer corridor (a corridor may consist of several tie-lines interconnecting the zones) between them. Finding the best location for the placement of a FACTS device aiming at enhancement of a corridor transfer capacity is a quite complex task. A majority of proposed methods, which can be found in the literature, rely on various sensitivity coecients, which point at the locations where the FACTS device would provide the best controllability. A majority of the proposed methods focus on the peak loading case only. We suggest, that these two assumptions might be inadequate because: a.) If the identied location (i.e. line) is the element, which outage would cause the overload of other element, i.e. limiting factor according to N-1 operation criteria, then the FACTS device could not be employed, because its outage would immediately mean a cascading failure. b.) As shown in gure 5.2, the peak loading conditions may occur very rarely, therefore it might be much more eective in terms of energy to optimize the performance of the FACTS device and the transmission corridor for
79
a lower loading. Therefore instead, it is proposed to select the location for the FACTS device after the rst part of the following comparison evaluation procedure. We propose an evaluation procedure for comparison of the power system performance in steady state conditions with and without FACTS as follows. For each load level it is computed: Minimizing the power generation in the zone B, while keeping N-1 security constraints. No FACTS device. This is equal to a basic energy market simulation. Minimizing the power generation in the zone B, while keeping N1 security constraints. The setting of the FACTS device must be the same for all considered post-outage states. Minimizing the power generation in the zone B, while keeping N1 security constraints. The setting of the FACTS device can be dierent for each considered post-outage state. In the last case we assume that the wide area control could modify the setting immediately after the contingency. An uniform employment of all generators is used as a starting point with the steady state sensitivities with respect to it. More details about the computations can be found in Appendices C.1 and C.2.
5.2.2
Simulation Results
The principles are demonstrated on the already mentioned test system described in the appendix A.1. We divided it into two zones as shown in gure 5.3. The northern zone A has dominantly generation character, whereas the southern zone B represents mainly consumption. The load values and variation are modeled according to [23], whereas the generation is computed as explained in the previous section. As the location for the FACTS device we have chosen line 22 connecting the nodes 13 and 23. We assume that the installed CSC would have a rating corresponding to 50 % increase or decrease of the reactance of the line where it is inserted. The installed PS would have a range from -20 to 20 degrees. Note that only one FACTS device is installed at the time. Figure 5.4 shows the situation when:
80
Zone A
G18
18 21
G21
G22
22
17
G16
16 19 20
G15
15 14
G14
FACTS device
G23
23
Zone B
G13
13 24 11 12
10
2 7
G1
G2
G7
Figure 5.3: Single line diagram of the IEEE Reliability Test System - 1996. Buses supplying connected loads are in the black color. Busses in the grey color contain no load. The system is divided to zones A and B respectively. The FACTS device is located between nodes 13 and 23.

1800
81
1600
1400 Transmitted Power [MW]
1200
1000
800
600
Export from Zone A to Zone B Available Power for Export from Zone A Zone B Demand
400 500
1000
1500
2000
2500
3000
System Load [MW]
Figure 5.4: Power transfer conditions unrestricted by the transmission system capabilities. The power generation in Zone A can entirely cover the consumption of the whole system up to 2200 MW (x axis). For larger values the generation also in Zone B has to be employed. There are no restrictions given by the transmission system. The amount of power to be exported from Zone A to Zone B is limited only by the generation capabilities in Zone A. An available power for export from Zone A is determined as a dierence between the installed generation capacity in Zone A and the power demand in Zone A. Applying the data from [23], we have obtained the result as shown in gure 5.4. However, this type of situation occurs very rarely in reality. The transmission system capabilities limits usually play a signicant role giving rise to congestion management. Therefore we have modied the limit of the four lines (down to 60 % of their original value, if not stated otherwise), which create the transfer corridor between the zones A and B. The computed transfer capacity values are shown in gures 5.5 and 5.6.
82

1800
1600
1200
1000
800 Export without CSC Export with CSC Available Power for Export from Zone A Zone B Demand
600
400 500
1000
1500
2000
2500
3000
System Load [MW]
Figure 5.5: Power transfer with and without CSC.
It can be observed, that the FACTS device brings the transfer capacity closer to the case unrestricted by transmission system as in gure 5.4. The increase of transfer capacity brought by the FACTS devices is quite signicant and amount of additional energy that can be transferred per year is calculated:
Nl
E=
i=1
Li .(TiF ACT S T i)
(5.1)
Where Li is the number of occurrence in the load level category i out of N l load levels . Ti is the transfer capacity without FACTS device and TiF ACT S with the FACTS device. For CSC E is 5.06 105 MWh (which is 6.77 % increase) and for PS it is 1.31 105 MWh (1.75 %). In this particular example the wide area control of FACTS brings only a small, almost negligible benet (only a few megawatts in addition to locally controlled FACTS device), explicitly no increase at all for CSC and for PS it is 1.35 105 MWh (1.81 %).
5.3. Security Benets of Optimal Control

1800
83
1600
1200
1000
800 Export without PS Export with PS Available Power for Export from Zone A Zone B Demand
600
400 500
1000
1500
2000
2500
3000
System Load [MW]
Figure 5.6: Power transfer with and without PS.
5.2.3
Remarks
It is a quite common situation that a power system consists of several zones with dierent characteristics (economical reasons-price dierence, technical reasons-distance between the hydro power sources and residential areas etc.), which give a rise to the heavy power transfers over certain lines/corridors connecting them. But very often these transfer corridors are loaded ununiformly due to the lack of controllability over the line power ows. Increase in controllability introduced by a FACTS device can therefore represent a better utilization. When analyzing the FACTS device impact under normal (undisturbed) conditions, the focus is on the power transmission capabilities that may be expressed in terms of MW or MWh, whereas security is the main measure of interest in emergency conditions, which is studied in the next section.
5.3
Security Benets of Optimal Control
If we assume that the FACTS device controlling the ow can be controlled quicker, security can signicantly benet from it. This section
84
shows how this can be demonstrated on the power system conguration from the previous section.
5.3.1
Since we used the N-1 criterion for the security constrained generation dispatch, we assume that the system can be endangered only by the outage of two or more elements, i.e. by double or multiple elements contingencies. Note that the scope is not to study the probability of their occurrence, only their consequences. Therefore only illustrative contingencies can be selected (system dependent) and the following defence strategies can be demonstrated, again, for each loading level and corresponding generation dispatch pattern obtained in the previous section: Power system is not equipped with any FACTS device. Line overload protection relays shall trip the overloaded lines if such a condition occurs. There is no coordination between the protection relays. Power system is equipped with one FACTS device. Line overload protection relays shall trip the overloaded lines if such a condition occurs. There is no coordination between the protection relays. Power system is equipped with one FACTS device and wide area control, which may modify the generated active power, FACTS device setting and initiate a load shedding in order to avoid a progression of cascading, i.e. subsequent overloading of other lines. Modelling details are given in Appendix C.3.
5.3.2
Simulation Results
For the emergency conditions we assume that the locally controlled FACTS device uses as a reference the power ow in the line where it is installed. As shown in the previous section, the installation of a FACTS device signicantly increases the utilization of the lines in the corridor. But this is achieved by the modication of the line ows pattern yielding a higher vulnerability of the power system to contingencies as shown in gure 5.7. However, when the wide area control is applied as a response to a contingency, this risk is minimized. It can be observed that when
5.3. Security Benets of Optimal Control

3000 No FACTS device, Local Protection PS, Local Protection PS, Wide Area Control
85
2500
2000 Unserved Power [MW]
1500
1000
500
0 500
1000
1500
2000
2500
3000
System Load [MW]
Figure 5.7: Load that cannot be served after the actions of local relays or wide area control employing PS. Initial event is tripping of the lines 19 (from node 11 to 14) and 27 (from node 15 to 24). the FACTS device is available as a control means for the wide area control, it is very eective.
5.3.3
Remarks
Based on the last two sections, we suggest that: Installation of the FACTS device can signicantly increase power transfer through the corridor. This can be also quantied for example for one-year period, by e.g. the presented method. If the FACTS device follows only locally determined reference value, it may even contribute to the cascading process and thus decrease the security of the system. If the FACTS device is controlled and coordinated with other controls as a part of the wide area control system, it strengthens the security.
86
Chapter 5. Benets of Better Power Systems Control Fast response of FACTS devices is required for the post contingency coordinated control. Traditional devices like Phase Shifting Transformers would be very limited in these cases.
Statements and suggestions are supported by the simulation examples. We believe the wide area coordination and control of several FACTS devices brings even more benets both in the security and transfer capacity sense.
5.4
Considerations
In previous sections studying FACTS devices we made several assumptions, e.g: The participation of the generators on the redispatch both in normal and emergency conditions was limited only by their capacity (not willingness etc.) No changes of the generation pattern due to generators scheduled outages (i.e. maintenance) were considered. In this section we want to outline possible eects of relaxing above assumptions. To do so, we will use the same test system, but in its original form (see gure A.1), i.e. FACTS device is not present in the system anymore. In addition, a methodology will naturally dier from the one used up to now.
5.4.1
In the previous sections generators unavailability was neglected, thus the analysis of the system conditions variation could have been simplied by considering only dierent system load levels as main drivers of the system behavior. Thus, analysis consisted only from a fairly limited number of possible scenarios that were studied. However, inclusion of generators scheduled outages increases the number of possible system states variation signicantly. Therefore in order to handle this increased complexity, probabilistic methods have to be employed for the assessment. Since an analytical approach is not possible, other methods have to be applied to provide an approximate result. The most widely applied
5.4. Considerations
87
method is Monte Carlo simulation. Fundamentally, there are two variants of this method according to the way of input samples generation [9]: time sequential sampling unsequential sampling For a better physical interpretation and insight into the analysis we have chosen the time sequential sampling. In our modeling and study we restrict ourselves to the pure line rating violation. We do not put the operational limits on the voltage magnitudes, although the voltage collapse situations are considered. We dene an initial event as a simultaneous outage of K elements that would result in a transition into a cascading process. Here in this denition we do not consider how wide and deep the cascading process continues, only if the initiating event creates an overload of at least one other element. An one-year period of the system operation from [23] as shown in A.1 is simulated. We do it in two stages. The rst one can be executed once employing a strongly vectorized form of the Matlab code: 1. Preparation of the data for the analysis. Load magnitude oats up and down basically each hour during the year, depending on the weather conditions, character of the day (weekday vs. weekend day) etc. 2. Maintenance of the generators has been scheduled according to the usual practice, i.e. the way having a least impact on the active power margin. The resulting available generation capacity is shown in gure 5.8. 3. We make a simplication in the unit commitment policy, which will be reasoned later. So all available generators (the ones which do not undergo a maintenance in that particular sample) share the production according to their rating. The second stage is repeated then for every sample (i.e. representing an hour) within the studied period (one year, i.e. 8736 samples = number of hours in 52 weeks):
88

3500
3000
2500 Active Power [MW]
2000
1500
1000 Dispatched Generation Available Generation Capacity 500
1000
2000
3000
4000
5000
6000
7000
8000
9000
time [hour]
Figure 5.8: Available generation capacity obtained after a generators maintenance scheduling. 1. First we solve a nominal power ow problem with Newton-Raphson method. 2. Having all voltage magnitudes and angles, we compute line ows for the system with the intact topology. 3. The sensitivities of the line ows are computed with respect to the node injections of the active and reactive power. 4. Contingency Analysis, described more in detail in Appendix C. 5. For the outage of elements identied in Contingency Analysis we compute the consequences when only line overload relays are used in Cascading progression analysis. 6. Optimal Control Actions calculation.
5.4.2
Simulation Results
Simulations and computations performed according to the procedure described in the previous section yield the border for the cascading
5.4. Considerations
3000 2500 2000 1500 1000 500
89
System Load [MW]
1000
2000
3000
4000
5000
6000
7000
8000
9000
5 4 3 K 2 1 0 0
1000
2000
3000
4000
5000
6000
7000
8000
9000
time [hour]
Figure 5.9: Minimal size of an initiating event triggering a cascading. Obtained values larger than 4 have been rounded to 4, since their probability of occurrence is negligible and an accuracy of the modeling of their impact becomes too uncertain. expressed in terms of the minimal initiating event size as shown in gure 5.9. As it can be observed, the border shrinks to 2 not only in the load peak situations at the end of the period, but also and even more often in the middle of the simulated period, what represents summer season in the power system operation. This can be explained by the signicant modications in the line ows pattern, which is caused by the modication of the generation pattern due to power plant maintenance. Figure 5.10 shows that the studied system is quite sensitive to the disturbances of certain type, since it often progresses to a whole system blackout if no system-wide coordinated control approach is used. When studying the system-wide blackouts, it has been discovered that they are caused by voltage collapse. On the other hand, comparing the top and bottom parts of gure 5.10, it can be seen that quite often the initiating event creates an overloading of other line, which is tripped, but after that the cascading does not spread any further and no load is actually aected.
90

40 Lost Lines in Cascading without WAMC 30
20
10
1000
2000
3000
4000
5000
6000
7000
8000
9000
3000 2500 Lost Load [MW] 2000 1500 1000 500 0 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 without WAMC
time [hour]
Figure 5.10: Consequences of the cascading if no action is taken and only line overload relays are employed. The large values of the line outages (36) and lost power correspond to the entire system blackout. Note that the number of lost lines corresponds to the number of tripped lines after an initiating event, which is not included in this number. Figure 5.11 shows the required control eort of WAMC system to modify the line ows in such a way that an initiating event will not result into a cascading. However, it is important to note that when having the control objective to not allow an overload of a line, the load is sometimes shed and generation redispatched although the tripping of a line would not aect the load at all, it would only weaken the system. So the WAMC acts more often than necessary but it aects only a small portion of the load and prevents large load disconnections due to system blackouts, as shown in gure 5.14. Figure 5.11 helps in referring the required controls to the particular time instant. But to obtain a better, more intuitive feeling about the possible control actions, gures 5.12 and 5.13 are presented. They show how often it
5.4. Considerations
700 600 Shed Load [MW] 500 400 300 200 100 0 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 sr = 0 sr = 0.25 sr = 0.5 sr = 0.75 sr = 1
91
Redispatched Generation [MW]
800
600
400
200
1000
2000
3000
4000
5000
6000
7000
8000
9000
time [hour]
Figure 5.11: Actions of WAMC. Redispatch of the generation represents the amount of the power, which has to be both reduced and increased in some generators, i.e. sum of the absolute value of the generators output change. sr is the spinning reserve factor parameter. would be necessary to employ a control larger than a certain size to keep the border of the safety so that the cascading would not spread further after an initiating event. The case when sr = 0 represents the situation when the generation injection can not be increased, so the dash-dotted curve in gure 5.13 shows only the decrease of the generation in some locations. A very important observation from gure 5.12 and especially from gure 5.13 is, that there is a small dierence between the spinning reserves factors 0.25, 0.5, 0.75 and 1. The deviations are only in the very low occurrence area. This feature suggests that it dies not have to be significantly benecial to allocate more spinning reserves than corresponding to sr = 0.25, since there is an extremely small probability that it would be employed and thus it could become more expensive than a slightly increased amount of possible load shedding consequences. (However, the most suitable level of spinning reserves should be determined based
92

0.06
0.05
sr = 0 sr = 0.25 sr = 0.5 sr = 0.75 sr = 1
0.04
CDF
0.03
0.02
0.01
100
200
300
400
500
600
700
Shed Power [MW]
Figure 5.12: Cumulative Distribution Function (CDF) of the load shedding with the varying parameter sr. on a risk analysis considering possible cascading consequences.) Therefore we use sr = 0.25 (which corresponds here in average to 200 MW) as a control setting and compare its impact on the unserved ctitious load in gure 5.14. Note again, that these presented numbers refer to all possible cascading boundary situations in one-year network operation period. They are not a quantication of how often initial events occur.
5.4.3
Remarks
We pointed out that the generators maintenance modies the line ows pattern in a way reducing the safety boundary, besides the amount of the consumed and transferred power. We suggest that there is a system specic property - spinning reserve factor level, which might be a good indication of reasonable spinning reserve capacity allocation to be used in the emergency conditions. Similarly to ndings stated here, an importance of spinning reserves in cascading is stressed also in [10] and [12].
5.4. Considerations
0.06 sr = 0 sr = 0.25 sr = 0.5 sr = 0.75 sr = 1
93
0.05
0.04
CDF
0.03
0.02
0.01
100
200
300
400
500
600
700
800
Redispatched Generation [MW]
Figure 5.13: Cumulative Distribution Function (CDF) of the generation redispatch with the spinning reserve factor sr as a parameter.
0.015 no WAMC
0.01 CDF 0.005
0 2000
2100
2200
2300
2400
2500
2600
2700
2800
2900
Lost Power [MW] 0.06 with WAMC 0.05 0.04 CDF sr = 0.25 0.03 0.02 0.01 0 0 50 100 150 Shed Power [MW] 200 250 300
Figure 5.14: Cumulative distribution function (CDF) of the disconnected load. In the rst case the power that is lost as a consequence of the cascading. In the second case the load, which is shed by WAMC.
Chapter 6
Conclusions and Future Prospects

6.1 Summary
Security of electricity supply is absolutely crucial for the modern society. Therefore a substantial eort has been invested into development of a concept addressing power systems security from early stages of power systems evolution. However, recent trends have focused on the improvement of the economy of electricity supply, introducing signicant changes (Such as systems interconnections, deregulation, trading over large distances, distributed generation, large distances between load and generation centers etc.) in power systems operation. These changes have resulted into larger stresses of transmission systems and thus an inadequacy of security concepts designed for dierent conditions. In order to overcome this discrepancy, this dissertation suggests several modications of the existing security concept. In particular: 1. Keeping the hierarchical control concept, more responsibilities shall be delegated from the tertiary to the secondary control. Secondary control shall simultaneously cover several phenomena (Explicitly: control of frequency and voltage while avoiding unacceptable components stresses.) under both normal and emergency conditions. 95
96
Chapter 6. Conclusions and Future Prospects 2. Secondary control can employ Model Predictive Control. In order to reduce the computational complexity, trajectory sensitivities formulation can be used. 3. A power system could be made more robust against larger disturbances if a new, proposed index expressing the system security, as a complement and/or alternative to N-1, would be used.
Several examples throughout the dissertation show promising results, indicating the feasibility of the above listed proposed modications of the security concept.
6.2
Directions for Future Work
The dissertation deals with the selected parts of the overall very large topic - power systems security. Further research work should be pursued along the direction given by this thesis. The relevant research problems and questions, which should be answered, may be divided into two categories: 1. Problems elaborating further the material directly presented in this thesis. 2. Topics covering the missing pieces in the overall security concept and thus complementing the topics addressed by this thesis. The rst category can be represented by examples as follows: Derivation and studies of the use of a variable step size in the proposed MPC prediction and controls computation phase. This should yield a further improvement of the computational eciency. Also, more natural implicit switching of objectives importance (emergency vs. preventive mode) should be thus possible. The proposed security index should be transformed into the form of operation criterion. In an ideal case, a direct relation, or even better, complement of N-1 could be derived. To mention just a few examples for the second category of the possible future research activities:
6.2. Directions for Future Work
97
Studies on vulnerability of the proposed modied hierarchical control. Studies on coordination of secondary controllers. Studies on the economical context of the proposed control, i.e. transmission pricing. Investigations of new policies for generator maintenance scheduling. It might have a very positive impact on the system security to consider the location of the generators as well, not only to keep the largest possible active power reserve.
Appendix A
Test Systems Data

A.1 The IEEE Reliability Test System 1996
The system shown in gure A.1 is fully described in the reference [23]. The data contain all fundamental items (such as system parameters, failure rate of components, hourly, weekly, monthly and annual load variation etc.) needed for static analysis and power systems reliability evaluation employing probabilistic methods.
A.2
A Simple Radial Test System
This system has been introduced in [36] and [37] and it is shown in A.2. A simple radial system connected to a sti network feeds load with a dynamic recovery. The system comprises three lines and a transformer equipped with a tap changer. The tap changer has its own traditional controller with a time constant of 20 seconds. At the time equal to 10 seconds (from the beginning of the simulation), one of the parallel lines is tripped. As a consequence, the voltage in the node 3 drops. 99
100
Appendix A. Test Systems Data
G18
18 21
G21
G22
22
17
G16
16 19 20
G15
15 14
G14 G13
13
G23
23
24
11
12
10
2 7
G1
G2
G7
Figure A.1: Single line diagram of the IEEE Reliability Test System - 1996. Buses supplying connected loads are in the black color. Busses in the grey color contain no load.
A.3. The Nordic 26 Bus Test System
101
stiff
Figure A.2: Single line diagram of the simple radial test system. Bus supplying connected load is in the black color. Busses in the grey color contain no load. The dashed line between nodes 1 and 2 is the line tripped in the disturbance scenario.
A.3
The Nordic 26 Bus Test System
We use the version of this test system used rst in [35] and shown in the gure A.3. The authors have used it for a study of power oscillations excited in the Nordic (interconnected power system of Sweden, Norway, Finland and eastern part of Denmark). We have introduced small modications (e.g. changing the load character to constant power load instead of impedance load) allowing us to focus on the voltage related issues. Contrary to other test systems used in this thesis, according to our knowledge, no public source of data is available, therefore the detailed dynamic data are listed here as follows. The 4th order synchronous machine models represent generators and their parameters are given in the table A.1. Basic control circuits of generators are modeled too, e.g., governor controlling the turbine supplying torque and thus the active power - table A.2, Automatic Voltage Regulator (AVR) controlling excitation circuits of the generator and thus the generators terminal voltage and reactive power output - table A.3 and Power System Stabilizer (PSS) modulating the reference signal for the AVR a way damping oscillations measured in the system frequency - table A.4. Loads and lines parameters are listed in tables A.6 and A.5 respectively. All together, the system model in our representation has 822 variables of all types (i.e., dierential, algebraic, and discrete variables, including controls).
102
G8
24 4
G9 G6
25
G7
3 23 16 22 6
G1
17
G2
15 18 2
G3
14 19
G10
26 13 21 12 5 1
G4
11 20
G5
10 7
Figure A.3: Single line diagram of the Nordic 26 Bus Test System. Buses supplying connected loads are in the black color. Busses in the grey color contain no load. The dashed line between nodes 3 and 16 is the line tripped in the disturbance scenario.
Generator G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 Table A.1: Parameters of generators.
ra 0.000167 0.000167 0.000167 0.000200 0.000167 0.000167 0.000125 0.000167 0.000167 0.000167
xd 0.033333 0.066667 0.033333 0.04 0.05 0.05 0.025 0.033333 0.036667 0.033333
xd 0.006667 0.013333 0.01 0.012 0.013333 0.013333 0.005 0.00667 0.008333 0.006667
Tdo 5 6 5 5 5 5 5 5 5 5
xq 0.026667 0.05 0.05 0.06 0.05 0.05 0.02 0.026667 0.036667 0.033333
xq 0.006667 0.013333 0.01 0.012 0.013333 0.013333 0.005 0.00667 0.008333 0.006667
Tqo 999 999 999 999 999 999 999 999 999 999
2.H 1.145916 1.145916 1.432394 1.193662 1.432394 1.145916 1.527887 1.145916 1.145916 1.432394
D 0.286479 0 0.286479 0.238732 0.859437 0.286479 0.381972 0.286479 0.286479 0.286479
103
104
Generator G1 G2 G3 G4 G5 G6 G7 G8 G9 G10
R 0.6981 10.4720 1.0472 1.2566 10.4720 0.6981 0.2123 0.6981 0.6981 0.6981
Ts 27 60 27 27 60 27 27 27 27 27
Tc 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
T3 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
T4 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
T5 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
Pref 17 20 17.5 12 10 5 10 10 25
Table A.2: Governor Controls Parameters.
Generator G1 G2 G3 G4 G5 G6 G7 G8 G9 G10
TR 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04
KA 50 50 50 50 50 50 50 50 50 50
TA 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1
TB 5 5 5 5 5 5 5 5 5 5
TC 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5
Vref 1 1 1 1 1 1 1 1 1 1
Table A.3: Automatic Voltage Regulators (AVR) Parameters.
Generator G1 G2 G3
KP SS 10 10 10
TW 0.1 0.1 0.1
T1 5 5 5
T2 50 50 50
VP SSmax 0.2 0.2 0.2
VP SSmin -0.3 -0.3 -0.3
Table A.4: Power System Stabilizers (PSS) Parameters.
105
From 1 1 2 2 3 3 3 3 3 4 4 5 5 5 6 6 7 7 8 9 10 11 11 12 12 12 13 14 15 15 17
To 19 20 17 19 4 5 16 16 23 17 24 12 20 22 17 25 10 20 9 10 21 20 21 13 14 21 26 15 16 18 20
R [pu] 0.0003750 0.0009375 0.0003750 0.0003750 0.0037500 0.0005625 0.0001875 0.0075000 0.0001875 0.0018750 0.0009375 0.0011250 0.0028125 0.0003750 0.0009375 0.0003750 0.0015000 0.0015000 0.0003750 0.0003750 0.0003750 0.0018750 0.0018750 0.00084375 0.00103125 0.00065625 0.0001875 0.0007500 0.00046875 0.0009375 0.0009375
X [pu] 0.00500 0.01250 0.00500 0.00500 0.05000 0.00750 0.00250 0.10000 0.00250 0.02500 0.01250 0.01500 0.03750 0.00500 0.01250 0.00500 0.02000 0.02000 0.00500 0.00500 0.00500 0.02500 0.02500 0.01125 0.01375 0.00875 0.00250 0.01000 0.00625 0.01250 0.01250
Table A.5: Line parameters. Note that the line between buses 3 and 16 listed rst, is the outaged line in the disturbance scenario.
106
Bus 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 21
Plinit [pu] 30 10 40 10 1 10 4 -3.36 3.13 4 8 5 10 0.1 3 0.01 2
Qinit [pu] l 0 0 0 0 0 0 -0.87 -0.53 -0.24 -1 0.15 -1 0.1 0.1 0.1 0.1 0
Table A.6: Load parameters. All loads are constant power loads.
Appendix B
Computation of Trajectory Sensitivities

B.1 Trajectory Sensitivities
An ecient framework for modeling of nonlinear systems featuring discrete states (i.e. 4.2 and 4.3) has been presented in [36]. Omitting parameters from the original formulation (Note that here the system parameters are included directly in the listed equations. The original formulation was targeting, among others, inverse problems, in which initial values of parameters can be modied according to experimental results, yielding real parameters values.) from [36] and introducing control inputs u, we can write: x = f (x, y, z, u) 0 = g 0 (x, y, z, u) 0= g i (x, y, z, u) yd,i < 0 i = 1, , d g i+ (x, y, z, u) yd,i > 0 j {1, , e} j {1, , e} (B.1) (B.2) (B.3)
z + = hj (x , y , z , u ) ye,j = 0 z = 0 ye,j = 0 107
(B.4) (B.5)
108 where
Appendix B. Computation of Trajectory Sensitivities
yd = Dy ye = Ey
x X Rn z Z Rk
y Y Rm u U Rl
(B.6)
Dierential variables are denoted x, algebraic variables y and discrete state variables z. Switching of the status of discrete variables is governed by the equation (B.4) when the corresponding auxiliary variables of ye are equal to zero. Auxiliary variables yd determine the region of validity of the equations (B.3). In the power systems context this may be explained on an example of a line, which changes its status. When the line is in service, equations linking the owing current through the line and voltages at both ends of the line as well as line parameters (line impedance and shunt admittance) are valid. When the line is out of service (i.e., disconnected), current owing through it is zero. The auxiliary variable is in that case the dierence between the time and the instant when the line was tripped. Matrices D and E have normally a very sparse structure and their non-zero elements are equal to one on the positions aiming at the auxiliary variables. Flow (i.e., time evolution) of the system from its initial point can be characterized by the time evolution of its variables, e.g. for algebraic variables we can write: y (y0 , t) = y(t) (B.7)
Note that the initial state y0 is obtained by solving equations (B.2) (B.3) by substituting initial values of x(t), z(t) and u(t) by x0 , z0 and u0 , respectively. An impact of small changes of initial conditions on the system ow can be investigated by trajectory sensitivities. So the impact of manipulated inputs (i.e., controls) on algebraic states time evolution can be obtained by a Taylor expansion of (B.7). When neglecting higher order terms: y(t) = y(t) u0 + higher order terms yu0 (t)u0 u0 (B.8)
Note that trajectory sensitivities are generally time varying quantities.
B.2. Numerical Computations
109
B.2
B.2.1
Numerical Computations
Nominal Trajectory
Similarly to [36], due to convenient redenitions: x x= z u f f = 0 0 x hj = hj u
(B.9)
(B.10)
(B.11)
formulation (B.1) - (B.6) can be written even in a more compact form: x = f (x, y) 0 = g 0 (x, y) 0= g i (x, y) yd,i < 0 i = 1, , d g i+ (x, y) yd,i > 0 j {1, , e} j {1, , e} (B.12) (B.13) (B.14)
x+ = hj (x , y ) ye,j = 0 z=0 ye,j = 0
(B.15) (B.16)
A nominal trajectory between events can now be obtained by applying trapezoidal integration, yielding equations: 0 = xk+1 xk .{f (xk+1 , y k+1 ) + f (xk , y k )} 2 0 = g(xk+1 , y k+1 ) (B.17) (B.18)
where superscript k denotes the time instant. Equations (B.17) and (B.18) are sets of nonlinear equations that can be solved for the time instant k + 1 by applying Newton iteration procedure in the form:
110
Appendix B. Computation of Trajectory Sensitivities
0 = F () i+1 = i F (i )1 .F (i )
(B.19) (B.20)
Subscript i refers to the iteration number. Vector c consists of vectors x and y. The Jacobian matrix F has the structure:
2fx
F =
2fy
gx
gy
(B.21)
B.2.2
Trajectory Sensitivities
Away from events, the sensitivities xx0 and yx0 are derived by dierentiating (B.12) - (B.13): xx0 = f x (t).xx0 + f y (t).yx0 0 = gx (t).xx0 + gy (t).yx0 (B.22) (B.23)
Trapezoidal integration can be again applied on numerical solving of (B.22) and (B.23): k+1 k 0 = xk+1 xk0 .{f k+1 xk+1 + f k+1 yx0 + f k xk0 + f k yx0 } (B.24) x0 x x0 x y x x y 2
k+1 k+1 k+1 0 = gx xk+1 + gy yx0 x0
(B.25)
The sensitivities between events can be then easily computed after rearranging the above equations:
xk+1 x0 k+1 yx0
k+1 2fx k+1 gx
k+1 2fy k+1 gy
k {f k xk0 + f k yx0 } xk0 x 2 x x y 0 (B.26)
B.2. Numerical Computations
111
B.2.3
Remarks
A numerical approximation of trajectory sensitivities can be computed by solving (B.1) - (B.6) for an incremental change of each control input. But that would represent a large computation eort if many control inputs are considered. Note, that the Jacobian to be inverted in (B.26) is available from the calculations of the nominal trajectory (for the present time instant k + 1) as well as the elements of the second matrix (from the previous time instant k). So the computations of the trajectory sensitivities require only a little additional eort and no iteration technique has to be employed (contrary to the nominal trajectory calculations). A further reduction of the computation time can be achieved if numerical methods potential is further examined [11]. We discussed here only trapezoidal integration method. However, other integration methods can be used too and some probably even with a better accuracy, as suggested in [53]. More details about application of the modeling framework, presented in this chapter, on power systems problems can be found in [37] and [34], where a natural exible modular structure following power systems components classication has been adopted. E.g. power systems comprise of independent modules - generators, loads etc., which interact with each other through their interfaces. For example two neighboring modules bus and load share voltage. This can be expressed through the coupling of corresponding algebraic variables y. The modular structure provides an additional advantage of exibility to apply the described modeling and control framework to various control problems with only small adjustments, e.g. choosing the complexity of load models according to studied phenomena - i.e. voltage dependent load or dynamic recovery etc.
Appendix C
Benets of Better Control - Computations

C.1 Traditional Generation Dispatch
The objective in this particular situation is to minimize the generation in zone B (Or alternatively to maximize the generation in zone A.), which can be mathematically formulated as follows:
NgB uP g ,uF AT CS
min
P gi
i=1
(C.1)
Constraints remain same as listed in the subsection 3.1.4, i.e.:

Ng Nl
P gm =
m=1 n=1
P ln
(C.2)
|Fi | Fmax
(C.3)
0 P g P gmax 113
(C.4)
114 Appendix C. Benets of Better Control - Computations
Pg RNg Fi , Fmax RNb i = 1, 2, . . . , Nc
(C.5)
where Pg and Pl are the injected generation and load, respectively. Fi denotes post-contingency ows in the remaining branches (i.e. lines and transformers) after the outage of the branch i. Nc is the number of considered contingencies (here single element outages, thus number of branches), Ng number of generators and Nl number of loads. In this case, control input uP g corresponds to the generation vector P g and control input uF ACT S is naturally zero.
C.2
Traditional Generation Dispatch Including a FACTS Device
A dispatch procedure including a FACTS device can be essentially formulated the same way as a traditional dispatch in the previous section. The slight dierence is introduced by the control input uF ACT S . In the case of TCSC it is a variable series capacitance added to the reactance of the controlled line, in the case of PS it is phase angle dierence across the PS. The corresponding control input can be than summarized: uF ACT S = {u0 ACT S } F umin S u0 ACT S umax S F ACT F F ACT u0 ACT S R F (C.6) (C.7) (C.8)
C.3
C.3.1
Proposed Control Including a FACTS Device

Generation Dispatch
Here the main assumption is that the setting of a FACTS device can be modied immediately after a disturbance in order to keep the line ows within the line ratings. Thus, each post outage i state (Expressed by post outage ows Fi .) may be determined with a dierent FACTS device
C.3. Proposed Control Including a FACTS Device
115
control setting ui ACT S . The rest of the formulation of the dispatch F procedure is same as in two previous sections.
c uF ACT S = {u0 ACT S , u1 ACT S , . . . , uNACT S } F F F
(C.9) (C.10) (C.11) (C.12)
umin S ui ACT S umax S F ACT F F ACT ui ACT S F R i = 0, 1, . . . , Nc
C.3.2
Emergency Conditions
The objective under emergency conditions is to minimally modify controls, determined in the dispatch procedure according to the previous subsection, in order to keep the system within its boundaries, expressed by branch ows not exceeding ratings. In addition to generation dispatch and FACTS device, load shedding can be employed when necessary too. Since dynamic aspects are neglected here, formulation (4.6) (4.13) can be simplied and for this particular case it can be written:
Ng uP g ,uP l ,uF AT CS Nl
min
|RP g .P gm | +
m=1 Ng Nl n=1
|RP l .P ln |
(C.13)
P gm =
m=1 n=1
P ln
(C.14) (C.15) (C.16) (C.17) u0 ACT S F (C.18)
|F | Fmax P g P g P gmax P g P l P l 0 umin S F ACT u0 ACT S F uF ACT S umax S F ACT
P g RNg P l RNl uF ACT S R F, Fmax RNb
(C.19)
where RP g and RP l are weighting factors expressing preferences of employed respective controls.
Bibliography
[1] V. Ajjarapu and C. Christy. The continuation power ow: A tool for steady state voltage stability analysis. IEEE Transactions on Power Systems, 7(1), February 1996. [2] P. M. Anderson and B. K. LeReverend. Industry experience with special protection schemes. IEEE Transactions on Power Systems, 11(3), August 1996. [3] G. Andersson, P. Donalek, R. Farmer, N. Hatziargyriou, I. Kamwa, P. Kundur, N. Martins, J. Paserba, P. Purbeik, J. Sanchez-Gasca, R. Schulz, A. Stankovic, C. Taylor, and V. Vittal. Causes of the 2003 major grid blackouts in north america and europe, and recommended means to improve system dynamic performance. IEEE Transactions on Power Systems, 20(4):1922 1928, November 2005. [4] S. Arnborg. Emergency Control of Power Systems in Voltage Unstable Conditions. PhD thesis, Royal Institute of Technology in Stockholm, Department of Electric Power Engineering, Stockholm, 1997. [5] S. Arnborg, G. Andersson, D. J. Hill, and I. A. Hiskens. On inuence of load modelling for undervoltage load shedding studies. IEEE Transactions on Power Systems, 13(2), May 1998. [6] A. G. Beccuti, T. Geyer, and M. Morari. Temporal lagrangian decomposition of model predictive control for hybrid systems. In IEEE Conference on Decision and Control, Atlantis, Bahamas, December 2004. 117
118
BIBLIOGRAPHY
[7] A. Bemporad and M. Morari. Control of systems integrating logic, dynamics and constraints. Automatica, 35(3):407 427, March 1999. [8] A. Berizzi, P. Bresesti, P. Marannino, G. P. Granelli, and M. Montagna. System-area operating margin assessment and security enhancement against voltage collapse. IEEE Transactions on Power Systems, 11(3), August 1996. [9] R. Billinton and W. Li. Reliability Assessment of Electric Power Systems Using Monte Carlo Methods. Plenum Press, 1994. [10] B. A. Carreras, V. E. Lynch, I. Dobson, and D. E. Newman. Complex dynamics of blackouts in power transmission systems. Chaos: An Interdisciplinary Journal of Nonlinear Science, 14(3):643 652, September 2004. [11] D. Chaniotis, M. A. Pai, and I. A. Hiskens. Sensitivity analysis of dierential-algebraic systems using the gmres method - application to power systems. In IEEE International Symposium on Circuits and Systems, Sydney, Australia, May 2001. [12] J. Chen, J. S. Thorp, and I. Dobson. Cascading dynamics and mitigation assessment in power system disturbances via a hidden failure model. International Journal of Electrical Power and Energy Systems, 27(4):318 326, May 2005. [13] H. D. Chiang, A. J. Flueck, K. S. Shah, and N. Balu. Cpow: A practical tool for tracing power system steady-state stationary behavior due to load and generation variations. IEEE Transactions on Power Systems, 10(2), May 1995. [14] H. D. Chiang and R. Jean-Jumeau. Towards a practical performance index for predicting voltage collapse in electric power systems. IEEE Transactions on Power Systems, 10(2), May 1995. [15] P. Cote, S. P. Cote, and M. Lacroix. Programmable load shedding systems - hydro-qubecs experience. In IEEE Power Engineering Society Summer Meeting, 2001. [16] T. Van Cutsem. An approach to corrective control of voltage instability using simulation and sensitivity. IEEE Transactions on Power Systems, 10(2):616 622, May 1995.
BIBLIOGRAPHY
119
[17] T. Van Cutsem and R. Mailhot. Validation of a fast voltage stability analysis method on the hydro-qubec system. IEEE Transactions on Power Systems, 12(1), February 1997. [18] T. Van Cutsem, C. Moors, and D. Lefebvre. Design of load shedding schemes against voltage instability using combinatorial optimisation. In IEEE Power Engineering Society Winter Meeting, New York, January 2002. [19] T. Van Cutsem and C. Vournas. Voltage Stability of Electric Power Systems. Kluwer Academic Publishers, 1998. [20] B. Delno, S. Massucco, A. Morini, P. Scalera, and F. Silvestro. Implementation and comparison of dierent under frequency loadshedding schemes. In IEEE Power Engineering Society Summer Meeting, 2001. [21] I. Dobson and L. Lu. New methods for computing a closest saddle node bifurcation and worst case load power margin for voltage collapse. IEEE Transactions on Power Systems, 8(3):905 911, August 1993. [22] D. Ernst and M. Pavella. Closed-loop transient stability emergency control. In IEEE Power Engineering Society Winter Meeting, 2000. [23] C. Grigg et al. The ieee reliability test system - 1996. IEEE Transactions on Power Systems, 14(3), August 1999. [24] H. Breulmann et. al. Analysis and damping of inter-area oscillations in the ucte/centrel power system. In CIGRE, 38-113, 2000. [25] B. Fardanesh. Future trends in power system control. IEEE Computater Applications in Power, 15(3):24 31, July 2002. [26] P. A. Fedora and G. E. Haringa. Assessment of npcc area interconnection reliability benets. In Proceedings of the 38th Hawaii International Conference on System Sciences, Hawaii, Januar 3-6 2005. [27] R. Findeisen, M. Diehl, Z. Nagy, F. Allgwer, H. G. Bock, and J. P. o Schlder. Computational feasibility and performance of nonlinear o model predictive control schemes. In European Control Conference, Porto, September 2001.
120
BIBLIOGRAPHY
[28] A. J. Flueck and H. D. Chiang K. S. Shah. Investigating the installed real power transfer capability of a large scale power system under a proposed multiarea interchange schedule using cpow. IEEE Transactions on Power Systems, 11(2), May 1996. [29] T. Geyer, M. Larsson, and M. Morari. Hybrid emergency voltage control in power systems. In Proceedings of the European Control Conference, Cambridge, UK, September 2003. [30] P. Gomes, A. P. Guarini, A. C. S. de Lima, and M. S. Rocha. Guidelines for power system restoration in the brazilian system. In IEEE Power Engineering Society Winter Meeting, New York, January 2002. [31] F. Gubina and B. Strmcik. Voltage collapse proximity index determination using voltage phasors approach. IEEE Transactions on Power Systems, 10(2), May 1995. [32] M. Hemmingsson, O. Samuelsson, K. O. Helgesen Pedersen, and A. H. Nielsen. Estimation of electro-mechanical mode parameters using frequency measurements. In IEEE Power Engineering Society Summer Meeting, 2001. [33] G. T. Heydt, C. C. Liu, A. G. Phadke, and V. Vittal. Solutions for the crisis in electric power supply. IEEE Computer Applications in Power, 2001. [34] I. A. Hiskens. Power system modeling for inverse problems. IEEE Transactions on Circuits and Systems - I: Regular Papers, 51(3):539 551, March 2004. [35] I. A. Hiskens and M. Akke. Analysis of the nordel power grid disturbance of january 1, 1997 using trajectory sensitivities. IEEE Transactions on Power Systems, 14(3):987 994, August 1999. [36] I. A. Hiskens and M. A. Pai. Trajectory sensitivity analysis of hybrid systems. IEEE Transactions on Circuits and Systems - I: Fundamental Theory and Applications, 47(2):204 220, February 2000. [37] I. A. Hiskens and P. J. Sokolowski. Systematic modeling and symbolically assisted simulation of power systems. IEEE Transactions on Power Systems, 16(2):229 234, May 2001.
BIBLIOGRAPHY
121
[38] I. Kamwa. Using mimo system identication for modal analysis and global stabilization of large power systems. In IEEE Power Engineering Society Summer Meeting, 2000. [39] I. Kamwa, R. Grondin, and Y. Hebert. Wide-area measurement based stabilizing control of large power systems - a decentralized/hierarchical approach. IEEE Transactions on Power Systems, 16(1), February 2001. [40] G. G: Karady, A. A. Daoud, and M. A. Mohamed. On-line transient stability enhancement using multi-agent technique. In IEEE Power Engineering Society Winter Meeting, New York, January 2002. [41] D. S. Kirschen, D. Jayaweera, D. P. Nedic, and R. N. Allan. A probabilistic indicator of system stress. IEEE Transactions on Power Systems, 19(3):1650 1657, August 2004. [42] D. S. Kirschen and D. P. Nedic. Consideration of hidden failures in security analysis. In Power Systems Computation Conference, Sevilla, June 24-28 2002. [43] M. Koaizawa, K. Omata, M. Nakane, and Y. Kokai. Actual operating experience of on-line transient stability control systems (tsc systems). In IEEE Power Engineering Society Winter Meeting, 2000. [44] S. Kolluri, K. Tinnium, and M. Stephens. Design and operating experience with fast acting load shedding scheme in the entergy system to prevent voltage collapse. In IEEE Power Engineering Society Winter Meeting, 2000. [45] P. Korba, M. Larsson, and C. Rehtanz. Detection of oscillations in power systems using kalman ltering techniques. In 2003 IEEE Conferenc eon Control Applications, pages 183 188, Istanbul, Turkey, June 23-25 2003. [46] D. N. Kosterev, C. W. Taylor, and W. A. Mittelstadt. Model validation for the august 10, 1996 wscc system outage. IEEE Transactions on Power Systems, 14(3), August 1999. [47] P. Kundur. Power System Stability and Control. Mc Graw-Hill Professional Publishing, 1994.
122
BIBLIOGRAPHY
[48] P. Kundur, G. K. Morisson, and L. Wang. Techniques for online transient stability assessment and control. In IEEE Power Engineering Society Winter Meeting, 2000. [49] P. Kundur, J. Paserba, V. Ajjarapu, G. Andersson, A. Bose, C. Ca nizares, N. Hatziargyriou, D. Hill, A. Stankovic, C. Taylor, T. Van Cutsem, and V. Vittal. Denition and classication of power system stability. IEEE Trans. on Power Systems, 19:13871401, 2004. [50] M. Larsson. Coordinated Voltage Control in Power Systems. PhD thesis, Lund University, Lund, 2000. [51] G. Ledwich and E. Palmer. Modal estimates from normal operation of power systems. In IEEE Power Engineering Society Winter Meeting, 2000. [52] J. D. McCalley. Operational defence of power system cascading sequences; probability, prediction & mitigation. Technical report, PSERC, October 2003. Seminar Presentation. [53] S. Meliopoulos, G. Cokkinides, and G. Stefopoulos. Improved numerical integration method for power electronic simulation based on three-point collocation. In IEEE Conference on Decision and Control, Seville, Spain, December 12-15 2005. [54] T. Minakawa. The required technological breakthrough in developing universal emergency control systems in terms with transient and dynamic stability. In IEEE Power Engineering Society Winter Meeting, 2000. [55] M. A. Mitschell, J. A. Pecas Lopes, J. N. Fidalgo, and J. D. McCalley. Using a neural network to predict the dynamic frequency response of a power system to an under-frequency load shedding scenario. In IEEE Power Engineering Society Summer Meeting, 2000. [56] C. Moors, D. Lefebvre, and T. Van Cutsem. Design of load shedding schemes against voltage instability. In IEEE Power Engineering Society Winter Meeting, 2000. [57] F. J. Nogales, F. J. Prieto, and A. J Conejo. A Decomposition Methodology Applied to the Multi-Area Optimal Power Flow Problem. Kluwer Academic Publishers, 2003.
BIBLIOGRAPHY
123
[58] J. Rawlings. Tutorial overview of model predictive control. IEEE Control Systems Magazine, 20(3):38 52, June 2000. [59] C. Rehtanz. Autonomous Systems and Intelligent Agents in Power System Control and Operation. Springer Verlag, 2003. [60] N. Dag Reppen. Increasing utilization of the transmission grid requires new reliability criteria and comprehensive reliability assessment. In 8th International Conference on Probbabilistic Methods Applied to Power Systems PMAPS, Ames, Iowa, September 12-16 2004. [61] K. S: Shim, H. K. Nam, S. G. Song, Y. G. Kim, and K. Y. Lee. Application results of the eigen-sensitivity theory of augmented matrix to small signal stability analysis of large power systems. In IEEE Power Engineering Society Summer Meeting, 2000. [62] A. F. Snyder, D. Ivanescu, N. HadjSaid, D. Georges, and T. Margotin. Delayed-input wide-area stability control with synchronized phasor measurements and linear matrix inequalities. In IEEE Power Engineering Society Summer Meeting, 2000. [63] G. N. Taranto, N. Martins, D. M. Falcao, A. C. B. Martins, and M. G. dos Santos. Benets of applying secondary voltage control schemes to the brazilian system. In IEEE Power Engineering Society Summer Meeting, 2000. [64] C. W. Taylor. Power System Voltage Stability. Number ISBN: 0070631840. McGraw-Hill, December 1993. [65] L. Y. Taylor and S. M. Hsu. Transmission voltage recovery following a fault event in the metro atlanta area. In IEEE Power Engineering Society Summer Meeting, 2000. [66] J. S. Thorp and H. Wang. Computer simulation of cascading disturbances in electric power systems - impact of protection systems on transmission system reliability. Technical report, PSERC, May 2001. Final Report. [67] G. Trudel, S. Bernard, and G. Scott. Hydro-qubecs defence plan against extreme contingencies. IEEE Transactions on Power Systems, 14(3), August 1999.
124
BIBLIOGRAPHY
[68] K. Uhlen, M. Palsson, T. R. Time, O. Kirkeleuten, and J. O. Gjerde. Raising stability limits in the nordic power transmission system. IEEE Transactions on Power Systems, 2001. submitted. [69] L. Vargas, V. H. Quintana, and R. Miranda. Voltage collapse scenario in the chilean interconnected system. IEEE Transactions on Power Systems, 14(4), November 1999. [70] H. Vu, P. Pruvot, C. Launay, and Y. Harmand. An improved voltage control on large-scale power system. IEEE Transactions on Power Systems, 11(3):1295 1303, August 1996. [71] K. Vu, M. Begovic, D. Novosel, and M. M. Saha. Use of local measurements to estimate voltage-stability margin. IEEE Transactions on Power Systems, 14(3), August 1999. [72] L. Wang and A. A. Girgis. On-line detection of power system small disturbance voltage instability. IEEE Transactions on Power Systems, 11(3), August 1996. [73] X. Wang, Y. H. Song, and Q. Lu. Lagrangian decomposition approach to active power congestion management across interconnected regions. IEE Proceedings - Generation, Transmission, Distribution, 148(5):497 503, September 2001. [74] Y. J. Wang, C. W. Liu, L. D. Sue, and W. K. Liu. A remedial control scheme protects against transient instabilities based on phasor measurement units (pmus) - a case study. In IEEE Power Engineering Society Summer Meeting, 2000. [75] P. Zhang, S. T. Lee, and D. Sobajic. Moving toward probabilistic reliability assessments methods. In 8th International Conference on Probbabilistic Methods Applied to Power Systems PMAPS, Ames, Iowa, September 12-16 2004. [76] L. Zongxiang, M. Zhongwei, and Z. Shuangxi. Cascading failure analysis of bulk power system using small-world network model. In 8th International Conference on Probbabilistic Methods Applied to Power Systems PMAPS, Ames, Iowa, September 12-16 2004.
Curriculum Vitae
EDUCATION 2002 - 2005 Swiss Federal Institute of Technology (ETH), Zurich, Switzerland, PhD studies in Electric Power Systems Royal Institute of technology (KTH), Stockholm, Sweden, MSc studies in Electric Power Engineering Slovak University of Technology, Bratislava, Slovakia, BSc studies in Electric Power Engineering Secondary School of Jozef Gregor Tajovsk, y Bansk Bystrica, Slovakia a
1999 - 2000
1995 - 1999 1991 - 1995
EXPERIENCE 2005 - present 2001 - present 2000 1998 1997 - 1999 Swiss Federal Institute of Technology (ETH), Zurich, Switzerland, External lecturing ABB Switzerland Ltd, R&D Protection and Substation Automation ABB Corporate Research Center, Vsteras, Swea den, R&D Power Electronics, Master thesis Finnish Power Grid (FINGRID), Hmmeenlinna, a Finland, Summer traineeship International Association for Exchange of Students for Technical Experience (IAESTE), Bratislava, Slovakia, chairing and PR & HR activities
125

Contribucion A La Seguridad de Los SEP

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Contribucion A La Seguridad de Los SEP

Caricato da

Copyright:

Formati disponibili

Diss. ETH No.

Contributions to Security of Electric Power Systems

Last Decade Trends . . . . . . . . . . . . . . . . . . . . Shortcomings of Present Security . . . . . . . . . . . . .

10 3.1.7 3.1.8 3.1.9 3.2 3.3

Contents Proposed Security Index . . . . . . . . . . . . . . Example . . . . . . . . . . . . . . . . . . . . . . . Remarks . . . . . . . . . . . . . . . . . . . . . . . 45 48 48 49 51 55 55 57 57 58 60 62 62 67 73 73 76 77 79 83 83 84 84 85 86 86 88 92

Proposed Control Philosophy Essentials . . . . . . . . . Proposed Control Structure . . . . . . . . . . . . . . . .

Security Benets of Optimal Control . . . . . . . . . . .

Positioning and Goal of the Work

1.2. Dissertation Outline

Contributions of the Work

1.4. List of Publications

Power Systems Security

Threats of Power Systems Security

2.2. Present Power Systems Control

Present Power Systems Control

Chapter 2. Power Systems Security

2.2. Present Power Systems Control

Normal and Preventive Control

Chapter 2. Power Systems Security

2.2. Present Power Systems Control

Chapter 2. Power Systems Security

2.2. Present Power Systems Control

Chapter 2. Power Systems Security

Chapter 2. Power Systems Security

2.2. Present Power Systems Control

Chapter 2. Power Systems Security

2.2. Present Power Systems Control

Chapter 2. Power Systems Security

Last Decade Trends

2.3. Last Decade Trends

Chapter 2. Power Systems Security

Shortcomings of Present Security

2.4. Shortcomings of Present Security

Improvements in Power Systems Security

A New Security Index

3.1. A New Security Index

Chapter 3. Improvements in Power Systems Security

Operation Criteria N-1 Criterion

3.1. A New Security Index

Chapter 3. Improvements in Power Systems Security

System Load [pu]

Number of Optimization Constraints 0 1 K 2

Pg RNg Fi , Fmax RNb i = 1, 2, . . . , Nc

3.1. A New Security Index

Relevant Topics in Literature

Proposed Security Index

Chapter 3. Improvements in Power Systems Security

Probability of the trip 0 Fmax k.Fmax

3.1. A New Security Index

|Fi | Fmax + si |Fnom | Fmax

0 si 0 Pg Pgmax Pg RNg W RNb Nb si , Fi , Fnom , Fmax RNb i = 1, 2, . . . , Nc

Chapter 3. Improvements in Power Systems Security

3.2. Proposed Control Philosophy Essentials

Injected Power [pu]

Proposed Control Philosophy Essentials

Chapter 3. Improvements in Power Systems Security

Cummulative Distribution Function

0.1 Proposed dispatch 0.05

15 20 25 Lost Power [pu]

3.3. Proposed Control Structure

Proposed Control Structure

3.3. Proposed Control Structure

MPC Employing Trajectory Sensitivities