IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008 1271 A Real-Time Voltage Instability Identification Algorithm Based on Local Phasor Measurements Sandro Corsi, Member, IEEE, and Glauco N. Taranto, Senior Member, IEEE Abstract—This paper proposes a new voltage instability risk indicator based on local phasor measurements at fast sampling rate. The effectiveness of the indicator is analyzed at EHV load and “transit” buses. The risk criterion is based on the real-time computation of the Thevenin equivalent impedance of the classic electrical circuit given by an equivalent generator connected to an equivalent load impedance through an equivalent connecting impedance. The main contribution of the paper is the innovating algorithm utilized on the real-time adaptive identification of the Thevenin voltage and impedance equivalents. The algorithm performance is shown through a detailed sensitivity analysis. The paper presents important numerical results from the actual Italian EHV network. Index Terms—Dynamic simulation, long-term voltage stability, online VSA, phasor measurement unit (PMU), Thevenin equivalent, voltage stability. I. INTRODUCTION V OLTAGE instability in the transmission networks has led or significantly contributed to some major blackouts around the world. Its timely recognition is very crucial to allow effective control and protection interventions. On this concern, a worldwide interest in defining an effective real-time voltage stability indicator, under the assumption of very fast measurements of system electrical variables (use of PMU—phasor measurement unit [1]—or devices computing phasors in comparable speed), is growing. By measuring the local voltage and current phasors at an EHV bus, the voltage instability analysis can be performed by considering the classic Thevenin equivalent “seen” from the bus [2]–[5]. As generally understood, the instability is linked to the condition of equality between the absolute value of two equivalent impedances—the load and the Thevenin impedances—with this equality corresponding to the maximal power transfer. One of the first works on the subject [2] refers to the two-bus Thevenin equivalent system, as depicted in Fig. 1. The identifiand is based on a recursive least-square idencation of tification method applied to the local voltage and current phasor measurements. Results are presented for -constant loads as a sequence of power flow solutions, resembling a continuation power flow. Manuscript received January 5, 2007; revised February 15, 2008. The work of G. N. Taranto was supported in part by CNPq and and in part by CESI during his sabbatical leave of absence. Paper no. TPWRS-00910-2006. S. Corsi is with CESI Spa., Milano, Italy (e-mail: corsi@cesi.it ). G. N. Taranto is with the Federal University of Rio de Janeiro, COPPE/UFRJ, 21945-970 Rio de Janeiro, Brazil (e-mail: tarang@coep.ufrj.br ). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TPWRS.2008.922586 Fig. 1. Two-bus Thevenin equivalent circuit. The works [5], [6] differ in the identification method utilized, which is simplified to achieve a high-speed voltage instability risk evaluation. The work [5] uses the concept of insensitivity of the apparent power at the receiving end of the transmission line to infer the voltage instability proximity, whereas the work [6] uses the same concept of Thevenin equivalent and relies on the Tellegen’s theorem to identify the Thevenin parameters. In [3], the authors extend the previous analyses to ZIP loads and present a mechanism to include the -constant and -constant portions of the load into the equivalent Thevenin impedance, allowing to conclude that maximum loadability and voltage instability occur at the same point. They also propose to monitor the status of the over-excitation limiters (OELs) of nearby generators and use the information for voltage instability proximity indication. In practice, the main critical aspects of the methodologies based on real-time measurements at a given bus are: • computing uncertainty of the equivalent grid parameters depending on the identification method, and their high sensitivity to small changes of local measurements at fast sampling rate; • computing time which often does not allow enough speed for real-time applications; • significant performance differences of the real system with respect to the simple electrical model of the considered equivalent circuit, when approaching the voltage instability condition; • unknown parameters of the ZIP load required in [3] when applied in the field; • absence of EHV “transit” buses from these analyses. “Transit bus” is a bus with no load directly connected to it (in the sense of a one-line diagram). It normally represents important EHV buses in bulk transmission corridors or nearby load centers, having weak or no parallel paths. 0885-8950/$25.00 © IEEE 1272 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008 This paper has the objective to overcome the above-mentioned criticisms by defining a real-time indicator, introduced in [7], and able to effectively support practical applications. The paper presents a new algorithm for fast-tracking the Thevenin parameters (voltage and reactance) based on the local voltage and current phasor measurements. Contrary to the least-squaretype identification methods, that generally need a large data window to suppress oscillations, the proposed algorithm has a good feature to have these oscillations filtered without significantly delaying the identification process. For voltage stability studies, some basic and fundamental considerations should be put in evidence [8]. The first point is that the Thevenin equivalent has to represent a detailed dynamic model. Therefore, its parameters change continuously and quickly while approaching the maximum loadability. The second refers to the local area OELs and on-load tap changer (OLTC) transformers whose dynamics strongly impact on the bus maximum loadability and voltage instability [9]. The third refers to the P-constant load analysis, which is not adequate for a correct identification in real systems, where a mixed ZIP load is usually adopted. The fourth point distinguishes the maximum loadability point (PV “nose” curve) from the real instability point, which in general occurs at very low voltage. The last basic consideration refers to the high speed the PV curve equilibrium point moves from the PV curve nose tip to the first unstable point at lower voltage. This fact confirms the practical importance of the nose tip identification to prevent against voltage instability, as well as the need of a very fast identification procedure when approaching the voltage instability limit. According to these basic considerations the proposed new way is capable to identify in real-time the Thevenin equivalent, with higher precision in proximity of the voltage instability where the OELs and OLTCs operate. This faster-change dynamic period also corresponds to the relevant simultaneous and in opposite direction variations of the load and the Thevenin equivalent impedances. The proposed algorithm is able to distinguish these critical variations from those happening during the normal operating conditions, and therefore, to send a timely alarm at the beginning of the voltage degradation process. We remark, however, that the proposed identification algorithm is not suited for short-term voltage instability occurring straight after large perturbations. The proposed algorithm is tested through time-domain simulations performed in a large and realistic representation of the EHV Italian network. Results on both load and “transit” buses are presented. II. BASICS OF THE PROPOSED ALGORITHM Considering the electrical circuit in Fig. 1, the computation and , based on the phasors and measured at of the load bus, is the objective to be pursued. From Kirchoff’s law (1) with . Fig. 2. Phasor diagram of the two-bus equivalent circuit. This equation gives evidence that the considered problem has infinite solutions. In principle, two subsequent phasor measureand can be used to compute and ments of the pair , under the assumption that they do not change during the time interval between the two subsequent measurements. According to that, the corresponding matrix equation takes a high risk of being singular when considering the required very short sampling-time interval. Therefore, computations as such, are very risky due to the numerical difficulties, and the identification procedures based on this approach usually requires time (a large data window) to converge. Only in the presence of significant system variations between two subsequent measurements it is possible to achieve an acceptable result. This happens when the system is collapsing; therefore, it could arrive too late for activating preventive controls and special protections. Because of that, it is proposed a new algorithm to speed up the identificaand . tion of the Thevenin equivalent parameters III. DESCRIPTION OF THE PROPOSED ALGORITHM The maximum power transfer to the load in the electric circuit shown in Fig. 1 occurs when (2) . with This circuit represents the entire network “seen” from the considered bus in an equivalent way. According to the phasor diagram show in Fig. 2 the following relationship holds: (3) (4) , and . with Separating (4) into real and imaginary parts, yields (5) (6) For the equivalent Thevenin impedance “seen” from an EHV , and the assumption of is bus, we have very reasonable. Therefore, an initial estimation for is given CORSI AND TARANTO: REAL-TIME VOLTAGE INSTABILITY IDENTIFICATION ALGORITHM by (7) and are measured quantities taken from the PMUs, Since the initial estimation of still depends on . The admissible must be in agreement with the electric circuit range for laws. Up to the maximum power transfer (MPT) and consid) corresponds ering an inductive load its minimum value ( to the load voltage, and its maximum value ( ) corresponds (with ). In normal opto the voltage when erating conditions the load impedance is much higher than the equivalent Thevenin impedance: a good start estimation value is the arithmetic average of its extreme values given by for (8) and , with obwhere tained from: Even being inside of its admissible range, as the free variable of the problem, should be updated towards its correct physical value. The appendix shows the proposed theorem and a simple numerical example that elucidates to which direction should be updated at each sampling time in order to speed identification convergence. The proposed algorithm up the and are constant in the brief “ assumes that ” interval of their identification, which requires a very short sampling time. In brief, the proposed adaptive algorithm will when the variations of and the estimated reduce have the same direction, otherwise it will increase . This is evident from the following (see the Appendix (A8)) where the has to be reduced to avoid the error on increases error when the current reduces: (9) The simple numerical example given in the Appendix helps to further clarify this concept. Knowing the direction should be updated at each sampling time, we need now to establish how large this variation should be. This quantity is calculated as follows: with (10) (11) (12) (13) with being a pre-specified parameter chosen in such way to constrain the identification error within narrower bounds, and being the corresponding time step. Most of the times drives the identification process, so its specification has a major impact and are active only when in the process. The quantities the estimated is close to the edges of its feasible range. In the following frame the adaptive algorithm that tracks the corto identify is presented. One possible rect value of . variant of the algorithm is with respect to the calculation of At Step 4, the calculation of and can be done by using 1273 , calculated over a window of approa moving average of at priate size ( ), instead of the instantaneous value of iteration . This variant has the advantage to filter the identified variables paying the price of a slower identification process. In value synthesis, the proposed algorithm will identify the identification is reached, by imposing an osas soon as the cillation with small amplitude around their correct values with frequency half of the sampling rate. Algorithm to identify Step 1) Estimate initial values for according to (7) already considering according to (8) and . Step 2) Calculate from (6). Step 3) Calculate according to the conditions: If load impedance variation is negative do If then If then If load impedance variation is positive do If then If then If load impedance is constant Step 4) Calculate and from (7) and (6), respectively. Step 5) Increment and go to Step 3. OBS: is an intermediate evaluation of that takes into account the present values of the voltage and current phasor and . measurements and the previous values of IV. SENSITIVITY ANALYSIS OF THE PROPOSED METHOD Generally the degree of success of identification methods relies on their tuning parameters. The proposed method has only one key parameter that has a major impact on the identification process, the parameter . This section starts showing a sensitivity analysis on the parameter . The results were obtained from dynamic simulation tests performed on a large power system considering a sampling rate of 20 ms. The test system will be further explained in the next section dedicated to the assessment of voltage instability risk based on the proposed for different method. Figs. 3 and 4 show the estimated values of the parameter . It can be noted that for larger values of the estimation of is faster (Fig. 3) but more oscillatory (Fig. 4). Our experience after many runs of the algorithm says that good results are obtained when is in the range from 0.01% to 0.1% of the nominal bus voltage. One good practice is to adopt larger values of during the initialization phase (first seconds or minutes) of the identification process, and then switch it to a lower value as the value of approaches to a steady value. Fig. 5 shows the corresponding estimated as a function of the parameter . Fig. 6 shows 1274 Fig. 3. Identification of E Fig. 4. Identification of E IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008 Fig. 5. Identification of X as a function of k . Fig. 6. Convergence of E as a function of its initial value. Fig. 7. Identification of X as a function of m. as a function of k . as a function of k . the convergence of the estimated parameter by increasing the difference of the initial values of with respect to the correct estimation. Similar convergence curves are obtained for . The results show a good performance of the proposed alwas utilized in gorithm during the initial phase. A the results of Fig. 6. Fig. 7 shows the sensitivity with respect , for the cases with to the parameter . Fig. 7(a) shows , 4, 10, and 100, in a period of time where there is a , and Fig. 7(b) shows change in the identified value of at the end of the simulation when reaching a voltage instability. In both cases, it is clear to see higher delays in the identificaas gets larger. Fig. 8 shows the convergence of tion of as a function of the signal-to-noise ratio (SNR) when white noise is added to the voltage and current phasor measurements. In the absence of a due filter, observe that when the power of the noise increases with respect to the power of the phasors, the algorithm starts to drift from the correct value. Therefore, the required filter on the phasor signals has to be checked not only on the cut-off frequency tuning but also on the SNR value effects. The maximum frequency of the white noise added to the signal in Fig. 8 coincides with the frequency of the phasor measurement sampling, i.e., 50 Hz. Fig. 9 shows the identification of the equivalent Thevenin voltage, for the case when SNR dB, as a function of the cut-off frequency of a low-pass filter. It can be noted that the performance of the proposed algorithm increases as the cut-off frequency of the low-pass filter decreases. When the cut-off frequency is half of the measurement frequency the algorithm is already able to track the correct parameters. Fig. 8. E Convergence as a function of SNR without signal conditioning. The choice of the filter cut-off frequency is not critical, as long as the system dynamics under analysis stay lower than 25 Hz (half of the sampling-rate frequency). Thus, Fig. 10 proceeds with a sensitivity analysis of the identification algorithm, CORSI AND TARANTO: REAL-TIME VOLTAGE INSTABILITY IDENTIFICATION ALGORITHM 1275 Fig. 9. Convergence of E as a function of the cut-off frequency of the lowpass filter, for the case when the SNR = 80 dB. Fig. 12. The 380-kV Italian network. is not small enough to track a problem occurring, taking the simulation of Fig. 11 as an example, in a timeframe of 1200 s. V. SIMULATION RESULTS ON A LARGE-SIZE SYSTEM Fig. 10. Convergence of E as a function of SNR with signal conditioning. This section shows the application of the proposed method to the Italian system. The Italian grid analyzed contains the 380-kV and 220-kV networks (Fig. 12 depicts the 380-kV network only). The system configuration has 2549 buses, 2258 transmission lines, 134 groups of thermal generators, and 191 groups of hydro generators. Dynamic models for the generator, AVR, OEL, governor and OLTC are considered. The system load is approximately 50 GW, represented as a and in (14). The system is static model with under primary voltage and frequency control only (14) Fig. 11. Identification of X as a function of the sampling rate (s.r.). for varying SNR, when the phasor signals filtered by low-pass filter with a cut-off frequency of 5 Hz. The results show that the algorithm has an acceptable performance for SNR equals or as a higher than 40 dB. Fig. 11 shows the identification of function of the sampling rate (s.r.) of the phasor measurements. It can be noted that when the algorithm is utilized with a s.r. is tracked with more precision, although being of 20 ms, more corrupted with high frequency noise. With a s.r. of 2.56 s is sluggish, and it can be noted that before the tracking of , the system undergoes voltage finding a steady value for instability. This is a very interesting and counter intuitive fact, because it is difficult to say that a sampling rate of few seconds All the simulations assume a sampling rate of 20 ms, with the identification updated at every 20 ms. The identified value 20 ms (for tests on is based on the sampled data of the last the Italian system was set equal to 4). The parameter was set within the range of 0.01% to 0.1% of the nominal voltage. Two sets of tests were performed, one at a load bus and the other at a “transit” bus. A. Application to a Load Bus The tests performed on a load bus consisted of increasing the load, while maintaining the power factor constant, in the neighboring buses at a given rate and in the considered load bus at a higher rate. Our objective in testing the proposed method was not to find which bus would first exhibit a voltage instability problem, instead, how the method would behave as the voltage instability point was reached at the bus most prone to instability. The tests were performed in Brugherio 380-kV bus at the Milano Area (Region A marked in Fig. 12). 1276 Fig. 13. Identified Thevenin reactance. IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008 Fig. 15. OEL indicators of generators near Brugherio bus. Fig. 16. Voltage instability risk indicator for Brugherio load bus. Fig. 14. Tap evolution of OLTCs near Brugherio bus. The analysis performed in the Milano Area consisted of increasing the local area load by a rate of 10%/min maintaining constant the power factor. The load at the Brugherio bus was increased by a rate of 40%/min. Fig. 13 shows the identified Thevenin reactance and the load impedance. The curves in all the figures of this section are drawn up to the voltage instability point (considered as the point where numerical difficulties start to occur in the simulation program). Fig. 14 shows the tap variation of the OLTCs located electrically close to Brugherio bus. The heavy solid lines represent the three OLTCs that feed the local loads at the Brugherio bus. A continuous tap model represents the OLTCs. Tests with a discrete model were performed without any relevant impact on the identification process. Fig. 15 shows the OEL indicators of some important group of generators. The heavy lines correspond to three groups of generators (Turbigo and Tavazzano 1–2) electrically close to the Brugherio bus. When the indicator reaches the value zero, the OEL starts to operate. It can be noted that when these three OEL start to operate at around 550 s, the identified Thevenin reactance, as shown in Fig. 13, increases substantially. Fig. 16 shows the proposed voltage instability risk indicator defined as the ratio between the identified Thevenin ) and the load impedance ( ). Since the loads reactance ( are realistically represented with a mixed ZIP model, the voltage instability occurs [9] a little later than the MPT (when the impedances are equal). Fig. 17. Equivalent Thevenin voltage seen from Baggio bus. B. Application to a “Transit” Bus Not only due to practical interest, but also for being the most onerous for testing the identification procedure, the proposed method was checked on a “transit” bus. The “transit” bus is modeled as a load bus according to the assumptions made in [7], where the transmission lines with power leaving the bus represent the “equivalent load,” and those with power entering the bus represent the Thevenin equivalent. The algorithm was tested in the Baggio 380-kV “transit” bus also located at the Milano Area. Figs. 17 and 18 show the time simulation of the equivalent Thevenin voltage and reactance, respectively. Fig. 17 also shows the theoretical limit values for the equiva, ) and Fig. 18 also shows the lent Thevenin voltage ( equivalent load impedance seen from Baggio bus. The discontinuities seen in the curves of the Thevenin voltage, Thevenin CORSI AND TARANTO: REAL-TIME VOLTAGE INSTABILITY IDENTIFICATION ALGORITHM Fig. 18. Equivalent Thevenin reactance seen from Baggio bus and equivalent load impedance. Fig. 21. Identification of 1277 X for load steps having 20% slope in its flat part. The searching for a best-performance indicator is out of the focus of this paper. C. Application in the Case of Large Step Sequence of Load Perturbations With 20% Slope in the Flat Part Fig. 19. The voltage instability risk indicator for Baggio “transit” bus. Here it is shown a very difficult condition to the algorithm where we combine high frequency large load stepping (50 s-duration pulses) and ramps with small slope. Fig. 21 shows the identification of the equivalent Thevenin voltage and reactance, respectively, for a load variation at P. A. Caiano, consisting of steps having a 20% slope in its flat part. The performance of the algorithm is satisfactory also in this case. VI. CONCLUSIONS Fig. 20. Voltages at Brugherio load bus and Baggio “transit” bus. reactance, and indicator (Fig. 19) represent the start of operation of nearby OELs. According to the simulation results for Baggio bus, closer attention should be paid after 400 s. It can be noted that before 400 s the indicator is reasonably flat, changing to a steeper increase afterwards. Fig. 20 shows the voltage magnitudes of Brugherio and Baggio buses. At round 580 s (100 s before the instability) the voltages at both buses are at 0.9 p.u. Thus, solely relying the voltage instability proximity on voltage magnitudes may be too late. Figs. 16 and 19 show a better and earlier insurgence view of the voltage instability process. We think that other more sophisticated indicators could be deand . fined by having the real time updated values of The paper proposes a real-time voltage instability algorithm to recognize in a very fast way the raising of the voltage instability phenomenon seen by an EHV bus provided with PMU. The first contribution is the description of the intrinsic difficulties to identify in real time the Thevenin’s equivalent circuit seen by the considered EHV bus. The proposed algorithm is described and deeply checked through a complete sensitivity analysis, based on large power system data with ZIP load characteristics at the sampling rate of 20 ms, with very satisfactory results in terms of reliability, robustness and speed. The effectiveness of the algorithm was also confirmed by the dynamic tests performed on the Italian large power system through a detailed simulation model, both on load buses and on “transit” buses. The results are very clean and precise, notwithstanding the continuous changing of the real system data (50-Hz band) and the ) imposed. The deidentification high speed ( scribed method allows computing a variety of real-time reliable indicators to be used in practice, very simple and at low comand putation cost, mainly based on the distance between or and , as well as on the slope of the variables under identification [10]. APPENDIX The identification problem covered in this paper is basically a case of an underdetermined system, where a free variable is arbitrarily selected and the system is solved based on the selection and then of the free variable. The free variable chosen is is calculated. The here proposed theorem is based on the circuit shown in Fig. 1, and it establishes to which direction (i.e., 1278 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008 higher or smaller value) the selection of the free variable should be updated in order to approach its correct physical value. The mechanics of the theorem are based on the variation of the load and are conimpedance, and in the assumption that ” interval of their identification. stant in the brief “ Therefore, the correct application of the Theorem does require a very short sampling time to assure its thesis. At the end of this appendix the reader can find a simple numerical example that clarifies the basic concept behind the theorem. Theorem: In a two-bus Thevenin equivalent circuit in Fig. 1 subject to load variation, the link between the estimation errors ( and ) of the Thevenin parameters modules is always given, under the following assumptions, by the here after thesis: Definitions: the variation on the load impedance; • the initial identification error on ; • • • • • • • • the complex error on the phasor ; the given identification error on . Assumptions: , that is ; and the correct values of the Thevenin parameters, be constant; is an inductive impedance; at each step of load variation. Thesis: , If the load increases: with then: with then: or analogously: • If the load reduces: , with then: with then: . Proof: With reference to Figs. 1 and 2, the circuit current measured in the load bus is associated to the last estimated system model that generally includes parameters errors. On the other hand, that current is coherent with the correct model (without errors on the parameters). Therefore the following two equations are consistent: (A1) (A2) Now, without loss of generality in the real axis. According to that Note that, with the load held constant, there is no way to identify the two errors. Conversely, considering a load change during two subsequent ) and ( ) and maintaining constant sampling times ( between the two, it is possible to write (A6) (A7) Subtracting (A6) from (A7) yields (A8) (A9) Equation (A9) confirms the thesis statement. , is also overIn words, in case have overestimated estimated (A4) and increasing the load, changes, assuming lower values. On the contrary, in case have underesti, is also underestimated and increasing the load, mated assumes higher values. Analogous conclusions are inferred from the case of reducing the load. Results presented in [7] showed that the simplification of conas a fixed value during load build-up, resulted on sidering critically dependent on the value fixed an identification of for and able to achieve the correct value only at the MPT (we always start from ). In agreement with the results presented in [7], the following corollary is given to show that at the MPT the errors on both and cancel out, and a precise identification is possible. and , that is Corollary: If , or if and , that is , then there will be a sampling interval where: • ; • this condition corresponds to Thevenin’s equivalent maximum loadability; • is the value of the load bus impedance at the . reached maximum loadability point, as if Proof: From (A5), giving evidence to the load impedance change: is being considered to lie (A10) (A3) (A4) A4 links the identification errors at any sampling time and shows that the two errors have the same sign. Moreover, from the equality of the current in the first two equations (A5) In Fig. 1, the MPT is reached (with ) when . At this limit condition, reachable under the given hypotheses when , the following result is obtained from (A10) that refers to the analysis with the estimation errors: (A11) CORSI AND TARANTO: REAL-TIME VOLTAGE INSTABILITY IDENTIFICATION ALGORITHM 1279 ation of the impedance load and the variation of the Thevenin impedance have the same direction, the value of the Thevenin voltage should be reduced, otherwise, it should be augmented. A similar analysis could have been done for the case when the load impedance increases. In this case, the conclusions would be the opposite, i.e., when both impedance variations have the same direction, the value of the Thevenin voltage should be augmented, otherwise, it should be reduced. TABLE I CIRCUIT VARIABLES WHEN OVER-UNDER ESTIMATING E The correspondent circuit current is ACKNOWLEDGMENT The authors would like to thank F. Parma, M. Salvetti, and M.Pozzi for their contributions regarding the SICRE simulator. (A12) This confirms the thesis that the MPT happens when , even in the presence of estimation errors. In words the corollary states that at the sampling time , reaches a condition that combined with is equivalent to say that . In this condition, it is also . Therefore, verified the MPT corresponding the approaching of the MPT corresponds to cancel the errors on and . At , the load bus phasor measurements allow . That also means cannot be the correct estimation of maintained constant during the identification process otherwise the correct parameters identification can be achieved at the MPT only. A Simple Numerical Example: A very simple numerical example shows the basis of the parameters adaptive identification. Referring to Fig. 1, we assume that the correct values for the Thevenin voltage and impedance are 20 V and 1 , respectively. To explain the core feature of the proposed method, we suppose that these values are not known. To actually estiand be mate them we need to assume the values of constant between two subsequent measurements (Considering a sampling rate of 20 ms this assumption is very reasonable). Starting the analysis with load impedance equal to 9 the circuit current is 2 A, and the voltage across the load impedance is of 18 V. In the next step the load impedance decreases to a value equal to 8 . Without changing the logic of our analysis, we could also assume that we had measured the load voltage and current, and then calculated the load impedance. Now we separate the analysis in two ways—one for the case when we overestimate the generator voltage value, and one when we underestimate it. Overestimation—Table I, when estimating 1) shows the values of the circuit variables (columns 5 and 6). Note that when we over, a decrease in is accompanied by a estimate decrease in . —Table I, when estimating 2) Underestimation of shows the values of the circuit variables ,a (columns 7 and 8). Note that, underestimating decrease in is accompanied by an increase in . In conclusion, when the load impedance is decreasing, it can be inferred from this simple analysis, that when the vari- REFERENCES [1] A. G. Phadke, “Synchronized phasor measurements in power systems,” IEEE Comput. Appl. Power, vol. 6, no. 2, pp. 10–15, Apr. 1993. [2] K. Vu, M. M. Begovic, D. Novosel, and M. M. Saha, “Use of local measurements to estimate voltage-stability margin,” IEEE Trans. Power Syst., vol. 14, no. 3, pp. 1029–1035, Aug. 1999. [3] B. Milosevic and M. Begovic, “Voltage-stability protection and control using a wide-area network of phasor measurements,” IEEE Trans. Power Syst., vol. 18, no. 1, pp. 121–127, Feb. 2003. [4] M. Zima, M. Larsson, P. Korba, C. Rehtanz, and G. Andersson, “Design aspects for wide-area monitoring and control systems,” Proc. IEEE, vol. 93, no. 5, pp. 980–996, May 2005. [5] G. Verbic and F. Gubina, “A new concept of voltage-collapse protection based on local phasors,” IEEE Trans. Power Del., vol. 19, no. 2, pp. 576–581, Apr. 2004. [6] I. Smon, G. Verbic, and F. Gubina, “Local voltage-stability index using Tllegen’s theorem,” IEEE Trans. Power Syst., vol. 21, no. 3, pp. 1267–1275, Aug. 2006. [7] S. Corsi and M. Pozzi, “A real-time EHV bus-bar indicator of local voltage instability,” in Proc. IFAC Symp. Power Plants and Power Systems Control, Calgary, AB, Canada, Jun. 2006. [8] C. W. Taylor, Power System Voltage Stability. New York: McGrawHill, 1994. [9] S. Corsi and G. N. Taranto, “Voltage instability – The different shapes of the “Nose”,” in Proc. IREP Conf. Bulk Power Systems Dynamics and Control VII, Charleston, SC, Aug. 2007. [10] S. Corsi, G. N. Taranto, and L. Guerra, “New real-time voltage stability indicators based on phasor measurement unit,” in Proc. CIGRE Conf., 2008. Sandro Corsi (M’89) received the Doctorate degree in electronics from the Polytechnic of Milan, Milan, Italy, in 1973. In 1975, he joined ENEL’s Automation Research Center, where his main interests were power system voltage control, generator control, power electronics devices, advanced control technology, and power system automation. He was a Research Manager at ENEL S.p.A. Ricerca beginning in 1997, in charge of the Power System Control and Regulation Office. After the ENEL reorganization, he was a Research Manager at CESI, in charge of the Network and Plant Automation Office from 1999 and of the Electronics and Communications Office in the Automation Business Unit from 2000. Since 2001, he has been with CESI R&D and Sales. He is author of more than 80 papers on system control. Dr. Corsi is a Member of CIGRE, IEEE-PES, CEI, and the IREP Board. Glauco N. Taranto (S’92–M’96–SM’04) was born in Rio de Janeiro, Brazil, in 1965. He received the B.Sc. degree in 1988 from the State University of Rio de Janeiro, the M.Sc. degree in 1991 from the Catholic University of Rio de Janeiro, and the Ph.D. degree in 1994 from Rensselaer Polytechnic Institute, Troy, NY, all in electrical engineering. In 2006, he was on sabbatical leave as a Visiting Fellow at CESI, Milan, Italy. Since 1995, he has been with the Electrical Engineering Department, Federal University of Rio de Janeiro /COPPE, Brazil, where he is currently an Associate Professor. His research interests include power system dynamics and controls, intelligent control, and robust control design. Dr. Taranto is a member of the IEEE PES, CSS, PSDP Committee, and CIGRE.