Mining Wide Area Frequency Measurements in Power Systems

Transcription

1 Mining Wide Area Frequency Measurements in Power Systems A thesis submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy by Gopal R. Gajjar Roll No: 9473 Under the guidance of Prof. S. A. Soman Department of Electrical Engineering Indian Institute of Technology Bombay Mumbai

2

3 Dedicated to

4

5

6

7

8

9 INDIAN INSTITUTE OF TECHNOLOGY BOMBAY CERTIFICATE OF COURSE WORK This is to certify that Mr. Gopal R. Gajjar was admitted to the candidacy of the Ph.D. degree on August 21, after successfully completing all the courses required for the Ph.D. degree programme. The details of the course work done are given below: Sr. Course No. No. Course Name Credits 1 EE635 Applied Linear Algebra 6 2 EE659 A First Course in Optimization 6 3 EE636 Matrix Computations 6 4 EE651 Digital Protection of Power Systems 6 5 EE713 Circuit Simulation in Power Electronics 6 6 IE718 Networks, Games and Algorithms 6 7 EE658 Power System Dynamics and Control 6 8 EE722 Restructured Power Systems AU 9 IE611 Introduction to Stochastic Models AU 1 EES81 Seminar 4 11 HS699 Communication & Presentation Skills PP Cumulative Credits 46 I.I.T. Bombay Date: Dy. Registrar (Academic) vii

10

11 Abstract This thesis presents various methods of utilization of frequency measurements obtained from wide area measurement systems to monitor the power systems. We start with description of the wide Area Frequency Measurement Systems (WAFMS). This system measures frequency with synchronized time stamps using Network Time Protocol (NTP). Frequency measurement instruments are deployed at several places wide apart geographically, they use the internet for communicating their measurement samples in real time to a central server. Domestic supply a single phase 23 volt plug point is used as a source of frequency measurements. Applications developed using WAFMS can be extended to Wide Area Monitoring System (WAMS) using Phasor Measurement Units (PMU). Frequency is one the parameters that can be accurately measured and the latency and transient responses of PMU are compatible with applications built around frequency measurements. The thesis focuses on using the frequency measurement from the WAFMS for developing power system monitoring applications. WAFMS and WAMS capture continuous stream of data in real time. It is humanly impossible to monitor and observe all the data in real time. At the same time it is equally difficult to review the stored data in historian database for any event. Some automated ways are required to analyse the measured data in real time and the data stored in historian. Basic data mining techniques like classification, clustering and pattern recognition can be used for the purpose. Data measured is stored in database in a time series form and identified by the point of measurement. i.e. we can obtain data of a location, say some substation name, and between some well defined time i.e., specify date and time up to resolution of milliseconds, to get the data. But database queries does not go beyond fetching the data. It cannot revel if fetched data pertains to a system that was in steady state or if any transient was ongoing while data was measured. Data mining techniques are employed to further analyse the data and may be classify it into different categories. The first part is dedicated to automatic identification of power system events. Here we developed methods to continuously monitor frequency measurements and identify any power system events in near real time. Sensitivity of the system and the accuracy in classifying the type of event was tested using real (i.e., not simulated) measurements. The event identification system was further extended to also identify the exact time ix

12 instance of occurrence of an event. This is an useful feature in automating the monitoring applications. Further studies were done on identifying the modes of oscillation in power systems from the frequency measurements. During small system disturbances due to, random, load and generation variations the power system can be modeled as a Linear Time Invariant (LTI) system. The properties of LTI system were reviewed and guidelines developed to apply, and obtain reliable results from, the oscillation mode identification methods. TLS- ESPRIT method was applied to identify the oscillation modes. During the identification process, both the event based identification methods and mode identification from ambient frequency measurement were studied. The results of oscillation mode identification were obtained from real measurements and the event identified using the methods developed for event identifications. The study of oscillation mode identification also revealed existence of common mode oscillation. This mode was further studied and characterized. The theory behind existence of such modes was studied and it was shown by simulations that traditional power system controllers like governors and Automatic Generator Control (AGC) do contribute in control the common mode oscillation in some range of frequencies. This assumes significance in context of India power systems. In India the AGC is not deployed and the primary control through governors actions are in limited scale. The common mode oscillations observed in Indian system is more pronounced. This was validated using actual data obtained from various real life systems. The study of common mode oscillation is useful in avoiding pit fall during oscillation mode identifications. Apart of the usual power system oscillations there are some unusual oscillations observed in real measurements. These are termed as abnormal oscillations. Analysis of such abnormal oscillations was performed through simple case studies that simulates the similar observations. It was noted that a generalized study and development of guidelines to avoid such oscillations would require further effort and this could be a target for future research. Through the studies of frequency measurements and other signals from PMUs it was felt that, to develop and study such applications, the existing simulation tools required further development. The present tools are inadequate at performing large numbers of long term transient studies. A proposal was made for extending the transient simulation tool to make it more compatible with requirements of Wide Area Protection and Control applications. Full fledged development of such a tool could also be the focus of a future research. Finally in depth analysis of the PMU, to check its adequacy for the applications around the frequency monitoring, was performed. The complete signal processing chain was analyzed from sampling right up to the output data frames of PMU. The accuracy of measurements, its steady state and transient response and latency involved in PMU x

13 measurements was studied. With this knowledge the guidelines were framed for further applications of PMU data. A proposal for a distributed PMU architecture to enable usage of PMU data locally at substation level for protection and distributed state estimation applications was made. This thesis present a comprehensive study of data mining of frequency measurements obtained through WAFMS. The study scope range from time duration of few seconds for small events like tripping of a line to few hours like ambient measurements. In terms of frequency range the thesis present analysis of frequency measurements from ultra slow common mode oscillations of below.1 Hz to fast abnormal oscillation of few Hertz. xi

14

15 Contents Abstract List of Figures List of Tables ix xx xxi 1 Introduction Theme Applications of WAFMS measurements Data Mining and WAMS data analytic PMU data example PMU data analytic Introduction To Indian Power System Auto Detection of Power System Events Introduction Ellipsoid Method Principal Component Analysis Error Ellipsoid Coarse Event Detection Steps for event detection Detection of small events Detection of major events Detection of load or generation tripping Identification of Event Instance Kalman Filter for df/dt estimation Event instance identification Summary and Conclusion Power System Oscillation Modes Identifications: Post Events, Ambient and Common Mode Introduction xiii

16 3.2 Power System Oscillations Response of Stochastic Linear Time Invariant System Validity of white noise assumption Output of LTI system when excited by white noise Some important aspects of response of stochastic LTI systems Methodology Preprocessing Mode Identification Results And Discussions Event Detection Algorithm applied to WAFMS data Post event mode identification Ambient Oscillation Mode Identification applied to WAFMS data Power System Oscillations Simulation Model Common Mode Characterization of the low frequency mode Effect of Primary and Secondary Control on Common mode Frequency Simulation of 39 Bus system Practical Observations Conclusion Abnormal Oscillations in Power Systems and Its Identifications Introduction Abnormal Oscillations Due to PSS limiters Abnormal Oscillations due to Governor Dead Band Abnormal Oscillations due to AVR limits Analysis of Abnormal Oscillations Some Discussion and Future Scope of Identification of Abnormal Oscillations from Frequency Measurements Conclusions Simulation of Power Systems for WAMS Applications Introduction WAMS Compared with Conventional SCADA Requirement of WAMS simulation Modeling of Power Systems for Quicker Simulations Fast Convolution Simulation Summary Case Study of WAMS Simulation xiv

17 5.4 Simulation of Cascade Events Conclusion Synchrophasor Measurement Characteristics Introduction The Phasor Measurement Unit Synchrophasor Measurement Synchrophasor Estimation at Off Nominal Frequency Power System Signal Model Phasor Estimation Through DFT analysis of the power system signals Phasor Reporting Rate and Swing Monitoring Timetags Implications of Synchrophasor Measurement Methods PMU as distributed system Conclusion Summary and Conclusions Summary Conclusions Autonomous Event Detection and Classification Method Oscillation Mode Identification from Ambient Frequency Measurements Existence of Common Mode Oscillations Abnormal Oscillations Distributed Architecture for PMU Future Scope of Research A Wide Area Frequency Measurement System 141 A.1 NTP based WAFMS A.1.1 Frequency measurement devices A.1.2 NTP synchronized client side computer with RTAI A.1.3 Server side computer B Solution of Linear Time Independent Systems 149 C TLS-ESPRIT 153 D New England 39 Bus System 157 E Multimachine Model 163 xv

18 Bibliography 165 Publications Based on This Work 175 Other Publications 177 xvi

19 List of Figures 1.1 Voltage at AGRA Bus, time 213/2/3 16:: Current at HISSAR Bus for line HISSAR to MOGA, time (IST) 213/2/3 16:: Positive Seq Voltage Angles Diff at Buses, time (IST) 213/2/3 16:: Positive Seq Voltage Freq. at Buses, time (IST) 213/2/3 16:: Frequency and Scatter Plot of measurements at Mumbai and Kanpur on :57: Scatter and Ellipsoid for frequency measured at Mumbai and Kanpur on :57: Error Ellipsoid - Lengths of major and minor axes Event :4:31 Sensitivity for small event detection Event :7:31 Detection of Major Event Detection of load shedding event using two ellipsoids Event :4:31 Detection of generation tripping event Performance of Kalman Filter in estimation of df/dt, using frequency data of Kanpur on :4:31. Notice that Kalman Filter eliminates the spikes in df/dt, while FIR filter does not Measured df/dt for the three events Detection of event instance Impulse Response of the simple LTI system (eqn. 3.3) Step Response of the simple LTI system (eqn. 3.3). Step response settles to steady state value of White Noise Response of the simple LTI system (eqn. 3.3) LTI excited by white noise of different bandwidths Auto covariance for response with different bandwidths Effect feed-forward on response of LTI, D = [.2] Effect of feed-forward on auto covariance of response of LTI, D = [.2] Output y and Y cov for LTI system with λ =.1 + j2π.4, ζ = Output y and Y cov for LTI system with λ =.1 + j2π.4, ζ = xvii

20 3.1 Theoretical impulse response for LTI systems λ =.1+j2π.4, ζ = and λ =.1 + j2π.4, ζ = Preprocessing Stages Comparison of two types of preprocessors Raw input data Covariance of an actual frequency signal Event-1. Auto detection of a 1 MW Generation trip event :57: Event-2. Auto detection of a fault cleared by back-up protection :35: Event-3. Auto detection of a fault cleared by primary protection :18: Modes Identified at Mumbai Through WAFMS Modes Identified at Kanpur Through WAFMS Modes Identified Through PMU measurements Frequency components of input signals Simulated common mode frequency Frequency of NEW measured by PMUs Preprocessed Frequencies of NEW grid measured by PMUs Frequency of SR measured by PMUs Preprocessed Frequencies of SR grid measured by PMUs Effect of Governors and AGC on Common Mode Frequency for IEEE 39 bus system Frequency of Western Interconnection, US measured by Frequency Meters Preprocessed Frequency of Western Interconnection, US measured by Frequency Meters Frequency of ENTSO-E, Central Europe measured by PMU Preprocessed Frequency of ENTSO-E, Central Europe measured by PMU Sustained oscillation of 2.11 Hz mainly seen at Kharagpur :54: Sustained oscillation of.38 Hz. in Kanpur and Mumbai Frequencies :11: Abnormal Oscillations observed by WAFMS frequency meters Stable oscillations due to PSS after a generation throw off event in 39 bus system. PSS limit -.1 pu Abnormal oscillations due to PSS after a generation throw off event in 39 bus. PSS limit -.4 pu. Observe the growth and decay of the oscillations in frequency signal xviii

21 4.6 Unstable oscillations due governor after fault clearing in 39 bus system. Governor dead band not applied Stable oscillations due to governor after fault clearing in 39 bus system. Governor dead band.3 Hz Abnormal sustained oscillations due to governor after fault clearing in 39 bus system. Governor dead band -.12 Hz Abnormal sustained oscillations due to AVR after opening of line:13-14 in 39 bus system. E fd limit 1.8 pu Abnormal sustained oscillations due to ARV after opening and restoration within.1 s of line:13-14 in 39 bus system. E fd limit 1.8 pu Abnormal oscillations due to AVR after opening of line: in 39 bus system. No limit of E fd Subcritical Hopf Bifurcation Supercritical Hopf Bifurcation Supercritical Hopf Bifurcation Random variation in Load at one of the Bus Internal angle response of one of the generator ω Slip of one of the generator Terminal current of one of the generator Bus Voltage at one of the bus Bus Voltage Magnitude Transmission Line Active Power Flow Signal Model Signal Spectrum Frequency Response of Single Phase Phasor Estimation Frequency Response of Positive Sequence Phasor Estimation of Three Phase Balanced Signals Frequency Response of 6 th order Butterworth filter Amplitude step response of the filter with group delay compensation Phase step response of the filter with group delay compensation Distributed PMU architecture PMU Signal Processing Block Diagram A.1 Location of the Synchronized Frequency Measurement Devices A.2 Architecture of WAFMS. All communication is through the internet A.3 Block diagram of Frequency measurement device A.4 Architecture of RTAI A.5 Hierarchy showing various levels of NTP clock sources xix

22 D.1 New England 3 Bus System xx

23 List of Tables 1.1 Mean voltage magnitude for each hour in kv Standard deviation in voltage magnitude for each hour in kv Output covariance for varying λ Mode Identification Through Prony Method Modes Statistics for 12 hours from : Modes Statistics for 12 hours from : Modes Statistics for 12 hours from : Modes Statistics for 12 hours from : from PMU data xxi

24

25 Chapter 1 Introduction THE power system operations and control is going through a revolutionary change due to integration of information and communication technology (ICT) along with the conventional power system network. The demand for higher level of efficiency in utilization of the existing power network is driving force behind these changes. Expansion of the power network is constrained by environmental concerns and economics considerations, at the same time it is called upon to cater to newer scenario like significant distributed generation and unpredictable renewable resources. It is envisaged in future that there will be ubiquitous metering of power systems quantities and two way communication between the consumers of electricity and the entities responsible for operation of the power system network. The aim of such metering is to enable power system network to operate at its fullest capacity. A situational awareness on the performance of the network enabled through the real time measurements, would enable the operators to take informed decisions in a timely manner to maintain power system security and integrity. Wide area monitoring, control and protection (WAMPAC) system is one of system that integrates ICT and power systems to deliver real time, high quality, synchronized measurements at a central location. These measurements can be used for either monitoring, control and protection of the system. WAMPAC system is built around Phasor Measurement Units (PMU). The output of a PMU is phasor representation of the measured sinusoidal quantities with a precise timestamp. The binary signals too are given out with a timestamp attached. These outputs are generally communicated to phasor data concentrator (PDC), where such outputs from several other substations are collated. In several places the PMU is considered equivalent to a disturbance recorder which measures the phasors and stores it locally as well as transmit it outside the substation. PMUs require precise and accurate time synchronization, usually through, a global positioning system (GPS) receiver. The PMU data is communicated through fiber optic links which meet the stringent requirements of bandwidth and, more importantly, the data latency. Deployment of WAMPAC system based on PMUs require extensive investment in PMU 1

26 and PDC hardware as well as recurring cost on communications. Because of this, PMUs are generally installed at important high voltage substations and the data is communicated to energy control centers. There is a requirement to extend this system to cover much more system than it is economically justifiable today. Moreover, before huge investment is committed for WAMPAC system, the decision makers require assurance of effectiveness of such systems in addressing their problems of improving system security and availability. Frequency is one parameter that is measurable at every point in a power system. Frequency measured at a low voltage level will also provide the measurements of the electromechanical modes that are observable in the high voltage bus of that area [Salunkhe and Kulkarni, 21, Zhong et al., 25]. Unlike phase angle measurement that require highly accurate clock synchronization, the frequency measurement is tolerant to an error of few milliseconds in clock synchronization. So a future can be envisaged where there is extensive deployment of just wide area frequency measurement system (WAFMS) which measures the frequency of the AC system. Such system can be built around frequency measurement devices (FMD) which can be connected to general purpose internet from where it takes the time reference and where it can communicate its measurement outputs. The source of the frequency is normal power outlet in domestic supply. In Appendix A we describe a WAFMS system built and deployed by IIT Bombay, Power Systems lab. The project is operational since mid 29, and at least several years of measurement archive is available now. This thesis uses the data collected through this system for further analysis. 1.1 Theme The future of power systems operations will be dominated by availability of several kind of measurements. With the advent of concepts of Internet of Things (IoT) [Zheng et al., 211] created around intelligent interconnections of diverse objects in the physical world there is torrent of data availability with the power system operators. IoT utilize low cost sensors to gather information that facilitate fast paced interactions between entities at any place and at any time. The FMD of WAFMS represent a type of IoT device. It is embedded in the system, constantly monitoring the frequency and communicating it. The data gathered by systems like WAFMS is unlike any other data seen by the power system operation personnel. First of all this data arrives at a fast rate of around 5 samples per second, the other is that there would be hundreds of such data streams sending the measurements simultaneously. Such a scenario requires a new way of thinking about concepts like situational awareness [Panteli et al., 213]. A sufficient situational awareness is achieved when the power system operators are able to receive, and interpret 2

27 correctly and in timely manner, the required information. This allows them to react effectively to an electrical disturbance, preventing its spread and minimizing its impact. The WAFMS solves the data receiving aspect of the situational awareness. In this thesis we try to address the part of correct interpretation of information in a timely manner. We present applications that convert the raw measurements into appropriate information regarding the state of power system operation, which enable the system operators to take right actions. There are three attributes of situational awareness viz., perception, comprehension and projection. Perception means obtaining and displaying the information in real time. WAFMS does this part with the frequency measurements. By comprehension, it is meant that understanding of what the perceived data means with respect to the objective of the operators. The concepts developed in this thesis are directed towards enhancing the comprehension of the operators. The extension of this work can be on the projection aspect, where the operators get a clear idea about the impact of any decisions they take based on their perception and comprehension of the power system situation. 1.2 Applications of WAFMS measurements We concentrate on utilization of the frequency measurement data collected through WAF- MS. These applications gives us insights into operation of power systems and its state i.e., whether the system is in steady state or it is experiencing an event. If it is in steady state, then whether it is in its usual state or some state that represent insecure operation or depleted network operation. In case of events we can identify the type of event, its instance of occurrence, location of event, severity of event etc. Moreover, we can also deduce if the event is normal power system event like fault clearance, tripping of generation or load etc., or whether the event represent some abnormal operation of the system. There are several incidents of power system oscillations measured through this system, some of them are reported in [Salunkhe and Kulkarni, 212]. The swings arising out of the large disturbances are of great interest and provide important information regarding the system behaviour. In chapter 2, we propose an analysis to detect power system events using the wide area frequency measurements. Most of the time the power system is in an ambient condition. However, there are always several events per day occurring in the system. This chapter presents a method to automatically detect such events. FMDs measure bus voltage frequency and communicate them to a central location in real time at rate of one sample per power system cycle. These measurements are fed to the auto event detection algorithm. The method of detection of event, event classification and exact instance of event occurrence is presented. The issues encountered in implementing such a scheme are discussed and solutions proposed. 3

28 Chapter 3 concentrates on utilization of the frequency measurements to identify the oscillation modes of the power system. The frequency, amplitude and the damping factor of these oscillations provide information regarding security of power system network. Poor damping may mean insecure operation, where any further tripping of lines can lead to instability. Similarly there are usually few well defined inter area oscillation modes with a known frequency of oscillation. Deviation of this value may indicate unusual operation may be with weak network or changed generation mix. There are two ways of oscillation mode identification, the first one takes short duration of frequency data during post event period. This is called event based oscillation mode identification. The post event scenario presents a condition when there are several electromechanical oscillation modes excited. Observing and quantifying the frequency, amplitude and damping ratio with good accuracy is possible by applying the methods described in chapter 3. However, this method has disadvantage that one has to wait for the event like fault or load or generation tripping, such events are inherently infrequent. Another disadvantage is that one has to automatically identify any event, carefully find the instance of event and select a small portion of the relevant signal that can be used for oscillation mode identification. The techniques proposed in chapter 2 are useful in alleviating this shortcoming. We can also monitor the power system electromechanical oscillation modes through observation of frequency variation due to ambient disturbances. This compliments the event based oscillation mode identification, by giving continuous tracking of the oscillation mode frequency and its amplitude. The drift in either frequency of mode or its amplitude from its usual values may indicate the changed state of operation. Calculation of these oscillation mode from ambient frequency measurements is also discussed in Chapter 3. We present the theory behind these mode identification method and also discuss the practical aspects of applying the method on the measurements obtained through WAFMS. A systematic approach in preprocessing is required for a reliable, accurate and consistent calculation of electromechanical oscillation modes through ambient deviations in frequency measurements. It was observed that the real frequency measurements had some components of very low frequency oscillations. These oscillations are the common mode oscillations caused by the random variations in the load and generation imbalance in the system. They are seen throughout the system. If not filtered out, these oscillations may mask the actual electromechanical oscillations. Chapter 3 also provides discussion in causes and characteristics of these common mode oscillations. Sometimes due to nonlinearities in the control systems, the power system may enter into a sustained oscillatory state. These sustained oscillations are also reflected in the frequency measurements. We call such sustained oscillations as abnormal oscillations. In chapter 4, we present some real observations of abnormal oscillations in WAFMS and also show simulation results that replicate the nonlinear oscillatory behaviour of the power 4

29 systems resulting in to abnormal oscillations. There is a future scope for research in identifying the cause and mitigation of such oscillations. During our studies of the utilization of WAFMS measurements, it was felt that there is a requirement of some simulation tool that can provide simulated output that is similar to the one obtained from WAFMS system. The data obtained from actual WAFMS is usually from a system that is operating in a secure state. The applications like event detection and oscillation mode identification would required to be tested for many different states of power systems. A simulation tool could provide a wide variety of such data through simulations. Such a tool would be useful for testing and developing the applications that would ultimately be used with WAFMS outputs. Formulation and preliminary development of such a simulation tool is presented in chapter 5. Finally we look at the PMUs in chapter 6. Ultimately all the analytic developed in this thesis can also be extended to the frequency measurements obtained for the PMUs. But, PMUs can do much more than just frequency measurements, hence, here we explore the signal processing aspects of the PMU measurements. This study helps us in understanding the capabilities of PMU measurements and also get a perspective on their limitations. This would help in focusing future research similar to that presented in this thesis in building applications that use PMU measurements. During course of this analysis, an proposal for developing an open architecture PMU also emerged, which is discussed at the end of the chapter. We summarize the complete thesis and bring out the salient points and contribution of the work in the final chapter. We end with conclusions based on our studies and present the directions for future scope of research. Before going into the main applications and algorithms on WAFMS we present some background on the general WAMS data analytic and data mining. All the frequency measurements used in the thesis are obtained from Indian power system. A brief introduction to the system is therefore presented next. 1.3 Data Mining and WAMS data analytic WAFMS and WAMS capture continuous stream of data in real time. It is humanly impossible to monitor and observe all the data in real time. At the same time it is equally difficult to review the stored data in historian database for any event. Some automated ways are required to analyse the measured data in real time and the data stored in historian. Basic data mining techniques like classification, clustering and pattern recognition can be used for the purpose. Data measured is stored in database in a time series form and identified by the point of measurement. i.e. we can obtain data of a location, say some substation name, and 5

30 between some well defined time i.e., specify date and time up to resolution of milliseconds, to get the data. But database queries does not go beyond fetching the data. It cannot revel if fetched data pertains to a system that was in steady state or if any transient was ongoing while data was measured. Data mining techniques are employed to further analyse the data and may be classify it into different categories. The classification applications could be mapping the data in a predefined groups. For example, the measured voltage could be classified as under voltage, normal voltage or over voltage. We can calculate measures like duration of under voltage, and whether it is continuous or intermittent. Similar analysis could be done for calculating other aspects like voltage unbalance, active and reactive power flow over transmission lines etc. Classification algorithms require that the classes be defined based on data attribute values [Dunham, 23, Sec , page 5]. Further, if the data is classified not by any predefined rules but just by their attributes then it is called clustering. Finally any particular instance or segment of data can be isolated through pattern recognition algorithm [Trachian, 21]. Known event patterns can be used to train an agent to recognise specific pattern and ignore noise and ambient data. This agent than can be used in real time or batch process to work on historian PMU data example Actual PMU measurement data was obtained from Indian Power Network. The data pertains to five 4 kv bus voltages and currents of two 4 kv overhead transmission lines. The basic characteristics of the data is mentioned below. Data pertains to Five 4 kv Buses and Two 4 kv Transmission Lines. The data consist of synchrophasors of the three phase voltages and three phase line current and bus frequencies. The data is sampled at 4 ms. i.e. 25 phasors per second. Total duration of data is eight hours in separate files for each hour. When stored in ASCII format the total memory is 47 MB. Same data stored in scaled 16 bit fixed point (COMTRADE) format occupy 84 MB. Actual data is never clean and clear. There is always some missing data, some outliers etc. The values corresponding to missing data frames need to be handled appropriately If represented by NaN or they create problems in unwrap, mean and standard deviation calculation. The analysis of the data in this manner is mentioned below. There are missing data in all signals for all hours. 6

31 Table 1.1: Mean voltage magnitude for each hour in kv Location Signal 16: 17: 18: 19: 2: 21: 22: 23: Van Agra Vbn Vcn Van Bassi Vbn Vcn Van Hissar Vbn Vcn Van Kishenpur Vbn Vcn Van Moga Vbn Vcn The best availability per hour is 99.8 % i.e. 182 missing frames per 9 frames of data in one hour. The worst availability per hour is % i.e missing frames per hour. The next step is to summarize the measurement. The voltage measurement is taken as an example to demonstrate how basic classification can be performed. The voltage magnitude varies from 1.6 pu (Max) to.98 pu (Min). Possible effects of switching of shunt reactors observed, see Agra, 18: and 19: hour data. Table 1.1 gives basic summary of voltage magnitude. The mean voltage magnitude measured for each hour for eight hour period is calculated. Table 1.2 shows the standard deviation of the voltage during that period. It can be observed that voltage magnitude at Kishenpur bus has a higher standard deviation compared to all others. Similar exercise could be done for the current over transmission lines. Further we can derive sequence voltage and currents, voltage unbalance and active and reactive power over transmission line. The voltage and current angle, power factor, frequency and rate of change of frequency can also be calculated. Figs. 1.1, 1.2, 1.3 and 1.4 shows plots of data obtained by PMU measurement for one hour duration. These are typical of any PMU data. It can be noted that some data mining is required to extract meaningful information from these raw measurement. 7

32 Table 1.2: Standard deviation in voltage magnitude for each hour in kv Location Signal 16: 17: 18: 19: 2: 21: 22: 23: Van Agra Vbn Vcn Van Bassi Vbn Vcn Van Hissar Vbn Vcn Van Kishenpur Vbn Vcn Van Moga Vbn Vcn PMU data analytic The PMU data could be used to develop some analytic that can help in better operation and control of the power systems. Some applications may be offline while other can be in realtime. Typical analytic could consist of: 1. Vulnerability check of protection system. This is an offline non real time application. The report on vulnerable relays after any major event can be generated through this application. This tool can be set up in continuous polling mode for all relays. There is a well known problem of the distance protection relays that work on impedance measurement principle being vulnerable to false tripping due to power swings and heavy load conditions. The issue is more pronounced on the relays that are installed on long lines carrying large power. Usually such long lines are important tie-lines connecting two bigger power system areas. Unexpected loss of such lines pose a risk in secure power system operation. The PMUs measure bus voltage and line currents phasors and communicate them to a central location in real time at rate of one sample per power system cycle. These measurements are be fed to a mimic of the distance relays of the lines to continuously monitor on-line the vulnerability of the relay to false tripping due to power swings and load encroachment. This system is built at the system control center and is simultaneously monitoring several lines. Any event is immediately logged and an alarm is raised in real time for some important events to bring it to notice of the system operators immediately. The consolidated log of the events is available 8

33 Voltage at AGRA Bus, time 213/2/3 16:: Van Vbn Vcn Voltage Mag Van, Vbn, Vcn (kv) Time (sec) Figure 1.1: Voltage at AGRA Bus, time 213/2/3 16:: Current at HISSAR Bus for line HISSAR to MOGA, time (IST) 213/2/3 16:: Ia Ib Ic Current Mag Ia, Ib, Ic (A) Time (sec) Figure 1.2: Current at HISSAR Bus for line HISSAR to MOGA, time (IST) 213/2/3 16:: 9

34 Positive Sequence Voltage Angle, Van (deg) Positive Seq Voltage Angles Diff at Buses, time (IST) 213/2/3 16:: AGRA BASSI HISSAR KISHENPUR MOGA Time (sec) Figure 1.3: Positive Seq Voltage Angles Diff at Buses, time (IST) 213/2/3 16:: with the system operators and planners, with the snapshot of the complete system condition at time tag of the event. The analysis of this log is important from point of view of a.) updating the relay settings, b.) identifying insecure system operation conditions and c.) and future planning of the network. 2. PMU Based CT/VT and Line parameter calibration. This is an offline exercise. Transmission line parameter estimation is one of the first applications that was built around PMU data. The line parameter estimation concerns with derivation of positive sequence π-equivalent transmission line parameters like series impedance and shunt admittance using the measurements obtained by PMUs connected at the both ends of the line. The PMUs must provide synchronised data of voltage and current of both the ends of the lines. There are several methods reported in literature that can be used for this calculations. The main approach of all these methods is to treat the PMU measurements as accurate. If at all some errors are considered they are modeled in the instrument transformers i.e. Current Transformers and Capacitive Voltage Transformers. In most cases the errors are treated as zero mean normally distributed noise. The standard deviations of the distribution are taken from the accuracy limits prescribed in standards. Experience with all these methods with actual PMU measurement data suggest that the assumption of just zero mean noise error is not very well supported. It can be 1

35 Positive Sequence Voltage Freq., Van (Hz) Positive Seq Voltage Freq. at Buses, time (IST) 213/2/3 16:: AGRA BASSI HISSAR KISHENPUR MOGA Time (sec) Figure 1.4: Positive Seq Voltage Freq. at Buses, time (IST) 213/2/3 16:: shown that actual data also has significant systematic error. In other words the error is not zero mean but has some bias. Under such conditions the performance of the line parameter estimation degrades very quickly. The differences between approaches of the different algorithms also get highlighted. Some algorithms are more tolerant to systematic errors than other. It is also noted that the errors in the estimated parameters is different for each parameters. The estimation of line series resistance is very sensitive to any error, while line series reactance is relatively tolerant to measurement errors. It must be noted that for a single transmission line parameter estimation we require measurements from twelve channels i.e., three phase voltages and currents from both ends. Each channel is independent from each other. We must also consider possibility of small percentage of unbalance in line currents and bus voltages. In this paper we develop methods of calculation uncertainty in the estimation of the line parameters due to normal and systematic errors. 3. PMU Based Linear State Estimator. Power systems are operated by system operators from the load dispatch centers. The system operator maintains the system in the normal secure state as the operating conditions vary during the daily operation. In order to do that, the system condition needs to be monitored and the system state identified on a continuous ba- 11

36 sis. The system may move into one of the four possible states, namely secure, alert, emergency and restorative as the operating conditions change [Dy Liacco, 1974]. In case the system is found to be insecure, proper preventive actions are taken to bring the system back to its secure state. This sequence of actions is referred to as the security analysis of the system. The first step of security analysis is to monitor the current state of the system. This involves acquisition of measurements from all parts of the system and then processing them in order to determine the system state. Substations are equipped with devices called remote terminal units (RTUs) which collect various types of measurements from the field and are responsible for transmitting them to the control centre. These devices are connected to a local area network (LAN), and a SCADA host computer at the control centre receives measurements from all the monitored substations via one of many communication links such as fiber optics, satellite, microwave etc [Abur and Exposito, 24]. In a conventional state estimator also known as the static state estimator, measurements received at the control centre includes line power flows, bus voltage and line current magnitudes, generator outputs, loads, circuit breaker and switch status information, transformer tap positions and switchable capacitor bank values. These raw data and measurements are processed by the state estimator in order to filter the measurement noise and detect errors. Hence, a state estimator, essentially, removes the impurities from the measurements and converts them into states (positive sequence voltage phasors), which the computer of the control centre can make use of, to make decisions on system economy, quality and security [Schweppe and Handschin, 1974]. In a conventional state estimator, the algorithms use asynchronous measurements of real and reactive power flows and voltage magnitudes. The system states i.e., the voltage phasors are calculated from these measurements. The equations relating these asynchronous measurements and the system states are non-linear in nature. In order to calculate the states, iterative weighted least squares methods are used. Although these state estimators are capable of solving very large systems, they are prone to convergence errors when the system is in a stressed state. As such, these state estimators are unreliable under certain conditions. With the advent of synchronized phasor measurement units (PMUs), these problems can be eliminated from the existing state estimators [Phadke et al., 29]. PMUs contain devices that are synchronized by a common timing signal, eg. from Global Positioning System (GPS) in order to produce phasors that are synchronized over a wide area. Since the PMUs directly measure positive sequence phasors of voltages and currents on a common reference, the system states can be estimated using a linear least square state estimator. 12

37 WAMS technology has been in use for more than a decade now. One of the most important features of this technology is that the measurements are timestamped with high precision at the source, so that the data transmission speed is no longer a critical parameter in making use of this data. PMU measurements having the same time-stamp are used to estimate the state of the power system at the instant defined by the timestamp. The timetags associated with the phasor data provide an indexing tool which helps create a coherent picture of the power system out of such data. Apart from the state estimation calculations, the state estimator also includes the three important functions, network topology processor, observability analysis and bad data detection. 4. PMU Based supervised Zone-3 Protection Scheme. Distance relays are widely used for transmission line protection. These relays also provide remote backup protection for transmission lines. However, there are a few issues with backup protection as provided by distance relays Zone-3 based remote backup protection schemes are dependable but not secure. A relay mal-operation can act as a catalyst or even trigger a system collapse situation. Incorrect Zone 3 relay operation may be a consequence of either (a) quasi-stationary events like load encroachment, overload, undervoltage etc., or (b) electromechanical oscillations like power swings. To overcome the problems mentioned above, the secure remote backup protection of transmission lines using synchrophasors scheme is developed. It is an adaptive remote backup protection scheme using output of the linear least square state estimator. This subject has been extensively researched and details are given in [Navalkar and Soman, 211]. In this module, decision making at the analytic engine as well as closing the loop implementation is considered. However, this activity may involve cooperation from other SCADA, substation automation and relay suppliers. It also poses a stringent requirement of latancies in communication. 5. Schemes for controlling angular instability. Load encroachment and out of step conditions can lead to mal operation of distance relays. Load encroachment refers to reduction in the magnitude of the impedance seen by distance relay due to either large load currents or reduction of bus voltage. 13