Academia.eduAcademia.edu

Switching time bifurcations in a thyristor controlled reactor

1996, IEEE Transactions on Circuits and Systems I: Fundamental Theory and Applications

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/3322731 Switching time bifurcation in a thyristor controlled reactor Article in IEEE Transactions on Circuits and Systems I Fundamental Theory and Applications · April 1996 DOI: 10.1109/81.486445 · Source: IEEE Xplore CITATIONS READS 64 43 4 authors, including: Ian Dobson G. Venkataramanan 247 PUBLICATIONS 7,491 CITATIONS 180 PUBLICATIONS 4,158 CITATIONS Iowa State University SEE PROFILE University of Wisconsin–Madison SEE PROFILE All content following this page was uploaded by Ian Dobson on 25 January 2013. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document and are linked to publications on ResearchGate, letting you access and read them immediately. IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS-l: FUNDAMENTAL THEORY AND APPLICATIONS,VOL. 43, NO. 3, MARCH 1996 209 Switching Time Bifurcations in a Thyristor Controlled Reactor Sasan Jalali, Ian Dobson, Robert H. Lasseter, and Giri Venkataramanan Abstract-Thyristor controlled reactors are high power switching circuits used for static VAR control and the emerging technology of flexible ac transmission. The static VAR control circuit considered in the paper is a nonlinear periodically operated RLC circuit with a sinusoidal source and ideal thyristors with equidistant firing pulses. This paper describes new instabilities in the circuit in which thyristor turn off times jump or bifurcate as a system parameter varies slowly. The new instabilities are called switching time bifurcations and are fold bifurcations of zeros of thyristor current. The bifurcation instabilities are explained and verified by simulation and an experiment. Switching time bifurcations are special to switching systems and, surprisingly, are not conventional bifurcations. In particular, switching time bifurcations cannot be predicted by observing the eigenvalues of the system Jacobian. We justify these claims by deriving a simple formula for the Jacobian of the Poincare map of the circuit and presenting theoretical and numerical evidence that conventional bifurcations do not occur. Fig. I. Single phase static VAR system. Since the mid-1970’s, thyristor controlled reactors have been used at the loaded ends of transmission lines to control the reactive power supplied from a fixed shunt capacitor so that voltage can be maintained when system loads or transmission line configurations change [ 111, [20]. More recently, thyristor I. INTRODUCTION controlled reactors have been used as one of the economical E study the stability of a high power switching circuit alternatives for the emerging technology of flexible ac transwith a thyristor controlled reactor. This nonlinear cir- mission (FACTS) [2]. The expected benefits of flexible ac vv cuit can exhibit novel bifurcation instabilities and our main transmission include increased and controllable power flows objective is to explain and verify the new instabilities using on transmission lines and’the enhancement of power system simulation and experiment. A state space approach is used to stability. derive a simple formula for the Jacobian of the Poincare map Despitethe significance of high power circuits with thyristor of the circuit and confirm the nature of the new instabilities. controlled reactors, little nonlinear theory has been developed A thyristor controlled reactor is a fixed inductor in series for their analysis. The standard approach is to replace the with two oppositely poled thyristors as shown in the right thyristor controlled reactor by an average inductor model most branch of Fig. 1. C.ontrolling the firing (switch on) of and then apply linear techniques to the resulting circuit [l 11, the thyristors controls the thyristor conduction time and hence [20]. While this average inductor approximation is sometimes the proportion of time for which the inductor is included effective for predicting steady state behavior, it fails to capture in the circuit. One approximation which is often used to much of the circuit nonlinearity and it breaks down when large understand the thyristor controlled reactor in a periodic steady harmonic distortions occur. Operating conditions with large state is that it acts as a continuously variable inductor. The harmonic distortions are documented in [l], [12], [27], [13]. value of the inductance depends on the thyristor conduction While detailed time domain simulation is valuable in analyzing time. The thyristor controlled reactor is often combined with nonlinear effects in thyristor controlled reactor circuits, there is a fixed capacitor in parallel so that varying the thyristor also a need to develop mathematical concepts and approaches conduction time varies the effective impedance of the parallel so that the simulated or actual nonlinear phenomena may be combination. understood and predicted. This paper describes new’ instabilities of a thyristor conManuscriptreceived June 5, 1993. This work was supported in part by trolled reactor circuit called switching time bifurcations in the Electric Power Research Institute under Contracts RP4000-29, RP8010which thyristor switch off times jump or bifurcate as a 30, RP8050-03, RP8050-04, WO8050-03 and by the NSF Presidential Young Investigator Grant ECS-9 I57 192. This paper was recommended by Associate parameter is slowly varied. The thyristor controlled reactor Editor M. Ilic. circuit and its classical operation are described in Sections S. Jalali is with Siemens Energy and Automation, Atlanta, GA 30202 USA. II and III. The large harmonic distortions asfociated with I. Dobson and R. H. Lasseter are with the Department of Electrical and Computer Engineering, University of Wisconsin, Madison, WI 53706 USA. switching time bifurcations are briefly summarized in Section G. Venkataramanan is with the Department of Electrical Engineering, IV. Sections V and VI explain that switching time bifurcations Montana State University. Bozeman, MT 59707 USA. are fold bifurcations of the zeros of the thyristor current. SecPublisher Item Identifier S 1057.7 122(96)01665-O. 1057-7 I22/96$05.00 0 1996 IEEE 210 IEEETRANSACTlONSONClRCUlTS ANDSYSTEMS--I:FUNDAMENTALTHEORYANDAPPLICATIONS,VOL.43,NO.3,MARCH tion VII presents simulations of a thyristor controlled reactor circuit for static VAR control which shows switching time bifurcations in detail: The circuit was developed by Bohmann and Lasseter to study harmonic interactions and distortions [l]. This simulation evidence for switching time bifurcations is followed in Section VIII by experimental work .showing switching time bifurcations on a single phase equivalent of the static VAR compensator installed near Rimouski, Quebec cw. The occurrence of switching time bifurcations raises expectations that they should.be related to well known generically occurring’ bifurcations. It is interesting that this is not the case and that switching time bifurcations appear to be a novel mechanism for instability. In particular, the occurrence of the switching time bifurcations cannot be predicted from eigenvalues of the system Jacobian. The remainder of the paper is devoted to explaining how the special properties of the thyristor controlled reactor circuit precludes conventional bifurcations. Sections. IX, X, and XI compute the. Poincare map [IO], [25] of the circuit and the Jacobian of the Poincare map. The formula for this Jacobian is simplified considerably by the special properties of the thyristor turn off and is the same formula that would be obtained for fixed turn off times. Section XII computes the eigenvalues of the Jacobian to show that conventional bifurcations do not occur and explains how the Jacobian simplification ensures that switching time bifurcations are not predicted by the eigenvalues of the Jacobian. This paper differs from the initial conference paper [14] by including experimental results and reworked theory. We briefly review other approaches to bifurcations in switching circuits. Switching circuits with high switching rates and ideal switches are well approximated using averaging methods [19], [17], [23] and the stability of the averaged system can be investigated using bifurcation theory [21], [24]. The high power switching circuits addressed in this paper have switching rates comparable to the 60 Hz frequency of the voltage sources and averaging methods have not been shown to be applicable. There are also simple low power switching circuit models that exhibit bifurcations and chaos when the nonlinear junction capacitance of a diode is modeled [4], [3]. In deriving the Jacobian of the Poincare map, we use a state space analysis of switching circuits which overlaps with contributions of other authors. The fundamental work of Louis [18] computes PoincarC maps for switching circuits including controls. The varying dimensions of the state vector and switching conditions are discussed and formulas for the propagation of first order deviations through switchings are stated. The formulas show that Louis had used the Jacobian simplification which is highlighted in Section X. Louis computes as an example the Jacobian of the Poincare map of an acfdc convertor with a current regulator. Verghese et al. [26] give a general approach to computing Poincare maps and their Jacobians for switching circuits. Circuit controls and symmetries and the automation of the computations are discussed but the Jacobian simplification and the varying dimension of the state vector are not treated. The linearized dynamics of a series resonant convertor are computed and studied. Grotzbach and Lutz [9] computed Jacobians of Poincare maps of switched Fig. 2. 1996 Classical operation of thyristor controlled reactor. circuits including control actions. They develop Newton algorithms for computing steady state solutions and compute eigenvalues for ac/dc convertors with controls. The extension to nonlinear circuits and the derivation of averaged circuit models are discussed. Dobson [6] derives the Jacobian of a general switching circuit with RLC elements and ideal diodes or thyristors with an emphasis on the Jacobian simplification. The work in this paper has stimulated further work on the thyristor controlled reactor circuit such as its novel transient dynamics [22] and analyzing damping and resonance phenomena [5]. The effects of thyristor firing synchronization schemes on the Jacobian are computed and presented in [ 161, [ 151. II. SYSTEM DESCRIPTION Fig. 1 shows a single phase static VAR compensator consisting of a thyristor controlled reactor and a parallel capacitor. The controlled reactor is modeled as a series combination of an inductor L, and resistance R,. The static VAR compensator is connected to an infinite bus behind a power system impedance of an inductance L, and a resistance R, in series. The switching element of the thyristor controlled reactor consists of two oppositely poled thyristors which conduct on alternate half cycles of the supply frequency. A thyristor conducts current only in the forward direction, can block voltage in both directions, turns on when a firing pulse is provided and turns off when the thyristor current, becomes zero. (The phenomenon of thyristor misfire [22] is not addressed here.) The thyristors are assumed ideal so that detailed nonlinearities in the turn on/off of the thyristors are neglected. The dependence of the thyristor switch off times on the system state causes circuit nonlinearity. The firing pulses are supplied periodically and the system is controlled by varying the phase 4 (delay) of the firing pulses. The system is studied with the phase as an open loop control parameter. In practice a closed loop control would modify the firing phase. III. CLASSICAL ANALYSIS The classical, idealized operation of a thyristor controlled reactor is explained in Fig. 2. In this figure, the gray line represents the thyristor controlled reactor voltage Vc(t) (cf. Fig. 1) and the solid line represents the thyristor current. If the thyristors are fired at the point where the voltage Vc(t) is at a peak, full conduction results. The circuit then operates as if the thyristors were shorted out, resulting in a current which lags the voltage by 90’. If the firing is delayed from the peak voltage, the current becomes discontinuous with a reduced fundamental component of reactive current and a reduced thyristor conduction time 0. As the phase angle 4 ranges between 90’ and 180” the thyristor conduction time 0 JALALI et al.: SWITCHING TIME BIFURCATIONS (b) (b) I Fig. 3. time Appearance of a new thyristor current zero. (a) 4 < 4*. (b) 4 =. d* time ranges between 180’ and 0”. The classical analysis assumes that the voltage Vc(t) is a pure sinusoid. IV. HARMONIC DISTORTION The classical analysis is often applicable, but can, as demonstrated here and in [l] and [ 121, fail for certain circuit parameters and operating conditions. Under these conditions, both the voltage and the current waveforms become greatly distorted with large harmonic components. This distortion is associated with a resonance phenomenon in which the natural frequencies of the circuit, from when the reactor is fully on to when it is fully off, span an odd harmonic [l], [5]. This harmonic distortion can lead to instabilities as switching times suddenly change or bifurcate as follows. It is useful to recall that fold bifurcations (also called saddle node bifurcations) can either create or annihilate pairs of zeros of functions or vector fields [lo], [25]. Fig. 4. Disappearance of a thyristor current zero. (a) 4 <’ v. (b) I$ = 4’. Cc)4 > d-. zero disappears. As the phase delay of the firing pulses slowly increases, the fold in the dotted line rises until, passing through the critical phase 4*, the current zero disappears and a later zero of the thyristor current applies (see Fig. 4(b) and (c)). The switching off time of the thyristor has suddenly increased in a switching time bifurcation and stability has suddenly been lost. Note how the zero of the actual thyristor current coalesces with a zero of the virtual current indicated by the dotted line and disappears in a fold bifurcation. As soon as the switching time bifurcates and system stability is lost, a transient starts. VII. SIMULATION RESULTS The switching time bifurcations are illustrated with the single phase static VAR circuit of Fig. 1 described in [l]. On a 1 MVA, 1 kV base, the source impedance has a per-unit V. INSTABILITY WHEN A NEW magnitude of 7.35% with an angle of 89”, the reactor is a THYRISTOR CURRENT ZERO APPEARS series combination of 62.6% inductive reactance and a 3.13% Fig. 3 describes one way in which the system can lose resistance. The capacitor bank has a capacitive reactance stability. Suppose that harmonic distortion produces a fold or of 177%. Alternatively,, the per unit component values can dip in the thyristor current as shown in Fig. 3(a). As the phase be specified as L,$ =0.195 mH, R, =0.9 mR, L, =1.66 delay 4 of the firing pulses slowly increases, the fold lowers mH, R, =31.3 mR, and C =1.5 mF. The firing pulses are until, passing through the critical phase 4*, a new, earlier zero equidistant and the phase 4 is the relative phase of the firing of the thyristor current is produced by a fold bifurcation of pulses with respect to the stiff ac source u(t) = sin wt shown the thyristor current as shown in Fig. 3(b). (The new zero in Fig. 1. 0 denotes the conduction time of the thyristors. of the thyristor current in Fig. 3(b) is in fact a double zero.) In this example, the system impedance and the capacitor The thyristor switching off time has suddenly decreased and bank have a natural frequency which is 4.9 times the fundathe stable operation of the system at the previous periodic mental frequency 60 Hz. If the controlled reactor is included orbit has been lost. We call this qualitative change a switching in the circuit, the natural frequency shifts to 5.18 times the time bifurcation. As soon as the switching time bifurcates, a fundamental frequency. Thus the natural frequencies of the transient starts. circuit from when the reactor is fully on to when it is fully off span the fifth harmonic. This crossing of an odd harmonic VI. INSTABILITY WHEN A THYRISTOR indicates that system odd harmonics can be large and that CURRENT ZERO DISAPPEARS voltage and current wave forms can be significantly distorted Fig. 4 explains another type of switching time bifurcation Ul, [51. Fig. 5 shows how the conduction time 0 of periodic orbits in which the system loses stability as a thyristor current zero varies as the phase 4 is varied. To simplify the calculations, disappears. Fig. 4(a) shows a periodic solution for the thyristor only periodic orbits which are half wave symmetric are comcurrent with the solid line. The gray line shows the thyristor current that would occur if the thyristor did not switch off puted, using (16) and (17) from Section XI. As can be seen, for negative current. This part of the current is referred to two separate sets of periodic solutions are computed. One set starts at 0 = 180” and ends at 0 z 91”. The other starts at here as “virtual”. The virtual current does not occur in circuit operation but it is useful in understanding how the current 0 = 0” and ends at 0 M 60”. The classical model predicts 212 IEEE TRANSACTIONSON CIRCUITS AND SYSTEMS-I: FUNDAMENTAL THEORY AND APPLICATIONS,VOL.43.NO.3,MARCH h ti I..““.“..‘.‘.‘.“1 . 1996 Periodic orbit 1, 0 = 102 Deg. t....“.““‘..‘.‘..‘....j 3 . l . 1” . .. dl 5 r 45 l ~..*.~~*b****.~*~~...................~~ 0.. . 20 Fig. 5. 60 100 . . . . 140 Q (Degrees) Conduction time d versus firing delay ~5. only one set of solutions starting from (T = 0” and ending at u = 180”. In order to investigate how the system loses stability when stable periodic orbits disappear, the ElectroMagnetic Transient Program (EMTP) was used [7], [8]. Periodic orbits 1 through 6 in Fig. 5 were chosen to study in detail how harmonic distortion can cause periodic solutions to disappear in switching time bifurcations. The loss of a stable periodic solution at u near 91” is a switching time bifurcation in which a new thyristor current zero appears. The EMTP simulation in Fig. 6(a)-(c) shows periodic solutions of the thyristor current at periodic orbits 1, 2, and 3 respectively. Note that as we move toward periodic orbit 3, the harmonic distortion produces a fold in the thyristor current. Fig. 6(c) shows that as the phase delay of the firing pulses is slightly increased, the fold in the thyristor current lowers and a new, earlier zero of the thyristor current is produced and a transient starts. The loss of a stable periodic solution at (T near 60” is a switching time bifurcation in which a thyristor current zero disappears. Fig. 7 shows how the periodic solutions behave as the switching time bifurcation is approached. The plots in Fig. ‘7 are on an expanded time scale so as to closely observe the behavior of the thyristors as they turn off. The dotted lines show the virtual current which would have occurred if the thyristors did not turn off as the current decreased through zero. As the phase is decreased, the periodic orbit progresses through periodic orbits 4, 5, and 6 and the zeros of the virtual and the actual current approach each other, coalesce, and disappear. The stable periodic solution disappears when the actual current zero disappears and the switch off time of the thyristor suddenly increases. The initial portion of the transient which occurs when the stable periodic solution disappears is shown in Fig. 8. The system will eventually converge slowly to another stable periodic solution. Even though many systems enforce equidistant firing in steady state operation, the firing pulses may or may not I.. .,.. I :. 0.66 T;‘ ? 9 0.5 ‘; 2i k 0.3 :, 0.67 Periodic . . I.. 0.68 orbit !a?1 0.69 2, 0 = 96 Deg, a Pi 0.1 E -0.1 -0.3 -0.5 I . . ..I....I....I... 0.96 0.97 ‘I’...1 (b) 0.98 0.99 Q ? a : 0.6 5 :: B 02. Periodic orbit 1 1 33 g -0.2 I -0.6 t t 1 . . . . . . ..I......... 1.275 *.*....... 1.3 I . . . . . . !"?i 1.325 time (seconds) Fig. 6. A new thyristor current zero appears. (a) periodic orbit I. (b) Periodic orbit 2. (c) Periodic orbit 3 up to 1.3 s. Q is increased by 2’ at 1.3 s. be equidistant during transients. Therefore, the detail of the transient depends heavily on the assumptions used in modeling the control of the thyristor firing pulses. The intent of the simulation results is to show the.existence of the transient as a consequence of the switching time bifurcation rather than the detail of the transient. However, the computations of steady state periodic orbits as in Fig. 5 are valid for any firing control scheme that enforces equidistant firings in steady state. 213 JALALI e, nl.: SWITCHING TIME BIFURCATIONS 230 kV I l~MVA 230124 kV h ‘Is ‘yr^A x =ll% t 431 PF Fig. 9. Rimouski static VAR compensator. The actual and the-virtual current zeros . . . . , . . . . , . . . , , . . I . . . 0.133 0.134 0.135 0.136 -1.0 TABLE I RIMOUSKI AND EXPERIMENTAL CIRCUIT COMPONENTS Data 1%I. ( KL N ) Power Fig. 7. Thyristor current on an expanded time scale for periodic orbits 4, 5, and 6. Zbvse ecrease ? ? ,a4 5 ar” H u H 2 Degrees - A& (1.07 p.u.1 St,,(O.288 p.u.1 St (0.11 p.u.) z&(0.016 p.u.) Rimouski(34) 230 (132.8) kV 100 MVA 529 (1 566 Cl 152.3 (2 58.2 0 8.46 II Experiment( 14) 115.0 v 301.4 VA 43.88 i2 46.95 R 12.64 61 4.83 61 0.70 cl 2 0 b 1 zo- 1.. . I . 0.3 . . . 0.33 . . ..I.. . t \L 1 . I 0.36 time (seconds) Fig. 8. Two current zeros coalesce and disappear. Periodic orbit 6 up to 0.3 s. C$is decreased by 2’ at 0.3 s. VIII. 100 EXPERIMENTAL RESULTS This section gives experimental results showing switching time bifurcations in the single phase equivalent of the static VAR compensator installed near Rimouski, Quebec [20, ch. 61. This compensator is a three phase delta connected thyristor controlled reactor and ungrounded wye-connected capacitor banks interfaced to a 230 kV system through a step down transformer as shown in Fig. 9. This experiment used a 115 V, 60 Hz ac line which was assumed to be a stiff and harmonic free source. The circuit components were scaled as shown in Table I. The reactance to resistance ratio was measured as ~20 for the inductors and ~70 for the capacitor. The control circuit used a zero voltage detector synchronized directly across the ac line and a firing pulse generator to build a train of equally spaced firing pulses. 120 140 160 # (Degrees) Fig. 10. 6 versus 0. The firing pulses are delayed by a time delay generator and transmitted to the back-to-back thyristor modules through a pair of fiber optics. The thyristors used were Westinghouse model 707408 rated 1000 V and 200 A. Fig. 10 shows how the conduction time 0 varies as the phase d, is varied. The solid line is computed using (16) and (17) from Section XI and the triangles show the experimental measurements. The solution at point A- is lost when the thyristor firing is slightly increased and the system state converges to a new steady state at point A+. This instability occurs when the thyristor conduction time is. suddenly decreased due to the appearance of a new earlier current zero. The steady-state 214 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS-I: FUNDAMENTAL THEORY AND APPLICATIONS,VOL. 43, NO. 3, MARCH 1996 solution at point B- is lost when the thyristor firing phase is slightly decreased upon which the system converges to a new steady state at point B+. This instability occurs when the thyristor turn off time is suddenly increased due to the disappearance of the thyristor current zero. In this example, the switching time bifurcations are associated with the two natural frequencies of the circuit (the thyristor fully on and fully off) spanning the third harmonic. For example, when the short circuit MVA of the system is infinite’(X, = 0), the two natural frequencies do not span the 3rd harmonic and the bifurcation instabilities are absent. IX. POINCAR~ MAP The PoincarC map of the static VAR system can be computed by integrating the system equations and taking into account a change in coordinates when the switchings occur. The thyristor turn on time is controlled by the phase parameter 4 which specifies the delay of the firing pulse. The thyristor turn off occurs when the thyristor current decreases through zero. The thyristors are modeled as short circuits when on and open *circuits when off. The system state vector z(t) specifies the thyristor current, capacitor voltage and the source current: IT(t) K(t) ( 1s(t) ) x:(t) = The system input u(t) is the source voltage function shown in Fig. 1. u(t) is assumed to be periodic with period T. During the conducting time of each of the thyristors, the system dynamics are described by the following set of linear differential equations i=Ax+Bu (1) where A= -R,L,l -C-l 0 ( L,l 0 -L;l m$;;l) andB= (L;l). During the off time of each thyristor, the circuit state is constrained to -lie in the plane 1, = 0 of zero thyristor current. In this mode, the system state vector y(t) specifies the capacitor voltage and the source current: y(t)=(K(t) Is(t) > and the system dynamics are given by the linear system G = PAPty + PBu ON (A,B) 0 1 0 where P is the projection matrix P = 0 0 1) . ( Fig. 11 describes the system dynamics as the system state evolves over a period T. A thyristor starts conducting at time ~$0. This mode as described by (1) ends when the thyristor current goes through zero at time 40 + ~1. The nonconducting mode as described by (2) follows the conducting mode and continues until the next firing pulse is applied at time ~$1.This (i;) I 40 + Cl 40 41 1 (PfifpB) 41 +a 40 +T i”~ ------+-c------Tze J Fig. 11. System dynamics from time &O to do + T starts a similar on-off cycle which lasts until the next period starts at time 40 f T. The state at the switch on time do is denoted either by the vector y/(40) or by the vector ~($0). These representations of the state at the switch on time are related by (3) X(40) = ptY(do). Equation (3) expresses the fact that the z representation of the’ state at a switch on is computed from the y representation by adding a new first component which has value zero. The state at the switch off time ~$0+ g1 is similarly denoted either by ~(40 + ~1) or y(& + al) and these are related by YY(40+ al) = Px(40 (4) + 01). The matrix P in (4) may be thought of as projecting the vector x onto the plane of zero thyristor current. Given a time interval [sl, sz], it is convenient to write f(., ~1,‘s~) for the map which advances the state at s1 to ihe state at ~2. For example, a Poincark map which advances the state by one period T starting at the switch on time 40 may be written f(y(&), 40, $0 + T). (If the time s2 at the end of the interval is a switching time, then there is ambiguity about whether f(.,sl, ~2) evaluates to y(s2) or x(s2). We adopt the convention that f(x(s1),sl,s2) evaluates to x(s2) and f(y(sl), sl, ~2) evaluates to y(sz). For example, the PoincarC map f(~Y(do), 40~40 + T) evaluates to y(& + T).) If the thyristor is on during all of the time interval [sl, sq], we write f(z(si), sl, SP) as fon(x(si), ~1, sp) and if the thyristor is off during [a, 921, we w&e f(y(sl), ~1, ~2) as .fo~(y(sl), SI, SP). fon(x(sl),~lr 32) or fod~(sl),sl,s2) can be computed by integrating the corresponding linear system (1) or (2) over the time interval [sl, ~‘21. Now we construct in stages an expression for the Poincark map f(~(do), 40,40 + T) in term of fan and .f,,ff and the coordinate changes (3) and (4). At the start of the period at time 40, the state is expressed in the y coordinates as y/(40) or, using in the 2 coordinates. The (3), is expressed as x(40) = P”y(&) state x(&,+~l) is obtained by integrating the on linear system (1) with initial state Pty(&,) from time 40 to 40 + 01: x(40 (2) OFF 1 (PAP’,PB) + Cl) =fcn,(ptY(~o)> =eAol 40740 (P’y@,,) + 01) + lo’ CA’Bu(7 + (Oo)dr). (5) The switch off time s,ff = 40 4 o1 is dtitermined by an equation including P which constrains the thyristor current to be zero: 0 = (I - PtP)x(s&) = (6) 215 JALALI er al.: SWITCHING TIME BIFURCATIONS The coordinate change (4) at the switch off and integration of the off linear system (2) yields Y(h) = foff(P440 + m)140 + Olr41). (7) A half cycle map is given by combining (5) and (7): foff(%n(ptY(40)r do, 40 + m),40 . [x+2) + Df(x(tz), The Poincare map may now be written by composing two successive half cycle maps and then neglecting the gory details of the time arguments: STABILITY (8) (9) OF PERIODIC SOLUTIONS When the static VAR circuit is in steady state with a periodic trajectory of period T, the Poincare map has a corresponding fixed point. That is, f(~(40), 40, do + T! = ~(401. t2r t3) eAborr-tz) x(t2) + ePAP’(ts-sorr)pA(~ + tl)y(tl) + function( son, tl, t2 ). Differentiating with respect to y(tl) and keeping in mind that the thyristor turn on time son is a fixed quantity (recall that the thyristor firing is equidistant), we obtain Df(y(tl), tl, t2) = eA(t2--S,“)ptePAP’(s,,-tI). (10) .Let [t2, t3] be a fixed time interval including a thyristor turn off at time s0a and no other switchings. The map f(x(t2), t2, t3) advances the state x(t2) to the state y(tj) and we want to compute the Jacobian Df(x(tZ), t2, t3) which is a 2 x 3 matrix. f(z(tz),tz,t3) = foff(Pf,,,(x(ta),tP,Soff 1 rsoff,t3) eAct2 -) Bu(~)h s tz _ ptp) 1 DS,E (12) Note that the two terms associated with sOff in the limits of the two integrals of (11) cancel. The row vector DS,E is the gradient of s,tf with respect to x(t2). The second term of (12) may be written as ePAPt(t3-sorf)PA(I - PtP)x(s,R)Ds,tf which vanishes according to the constraint equation (6) so that we obtain the surprising and simple result t2, t3) = ePAPt(ts-s,rf)peA(~urr-ttz). (13) Result (13) implies that the switch off time may be regarded as constant when deriving the Jacobian. The map advancing the state over the combined interval [tl, t3] can be written as the composition f(y(t1),t1,t3) = f(f(y(t1),t1,tz),tz,t3) and the chain rule yields DfMh), tl, t3) = Df(x(tz), t2, kx)Df(dh), tl; t2) so that substitution from (10) and (13) gives Df(y(h),tl, t3) = In particular, let tl = son = 40, sol = 40 + 01, t3 = 41 and Tl = 41 - ~$0to obtain the Jacobian of the half cycle map: Df(y(qbo), = fon(ptf,ff(Y(tl),tl,s,,),Son,t2) = eA(t2-s,,)ptePAPt(s,,,- (11) = Soff Df(x(tz), The stability of the periodic orbit can be computed from the Jacobian of the Poincare map evaluated at the fixed point [lo], [25], [26]. In particular, the periodic orbit is exponentially stable if the eigenvalues of the Jacobian lie inside the unit circle. Since the thyristor turn off time s,,~ and the Poincare map are discontinuous at a switching time bifurcation [22], [6], we assume when computing the Jacobian in this section that the system is not exactly at a switching time bifurcation. To compute the Jacobian of the Poincare map, we first compute the Jacobian of maps which advance the state from the beginning to the end of a time interval containing one switching. Let [tl, t2] be a fixed time interval including a thyristor turn on at time s,, and no other switchings. The map f(y(tl), tl, t2) advances the state y(tl) to the state x(t2) and we want to compute the Jacobian Df(y(ti), ti, t2) which is a 3 X 2 matrix. f(Y(tl),tl;t2) I t3 ePAP”(t3--r)pBu(T)&. s sow ePAPt(t3--s,rr)peA(s,rf-tz) [ X. eA(t2-‘)Bu(r)d7 Differentiating with respect to x(h) and keeping in mind that the thyristor turn off time .sOeis a function of x(ta), we obtain + ~lr41). do, h), 41,40 + T) f(~l(40), do,40 + T) 7 f(.f(~($o), = f~ffPf”~P”f~ffPf~~~“(Y~~O)). + lIoff $o, 41) = ePAPL(T1-gl)PeAulPt. (14) Applying the chain rule to (8) and using (14) for the Jacobians of the half cycle maps yields the 2 x 2 Jacobian of the Poincare map: Df(y($o), 40,40 + T) =e PAPt(T*-al)peAuzptePAP1(T~-ul)peAulpt. (15) One of the interesting and useful consequences of (15) is that the stability of the periodic orbit only depends on the state and the input via g1 and ~72.It is remarkable that (15) is also the formula that would be obtained for fixed switching times 01 and 02; the varying switching times apparently introduce no additional complexity in the formula, but the nonlinearity of the circuit is clear since ~1 and 02 vary as a function of ~($0). 216 IEEE TRANSACTIONSON CIRCUITS AND SYSTEMS-L FUNDAMENTAL THEORY AND APPLICATIONS.VOL.43, NO. 3. MARCH 1996 XI. SIMPLIFICATION FOR SYMMETRIC PERIODIC ORBITS It is convenient to take advantage of symmetry when the periodic orbits are half wave symmetric. Half wave symmetry of a periodic orbit means that the thyristor firing pulses are sent every half cycle and the orbit is half wave symmetric. In particular, the two conduction times g1 = c2 = cr are equal. Moreover, the system states at the half cycle are equal in magnitude and opposite in sign to the system states at the beginning of the cycle. That is, if y(&) is the system state at the beginning of the period, then: f’(Y(do), $0, do + T/2) = -Y/(40). (16) The conduction time g of the conducting thyristor is given by the constraint equation 0 = (I - PP)x(& + LT). (17) The fixed points ~($0) corresponding to half wave symmetric periodic orbits can be computed by solving (16) and (17) simultaneously. In addition, the Poincark map Jacobian in (15) simplifies to: ~f(y(40),qb0,~0 + T) = (epAp’(T/2-n)reA”p’)z.(18) XII. CONVENTIONALBIFURCATIONS We review the instabilities expected from conventional bifurcation theory. The thyristor controlled reactor circuit becomes a discrete time nonlinear system when analyzed with the Poincark map. Conventional bifurcation theory (e.g., [lo], 1251) describes several typical ways in which a discrete time nonlinear system can become unstable as parameters vary. The system is assumed to be initially operating in a stable periodic fashion before the bifurcation occurs and the nature of the bifurcation is determined by where a critical eigenvalue of the Jacobian of the Poincare map crosses the unit circle. (The stable periodic orbit disappears in a fold bifurcation if an eigenvalue crosses the unit circle at 1, becomes modulated with another frequency or becomes unstable in a Niemark or secondary Hopf bifurcation if a complex conjugate pair of eigenvalues crosses the unit circle and period doubles if an eigenvalue crosses the unit circle at - 1.) Several authors (e.g., [4], [3], 1211)have investigated instabilities such as Hopf and period doubling bifurcations in averaged models of fast switching power electronic circuits. The Jacobian formula (18) can be used to rule out conventional bifurcations of half wave symmetric periodic orbits in the thyristor controlled reactor. circuit. The Jacobian in (18) is a complicated function of the initial state ~(40) and the input u(t), .but it only depends on ~($0) and u(t) via the switching time U. Thus to determine stability, it is only necessary to test the stability of the Jacobian in (IS) as fl varies over its range of 0 to 180”‘. Numerical computation of the eigenvalues of the Jacobian in (18) as 0 varies over this range as shown in Fig. 12 indicates that the absolute values of the eigenvalues are less than 0.98 for both the simulation and the experimental examples. This demonstrates that, for the given component values, the circuits do not lose stability in a conventional bifurcation. We conclude that the eigenvalues of Fig, 12. Eigenvalues of Jacobian as c varies. the Jacobian of the PoincarC map are strictly inside the unit ’ circle as a switching time bifurcation is approached; that is, the eigenvalues give no warning of the switching time bifurcation. The points in the eigenvalue locus of Fig. 12 at which an eigenvalue approaches the unit circle can be predicted and associated with resonance effects; see [5]. The switching time bifurcation can be detected as a zero gradient of the thyristor current at the switch off time; what may be surprising is that this zero gradient has no effect on the Jacobian. As a thyristor current zero disappears in a fold bifurcation, the gradient of the thyristor current evaluated at the current zero tends to zero and one might expect this to influence the Jacobian. In particular, the gradient of the thyristor current tending to zero implies that the sensitivity DSOff of the switch off time with respect to the initial state becomes infinite. If the Jacobian contained a term including Ds,tf (cJ: (12)), at least one eigenvalue of Df would leave the unit circle and a conventional bifurcation would occur before the switching time bifurcation was encountered. However, this analysis is wrong because the Jacobian simplification shows that the term in the Jacobian involving Dso~ vanishes (cf the vanishing of the second term of (12)). Thus the Jacobian simplification ensures that the fold bifurcation’ of the thyristor current is not expressed in the eigenvalues of the Jacobian and that the switching time bifurcation will occur. See [6] for a more rigorous explanation. In the case of a switching time bifurcation in which a new thyristor current zero appears, the gradient at the thyristor current zero before the bifurcation does not become zero as the bifurcation occurs and the Jacobian is unaffected by the bifurcation. XIII. CONCLUSION This paper studies instabilities in a thyristor controlled reactor circuit used for reactive power control of power systems in which switching times change suddenly, or bifurcate as a system parameter (phase of the thyristor firing) varies slowly. The switching time bifurcations are explained and their mechanisms are illustrated by simulation and experiment. In JALALI er rd.: SWITCHING TIME BI~RCATIONS 211 [12] S. G. Jalali and R. H. Lasseter, “Harmonic interaction of power systems particular, we have shown how distortion of current waveforms with static switching circuits,” Power Electronics Specialists Conj:, June can cause a thyristor switch off time to disappear or a 1991, M.I.T., MA, pp. 330-337. new thyristor switch off time to suddenly appear by fold “Harmonic instabilities in advanced series compensators,” in [ 131 -, EPRI FACTS Conf., Boston, MA, May 1992, in Electric Power Research bifurcations of the thyristor current. The consequence of the Institute Report .TR-101784. sudden change in switch off time is that stable periodic [ 141 S. G. Jalali, I. Dobson, and R. H. Lasseter, “Instabilities due to operation of the circuit is lost and a transient occurs. These bifurcation of switching times in a thyristor controlled reactor,” in Power Electronics Specialists Cont. Toledo, Spain, July 1992, pp. 546-552. switching time bifurcations are not explained by the analysis 151 S. G. Jalali, “Harmonics and Instabilities in thyristor based switching of thyristor controlled reactor circuits with average inductor circuits,” Ph.D. dissertation, Univ. Wisconsin-Madison, 1993. 161 S. G. Jalali, R. H. Lasseter, and I. Dobson, “Dynamic response of a models. Nor are the switching time bifurcations explained by thyristor controlled switched capacitor,” IEEE Trans. Power Delivery, the conventional theory of bifurcations in which stability is lost vol. 9, pp. 1609-1615, July 1994. when eigenvalues of the Jacobian matrix cross the unit circle. [17] P. T. Krein, J. Bentsman, R. M. Bass, and B. C. Lesteutre, “Un the use of averaging for the analysis of power electronic systems,” IEEE Trans. The thyristor controlled reactor circuit is a nonlinear, periPower Electron., vol. 5, pp. 182-190, Apr. 1990. odically operated circuit. The PoincarC map can be computed [18] J. P. Louis, “Non-linear and linearized models for control systems inby integrating the system equations and taking into account a cluding static convertors,” in Proc. Third Int. Federation Automat. Contr. Symp. Contr. Power Electron. Elect. Drives, Lausanne, Switzerland, change in coordinates when switchings occur. We have derived 1983, pp. 9-16. a simple formula for the Jacobian of the PoincarC map (also see [ 191 Sept. R. D. Middlebrook and S. Cuk, “A general unified approach to modeling [18], [6]). This Jacobian describes the stability of the steady switching Dower converter stage,” in IEEE Power Electron. Specialisfs Conj, 1976, pp. 18-34. state periodic operation of the circuit under an open loop [20] T. J. E. Miller, Ed. Reactive Power Control in Electric Systems. New equidistant firing pulse control. In particular, the Jacobian forYork: Wiley, 1982. mula is used to demonstrate numerically that none of the con- I211 _ _ S. Pavliasevic, D. Maksimovic, “Using a discrete-time model for largesignal analysis of a current-programmed boost convertor,” in Power ventional bifurcations occur in our circuit examples. For simuElectronics Specialists Con$, June 1991, M.I.T., MA, pp. 715-721. lation evidence of switching time bifurcations in a single phase [22] R. Rajaraman, I. Dobson, and S. Jalali, “Nonlinear dynamics and switching time bifurcations of a thyristor controlled reactor,” IEEE Intl. thyristor controlled series capacitor circuit see [13] and [15]. Symp. Circuits Syst., Chicago, IL, May 1993, pp. 2180-2183. This paper makes progress in describing and understanding [23] S. R. Sanders, J. M. Noworolski, X. 2. Liu, and G. C. Verghese, instabilities in a nonlinear thyristor controlled reactor circuit “Generalized averaging method for power conversion circuits,” IEEE Trans. Power Electron., vol. 6, pp. 251-259, Apr. 1991. with a simple topology. However, despite the simple topology, D. Tan and R. S. Ramshaw, “Instabilities of a boost converter system we expect that this circuit may well contain other novel [24] F. under large parameter variations, ” IEEE Trans. Power Electron., vol. 4, behaviors. One of our objectives is to develop new ways of pp. 442-459, Oct. 1989. [25] J. M. T. Thompson and H. B. Stewart, Nonlinear Dynamics und Chaos: understanding and computing instabilities of general switching Geometrical Methods for Scientists and Engineers. London! Wiley, circuits by first studying in detail the thyristor controlled 1986, reprinted 1987. reactor circuit. The genericity of fold bifurcations suggests [26] G. C. Verghese, M. E. Elbuluk, and J. G. Kassakian, “A general approach to sampled-data modeling for power electronic circuits,” IEEE Trans. that switching time bifurcations describe typical ways in which Power Electron., vol. 1, pp. 76-89, Apr. 1986. general switching circuits with thyristors become unstable. [27] R. Yacamini and J.W. Resende, “Thyristor controlled reactors as harREFERENCES [1] L. J. Bohmann and R. H. Lasseter, “Harmonic interactions in thyristor controlled reactor circuits,” /EEE Trans. Power Delivery, vol. 4, pp. 1919-1926, July 1989. [2] Current activity in flexible AC transmission systems, IEEE Publication 92 TH 0465-5 PWR, Apr. 1992. [3] J. H. B. Deane and D. C. Hamill, “Instability, subharmonics and chaos in power electronics systems,” IEEE Trans. Power Electron., vol. 5, pp. 260-268, July 1990. [4] J. H. B. Deane, “Chaos in a current-mode controlled boost dc-dc convertor,” IEEE Trans. Circuits Syst. - I, vol. 39, pp. 680-683, Aug. 1992. [5] I. Dobson, S. G. Jalali, and R. Rajaraman, “Damping and resonance in a high power switching circutt, ” in Systemsand Control Theory for Power Systems,J. H. Chow, P. V. Kokotovic, and R. J. Thomas, Eds., IMA vol. 64 in Mathematics and Its Applications. New York: Springer Verlag, 1995, pp. 137-156. [6] I. Dobson, “Stability of ideal thyristor and diode switching circuits,” IEEE Truns. Circuits Syst. - I, vol. 42, pp. 517-529, Sept. 1995. [7] H. W. Dommel, “Digital computer solutions of electromagnetic transients in single and multiphase networks,” IEEE Trans. Power Apparat. Syst., vol. PAS-88 pp. 388-399, Apr. 1969. [8] EPRI Report ELP454lPCCMP, vol. 1 and 2; The Electromagnetic Transients Program (EMTP), Version 1.0, RevisedRule Book, Apr. 1986. [9] M. Griitzbach and R. von Lutz, “Unified modeling of rectifier-controlled DC-power supplies,” IEEE Trans. Power Electron., vol. 1, pp. 90-100, Apr. 1986. [IO] J. Guckenheimer and P. Holmes, Nonlinear Oscillations, Dynamical Systemsand Bifurcations of Vector Fields. New York: Springer-Verlag, 1986. [I 1] L. Gyugyi, “Power electronics in electric utilities: Static var compensators,” Proc. ‘IEEE, vol. 76, pp. 483494, Apr. 1988. monic sources in HVDC converter stations and AC systems,” IEE Proc., Part B, vol. 133, pp. 263-269, July 1986. Sasan JaIali received the B.S., M.S., and Ph.D. degrees in electrical engineering in 1988, 1990, and 1993, respectively, from the University of Wisconsin-Madison. He joined Siemens Energy and Automation, Atlanta, GA, in 1993. His interests include nonlinear effects in utility power electronics and the analysis and development of flexible ac transmission systems. Ian Dobson received the B.A. degree in mathematics from Cambridge, England in 1978 and the Ph.D. degree in electrical engineering from Cornell University, NY in 1989. He worked for five years as a systems analyst for the British firm EASAMS Ltd.; this included simulation of switching circuits for plasma physics power supplies on contract to Culham Laboratory, UKAEA, England. He is now an associate professor of electrical engineering with the University of Wiscnsin-Madison. His current interests are applications of bifurcation theory and nonlinear dynamics, voltage collapse in electric power systems and utility power electronics. 7.18 IEEE TRANSACTIONSON CIRCUITS AND SYSTEMS-I: FUNDAMENTAL THEORY AND APPLICATIONS,VOL. 43, NO. 3. MARCH 1996 Robert H. Lasseter received the Ph.D. degree in physics from the University of Pennsylvania, Philadelphia, in 1971. He was a Consultant Engineer with the General Electric Company until 1980 when he joined the University of Wisconsin-Madison, His main interest is the application of power electronics to utility systems including hardware, methods of analysis and simulation. Dr. Lasseter became a Fellow of the IEEE Power Engineering Society in 1992. Giri Venkataramanan received the B.E. degree from the University of Madras, India, in 1986, the M.S. degree from the California Institute of Technology, Pasadena in 1987 and the Ph.D. degree from the University of Wisconsin-Madison in 1992, all in electrical engineering. Currently he is an Assistant Professor of electrical engineeering at Montana State University, Bozeman, His interests .include modeling, design and control of power conversion systems and the introduction of pragmatism into engineering education.