POWER SYSTEM STABILITY Analysis by the Direct Method of Lyapunov M.A.PAI NORTH-HOLLAND PUBLISHING C¢ AMSTERDAM *NEW YORK *OXPORDISTH-HOLLAND PUBLISHING COMPANY, 1981 yarn recantengal i tuoi arhygctny wads, eet nie wieebernse fee cag er, at athens ISBN: 444 868108 Publishers. NORTH-HOLLAND PUBLISHING COMPANY AMSTERDAM * NEW YORK + OXPORD Sote dighribuions fur the US A. and Cand ELSEVIER NORTILILOLLAND, ING 52 VANUBEBILT AVENUT NEY YORK. N.Y, 1uOT? Library of Cangevan Cataloging i Palilieation Daty obabi Tiny. syetemy and eotol gerd sae uyeben stability. Lt, Series. THERLANDS PRINTED IN THE NiPREFACE Research work in applying Lyapunov Stability theory to power systems has been pursued actively for the last decade and a half. The amount of literature is quite vast and there is now strong evidence that the method may find eventual application in planning and on-line dynamic security assessment. It was, therefore, felt that it is now appropriate to collate, unify and present in a cohesive conceptual framework both the theory and application aspects of the direct method of stability analysis for power systems. After a physical statement of the problem in the introductory chapter, Lyapunov's stability theory is reviewed in Chapter Two. We develop the mathematical models for multi-machine power systems in Chapter Three stressing both the center of angle formulation and the Luré formulation of the equations. The systematic construction of Lyapunov functions for single and multi-machine power systems is the theme of Chapter Four. It is shown that the energy function is one of the many Lyapunov functions that can be constructed. The computation of the region of attraction of the post-fault stable equili- brium point is the most difficult problem in Lyapunov stability analysis of power systems. The recent approaches which make use of the faulted dynamics of the system are emphasized in Chapter Five as opposed to earlier results which gave con- servative estimates of the critical clearing time. Two case studies are presented. in Chapter Six we discuss the stability of large scale power systems by decomposition (often termed as the vector Lyapunov function approach). Both the weighted sum scalar Lyapunov function and the comparison principle-based vector Lyapunov function approaches are discussed. The final chapter concludes with a survey of some recent results in this rather active area of research and some thoughts on problems still requiring a solution. In the Appendix we have derivedPREFACE the detailed and simplified models of the synchronous machine. thd book is aimed at a research audience as well as an appli- cation oriented power engineer who needs to understand the basis of direct methods of stability analysis in power systems. The book can be used in a graduate level course in power system stability or dynamics. In writing this book, I have been influenced considerably by the work of many other research workers in this field whose works are cited. In particular, I would like to mention the report "Transient Energy Stability Analysis" by Systems Control Inc. in Proceedings of the Engineering Foundation Conference on 'System Engineering for Power: Emergency Operating State Control' sponsored by the U.S. Dept. of Energy at Davos, Switzerland in Oct. 1979. I also wish to thank Professor A. N. Michel of Iowa State University for numerous interesting discussions on the stability of large scale dynamical systems. I would like to thank my former graduate students at I.I.T. Kanpur who got excited by this area of research and worked with great enthusiasm. My gratitude to I.1.T. Kanpur is an eternal one for nurturing an atmosphere of academic excellence and faculty support for research. I would like to thank the Council of Scientific and Industrial Research and the Dept. of Science and Technology, Govt. of India for supporting my research work at I.1.T. Kanpur. The writing of this monograph was completed during my stay at Iowa State University during 1979-81. I am very grateful to Professor J. O. Kopplin, Chairman, Dept. of Electrical Engineering, for his constant encouragement and unfailing support. The support of the Engineering Research Institute of Iowa State University is also acknowledged. I would like to thank Shellie Siders for her speedy and excellent job of typing the manuscript. Writing of this book has taken long hours away from my family, especially my son, Gurudutt, whom I thank for his understanding. M. A. Pai June 1981 viiiDedicated to Indian Institute of Technology, Kanpur, IndiaCHAPTER I INTRODUCTION 1.1 Introduction Any physical system that is designed or operated to perform certain preassigned tasks in a steady state mode must, in addi- tion to performing these functions in a satisfactory manner, be stable at all times for sudden disturbances with an adequate margin of safety. When the physical system is large and com- plex such as a typical modern interconnected power system, investigation of stability requires both analytical sophistica- tion in terms of techniques employed and practical experience in interpreting the results properly. In the last decade and a half and in particular after the famous blackout in N.E. U.S.A. in 1965, considerable research effort has gone into the stability investigation of power systems both for off-line and on-line purposes. At the design stage although the planner takes many contingencies into con- sideration, in subsequent operation and augmentation of the network, new considerations arise which were not foreseen by the planner. Hence, an entirely different pattern of system behavior can be expected under actual operation conditions. This is particularly true regarding the capability of the sys- tem to maintain synchronism or stability due to sudden unforeseen disturbances such as loss of a major transmission line, load or generation. The tools suitable for off-line studies such as simulation may not be suitable for on-line application since a large number of contingencies have to be simulated in a short time. A technique which offers promise for this purpose is Lyapunov's method. The appeal of this method lies in its ability to compute directly the critical2 POWER SYSTEM STABILITY clearing time of circuit breakers for various faults and thus directly assess the degree of stability for a given configura- tion and operating state. The critical clearing time can also be translated in terms of additional power disturbances that the system can withstand, thus offering a tool for dynamic security assessment. For off-line applications, the method can serve as a complementary role to simulation by selecting quickly the important faults which need to be studied in detail. The state of art in this field resulting from nearly two decades of research work has reached a certain degree of maturity and for the first time offers possibilities of appli- cation by the electric utilities. The purpose of this book is to present Lyapunov's method and its systematic application to power system stability. Perhaps for no other physical system has Lyapunov's stability theory been applied so extensively and vigorously as for power systems. It is difficult to precisely pinpoint historically the first application of Lyapunov's method for the investiga- tion of power system stability. Since Lyapunov's function itself is a generalization of the energy function, one can say that Lyapunov's method indeed was applied implicitly to the power system problem when the equal area criterion was first proposed. The earliest work on applying the energy function to determine the stability of a power system was by Magnusson in 1947 [1]. An energy integral criterion as applied for multi- machine system was proposed by Aylett in 1958 [2]. Lyapunov's method as we know it in control literature today was first proposed as a solution to the power system stability problem by Gless [3] and El-Abiad and Nagappan [4]. In Ref. [4], the method was put on a firm footing as a powerful tool for stability analysis by power engineers through a realistic example of computing the critical clearing time and giving an algorithmic procedure for computer implementation. The early research such as those by Willems [5], Fallside and Patel [6], Dharma Rao [7], Yu and Vongasuria [8], and Siddigee [9] concen- trated on applying different methods of constructing Lyapunov functions for power systems. Di Caprio and Saccomanno [16]I INTRODUCTION 2 and Luders [11] related the physical aspects of the problem to the mathematical model and constructed an energy type Lyapunov function for multimachine systems. In 1970, the systematic construction of Lyapunov function to power systems using Popov's criterion was proposed by Willems [12], Willems and Willems [13], Pai, Mohan and Rao [14] and Pai [15]. It was soon realized that the more important aspect of applying the method to practical problems, namely the computation of criti- cal clearing time, lay in computing accurately the region of attraction of the post-fault system. This has received considerable attention by Ribbens-Pavella beginning with her work in 1971 [16,17]. Since 1976, there has been an explosive growth in the research work on Lyapunov stability analysis of power systems. A number of state-of-the-art papers have appeared since then. In particular, the papers by Willems [18], Ribbens-Pavella [19], Saccomanno [26] and Fouad [21] summarize thoroughly the research efforts until 1975. Although the power industry evaluated the potential application of Lyapunov's method for large systems in 1972 [22], the method still had its drawbacks in terms of not giving consistently good results for critical clearing time for a wide range of fault conditions. Hence, largely theoretical aspects of the problem were pursued by the research workers in the 76's. Recently a significant breakthrough in obtaining excellent practical results has been achieved by Athay, Podmore and Virmani in U.S.A. [23], Ribbens-Pavella in Europe [24] and Kakimoto, Ohsawa and Hayashi in Japan [25]. Hence, what once appeared to be a mere academic exercise and viewed with skepticism by the power industry because of its conservative nature of results is today gaining acceptance both as an off- line tool for planning purposes as well as for on-line security assessment schemes. An alternative method of investigating stability of large scale power systems is through decomposition and subsequent aggregation. The concept of vector Lyapunov function proposed by Bailey [26] and extended to a wide class of composite systems by Michel and Porter [27], Michel and Miller [28], and Siljak [29,36] was first applied to power systems by Pai and Narayana [31].4 POWER SYSTEM STABILITY Improvements in decomposition techniques in terms of reducing the number of subsystems as well as handling transfer con- ductances were proposed by Jocic, Ribbens-Pavella and Siljak [32]. However, results from a practical viewpoint are still conservative. In the future, this promises to be an area of active research and any breakthrough will have a major impact on the stability analysis of large scale power systems. There are two other promising research areas. One consists in stability analysis by retaining the topology of the trans- mission network [33] thus offering the possiblity of a better understanding of the phenomenon of loss of synchronism and also accomodating nonlinear load representation [34]. The other direction in which research is being pursued is incor- porating accurate models of synchronous machines, excitation systems, etc. [35,36]. 1.2 Statement of the Problem The intuitive idea of stability of a physical system is as follows: Let the system be in some equilibrium state. If on the occurrence of a disturbance, the system eventually returns to the equilibrium position, we say the system is stable. The system is also termed stable if it converges to another equilibrium position generally in the proximity of the initial equilibrium point. If the state of the system "runs away" so that certain physical variables go on increasing as t + ~, then we say the system is unstable. These intuitive ideas are also applicable to a power system. Suppose at a given time t = t,, the power system is operating in a synchronous equilibrium and frequency equilibrium, i.e. the rotor angles of the different machines with respect to a rotating synchronous reference frame are fixed and the system frequency is constant. The constant nature of frequency implies that the system as a whole is at rest or the inertial center of the system is not accelerating. The occurrence of a disturbance tends to momentarily alter the synchronous equili- brium and the frequency equilibrium. The disturbance may either be small or big and the system may become unstable inI INTRODUCTION 5 either event depending on the operating condition at t = to: The study of stability in the presence of small disturbances constitutes what is known as "static stability" or "dynamic stability" analysis in the literature. The mathematical model for such a study is a set of linear time-invariant differential equations. When the disturbances are large, the nonlinearities inherent in the power system can no longer be ignored and the study of stability under such circumstances constitutes what is known as "transient stability" analysis. The mathematical model for such a study is a set of nonlinear differential equations coupled with a set of nonlinear algebraic equations. The dimensionality of the mathematical model both for dynamic and transient stability can be very large, for even moderately sized systems. We shall restrict ourselves to the transient stability aspects and now look at the physical phenomenon. Assume that the system is in steady state with a frequency @ and the rotor angles 8 of the various machines with respect to the synchronously rotating reference frame (i.e., 85 = a, 7 wot where as is the angle with respect to a fixed reference) assuming constant values. Under these conditions, the power supplied by the generators exactly matches the power absorbed by the loads plus the electric power loss in the transmission system. At t = tor suppose a large perturbation in the system occurs, such as the loss of a major line or a fault at one of the buses. This disturbance upsets the energy balance existing prior to the disturbance. For example, a fault on a transmission line blocks the transfer of power to loads from certain nearby generators and will try to absorb power from other generators. This results in an excess or deficit of mechanical power supplied over the electrical power at each generator in the system resulting in acceleration or deceleration of the rotors. The rotor angles 83 with respect to the synchronously rotating reference frame %% either increase or decrease as a function of time. A typical behavior or rotor angles is shown in Fig. 1.1. When the fault is removed by isolating the faulted line so that an energy balance is again possible, we have the problem of the6 POWER SYSTEM STABILITY redistribution of the excess kinetic energy acquired by the system until the instant of fault clearing. If the system after the fault is cleared can absorb this energy, the system is considered to be transiently stable; if the system cannot absorb all of this energy, then instability results. Figure 1.1 shows a system which is transiently stable and Fig. 1.2 an unstable system. When the system is transiently stable, it may recover to its original frequency or settle down to a new frequency. Knowledge regarding transient stability of the system is important both to the system planner and the opera- tion engineer. The system planner has to coordinate the relay settings so that a given fault is cleared in time so as not to result in instability. A valuable piece of information in this respect is the critical clearing time Tue of circuit breakers which isolates the faulty portion from the rest of the system. A fault cleared before tor will result ina stable system and a fault cleared after toy will result in the loss of synchronism. In order to compute Fart the system planner will have to conduct several stability runs each time fixing the clearing time te and simulating the entire system. If the system is stable, ee is increased and the simulation is repeated. This is done until at some time when t., = ty, an inspection of the swing curves reveals instability. This is a tedious process besides being expensive in terms of computer time. Considering the fact that this procedure has to be repeated for different faults in the system, the need is evident for some direct method of computing tor This is precisely sought to be done by Lyapunov's method. In on-line applications, it is necessary to convert the infor- mation about the critical clearing time into some kind of a security index for interpretation by the operator. For example, the difference between the actual clearing time and the critical clearing time can be converted into a transient security index [37]. It is also possible to consider the effects of the outages on the transient stability. The appli- cation of Lyapunov's method to transient security monitoring is a relatively new area of research [38,39,40].i ROTOR ANGLE 6, i ROTOR ANGLE 6. I INTRODUCTION i i ne a t Fig. 1.1. Stable system. t Fig. 1.2. Unstable system.8 POWER SYSTEM STABILITY in Lyapunov's method, a certain scalar function of system state variables known as Lyapunov function V(x) is computed first analytically. In many cases, the system energy is a good enough candidate for V(x). We also compute a region of stability around the post fault stable equilibrium point denoted by the expression V(x) < Voy where Vor is the critical value of the Lyapunov function. In terms of energy, Yow is the critical energy of the system which if exceeded during the fault-on period, the system will not be able to absorb it subsequently. ‘There are many ways of computing V,, and nore recently, to get more accurate estimates of critical clearing time, the computation of Vows is made fault dependent instead of being valid for all faults in the system for a given post- fault configuration. Also considerable research effort has been directed towards constructing better Lyapunov functions, accommodating accurate models of synchronous machines and extending the methodology for large scale systems through decomposition and vector Lyapunov function approach. All the collective research work done so far holds promise for Lyapunov's method being used as a tool for planning as well as on-line security assessment. As a planning tool, Lyapunov's stability analysis can be performed on a large number of faults to compute tor and consider for detailed study only those which are relatively less stable. In real time operation, the difference between Vise and the value of V(x) at t = to can be converted into stability margins for different contingencies and have the information displayed to the operator in a meaningful manner. 1.3. Chapter outline We first give a basic review of Lyapunov stability theory in Chapter II. Various methods of constructing Lyapunov functions are covered. Lyapunov stability theory is discussed in slightly greater detail so as to serve as an independent and self-contained treatment of the subject. It is then followed in Chapter III by developing the mathematical model of multi- machine power systems. Both the center of angle reference frame and machine angle reference models are developed. TheI INTRODUCTION 9 systematic construction of Lyapunov function for both single and multimachine models is then emphasized in Chapter IV using control theoretic as well as physical reasoning. For the single machine case, the region of attraction is computed. Computation of the region of stability for the multimachine case is the single biggest impediment to the method being used for on-line purposes. This is discussed in Chapter V. The transient energy function approach which takes into considera- tion the effect of faulted dynamics in the computation of Veg and which has resulted in the accurate computation of toe is stressed. In Chapter VI, the vector Lyapunov function approach via decomposition is discussed. In the final chapter, some of the recent extensions and refinements in power system stability analysis by Lyapunov's method are briefly reviewed.1¢ POWER SYSTEM STABILITY References Less 2. ic. il. 12. 13. 14. 15s Magnusson, P. C., "Transient Energy Method of Calculating Stability", AIEE Trans., Vol. 66, 1947, pp. 747-755. Aylett, P. D., "The Energy Integral-Criterion of Transient Stability Limits of Power Systems", Proceedings of IEE (London), Vol. 165C, No. 8, Sept. 1958, pp. 527-536. Gless, G. E., "Direct Method of Lyapunov Applied to Transient Power System Stability", IEEE Trans. PAS, Vol. 85, No. 2, Feb. 1966, pp. 159-168. El-Abiad, A. H. and Nagappan, K., "Transient Stability Regions of Multimachine Power Systems", IEEE Trans. PAS, Vol. 85, No. 2, Feb. 1966, pp. 169-178. Willems, J. L., "Improved Lyapunov Function for Transient Power System Stability", Proceedings of IEE (London), Vol. 115, No. 9, Sept. 1968, pp. 1315-1317. Fallside, F. and Patel, R., "On the Application of Lyapunov Method to Synchronous Machine Stability", International Journal of Control, Vol. 4, No. 6, Dec. 1966, pp. 561-513. Rao, N. D., “Generation of Lyapunov Functions for the Transient Stability Problem", Trans. Engineering Institute of Canada, Vol. 11, Rep. C-3, Oct. 1968. Yu, Y. N. and Vongasuriya, K., "Nonlinear Power System Stability Study by Lyapunov Function and Zubov's Method", IEEE Trans. PAS, Vol. 86, No. 12, Dec. 1967, pp. 1486- 1485. Siddigee, F., "Transient Stability of AC Generator by Lyapunov's Direct Method", International Journal of Control, Vol. 8, No. 2, Aug. 1968, pp. 131-144. Di Caprio, U. and Saccomanno, F., “Application of Lyapunov's Direct Method to the Analysis of Multimachine Power System Stability", PSCC Proceedings, Rome, Italy, 1969. Luders, G. A., "Transient Stability of Multi-Power System via Direct Method of Lyapunov", IEEE Trans. PAS, Vol. 1, Jan./Feb. 1971, pp. 23-32. Willems, J. L., “Optimum Lyapunov Functions and Stability Regions for Multi-Machine Power Systems", Proceedings of IEE (London), Vol. 117, No. 3, March 1970, pp. 573-578. Willems, J. L. and Willems, J. C., "The Application of Lyapunov Methods to the Computation of Transient Stability Regions of Multi-Machine Power Systems", IEEE Trans. PAS, Vol. 89, No. 5, May/June 1976, pp. 795-861. Pai, M. A., Mohan, M. A. and Rao, J. G., "Power System Transient Stability Region Using Popov's Method", IEEE Trans. PAS, Vol. 89, No. 5, May/June 1976, pp. 788-794. Pai, M. A., "Power System Stability Studies by Lyapunov - Popov Approach", Proc. 5th IFAC World Congress, Paris, 1972.16. 17. 18. 19. 26. 21. 22. 23. 24. 25s 26. 27. 28. I INTRODUCTION 11 Ribbens-Pavella, M. "Transient Stability of Multimachine Power Systems by Lyapunov's Method", IEEE Winter Power Meeting, New York, 1971. Ribbens-Pavella, M. and Lemal, B., "Fast Determination of Stability Regions for On-line Transient Power System Studies", Proceedings of IEE (London), Vol. 123, No. 7, July 1976, pp. 689-696. Willems, J. L., "Direct Methods for Transient Stability Studies in Power System Analysis", IEEE Trans. Automatic Control, Vol. A6-16, No. 4, Aug. 1971, pp. 332-341. Ribbens-Pavella, M., "Critical Survey of Transient Stability Studies of Multi-Machine Power Systems by Lyapunov's Direct Method", Proceedings of 9th Annual Allerton Conference on Circuits and System Theory, Oct. 1971. Saccomanno, F., "Global Methods of Stability Assessment", Proceedings of Symposium on Power System Dynamics, The University of Manchester Institute of Science and Technology, Manchester, England, Sept. 1973. Fouad, A. A., "Stability Theory-Criteria for Transient Stability", Proc. Engineering Foundation Conference on "System Engineering for Power", Henniker, New Hampshire, 1975, pp. 421-4506. Willems, H. F., Louie, S. A. and Bills, G. W., "Feasi- bility of Lyapunov Functions for the Stability Analysis of Electric Power Systems Having up to 66 Generators", IEEE Trans. PAS, Vol. 91, No. 3, May/June 1972, pp. 1145- 1153, Athay, T., Podmore, R. and Virmani, S., "A Practical Method for Direct Analysis of Transient Stability", IEEE Trans. PAS, Vol. 98, No. 2, March/April 1979, pp. 573-584. Ribbens-Pavella, M., Gruijc, Lj T., Sabatel, J. and Bouffioux, A., "Direct Methods for Stability Analysis of Large Scale Power Systems", Proc. IFAC Symposium on "Computer Applications in Large Scale Power Systems", New Delhi, India, Aug. 16-18, 1979, Pergamon Press, U.K. Kakimoto, N., Ohsawa, Y., and Hayashi, M., "Transient Stability Analysis of Electric Power System vis Lure* Type Lyapunov Function", Parts I and II, Trans. IEE of Japan, Vol. 98, No. 5/6, May/June, 1978. Bailey, F. N., "The Application of Lyapunov's Method to Interconnected Systems", Journal SIAM (Control) Ser. A., Vol. 3, No. 3, 1966, pp. 443-462. Michel, A. N. and Porter, D. W., "Stability of Composite Systems", Proc. Fourth Asilomar Conference on Circuits and Systems, Monterey, California, 1976. Michel, A. N. and Miller, R. K., "Qualitative Analysis of Large Scale Dynamic Systems" (Book), Academic Press, New York, 1977.12 29. 30. sl. 32, 33% 34. 35. 36. 37. 38. 396 40. POWER SYSTEM STABILITY Siljak, D. D., "Stability of Large Scale Systems", Proc. 5th IFAC World Congress, Paris, 1972. Siljak, D. D., "Large Scale Dynamic Systems - Stability and Structure" (Book) North Holland, New York, 1978. Pai, M. A. and Narayana, C. L., "Stability of Large Scale Power Systems", Proc. 6th IFAC World Congress, Boston, 1975.5 Jocic, Lj B., Ribbens-Pavella, M., and Siljak, D. D., "On Transient Stability of Multi-Machine Power Systems", IEEE Trans. Automatic Control, Vol. AC-23, 1978, pp. 325-332. Bergen, A. R. and Hill, D. J., "A Structure Preserving Model for Power Systems Stability Analysis", IEEE Trans. PAS, Vol. 106, Jan. 1981, pp. 25-35. Athay, T. M. and Sun, D. I., "An Improved Energy Function for Transient Stability Analysis", Proc. International Symposium on Circuits and Systems, April 27-29, 1981, Chicago. Sasaki, H., "An Approximate Incorporation of Field Flux Decay into Transient Stability Analysis of Multimachine Systems by the Second Method of Lyapunov", IEEE Trans. PAS, Vol. 98, No. 2, March/April 1979, pp. 473-483. Kakimoto, N., Ohsawa, Y. and Hayashi, M., "Transient Stability Analysis of Multi-machine Power Systems with Field Flux Decays via Lyapunov's Direct Method", IEEE Trans. PAS, Vol. 99, No. 5, Sept./Oct. 1986, pp. 1819- 1827. Teichgraeber, R. D., Harris, F. W., and Johnson, G. L., "New Stability Measure for Multimachine Power Systems", IBEE Trans. PAS-89, Vol. 2, Feb. 1970, pp. 233-239. Ribbens-Pavella, M., Calavaer, A. and Ghewry, A., "Transient Stability Index for On-line Evaluation", IEEE PES Winter Power Meeting, 1980. Saito, O., Kinnosuke, K., Muneyuky, U., Masahiro, S., Hisao, M. and Toshihiko, T., "Security Monitoring System Including Fast Transient Stability Studies", IEEE Trans., PAS-94, Sept./Oct. 1975, pp. 1789-1805. Fouad, A. A., Stanton, S. E., Mamandur, K.R.C., Kruempel, K. C., "Contingency Analysis Using the Transient Energy Margin Technique", Paper 81 SM 397-9, IEEB-PES Summer Power Meeting, Portland, Oregon, July 26-31, 1981.CHAPTER II REVIEW OF LYAPUNOV STABILITY THEORY 2.1 Introduction The theory of stability of linear time-invariant systems is well known in the literature through the methods of Nyquist, Routh-Hurwitz, etc. which provide both necessary and sufficient conditions. However, in the case of nonlinear systems, no such systematic procedures exist. Closed-form solutions of non- linear differential equations are exceptions rather than the rule. Although the digital computers can numerically inte- grate the differential equations, they do not provide an insight into the qualitative behavior of the differential equations. Consequently, the problem of finding the necessary and sufficient conditions for the stability of nonlinear systems is a formidable one and as yet an unsolved problem. All efforts in this direction relate directly or indirectly to the work of A. M. Lyapunov [1] who in 1892 set forth the general framework for the solution of such problems. Lyapunov dealt with both linear and nonlinear systems. He outlined two approaches to the problem of stability, known popularly as Lyapunov's "first method" and the other as the "second method of Lyapunov" or the "direct" method. The distinction is based on the fact that the "first method" depends on finding approximate solutions to the differential equations. In the “second method", no such knowledge is necessary. This is a great advantage in the case of nonlinear systems. We will be mainly concerned with the "second method of Lyapunov" or the "direct method" as applied to nonlinear systems. A second major breakthrough in nonlinear stability theory came through the work of V. M. Popov [2] who in 1962 outlined a 13II REVIEW OF STABILITY THEORY is criterion in the frequency domain for certain classes of non- linear control systems. This method is particularly attractive because of the ease with which it can be applied in much the same way as Nyquist's criterion is applied to linear systems. The mathematical model of power systems is amenable to analysis by Popov's method under certain simplifying assumptions. A literature survey of the area of nonlinear stability using Lyapunov's method is practically impossible because of the vast amount of research surrounding this topic. There are a number of texts devoted to this topic constituting a good source of reference material on various aspects of stability theory (Hahn [3], LaSalle and Lefschetz [4], Willems [5], Vidyasagar {6]). What follows is a self-contained treatment of the topic with an emphasis on applications. 2.2 Mathematical Description of Systems Most physical systems are described by their mathematical models for purposes of analysis. The following are some of the descriptions that are possible in continuous time systems. xX represents the n dimensional state vector, u represents the x dimensional input vector and t represents the independent time variable. 1 £ (x,u,t) Nonlinear, time varying and forced 2 £ (x,t) Nonlinear, time varying and force free 3 £ (x,u) Nonlinear, time invariant and forced 4 £ (x) Nonlinear, time invariant and force free (autonomous) The mathematical model of power systems for stability analysis results in an autonomous system of type (4) and fortunately for such type of systems a number of systematic procedures exist for constructing Lyapunov functions. In this chapter, we restrict the treatment to autonomous systems. Consider the autonomous system16 POWER SYSTEM STABILITY x = £(x) (2.1) i.e. = £5 ey X gre XQ) k= £4 (Kp Ryne ee Ky) Ry FL OX gr eee X) Let x, be a solution of (2.1), i-e. R= E(x.) (2.2) If x, = constant, then G = £( (2.3) Let at some instant of time, which can be designated as t = 0 without loss of generality, the system be disturbed instan- taneously to a position x, in the state space. The motion of 0 the system for t > 0 is denoted by x(t). Define y = x - Xo Hence (2.1) becomes in view of (2.2) - £(x,.) (2.4) with y =x It is evident that y = 0 satisfies (2.4) identically. Hence, instead of examining the stability of x, in (2.1) we may equivalently examine stability of the origin of the equation y = gly). This is often referred to as change of coordinates or transferring the equilibrium solution to the origin. Henceforth, we shall assume that this has been done and con- sider the stability of the origin (equilibrium point) of the equations k= f(x), £(0) =o (2.5) The origin is equivalently called the trivial solution, singular point or the equilibrium point. For the sake of simplicity, let us assume that the origin is the onlyII REVIEW OF STABILITY THEORY 17 equilibrium point of (2.5). Furthermore, we will assume that (2.5) satisfies the so-called Lipschitz condition so that the solution of (2.5) exists and is unique for a given initial condition x). 2.3 Definitions of Stability We are all familiar with the intuitive idea of stability discussed in Chapter I. But precise definitions are needed before we can formulate the problem in more quantitative terms. In the mathematical literature on this subject, there are scores of definitions each differing from the other in a subtle way. However, for engineering purposes, we need to consider only about the stability of a certain motion of the system (2.1). An undisturbed motion Xe of (2.1) satisfying (2.2) is con- sidered to be stable if by giving a small disturbance to it, the disturbed motion remains close to the unperturbed one for all time. More specifically: a) If for small disturbances, the effect on the motion is small, one says that the undisturbed motion is 'stable’ b) If for small disturbances, the effect is considerable, the undisturbed motion is termed ‘unstable’. c) If for small disturbances, the effect tends to dis- appear, the undisturbed motion is ‘asymptotically stable’. d) If regardless of the magnitude of the disturbances, the effect tends to disappear, the undisturbed motion is ‘asymptotically stable in the large’. The above qualitative discussion will be put in more quantita- tive terms. Consider the autonomous system £(x) (2.6) where the origin representing the undisturbed motion is the equilibrium point in the state space, i.e. = £(0)18 POWER SYSTEM STABILITY 2.3.1 Stability Definition 1: The origin is said to be stable in the sense of Lyapunov or simply stable if for every real number « > 0 and initial time t, > 0 there exists a real number 6 > 0 depending on ¢ and in general on t such that for all initial conditions satisfying the inequality, Ilxgll < 8 the motion satisfies x(t) || < © for all t > ty. The symbol ||-|| stands for the norm. In the Euclidean space, the norm can be interpreted to mean the Euclidean norm, i.e., 2 1/2 x) ius sll = i=l Note that stability in the sense of Lyapunov is a local concept. It does not tell us apriori how small 6 has to be chosen in the definition. The origin is termed unstable if for a given € as above and for sufficiently small 6 there exists one solution which does not satisfy the inequality, |]x(t) ||] < ©. The above definition is unsatisfactory from an engineering point of view since one is interested in a strong type of stability, namely that the solution must return to the origin as t 7 @, 2.3.2 Asymptotic Stability Definition 2: The origin is said to be asymptotically stable if it is stable and every motion starting sufficiently close to the origin converges to the origin as t tends to infinity, i.e. lim ||x(t) || > 0 tr ©II REVIEW OF STABILITY THEORY 19 (a) Stability (b) Asymptotic stability (c) Asymptotic stability in the large Fig. 2.1. Various types of stability.20 POWER SYSTEM STABILITY A deficiency of the above definition is that it does not indicate how big the disturbances can be in order that all subsequent motions will converge to the origin. In that sense, it is again a local concept. 2.3.3 Asymptotic Stability in the Large Definition 3: The origin is said to be asymptotically stable in the large if it is asymptotically stable and every motion starting at any point in the state space, returns to the origin as t tends to infinity. This is a very desirable type of stability in engineering systems since one does not have to worry about the magnitude of disturbances. The above three definitions can be illustrated geometrically as in Fig. 2.1 for a two-dimensional case. 2.3.4 Absolute Stability Consider the nonlinear control system with zero input shown in Fig. 2.2(a). The nonlinearity is known to pass through the origin, lie within the first and third quadrants and is restricted to a region between the horizontal axis and a line of given slope k. Such a type of nonlinearity is said to satisfy a sector condition. The linear transfer function G(s) is fixed. If for the class of nonlinearities satisfying the stated restrictions, the origin is asymptotically stable in the large, we say that the nonlinear system is absolutely stable. For multiple nonlinearities, we can extend the concept to a transfer function matrix W(s) (Fig. 2.2(b)) with a vector nonlinearity £(g) such £,(g) = £,(0,) and each £,(0,) satisfies the sector condition. 2.4 Region of Asymptotic Stability In many physical situations, the origin may not be asymp- totically stable for all possible initial conditions but only for initial conditions contained in a region around the origin. Such a region is called the region of asymptotic stability or finite region of attraction. If such a regionII REVIEW OF STABILITY THEORY 21 (a) Single nonlinearity (b) Multiple nonlinearity with f,(o) = f,(0;) Fig. 2.2, Single and multi-nonlinearity type feedback systems.22 exists, POWER SYSTEM STABILITY computation of such a region is of great interest to the system designer. For example, in a multi-machine power system, if a fault is cleared too late, the post-fault system may not return to a stable state, This can be interpreted equivalently as the computation of a region of attraction around the post-fault stable state and determining if the state of the system at the point of clearing the fault lies within this region or not. Examples i) ii) iii) iv) Stability: Consider the harmonic oscillator equation. hx. ay ay The origin is stable since given an ¢ within which we want the trajectory to lie, we can always give such an initial condition x, lying within a circle of radius 6, (8 0. The system has negative damping for small x and positive damping for large x. It has a stable limit cycle. The origin is unstable but the motion of the system is bounded. Stability: Consider the linear damped system. Ry = % Ky = ok 7 XD The origin is asymptotically stable in the large. Region of Asymptotic Stability: Consider Vanderpol's equation with - e(x2-1)k + x = 0 The origin is asymptotically stable but not in the large. There is a finite region of asymptotic stability. The limit cycle is, however, unstableII REVIEW OF STABILITY THEORY 23 since a small perturbation around the limit cycle will take the trajectory away from the limit cycle. 2.5 Basis of Lyapunov's Method The "direct" method of Lyapunov represents a philosophy of approach to the problem of stability and yet it is a very basic concept. The principle idea of the method is contained in the following reasoning. ‘If the rate of change cE of the energy E(x) of an isolated physical system is negative for every possible state x except for a single equilibrium state Ker then the energy will continually decrease until it finally assumes its minimum value E(x,)-' In our formulation x, = 0. The above concept was developed into a precise mathematical tool by Lyapunov. When the description of the system is given in a mathematical form, there is no way of defining energy. Hence, the energy function of E(x) is replaced by some scalar function V(x). If for a given system, one is able to find a function V(x) such that it is always positive except at x = 0 where it is zero and its derivative V(x) is < 0 except at x = 0 where it is zero, then we say the system returns to the origin if it is disturbed. The function v(x) is called the Lyapunov function. In what follows, we shall state somewhat more precisely Lyapunov's theorem on stability. Note that in the computation of V(x) , the system equations come into the picture as shown below: Consider the autonomous equation = 9 joy = Ov ML, av M2 oy xy dt x5 © ox, dt = aX x > = < Grad V, & > = < Grad V, £(x) > 2.6 Sign Definite Functions The sign definite properties of a scalar function V(x) will now be considered.a4 POWER SYSTEM STABILITY befinition 4: A function V(x) is called positive definite (negative definite) if V(Q) = 0 and if around the origin Vix) > @ (<0) for x ¥ 0. Definition 5: A function V(x) is called positive semi- def te {negative semi-definite) if V(G) = 0 and if around the origin W(x) > 0 (<0) for x # 0. Examples: i) Por n = 3 V(x) = 3 + % + x is positive definite di) Fern = 3 2 2. was . Vix) = XL + (%y#X4) is positive semi-definite since on x, + x, = 0 and x, = 0, V(x) = 0 iii) For n = 2 = y2 2 44, tae wo V(x) = x] + x3 - (x)+x,) is positive definite near the origin. In general, it is sign indefinite. More precisely, it is positive definite inside the square defined by |x,/ < 1, [x5 | el. 2.7 Stability Theorems In this section, we state the theorems on stability of the origin of the autonomous equation defined by (2.6). 2.7.1 Theorem on Stability The origin of (2.6) is said to be stable if there exists a scalar funetion V(x} > 0 in the neighborhood of the origin such that ¥(x) < 0 in that region. Such a function is known as the LyII REVIEW OF STABILITY THEORY 25 Example: Consider the equaLion corresponding to the harmonic oscillator. The ovigin is the only equilibrium point as can be seen by inspection. Choose V(x) = x] + =} which is positive definite. boo = av SL, av Sez "= ox, dt! ax, GE = Axx, - 2xyx, =0 Hence, the condition U(x) < 0 is satisfied and v(x) > 0 Theretore, the origin is stable. If we associate the physical system with the differential equations, namely, the harmonic oscillator, it is easily verified that V(x) in fact represents the total energy of the system, The fact that V(x) = 0 merely indicates that it is a conservative system and the total energy remains constant. Example: Consider the system of equations 2 Ry, = Ry > Buy ey (x,-2x,) oa _ 2 Ry = 2x, ia 3x3 Hy (81183) Ky oa 2x, ~ Ry 7 Ry choose oe) V(x) = ax) + xi + 3x _ 26. coy 12 _ ou2 _ 2 W(x) = waxy (x5 2x4) 2x5 (xy tx) 6x326 POWER SYSTEM STABILITY Vix) is negative semi-definite, being zero for 4) =X, = 0 and xy #0. Whe origin is therefore stable. 2.7.2 Theorem on Asymptotic Stability The origin of (2.6) is asymptotically stable if there exists a scalar function V(x) > 0 such that W(x) <0. If such a function exists, it is called a Lyapunov function for the system or simply the V function. Example Consider the system of equations _ _. 2,2 Ry = Xp 7 ax, (x5) ~ oe 2,2 = ax, (xj +15) where 'a' is a positive constanL. Choose 2,2 Vix) = xt + x) Then V(x) = -2aGxt+x5)? = -2a v? (x) Hence, V(x) Gonstitutes a Lyapunov function for the system and the origin is asymptotically stable. Example: say _ 2 eo x, = (xy Bx) (axy+bx5 1) = (ox, +4) (axftbxd-1) choose xe 2 2 V{x) = ax, + bey . _ De Be ge vO = 2 (ax) +B) (ax) tbx5 Eis If a, b, u, & are all positive, then V(x) is positive Gefinite and V(x) is negative definite in the domain ax” + bx” ¢ 1. The origin is asymptotically stable for all initial conditions in this domain. Mote that in this instance by examining V(x) we are able to get an estimate ofIl REVIEW OF STABILITY THEORY 27 the domain ef attraction. ‘The exact region of asymptotic stability is bigger than this since Lyapunov's theorem gives only sufficient but not necessary and sufficient conditions. 2.7.3 Theorem on Asymptotic Stability in the Large The origin of (2.6) is asymptotically stable in the large if there exists a scalar function V(x) such that i) V(x) » 0 ii) Vix) < 0 iii) Vix) + © as ||x]| 9 © ives, Vix) is radially unbounded, Condition (iii) ensures that Vix) = ¢ (C is a constant) represents closed surfac for all C in the entire state space. An example of a radially unbounded function is x2 + ax? Vix) = f + 3 whereas V(x) = page x) + 2 is not radially unbounded. Theorem 2.7.3 is difficult to apply in practical situations because it is difficult to find Ulex) » 6 with Vix) <0 in the entire state space. Hence, we have an alternative theorem, 2.7.4 Alternative Theorem on Asymptotic Stability in the Large In Theorem 2.7.3, Condition (ii) can be replaced by ii) (a) VGN «6 ii} (b) V(x) does not vanish identically for t » 0 along any other solution excepting x = 0. Stated in another manner, (ii) (5) implies that at vex) = 0 at points other than the origin, then these points should not constitute a solution of the system. example: Consider the series RLC circuit. The differential equation is28 POWER SYSTEM STABILITY ‘ di ; +L S+ dt = ik L at fidt 6 If R is nonlinear resistor, then R = fF{i) the Suppose we choose energy as the Lyapunov function V0) 5) # “ + 3 L xg Then Vor ety) +L xk, = Baus - $x, FAR) = =x} £ (x5) Assume positive values of resistance only so that E(x) > 0. Hence, V(x) > O and Vix) < 0. Alse V(x) + as |jx|| +, We now invoke the alternative fo: 4@.s.i.1. and verify that V(x) does not vanish along any solu- ulation of the theorem on tion other than the origin. V(x) = 0 for x, = 0 but x, @s not constitute a =O. Variable i.e. the xy axis. But it 4 solution of the system or a trajectory except x, = x 1 Therefore, we conclude that the origin is a.s.i.l. 2 In many physical systems, it is much casier to find V(x) > 0 and W(x) < 0 rather than a strict nedative definite V(x) - Hence, Theorem 2.7.4 will be found to be most readily amenable for engineering applications. The fact that Vix) # 0 along any solution other than origin should be argued out from physical considerations as in the preceding example.II REVIEW OF STABILITY THEORY 29 2.7.5 Theorem_on Region of Asymptotic Stability Consider the autonomous equation k= f(x), £(G) = 0 Let Vix) be a scalar function. let designate a region where Vix) > 0. Assume that {1 is bounded and that within i) Vix} > 0 ii) Vig) <0 when the origin is asymptotically stable and all motions starting in & converge to the origin as t + ©. Condition (ii) can be replaced by (ii)a and (1i)b as in Theorem 2.7.4. From the above theorem, a rule can be constructed for finding an estimate of the region of asymptotic stab: lity as follows: Choose a V(x) for all x and let V(x) be < 0 near the origin. Let V4, be the lowest value of V(x) on the surface ix) = 0. The region determined by vig) © Vin is contained in the region of asymptotic stability and there- fore provides an estimate of this region. Note that the region obtained by the inequality given above depends on the choice of v(x) and different choices of Vix) may yield different estimates of the region of asymptotic stability. Then the union of these regions is contained in the exact region of asymptotic stability. Thus, improved estimates can be obtained by taking different Lyapunov functions, Example: Consider the equations Choose V(x) = ax? ~ 2x x, + IxD30 POWER SYSTEM STABILITY It is easily verified by applying Sylvester's theorem that Vix) > 0. 2 3) e - 2_ 4444 .3y ay Wx) = -6R> xy +E ApH ts Near the origin V(x) < 0 but in general U(x) < 0. We proceed to find the region of asymptotic stability by examining the surface Vix) = 0. It is sketched in Fig. 2.3. Setting x, = 0, 2 it is seen that Vix) = 0 for x, = £1. Computations reveal the curves Vix) = 0 as open surfaces in Ry 7 Ry plane. We must find She on this surface. Computationally this can be Ain. quile difficult. > that the surface defined by W(x) < 2 Lies in e regi V< 0 and is shown in the figure. It is therefore an estimate of the region of asymptotic stability. Example: Censider Vanderpol's equation ap 2 . M+ e(x"-l)k +e = 0, € > 0 Define the state variables as xe ae e(e?-1jdu xy =e Then = xy # By = Xo og - xy) The above equetions are known to have a stable limit cycle. in) is an unstable point. I£ we ke The equilibrium point (ori put t = -t, then the origin will become a stable equilibrium point with an exact region of asymptotic stability yiven by the limit cycle. By using Theorem 2.7.5, we shall get an estimate of this region. Putting t = -tTI REVIEW OF STABILITY THEORY 31 Fig. 2.3. Computing region of asymptotic stability. Fig, 2.4, Geometrical interpretation of the stability eriterion.32 POWER SYSTEM STABILITY 2 Fs = -x, + ete - ®,) el "3 2 choose Lt Vig) = V(x) = xR) + Kako x? = MyXy + MZ (ox, tel HI) 2 x3 = 6x Cy - 1) ie Vix) £0 for x, < 3 + ax ¥oQ = 0 on the x, axis. On Lhe entire x, axis a 7 $e and hence the xy axis for xy # 0 does not constitute a solution of only solution is the origin. Since the system. Hence W(x) + 0 for |x,| < /3 the value of v_.. is evaluated by 2/ computing V(x} at ¥, = 0, x, = t ¥3. This is easily found to be 3/2. Henee an estimate of the region of asymptotic stahility is given by x + 2 Vin) = < 3/2 2,2 or x) + mo =e Ghia corresponds to a circle of radius ¥3 and the actual limit cycle, whatever the value of ¢, lies exterior to this. The conservative nature of the region of asymptetia stability is both due to the particular choice of the Lyapunov function ana its sufficiency nature.II REVIEW OF STABILITY THEORY 33 Example: Consider the equations By = ~2b xq - ax, - 3x7 We will examine the stability of the origin and try to con- struct a region of asymptotic stability. Take V(x) as x Th : 3 Vix) = bx> + a) tg Vix) > 0 in a region around the origin. In fact, it is indefinite in sign for values of x, < -b. V(x} is evaluated as Yon) = -axd V(x) < 0 around the origin. ‘The set of paints E for which V(x) = 0 is the entire x, axis. Examining 2 ey By = ay ax, X> it is “ outside the origin and zero at the point {-2b/3,0) ana (0,0). Hence, the equilibrium points are the origin and the point (-2b/3,0}. Lvaluating V(x) at the point (-2b/3,0), we 4p? at of asymptotic stability of the origin. get Lhe region V(x) < Gefined as an estimate of Lhe region From Lhe preceding three examples, we see that the region of attraction ¥(x) <~ C can be computed either by evaluating © on the curve V(x) = 0 and taking minimum value or evaluating ¥(x) at the nearest equilibrium point other than the origin. As it turns out, the latter method has been used extensively in power systems. 2.8 Proofs of Stability Theorems We have so far only stated the theorems with several illustra- tive examples. The proofs have been omitted since they involve34 POWER S¥STEM STABILITY a reasonable amount of mathematical rigor. However, a geometric interpretation of some of the theorems is found to be of great value. Consider ave fav au” 2, om, * Ak Consider the surface Vix) = C, © » 0 and some point on the surface. The quantities oa are the direction derivatives K along the different axis. The normal to the curve V(x) is given by the vector ¥#V whose components are ah . TE ay, k represents angle of the normal to x axis, then ave 8 , ax, 7 |¥vlcos ny, The positive direction of the normal is in the direction of inereasing V(x). Now Nn ay ax n dx dv ov Kk . -—= 5 = jvyv| © = cos nx, dt kel ax, at “2 ee at k n day The term 2 -g¢> cos na, represents the sum of the projections =1 1 Av of the velocity veetor components on the normal WW. If 52 < 0, it means trajectories move towards surfaces with smaller value of Cc This in turn implies that the trajectories intersect the families of surfaces inward since surfaces for small values of C are contained in surfaces with greater values of c. If dV/dt > 0, then the trajectories cross V(x) = C surfaces ° n dx. from inside. If % 7° cos nu, = 0, the trajectory lies on the surface V(x)= The preceding observations can be verified from Fig. 2.4 for a two-dimensional case. 2.9 Lyapunov Functions for Linear Ti e Invariant Systems In this section, we will consider the stability investigation ef linear time invariant (L.T.1.) systems by Lyapunov's method.J] REVIEW OF STABILITY THEORY 35) Consider the system ie = Ax (2.7) with origin as the only equilibrium point whose stability is to be investigated. The stability of (2.7) can be examined either by computing the eigenvalues of A to see if any of them lie in the right half plane or not, Alternatively, the Routh Hurwitz conditions can be applied to the characteristic equa— tion given by det[A - AI] = 0. lowever, both these methods do not reag@ily give an insight into the class of A matrices which are stable. This is possible through the construction of 4 Tyapunoy function of a quadratic form which also yields the necessary and sufficient conditions regarding stability of the A matrix. The following theorem regarding the solution of a matrix equation is of fundamental importance. Consider the matrix equation "p+ pas -g (2.8) where A is the given n%n matrix, P and Q are symmetric n*n matrices. If Q is given, the above matrix equation results in linear equations for the elements of P. Theorem: Agevseek, axe the eigenvalues of the matrix A, then the above equation has a unique solution for Pp if and only if +A, #0 for all i, 7 = 1,24....n This implies that for a unique solution to exist, A should have no sere eigenvalues and no real cigenvalues which are of opposite sign. he proof of this theorem is contained in Hahn [3]. 2 1 Lyapunov function for L.T.1. Systems We choose V(x) = x"P x where P is positive definite, as the Lyapunov function for (2.7). Thent4- 35(a) POWER SYSTEM STABILITY Vix) = ap x + a7 x = x"(A"P + PB ade = -x"Q x Since P is positive definite, by Theorem 2.7.3 the origin is asymptotically stable if Q > 0. Since origin is the only equilibrium point, the origin is also a.s.i.l. This leads us to a precise statement of the following theorem. 2.9.2 Stability Theorem for L Systems Suppose that Q is a positive definite symmetric matrix. Then for the system (2.7), A is a stable matrix, i.e. all eigen- values of A have negative real parts if the solution for the real symmetric matrix P of the matrix equation a"R +P A= -9 (2.9) is positive definite. Furthermore, V(x) = x"P x constitutes a Lyapunov function for the system (2.7), having a negative definite time derivative Von = -x79 x Equation (2.9) is commonly known as Lyapunov's matrix equation. We must clearly understand the implication of (2.9). Suppose we pick Q to be any p.d. matrix say Q =I, then if the equation has no solution or more than one solution, then the origin is not asymptotically stable. Suppose that (2.9) has a unique solution and P is not p.d., then again the origin is not asymptotically stable. On the other hand, if the unique solu- tion is such that P is p.d., then the origin is asymptotically stable in the large. The preceding theorem and discussion has not commented on what happens if A has eigenvalues on the imaginary axis. In terms of Lyapunov's stability theorem if there are distinct eigen- values on the imaginary axis, the origin is stable but not36 POWER SYSTEM STABILITY asymptotically stable. The question is partly answered by the following theorem due to B, D. O. Anderson [7]. vheorem: If the matrix A has no eigenvalues with positive real parts and has some eigenvalues with zero real parts which are distinct, then for a given 9 there exists a F > 0 m (2.9). Furthermore, > 0, satisfying the Lyapunov matrix equatiol x"p x and U(x) = -x7Q x < 0. the system is stable with V(x) = 2.10 Construction of Lyapunov Functions for Nonlinear Systems One of the main impediments to the application of Lyapunov's method to physical systems is the lack of formal procedures to sonsLruct the Lyapunov function for the differential equations describing the given physical system, We have seen in the previous section that for a linear time invariant system, systematic procedures exist via the Lyapunov matrix equation. However, the bulk of practical problems fall under the non- linear tegory Eor which systematic procedures are an exception rather than the rule. A number of investigators have worked on evolving some kind of general methods applicable to a certain class of problems, A good categorisation and historical summary of various methods of constructing Lyapunov functians is contained in a paper by Gurel and Lapidus 18]. We can broadly categorize the methods as 1) Methods based on first integrals 2) Methods based on quadratic forms 3) Methods based on solution to partial differential equation 4) Methods based on quadratic plus integral of non- dinearity type Lyapunov function 5) Miscellaneous methods 2.11 Methods Based on irst int This method of constructing Lyapunov functions is based on a linear combination of first integrals of the system equations. In turn, this is based on Lyapunov's original idea that total energy, in the case of a conservative system, could serve to define the stability of an equilibrium point.II REVIEW OF STABILITY THEORY a7 We shall first define what we mean by a first integral. Consider £; (xy eX gees Xy) (2.10) ie. = f(x), £(0) = 0 By a first integral (or simply integral) we understand a differentiable function G(x, ,%,,-..,¥,) defined in domain D of the state space such that when xis constitute a solution, G(x) eXyreeesX)) assumes a constant value C. Existence of a first integral can be used to define a conservative system. A necessary and sufficient condition for (2.10) to have a first integral is given by the condition Ms = 0 (2.11) diel B There are na general methods to construct the first integral except for second order systems. Example: Consider Ag Oy (2.42) = dx? - ax 1 1 It is easily verified that condition (2-11) is satisfied for this equation. Wow (2.12) can be manipulated as x By ad - aay Multiplying and integrating we get 2 x, x _ *2 1 asf Gla ky) = gt fy AtRy ey) dey as the first integral or energy integral of the system. since G(X) oy) = 0, and G(x ) is positive definite around the x ets origin, it constitutes a Lyapunov function. The origin is stable.38 POWER SYSTEM STABILITY The method of first integrals is limited to systems which are conservative and which in addition satisfy condition (2.11). 2.12 Method of Quadratic forms (Krasovskii's Method) [10] In this type of method, the Lyapunov function is of the form aD T ‘ x A(x)x or £°P(x)£. Consider the autonomous system be = £0), £(0) = 0 (2 Assume that £(x) has continuous first partial derivatives and define the Jacobian matrix a£ £ 1 3 got mr Be ag s : goo=age |: : (2.14) af of a a i Krasovskii's Criterion If there exists a constant positive definite matrix P such that the matrix Q(x) defined by Qlx) = Bole) + 2 Uxhe is negative definite, then the origin of (2,13) is asymptotically stable in the large. Consider the Lyapunov function va) = 272 £ which is positive definite in the f space, Since there is a one Lo one mapping between x-space and f-space, V(x) is alsa positive definite in the x-space. The derivative of V(x) is given by By chain rule EG) = JO)e ‘TOOL On)T1 REVIEW OF STABILITY THEORY 35. Hence 7 oe) = £797 ics Paget Bale Vix) ig negative definite since the term inside the brackets is negative definite, Hence, the origin is asymptotically stable in the large. Example: kh, = -ax, + x a>] (2.15) Choosing P = 1, we have J7(x) + J(x) = Qlx) where Za 2 The condition for a.s.i.l. of the origin is that Q(x) be negative definite i.e., -2a < 0 and da + 12ax} -4> 0. Since 1, both these inequalities are satisfied. Hence, the origin is a.s.i.l. 2.13 As the name implied, this method is based on the assumption of a vector ¥V with n undetermined components. An important feature of the gradient function is that both V and V may be determined from it. Thus dv ave. aves at ~ ca at Fosse 3x, By n40 POWER SYSTEM STABILITY x and Vi= Sq SVU, dx The upper limit x in the integral indicates that the line integral is to an arbitrary point in the x space and is independent of the path of integration. ‘The explieit form is i *o We for Wy ly. Oldy, +S gf) Wg (hp r¥grOe. May, Xy (2.16) # rae Jig’ OW yearns at, at ayy where Wg. is the ith component of the vector V¥- It is shown in standard texts on veetor calculus that for a scalar function V to be obtained uniguely from a line integral of a vector function TV, the following "%"*) equations must be satisfied avy. avy. 5 = (i, j=1,2...n) (2.17) These are also necessary and sufficient conditions so that the scalar function V be independent of the path of integration. In the three-dimensional case, Eqs. (2.17) are identical to those obtained by setting the curl of the vector VV equal to Zero. The procedure af constructing V(x) is to determine a gradient Gv such that i) The n dimensional curl of WV is sero. ii) V and ¥ is determined from YVv such that V > 0 and <0. If ¥ < 0 then we must ensure that it is not zero along any solution other than the origin, If such a 1V is determined, then the oriqin is a.s.i.l. The constructive procedure consists in assuming Vv of the formII REVIEW OF STABILITY THEORY 4l 5 (x) 979 (x) ee oy, 0 x WW = : : (2.18) Oy (x) 4 ee Oy GF) xy Sn2 ai4's are assume to consist of a constant term Hs pe Plus a variable term Sigye For V to be positive definite in the neighborhood of the origin, the diagonal terms a,, (x) are assumed to be of the form + = O., * = - + ie " . 94500 6,5 Oy) Soak iy Fy) (2.19) with aj, 2 0 and a,, (r,) being an even function of x,. In this manner after integration, the a,, coefficient will give rise to terms such as 2 x} xy ia Be NE Fg yg Oey OMy where u, is a dummy variable of integration. Further knowledge of the unknown coefficients in Vv is obtain- able from an examination of the generalized curl equations ao. « = bd é 7 (a) xy + eee 3x0 (ay + Keg Ge) J a ve a Go avy, Oa. i. ii g s Teo (s)R) + eres gy 105 ey + yay (He i i i aa, mea (x)a (2.20) From (2.17), the right-hand sides of (2.20) are equal. Since constant terms on either side must be equal, it follows (2.21)42 POWER SYSTEM STABILITY The other terms in VV are determined jointly from the curl equations (2.17) and dV¥/dt. Since aV/dt < 0, an attempt is made to make it negative semidefinite in as simple a way as possible. This may be accomplished if indefinite terms in dv/dt are eliminated. In the simplest case, we assume av _ 2 ae 7 RL (2.22) where k is initially assumed a constant. If dv/dt is con- strained as in the above equation, the remaining terms in dv/dt must be forced to cancel. ‘This is accomplished by grouping terms of similar state variables and choosing a.4°8 to force eancellation. The ays are assumed constants unless cancella- tion or the generalized curl equations require a more compli- cated form, Grouping of terms is governed by the restrictions on the a.,'s as stated above. For example, if in ie order system dV/dt contains the terms 944% 1¥%o" 2X2 and “x Xr the indefinite term -Xy XQ could not be grouped with 4% %2 since oy, can only be a function of X)- However, if -x,x3 were grouped with Oy 95 it could be eliminated by letting = x)x The choice of the bp to foree cancella- ig tit tion is not arbitrary because the generalized curl equations (2.17) must also be satisfied. If dV/dt cannot be made semi— definite in one state variable, then one may try to make it semi-definite in two or three variables. Another possibility is to make d¥/dt negative semi-definite in as large a region as possible around the origin in which case, we can try to find an estimate of the region of asymptotic stability by fitting the largest possible VW curve using Theorem 2.7.5. A formal five-step procedure for the application of the variable gradient method is given below: 1) Assume a gradient of the form (2.18). The a,,'s may be written as though they are constants until the need arises to allow them to be more complicated. 2) From the variable qradient, form aV/dt = . 3) In conjunction with and subject to the requirements of the generalized curl equations (2.17), constrain dvV/dt to be at least negative semi-definite.II REVIEW OF STABILITY THEORY 43 4) From the known gradient, determine V. 5) Use the necessary theorem to establish stability, asymptotic stability in the large, or region of a.s.i.l. In the above procedure, the most difficult step is step 3. Im fact, step 3 can be conceptually broken down inte smaller steps as 3(a) Constrain dv/dt to be negative semi-definite. ‘his will give some contraint on the coefficients. (b) Use the curl equations to determine the remaining unknown coefficients, {c] Recheck ¥ because the addition of terms required as a result of 3(b} may alter ¥. Example Consider the system of equations Re 3x, -— G(x) xy oy = 2x, + gta) xy The procedure outlined above will be carried out step by step. Step 1: The gradient function is assumed to be 1 “Sa e™s Step 2: av/dt is determined as av _ 2 . we = 7%] fey a Oey) - 89 Oy) + HR [-30) 4 = Oyo (%)) = 26g) a 559 08,)] 2 “Rg (thos P88 yg)a4 POWER SYSTEM STABILITY Step 3: The simplest manner in which aV/dt may be constrained to be negative semi-definite is if a = 0. Then 12" "21 av 20, 2 a ae TH yo Otay 2tao%_ + KRy(-39Q4) F AyyT(HL)) The indefinite terms in XyX5 May be completely eliminated if Thus, Lhe gradient is known to be cy 22 a7 ge) wv = MpoMe oe ave 2 2_. 2 and mt (x, ) ex] — Bey yxS step The gradient VV is integrated to yield xo x ~ jth S22 2 as Wied = Lg0 gm ayy) ¥y8t, t+ SQ” Bag To8¥g With a,, = 6 as a choice of scale factors to eliminate fractions vin = 26-4 gty)yjay, + 3x2 x 4g PSL 2 As long as *)/ 91%) lies in the first and third quadrant, V(x) is positive definite. If the integral goes to ™ as liasq II + om, then V(x) is not only positive definite, but also V(x) = c represents closed surfaces for all ¢ in the whole state space. av/fdt is negative definite in the entire state space. step 5 On the basis of Theorem 2.7.3 on a.s.i.l., the system is globally asymptotically stable.II REVIEW OF STABILITY THEORY 45 In these steps, particularly in step 3, ho apparent use was made of the genéralized curl equation. However, they were implicitly used, since By, Was a constant equal to zero and on the basis of the curl equations Mp, was also chosen as zero. g and VV, does not contain Also, since VV, does not contain x Byer 2.14 Zubov's Method [12,131 This method enables us not only to generate a Lyapunov function, but also construct a region of attraction or an approximation to it. The crux of the method lies in salving a linear partial differential eguation (p.d.e.) and when the solution to this p.d.e. is obtained in a closed form, we obtain a unique Lyapunov function and an exact stability region. When the solution is not possible in closed form, a series solution is possible and we get an approximation to the exact stability region. In particular, this method is amenable to computer solution, We first state Zubov's main theorem. Theorem: Iet U be a set containing the origin. Necessary and sufficient conditions for U to be exact domain of attraction is that two functions V(x) and (x) exist with Lhe following properties: a) V(x) is defined and continuous in U. 4 (x) is defined and continuous in the entire state space. b) 8(x) is positive definite for all x. ©) V(x) is positive definite in U with v(0) = 0. Also, the inequality 0 < Vix) < 1 holds in U. d) On the boundary of U, V{x) = 1. e) The following partial differential equation is satisfied £09 = -000 G-vGe)) +1E1l7) (2.23)46 POWER SYSTEM STABILITY Comme In the above theorem, if the domain of attraction is the entire state space, we then have qlobal asymptotic stability and the condition on V(x) is then Vix} > 1 as |]x|] + & Alternative Formulations of the p.d. (2.23): 1. Since the term 4 |e?) is positive, we may also define another positive definite function (x) as dix) = Oty 4|[E11) The p.d.e. then becomes av _ fe, 2108 = ean G-van) (2.24) 2. In another formulation of Zubov's method, the right hand side of the p.d.c. (2.23) appears as ~6 4x) (Lv) (Le |] £11?) 24? (2.25) 3. IE we define a change of variable V = - n(1-V), the p.d.e. (2.24) becomes = -o() (2. 26} and the Lyapunov function is 0. The region of attraction then becomes 0 < ¥ < ©. This form of p.d.e. is sometimes helpful jn obtaining V(x) in a series form which converges rapidly. Example: — 2,2 By OF ey + Ry + x, Gy tx5) 4 8 Pig kg = “ey + Ky + Hy Oe] Hx) Assume a function @(x) = 2 oeaxd) , Substituting in (2.24), ave i 2 av. 3,2 Ox, (x, Hey thep te po) + By (=Xy -%gtHG 1x] Xp) -2 (xP a8) (eve)II REVIEW OF STABILITY THEORY a7 It turns out that Vix) = x} + cd Satisfies the p.d.e. Hence, the required Lyapunov funetion is V(x) = x? + x and the stability boundary is given by 2 2 xy 1 Hy = 1 Example: Consider the system we 2 Ry = -#y + 2x] %5 ky =X Using the formulation in (2.25), the partial differential equation is aw 2 av ae (28D yo) + Gye bx) 1 2 = ae) {(1-v) 1 4 xd 4 (2Pegnxey) 71/25 x tog Let Ost) gg oe Ltx5 + (2xixy-x)) Clearly 9 Ory XQ) satisfies the condition of the theorem. The solution to the p.d.e, is given by -x? -x2 yn + BEBIMg™ ° Sa gl V(x,-%) vanishes at the origin and it is equal to 1 on the curve X)X) * 1. Hence, the region of stability is defined by the interior of the curve x,x, = 1. 2.14.1 Series Solution for Zubov's Method In general, we cannot expect a closed form solution to the partial differential equation in Zubov's method. If the non- linearity is expressible in an analytic manner, e.g, power4g POWER SYSTEM STABILITY series, then Zubow's method can be used to construct the Lyapunov function as well as approximate the region of attraction. Let x = f(x) be expanded as xe =Ax4+ g(x) (2.27) where g(x) contains terms of second degree and higher and A is the lincar part. Assume A to be stable i.e. A has all eigen- values with negative real parts. We choose #(x) toa be a positive definite quadratic form, The solution of the p.d.c. 2 oa Roget £y Gg) = oo Ce) CL-W (x) (2.28) i x = s s isl i is sought in the form Vix) = Vg 08) # VGN) Fees (2.29) where V,(x} is quadratic in x and V(x) (m=3,4,-+-) are homogeneous in degree m, i.e. Vi(y x) = y"v(x) for any constant y- To find Vf) (m=2,3,...) we substitute the system differential equations (2.27) and V(x) given by (2.29) in (2.28). Because of the assumptions on g{x) and Vite). Vi (x) is found as the Lyapunov functien for Lhe linear equation z= AX (2.30) v, (gl, (m=3,4,++.) are found from the recurrence relations n av lo gg- Ff, Go = RG) (2.31) i=l Ba which simplifies to Vin ay, bas 2 gol IA x] = Rye) med (2,32) R,(#) will be known if Vy (x)...¥,1 (4) have already been determined. Thus, V(x) is found as EV, (x) and the number of terms to be taken in V(x) depends upon the particular problem under consideration. The region of asymptotic stability obtained by taking more terms does not necessarily converge toII REVIEW OF STABILITY THEORY 49 the exact stability boundary uniformly. It is possible to substitute ¢(x) in (2.28) by means of a positive semi-definite quadratic function instead of a positive definite quadratic function. Because of the assumption on A we are still guaranteed to q¢et a positive definite Va (x). 2.14.2 Computation o: E Reg ign et The proceduré is outlined below with the computer implementa- tion in mind for computing the region of asymptotic stability. Compute V, (4) with x = Ax + g(x). Since A is stable, V, < 0. Let W, be the set of points on which Vj(x} = 0. Let a = 34 Min V(x) and 6 = Max V,(x) for xeW,. * eaenaing to Zubov, the curve defined by V,(x) = a lies entirely inside the region of attraction of the equilibrium point and the curve defined by Wy (x) = § lies entirely outside the region of attraction. These two assertions imply that the boundary of the reqion of asymptotic stability lies in the region defined by a © V(x) <6 We can extend the above reasoning to V, + Vy, V, + V3 + Vv +,etc. ‘To generalize, define arte vig) =v, b0 4 V;00 +o. 900 tet w'") (4) = set of all points on which 0!" tx) = 0. Specifically, we ensure that wi) zero HM) (x) comprises all points of (x) which define boundaries between regions of positive and negative 7!) (x). net a!) te the smallest value of V™ on wi?) ana a be the largest value ef v'") on (n) . Let a W (n} {n) afm We can then say that the curve yin) Ga) be the set of points x for which V {n) =u (x) s is wholly contained in the region of attraction and the curve yin Ge = a!" wholly lies outside the region of attraction. In implementing the procedure for a numerical problem, we (ny (mn) successively compute a One would be am and the region A tempted to infer that as n takes on large values,50 POWER SYSTEM STABILITY approximates better and better to the exact stability region. However, this is not necessarily true, i.e. (n41) aM an 2.14.3 Analytical Delails of es Solution Consider the system of differential equations (2.27) written in the summation form as n mm) my = ae ayy t py imps... om )x, x, ree # =m) + My ees Mm, (i=1,2,...-n) (2.33) where the aay and py (my) m)-- +2 mt) are real numbers. Also assume ¢ of the form dy + #, + ... where ¢; is a term homogencous of degree i, Then for Vy(x) we have aig To 2) (2.34) and the other recurrence relations corresponding to (2.32) becone n n e z = R(x) m=3,4,.... (2.35) , me isl jel where RO) is formed from #,, and the already known V's. If we compare like terms on both sides in (2.34) and (2.35), then a set of linear algebraic equations result which may he solved for cocfficients of Vi (m-2,3,...). The series is generally truncated up to a certain value of m. 2.14.3 Computation Example This example is adapted from Ref. [12]. Consider the modified form of the well known Vanderpol's equation H+ c(l-x7)k +x = 0 (2.36) This is obtained from the usual Vanderpol's equation by putting t = -t so that we have a region of attraction around the origin. Introduce the variables defined byIL REVIEW OF STABILITY THEORY 51 2.0 3 a Se 6 10 4 ACTUAL BOUNDARY "(ANALOG SOLUTION) -1.0 7 NOTED ’ VAN DER POL EQUATION a= 0.7 -2.0 LIENARD PLANE SOLUTION ~2.0 -1.0 O.0 x 1,0 2.0 3.0 ‘| Fig. 2.5. Stability boundary as given by approximate solutions to Zubov equation. (Reproduced from Ref. [12]. Hy = 3 (2.37) ake ee i and thus obtain 2 x. ig, Segue ve Shy 1 g TOE a (2.38) Bg ETRY In the state space description (2.38), we have the state traieetories which plot x against its (negative) time integral. This contrasts with the usual procedure of x versus its time derivative. The results can be transferred to the usual phase plane by solving for % from (2.37).52 POWGR SYSTEM STABILITY Assume V(x) ,/%,) of the form nj gis Vong) = FE ay x Fk x ® (2.39) j=2 k=0 7 The terms Vy (x) Xy) é Vz; (2%) 6% q) -e+. are computed recursively using the procedure outlined in the preceding section. The procedure will be illustrated for n=3. vi oye = (d +d xx gtd 2 207 : 21 22%) 3 3 + (dy 2 +d, x2x tay ox eet, 33 x5) (2.40) choose 9(%).%) = 2 + a. The p.d.e. (2.35) with v(x) = vy? now becomes a (egret gl Se lea ~ (xf48) v2) Gay (2.41) Equating terms of same deqree on both sides, we get av ay, . 2 _ 2,32 Bey (xg-ex)) 4 IX, (-x,) = (xpexs) (2.42) av avg ay (Ny eX)) + ao (-x,) = 0 aR] 2 L OMS 1 The appearance of 0 term on the right side of the second equation in (2.42) for terms of degree 3 is not surprising because there is no second degree term in the nonlinearity and, furthermore, (x) has been chosen as $y (x). In Ref. {12}, computations have been done by taking t)(x) = x? and region of attraction computed (Fig. 2.5) as discussed in Sec. 2.14.2. ce is taken as 0.7. For n=2, the stability region is found to be a circle of radius /3, Por n=6, the stability boundary is not improved, but contained in the curve for n=2. On the other hand, n=10 approximation encloses beth the n=2 and n=6TI REVIEW OF STABILITY THEORY 53 approximation and is in turn completely enclosed by the n=14 approximation. Thus, a surprisingly high order approximation must be used in order to improve upon the quadratic results. References [12] and [12] contain a good discussion regarding the numerical aspects of the solution. 2.15 Absolute Stability - Popov's Method so far, we have been considering general nonlinear systems without any restrictions on the nonlinearities. We also found that for such systems there is mo systematic method to construct Lyapunov functions in order to ascertain stability. In addition to the classical method based on first integrals of motion and Krasavoskii's method, two other popular methods were discussed, namely, variable gradient and Zubov's method. liowever, if the nonlinearity is such that it lies in the first and third quadrant or in a sector thereof, then a systematic procedure to construct the Lyapunoy function is possible. The problem therefore is to investigate the stability of such systems. In 1944, Luré [14] constructed a Lyapunov function for such systems consisting of the sum of a quadratic form in the state variables and an integral of the nonlinearity and derived therefrom a set of nonlinear equations called Lure's resolving equations. The existence of a set of solutions for such equations constituted sufficient conditions for global asymptotic stability of the nonlinear system. A significant breakthrough in this area of investigation came from the contribution of V. M Popov in 1962 [2]. Popov outlined a frequency domain criterion for the stability of a nonlinear syslLem containing a single nonlinearity lying in the first and third quadrant. Kalman [15] and Yakuboviteh [16] established the connection between Popev's frequency domain criterion and the Lyapunov function of the type of quadratic farm plus integral of the nonlinearity by proving that satisfaction of the former is a necessary and sufficient conditien for the existence af the latter, Subsequently, a number of research workers have extended the results to nonlinear systems with multiple nonlinearities as well as54 POWER SYSTEM STABILITY discrete time systems, ete. [17]. By now, the literature on the topic of absolute stability is vast and scattered both in the mathematical and the engineering literature. 2.15.1 Single Nonlinearity Case Consider the system in state space form E-Ax+ BE es pe) (2.43) - He Be where A is a nxn stable matrix, x, b, g are n-vectors. ¢(u)} is a nonlinearity which lies in the first and third quadrant. The system (2.43) can be cast in the block diagram as in Fig. 2.6 which consists of a linear Lransfer function with a nonlinearity in the feedback path. It can be verified from (2.43) that the transfer function of the linear part is o{s) _ a = Gis) = Pep > ge sr-w eb (2.44) If A is a stable matrix, i.e. all eigenvalues are in the left half plane, then G(s) has all poles with negative real parts. Special cases of interest are when G(s) has poles on the imaginary axis. We will consider only the simplest of these cases, namely, when G(s) has a single pole at the origin. In terms of the state space description, this means that the A matrix has a zero cigenvalue., The state space description then is of the form x Ao x b 7 al ee (2.45) g [& #8 6 L uo = <¢ (i) ge cle eae In addition to A being a Hurwitz matrix in both (2,43) and (2.45), we will further assume that (i) the pair (A,b) isIL REVIEW OF STABILITY THEORY 55 Fig. 2.6. Block diagram for the system (2.43), Fig. 2.7. Block diagram for the system (2.45).56 POWER SYSTEM STABILITY controllable, and (ii) (A,e") is observable. The nonlinearity satisfies the sector condition, i.e. 0 < blo) © ho” (2.47) for the system (2.43) and 0 « gla) < ke (2.48) for the system (2.45). 2.15.2 Popov's Theorem The system (2.43) or (2.45) is absolutely stable, i.e., the origin is a.s.i.l. for all nonlinearities satisfying the sector condition (2.47) or (2.48) respectively if there exists a finite real q such that i+ Re{ (1+joq)Gijo)} > 0 for all w > 0. (2.43) In the case of the system (2,45), an additional necessary condition, namely, d > 0 is to be satisfied. It is noted from the transfer function that d is the residue of G(s) at s = 0, This condition is physically equivalent to the con- dition that the system (2.45) is stable if ¢(o) = eafe > 0 and small). 2.15.3 Geometric Interpretation For the single nonlinearity case, a simple geometric inter- pretation can be given to the Popov criterion. Let (ju) = X(a) + 5 ¥l) Then (2.43) becomes + Rell+jwq} [X(w) + 3 ¥(m)] > 0 Rie + X(w) - qw Yu) > 0 (2.50) IP * * x (a) = Xd) and ¥'(w) = w ¥iw)IT REVIFW OF STABILITY THEORY 57 * Then we can define modified frequency response G (jm) as * * * G (wo) = X (w) #5 ¥ (a) Expression (2.50) becomes l * * E bX (uw) - gq ¥ (w) > O (2.51) Geometrically, we can interpret (2.5L} in the moditied frequency response plane as implying that there must cxist a line called the Popov line passing through the point [- z +0) having a slope a: so that ew lics to the right of this line (Fig. 2.8). Thus, absolute stability, i.e., a.s.i.l. of the origin tor nonlinearities satisfying the sector condiLion (2.47) or (2.48) can be ascertained in a simple manner analogous to the Nyquist's criterion for linear systems. An additional bencfit of Popov's criterion is that if we can establish absolute stability, then we also have a systematic procedure to construct a Lyapunov function. This Lyapunov function is of the quadratic plus integral of the nonlineariLy type. Systematic construction of the Lyapunov function is possible through the solution of a set of nonlinear algebraic equations (2.52) and (2.53} which are written for the infinite sector case, i.é. k =~, These equations form part of the Kalman-Yakubovitch Lemma [15,16] which is discussed in detail in Refs. [6,17,19]. For the System (2.43) T . AP + PA= Q-ue Bee 3 pate tag + (8 hp)t/? y (2.52) g* » U is n-vector, c > 0 and small. For the System (2.45 Ap +B aA=-2Q- uu Pb=h akg +a do + [ate bray! /? o (2.53)38 POWER SYSTEM STABILITY G* (ju) — Fa) Fig. 2.8. Polar plot of the modified frequency response and the Popov line. Fig. 2.9. Multivariable nonlinear system.II REVIEW OF STABILITY THEORY 59 where q = 38), © is » 0 and small and yp is (n-1) vector. Although the form of (2.52) and (2.53) appear similar, the former is not the special case of the latter. If Popev's criterion is satisfied, we know q. Then by an arbitrary choice of a, B and u, a p.d, solution for PF can be found from (2.52) or (2,53) and the Tyapunov function is then of the form For the system (2.43) Wx) = xR x +g If @lolde (2.54) For the system (2.45) T. VO = ate x 4 5 ae? 3 L + 4 fg olo)dao (2.55) In arriving at (2.54) and (2.55) the choice of a = 1 and a = 4 has boon made respectively so that q = B. 2.15.4 Kalman's onstruction Procedure The solution of nonlinear algebraic equations can be cireum- vented for the single nonlinearity case by what is known as the Kalman's construction procedure [15]. The various steps in the procedure are: 1. Find q from the Popov Criterion. 2. Write down the following polynomial of order 2(n-1) in w. Was) = Rel Ajqud Gore Gude (ju (2.56) where p(s) = det(sl-A). Factorize W(w) = Q(jw) O(-ja) (2.57) 4. Let qu = 2. Then @{2) will be a polynomial in z of we formal order n for the system (2.43) and (n-1) for system (2.45) with leading coefficient “rt where r= qelb for (2.43) and q(d+e™p) for (2.45). 5, Define v(z) = -O(z) + VE w(z)} and arrange this (2,58) polynomial of formal order 2n for the system (2.43) (2n-2) for the system (2.45) in ascending powers of z.60 POWER SYSTEM STABILITY 6. Define a real n-vector u for the system (2.43) or (n=1l)-vector for the system (2.45) with its components being the coefficients of the above viz). Solve for P using the Lyapunoy matrix equation ATR +P A> -u ye u (2.59) The Lyapunov function is then given by either (2.54) or (2.55). 2.16 Multivariable Popov Criterion There are several versions extending Popov's criterion to systems containing multiple nonlinearities. Perhaps the most popular among these is using the Moore-Anderson Theorem [18]. Consider the system in Fig. 2.9 where W(s) is an mem matrix transfer fanction such that Wie) = 0. The nonlinearity f(a) of dimension m is assumed to satisfy the properties i) fo) = £, (oy) ii) f,(odo, > 0 hel Qivcom (ii) implies that each nonlinearity lies in the first and third quadrant. Let A, B, € constitute the minimal realiza- tion of Wis), i.e. 1 Wis) = C(sI-A)""B The state space description of the system is given in the form x= Ax - B E(g) (2.60) g=c¢xX where A, B, C are nen, n*m and myn matrices respectively. Moore-Anderson Theorom: If there existsreal matrices N and Q such that Z(s) = (N 4 Q s)W(s) (2.61) is positive real, then the system (2.60) is asymptotically stable in the large providing (N + Qs) does not cause pole- zero cancellation with W(s).If REVIEW OF STABILITY THEORY 61 The conditions for 2(s) to be positive real are i) Z2(s) has elements which are analytic for Re(s) > 0. * * ii) 4 (s) = 2(s ) For Re s > Q. * iii) a (s ) + £(s) is positive semi-definite for Re (s) > 0. Conditions (i) and (ii) hold for most physical systems and condition (iii) implies 2 (-ja) + 2(ju) 2 0 for uw > o, (2,62) Analogous to the nonlinear equations (2.52) and (2.53) for the single nonlinearity case, we have a method of constructing the Lyapunov function for (2.60) if the Moore-Anderson theorem is satisfied. vTheorem 118] Given N and Q satistying the multivariable Popov criterion (2.61), then a Lyapunov function of the "quadratic plus integral of nonlinearity" exists of the form Vo) = xB ef o1a £7 (0) Q dg (2.63) and P js obtained as a positive definite matrix satisfying the Following set of nonlinear algebraic equations aTy 4 pas -E bT Pa=c+aTcTo-nw (2.84) ww = 9 ¢ B+ BtcTg where L and W are auxiliary matrices, Comments i) Unlike in the single nonlinearity case, an easy geometric interpretation of the multivariable Popov criterion is not possible, ii) Extensions of the above basic theorem to finite sector case are available in the literature [17].62 POWER SYSTEM STABILITY iii) The alqebraic equations are not amenable to a systematic solution procedure like Kalman's construc— tion procedure for the single nonlinearity cases unless we resort to techniques like spectral factoriza- tion, etc. Fortunately, as we shall see in Chapter 1V, for the power system case the mathematical model results in a solution of (2.64) for P easily. iv) Although not explicitly stated, the condition on A for the theorem to hold is that it be stable, The poles on the imaginary axis if they exist must be simple. In this chapter, we have discussed various methods of con-~ structing the Lyapunov function in 4 systematic manner. We have studied the basic stability definition in the sense of Lyapunov and theorems which give sufficient conditions for stability of various types. Popov's criterion, both for single nonlinearity as well as multinonlinear cases, have been discussed.12. 13. 14. Is, II REVIEW OF STABILITY ‘THEORY 63 References A. M, Lyapunov, "Problené Général de la Stabilité du Movement", Reprinted in Annals of Mathematical Studies Na. 17, Princeton University Press, Princeton, N.dJ., 1949 (Russian Edition 1892). V.M, Popov, "Absolute Stability of Nonlinear Systems of Automatic Control", Automation and Remote Control, vol. 22, 1962, pp. 857-875. W. Hahn, “Theory and Application of Tiapunov's Direct MeLhod", (Book) Prentice Hall, Englewood Cliffs, N.J., 1963. J. PD. LaSalle and S. Lefschetz, “Stability by iapunov's Direc Method with Applications", (look) Academic Press, New York, 1961. J. L. Willems, "Stability Theory of Dynamical Systems", (Book) Thomas Nelson 6 Sons, U.K., 1970. M. Vidyasagar, "Nonlinear Systems Analysis", (Book) Prentice Hall, Englewood Cliffs, N.J., 1978. B. D. O. Anderson and §. Vongpanitlerd, "Network Analysis and Synthesis - A Modern Systems Theory Approach", (Book) Prentice Hall, Unglewood Cliffs, N.J. O, Gurel and L. Lapidus, "A Cuide to the Generation of Lyapunov Functions”, Industrial and Engineering Chemistry, March 1969, pp. 30-41 N. G. Chetaev, "Stability of Motion", (Book) Pergamon Press, 1961 (Russian Edition, 1946-1950). N. N. Krasovskii, “Stability of Motion", (Book) Stanford uni Sity Press, Stanford, California, 1963 (Russian Fdition 1959). D. G. Schultz and J. BE. Gibson, "The Variable Gradient Method for Generating Liapunov Tunctions", AIEE Trans. Part Il, Appl. & Industry, Vol. 81, 1962, pp. 203-210. S. G. Margolis and W. G. Vogt, "Control Engineering Applications of V. I. Bubov's Construction Procedure for Lyapunov Functions", IEEE Trans. Automatic Control, Vol. AC-8, No. 2, April 1963, pp. 194-113. Discussion on Ref. 12 by FP. Fallside and M. R. Patel an Reply, IEEE Trans. Automatic Control, Vol. AC-10, No. 2, April 1965, pp. 220-222. A. I. Loré and V. N. Postnikov, "On the Theory of Stability of Control Systems", Prikl, Mat. i. Mehk, vol. & No. 3, 1944. R, E, Kalman, "Lyapunov Funetions for the Problem of Luré in Automatic Control", Prac. Nat. Acad. Sci., Vol. 49, February 1963, pp. 201-205.64 16, 18. 19. POWER SYSTEM STABILITY V.oA, Yakubovitch, "The Solution of Certain Matrix inequalities in Automatic Control", Dokl. Akad. Nauk. §.5.5.R., Vol. 143, 1962, pp. 1304-1307. K. 5. Narendra and J. H. Taylor, "Frequency Domain Criteria for Absolute Stability", (Book) Academic Press, New York, 1973. gd. B, Moore and B. D. 0. Anderson, "A Generalization of the Popov Criterion", Jour. Franklin Institute, Vol, 285, No. 6, June 1968, pp. 488-492. M.A, Aizerman and F. R, Gantmacher, "Absolute Stability of Regulator Systems", (Book) Holden-Day, Inc., San Francisco, U.S.A., 1964. (Revision Edition 1963)CHAPTER IIT MATHEMATICAL MODELS OF MULTI-MACHINE POWRR SYSTEMS. 3.1 Introduction As a preliminary step Lo the application of Lyapunov's method for power systems, we will develop the necessary mathematical model of a multi-machine power system in this chapter. ‘The bulk of literature in Lyapunov stability analysis of power systems rests on a mathematical model involving a number of simplifying assumptions. The mathematical model can be retained in the form of a set of second order nonlinear differ- ential equations or conveniently be cast in the Luré form with the nonlinearities in the first and third quadrant over a region around the origin. ‘This enables the systematic con- struction of the Lyapunov function and also the computation of the region of attraction. Both formulations are equivalent and are prevalent in the literature. Towards the end of the chapter, we present the mathematical model relaxing some of the simplifying assumptions and which give rise to multiplicative nonlinearities in the mathematical model. Since synchronism is essentially a physical phenomena associated with relative motion between synchronous machines, ome sort of reference angle other than the synchronous a ieference frame angle is necessary. ‘here are two possibil- ities here. We can take the angle of one of the machines as reference angle (generally the machine having the largest inertia) and measure angles of other machines with respect to this machine. The other possibility is to define a center of angle proportional to inertia weighted angles ef all the machines, a concept analogous to the center of mass in mechan- ical systems. All machine angles are then measured with 6566 POWER SYSTEM STABILITY respect to this center of angle. ‘he state space models based on both these notions are developed and compared. The correct dimension of the state space for the different models is derived from physical as well as centrel theoretic concepts of controllability, observability, minimal realization theory, etc. ‘he Luré form of equations are then derived which con- stitute the basis for systematic contruction of Lyapunov functions and also for applying the Popov stability criterion. Both the cases of zero and non-negligible transfer conductances are considered as well as zero and nan-zero damping, Finally, the mathematical model for the multi-machine case including the flux decay effect is considered. 3.2 Transient Stability Problem Restated By transient stability, we imply the stability of the system to maintain synchronism under sudden and major impacts. A fault on a high voltage transmission line, loss of a large generating unit, are examples of large disturbances. Let us first see what happens actually on the occurrence of a major disturbance. A major disturbance basically creates an imbalance between generation and load. Consequently, the power balance at each generating unit (i.c., mechanical input power minus the electrical output power) differs considerably from generator to generator. Some of them will have net accelerating power while others have decelerating power. hs a result, the retor angles of the machines accelerate or decelerate beyond the synchronous speed for t > 0. This is called “swinging” of the machines. JIf the rotor angles are plotted as a function of time, there exist two possibilities. (See Figs. 1.1 and 1.2) i) The roter angle increase together and swing in unison and ultimately settle to new angles. Since the rela~ tive reter angles do not increase, we say the system is stable and synchronism is preserved. ii) One or more of the machine angles accelerate faster than the rest of them, so that ultimately the rotor angles of the faster group continues to increase[11 MATHEMATICAL MODELS 67 indefinitely. ‘The relative rotor angles diverge. Such a mode of behavior is classified as unstable or losing synchronism. If the fault is cleared quickly, the tendency to separate may be arrested through the post-fault network by a redistribution of the energy gained by the system during the faulted period. The maximum time through which a fault can be allowed to remain without losing synchronism subsequently is called the critical clearing time t__. All of the above concepts which are mainly physical have to be translated into mathematical ideas before Lyapunov's theory can be applied. We do this first by developing the mathemat- ical model. 3.3 Mathem, al Model Under disturbed conditions, at each machine, we can write the swing equation as follows: d = _ i ac (We eal + Pai = Pai Pai (i51,2,....n) (3.1) where for the i‘? machine Ww. , = the kinetic energy of the rotor in Mega K.E,i Joules Pai = the damping power in MW Pai = the mechanical power input in MW Pai = the electrical power output in MW Equation (3.1) merely states that the accelerating power for each machine is balanced by the increase in kinetia energy of the rotor and the power absorbed by the damping forces. Let us denote the synchronous frequency by fo The kinetic energy is proportional to the square of instantaneous frequency fe of each machine. Hence,oe POWER SYSTEM STABILITY x " r sth where W2 4 ; is the kinetic energy of the i* machine at the -E, synchronous frequency £,. & (f.+At.) RAE. _ oO i x i WesEpd = Weed 2 Wepyi t=) fo ° since the deviation of ty from fy during a transient is small. Let 2; be the rotor angle with respect to a fixed reference and 85 the rotor angle with respect to & synchronously rotating reference frame both in electrical radians. Thus ay = os * we as, h i _ = we 2 = = 4 and ae W, = 2n(f,-£5) = 27 af, 2 uy w, is Lherefore the frequency deviation from the synchronous frequency. We shall make the assumption that the damping power Pal is propertional to the frequency deviation w,. Another form of damping is also present in the system which is due te the asynchronous torques between the machines and is proportional to the velocity difference between machines. Thus, actually we can write a Po, = Dew, + F Deu (a sm.) di ii j=l 13 (3.2) as, on b. “Page * Pa GEO at) Although subsequent trealmenL can be carried out with the qi’ for ease of mathematical analysis, n the D,; Lerms only and assume the asynchronous above representation of P we will ret torques to be zero. Substituting the expression for Wy Ei in 0, one for the faulted state The difference until t = t., and the other for t > t el ei" between these Lwo sets of differential equations is in the transmission configuration, this being reflected in the YayIIl MATHEMATICAL MODELS 75 parameters of the Yea matrix, i.e., it effects P_, only. We gi can symbolically therefore write Faulted State M. =P, - pt, ochet (3.14) 1 mi qi = ‘el . _ of _ of j = Ps Ped wakes = pet Becge = Pag ~ Pes EP bye Style F E i (3.15) determined from (3.14) i=l,2, n The problem of stability can be stated as follows. "Given el Tf it does, the system is then stable. Otherwise, the system t.,, does 6, (t) assume constant values and byt) > O as t > ©? is unstable." he problem of finding t,. is to find the maximum value of toy so that the system is stable. We at once notice the mathematical ramifications of the above stated problem. We have basically two sets of dse's (3.14) and (3.15) which have a discontinuity at t = tay and we are asking not for stability of the origin or the equilibrium point of (3.14) but rather boundedness of the trajectory defined by (3.14) and (3.15). This ie a rather difficult problem to solve except by simulation. To find toxe simulation has to be repeated for different values of t, So that we finally find at.) such that the system is stable for t,)-© and unstable for t. le where & > 0 and small. Such a value of tg, is called the critical clearing time t.y. Repeated simula- tion to Find L, for a single fault can be expensive computa— tionally since in a large system there are several faults to76 POWER SYSTEM STABILITY be considered. Typically, an experienced engineer does 3-4 repetitive simulations for a single fault before he can find B..18 cr Lyapunov's method addresses itself to the question of avoiding this repetitive simulation and having just one simulation to find tee directly. If toy were to be estimated thus accurately, it will mean a substantial savings in computer time for planning studies. Since initial conditions for the post fault system (3.15) fat L = t,)) correspond to the terminal values of 6, and a, from the solution of the faulted system (3.14) at t = t.,, it would seem logical to construct a region of attraction areund the post fault equilibrium point and for stability, examine if the terminal state of the faulted system (3.14) lies within the region of attraction of the post-fault system (3.15). theorem 2.7.5 provides a basis for estimating the region defined by the inequality V(x) < C for the post-fault system such that inside the region v(x) >0 and ¥(x) < 0 or Vix) < 0 providing the points where V(x) = 0 are not solutions of the post fault system. The faulted equations are then integrated until V(x) = C and the instant of time when this equality is satisfied is an estimate of top Since Lyapunov's method yields sufficient conditions, the region defined by W(x) < ¢ is contained in the exact region of stability and hence the computation will yicld in general values of tor less than the actual value. There are two issues to be resolved, he first one is how to construct a region of stability for a given fault (disturbance) and the manner in which it is cleared, i.e. with or without line switching. The second one is regarding the construction ef the Lyapunov function V{x) itself since different Lyapunov functions will yield different values of C, Computation of C, 80 that accurate estimates of toy may be obtained, has been a major bottleneck in the praclical implementation of Lyapunov's method. Initial research tended to compute a single value of C from (3.15) only for all fault locations and ignored the faulted equation (3.14) totally. Since the mode of instability is determinedIII MATHEMATICAL MODELS wD by the fault and the faulted dynamics (3.14), it is natural that a computation of C which ignored (3.14) will yield con- servative values of tor for certain faults and acceptable values for other faults. This puzzled research workers for a Jong time until very recently when the work reported in Refs. [23-25] of Chapter I removed this conservativeness. 3.4 State Space Models We now develop the appropriate stale space models for (3.13). the arameters will pertain cither to the faulted state or the post-fault state depending respectively on whether the equa- tions are to be integrated or a Lyapunov function has to be devloped. In the literature, there are several types of models proposed and often there is difficulty in reconciling results obtained from different models. One school of thought prefers to treat Eg. (3.13) as it is in which case the post-fault stable equilibrium point (SEP) is at a point other than the origin in the state space and the Lyapunov function is not zero at the origin. The other type of model transfers the post-fault SEP to the origin by a coordinate transformation and V(x) is zero at the origin. Moreover, the equations in this case can be put in the standard Luré form well known in modern stability theory and control theoretic approaches are possible. We follow the latter approach. In all power system problems, a reference angle is always required. We can have the state space medel based on machine angle as a reference (MAR) or center of angle as a reference (COA}, We discuss both types ef medels in this chapter and point out the connection between them. 3.4.1 State Space Model Using Machine Angle as Reference Define * v8 o x” = [ayigressie, 18) bgp 8] Then, Eq. (3.13) is equivalent to the following 2n equations.78 POWER SYSTEM STABILITY 41=1,2,...,n (3.16) This mode has 2n state variables with rotor angles 6, and roter speed deviations w, as state variables. This was the model used in earlier works in Refs. [13] and [15] of ChapterI in deriving the Lyapunov functions for multi-machine power systems using Moore-Anderson Theorem and Kalman's construction procedure respectively. While these Lyapunov functions are valid if we invoke the partial stability concept [4], they are not. valid in the state space defined by x if stability is considered in the sense of Lyapunov. This was pointed out by Sastry and Murthy [5] and Ribbens-Pavella (Ref. [19], Chap. 1). P,,'S are functions of angular differences only and, hence, we must take angular differences of machines with respect to a reference machine as the angle subvector while the velocity subvector remains unchanged. (Refs. [10,16], Chap. ©) Thus, the proper description of {3.13} is through the (2n-1) dimensional state vector “ ) 5 HT Ly Ware ee Wy by -8eS gS ee eB 8, t, where the n 5 machine has been taken as the reference machine. This results in the state space model Dd. i, eo 2 Pe = uy = i Oy +B By isl,2,.-..n (3.17) oy a ay 1=1,2,...,n-1 D. Now let us consider the case of uniform damping (q 4 for i all i) or zero damping case (D, = 0 for all i). ‘The state variables will now be x = [u,-w., = fay Wyre ee rly TWyt By o8 The state equations will be