14th PSCC, Sevilla, 24-28 June 2002 Session 24, Paper 4, Page 1 STOCHASTIC PREDICTION OF VOLTAGE DIPS USING AN ELECTROMAGNETIC TRANSIENTS PROGRAM Juan A. Martinez-Velasco Jacinto Martin-Arnedo Universitat Politècnica de Catalunya Barcelona, Spain martinez@ee.upc.es Abstract –This paper presents a summary of the work performed by the authors on voltage dip analysis using a time-domain tool, the ATP version of the EMTP. The document describes how to adapt this program to this type of studies and how to perform Monte Carlo-based stochastic predictions of voltage dips. A medium size distribution network is used to analyze the advantages of this approach, the convergence of the Monte Carlo method, the influence of the protection system and the importance of voltage dip indices. The current limitations of this tool are also discussed. Keywords: Power Quality, Voltage Dips, Monte Carlo Analysis, Digital Simulation, Transient Analysis, Stochastic Prediction, ATP/EMTP 1 INTRODUCTION A voltage dip is a sudden short duration drop of the rms voltage, followed by a recovery within 1 minute. A dip is usually characterized by the remaining (retained) voltage, its duration and the phase jump, that is the difference between the voltage phase before and after the dip [1]. The most severe voltage dips are caused by short-circuits, generally associated to bad weather conditions (i.e. lightning strokes). They can also be caused by transformer energizing, motor starting and sudden load changes. Although voltage dips are less severe than interruptions, they are more frequent; in addition, their consequences for sensitive equipment, such computers, adjustable speed drives or control equipment [2], can be as important as those by an interruption. Given the diversity of their causes and the difficulty of preventing all these causes, voltage dips are presently the most important power quality disturbances. These reasons have increased the interest on equipment aimed at preventing or reducing their effects [3], [4]. The voltage dip performance of a power network can be predicted by monitoring the network; however, the monitoring period that is needed to achieve a reasonable accuracy is too long [1]. Digital simulation can be then an alternative to predict this performance. In addition, other benefits can be derived from simulation, for instance the effect that some mitigation techniques can have on the network performance. Several methods have been proposed to predict the number of voltage dips originated in power networks. The most widely used are the “Method of Fault Positions” and the “Method of Critical Distances” [1], [5], [6]. The approach presented in this paper is based on the applica- tion of the Monte Carlo method and assumes that voltage dips are only caused by faults. A voltage dip is originated as a consequence of a transient in an electrical network. Although several type of tools have been used to simulate voltage dips, only those based in a time-domain solution can obtain with high accuracy the main characteristics of voltage dips and reproduce their effects, include the dynamic behaviour of power components and analyze the performance of mitigation techniques. This paper presents a new procedure for the stochastic prediction of voltage dips in power networks using the ATP (Alternative Transients Program) version of the EMTP (ElectroMagnetic Transients Program) [7]. The present work has been based on the development and implementation of new ATP capabilities created for this specific application. A stochastic prediction of voltage dips based on a Monte Carlo method requires several capabilities : a multiple run option to simulate the test system as many times as needed, the generation of random numbers that will be used to obtain the fault characteristics (location, resistance, initial time and duration, type of fault), modules for monitoring dips and calculating power quality indices [8]. After a short introduction to the Monte Carlo method, the document presents a summary of the main solution methods and capabilities of the ATP. All test studies presented in the paper are based on a distribution network of medium size (27 nodes, 26 lines). They show the results to be expected from ATP simulations, how the convergence of the Monte Carlo method is achieved, the influence of protective devices on the dip characteristics, and the calculation of voltage dip indices that can be useful to qualify the performance of a power network. A discussion about the limitations of the current ATP version for this type of studies is also included. 2 THE MONTE CARLO METHOD The Monte Carlo method is a widely used technique for analyzing multidimensional complex systems. It can be used for solving both stochastic and deterministic problems, although the first type is the most usual one. This technique is based on a iterative procedure that is repeated using in every new step a set of values of the random variables involved in the process, being these values generated according to the probability density function associated to each variable. 14th PSCC, Sevilla, 24-28 June 2002 Session 24, Paper 4, Page 2 The following paragraphs are aimed at clarifying this question, presenting some basics of the method and introducing some definitions related to its application [9]. Although all the comments are based on the assumption that only one variable is involved, the conclusions are also applicable to multidimensional systems. • A random variable, yn, converges in probability to Y, if for every γ >0, the probability that yn is outside an γ interval of Y converges to zero when n → ∞. • The sample average of a random variable converges in probability to the mean value of the variable. • A proper sample procedure for a given Probability Density Function (PDF), f(x), and a Cumulative Density Function (CDF), F(x), is an algorithm which generates an outcome x such that for any value x0 the condition Pr[x ≤ x0]= F(x0) is fulfilled. • The realization of a random variable generated by a proper sampling procedure cannot be distinguished from a random variable generated in a process controlled by the CDF itself. • The realization of a random variable, x, obtained from the solution of x ∫ f ( x')dx = F ( x) = ξ (1) −∞ being ξ a random number generated from a uniform distribution defined in the interval [0,1], is a proper sampling procedure from the original CDF. • The statistical error of a Monte Carlo solution converges to zero at a rate ( n ) −1 , and the PDF of the variable converges to a normal distribution. This means that the limit distribution of the sample average is independent of the estimator distribution. • A Monte Carlo solution has a slower convergence that a numerical solution, i.e. for calculation of integrals, but the convergence is independent of the dimension of the phase space. The goal of a Monte Carlo solution is to derive the response/performance of a system as a function of some stochastic input variables. Sampling is iteratively repeated until convergence is achieved. If the realization of the random input variables is generated by a proper sampling procedure, the solution converges as the number of samples n → ∞, being ( n ) −1 the rate of the statistical error convergence to zero. The convergence is independent of the phase space dimension. 3 ATP SOLUTION METHODS AND CAPABILITIES ATP is a time-domain circuit oriented tool based on the trapezoidal rule. This rule converts equations of the network components into algebraic equations involving voltages, currents and past values. These equations are assembled using a nodal approach. Although this tool is mainly intended for transients simulations, it can also be used in frequency-domain simulations. Steady-state phasor solutions can be carried out to establish initial conditions for transient simulations, to analyze harmonic propagation or to obtain a system impedance as a function of frequency. The trapezoidal integration rule is very simple and numerically stable; however, it uses a fixed time-step size and can originate numerical oscillations. Although ATP users can incorporate additional damping or snubber circuits to avoid these oscillations, both techniques do introduce simulation errors. In addition, the steady-state solution can be applied only to linear networks at a single frequency. Many procedures have been proposed up to date to obtain the steady state of nonlinear power systems, but no solution method has been yet implemented in the ATP. These drawbacks and the fact that a very small time step size is needed when variable topology converters are included in the test system represent serious limitations for some studies, for instance a stochastic prediction of voltage dips when custom power equipment is used to mitigate their effects. The ATP package is made of at least three types of tools [10] : a preprocessor (ATPDraw), the main processor (TPBIG), and a postprocessor. In general, the ATPDraw provides an interactive, mouse-driven, graphical user interface, with very flexible options for incorporating custom-made modules and new tools to the package. Some new TPBIG capabilities allow users to create very powerful modules by adding calculations with module arguments and internal variables. They have also expanded the applications of this tool, which can be currently used for performing sensitivity analysis and any type of statistical studies. The custom-made modules that have been developed for voltage dip studies include power components, which simplify the use of basic models and extend the capabilities of the package to more complex equipment models, protective devices, such as reclosers and fuses, and monitoring devices, to characterize voltage dip and calculate power quality indices. ATP capabilities have also been used to develop those modules needed to perform a stochastic prediction of voltage dips, using an approach based on the Monte Carlo method. 4 TEST SYSTEM The scheme of the test system is shown in Figure 1, it is a medium size distribution network with two radial feeders. The lower voltage side of the substation transformer is grounded by means of a zig-zag reactor of 75 Ω per phase. 14th PSCC, Sevilla, 24-28 June 2002 2 km 6 Session 24, Paper 4, Page 3 8 5 km 2 0 FEEDER 2 10 km 13 11 24 22 2 km 5 km 21 5 km 5 km 2 km 10 10 km 15 17 2 km 12 2 km 10 km 10 km FEEDER 1 16 5 km 5 km 23 27 25 5 km 5 km 26 2 km 1 9 14 5 km 2 km 5 km 4 3 2 km 2 km 5 7 2 km 18 20 19 2 km 2 km HV equivalent : 110 kV, 1500 MVA, X/R = 10 Substation transformer: 110/11 kV, 20 MVA, 8%, Yd Distribution transformers : 11/0.4 kV, 1 MVA, 6%, Dy Lines : Z1/2 = 0.61 + j0.39, Z0 = 0.76 + j1.56 Ω/km Figure 1: Diagram of the test system. 5 SIMULATION RESULTS 5.1 Introduction The approach presented in this paper is based on the random generation of disturbances, and assumes that dips are due only to faults originated within the distribution network. Since load are represented as constant impedances and they do not include any dynamic behaviour, voltage dips will be rectangular, and characterized by the remaining voltage, the duration and the phase angle jump. Figure 2 shows the rms voltages due to a line-to-ground fault produced at a radial distribution network, similar to that shown in Figure 1. The plot shows a case with a dip and a swell originated at the same node. A 20 C Voltage (kV) 15 10 5 0 20 40 60 80 100 Time (ms) 120 140 Figure 2: Voltage dip simulation. 160 The developed procedure can be summarized as follows. The test system is simulated as many times as required to achieve the convergence of the Monte Carlo method. Every time the system is run, faults characteristics are randomly generated using the following parameters • Fault location : the fault may occur at any point of the distribution system, so the location is selected by generating a uniform random number • Fault resistance: normal distribution, average value = 10 Ω, standard deviation = 1 Ω • Initial time of the fault: uniform distribution between 0.04 and 0.06 s • Duration of the fault: normal distribution, average value = 0.06 s, standard deviation =0.02 s • Probability of each type of fault: LG: 80%, 2LG: 17%, 3LG: 0%, LL: 2%, 3L: 1%. After every run, voltage dip characteristics at the nodes of concern are recorded. At the end of the simulation, this information is manipulated to obtain the voltage dip density function. Note that the number of variables involved in any voltage dip simulation is rather high, and they could be increased if the random nature of the loads was also represented. Although for radial networks like that simulated in this paper, a systematic approach could be used, in general a Monte Carlo method will be the best approach to solve this problem. Some of the characteristics values used in this study are not very realistic, and not applicable to most actual distribution networks. The fault durations are too short and no permanent faults are considered; a uniform distribution of the fault location could be a very crude approach in many actual networks; the fault resistance distribution could be far from the normal distribution assumed in this work. A permanent fault will obviously force the protection equipment to act, and will originate an interruption, which is not a subject of this study. However, if reclosers have been installed, their operation could produce typical reclosing sequences with two or more reclosing intervals, which could become two or more voltage dips at some nodes. The main goals of this study are, rather than simulating actual systems, to check the convergence of the Monte Carlo approach and test the ATP capabilities developed for this application. On the other hand, it is important to emphasize digital simulation advantages. As mentioned in the Introduction, realistic data can be only obtained by monitoring; however, a very long period can be needed to obtain accurate information [1]. Digital simulation can be useful to determine what is the influence of the most important parameters by performing parametric studies (sensitivity analysis) and find out what system parts/components do need to be improved. The following sections are dedicated to analyze the convergence of the Monte Carlo method; the influence that the protective devices can have on the voltage dip density function and the usefulness of voltage dip indices. 14th PSCC, Sevilla, 24-28 June 2002 Session 24, Paper 4, Page 4 • Fault initial time : The time span, 20 ms, was divided into intervals of 2.5 ms, so each interval represents a 12.5% of the span After 1000 runs : Standard deviation = 0.9695 Confidence interval = 0.8105 After 5000 runs : Standard deviation = 0.4509 Confidence interval = 0.3770 • Fault type : The distribution of fault types after 1000 and 5000 runs was as shown in the following table Fault type L-G 2L-G 3L-G L-L 3L 1000 runs 80.40% 16.00% 0% 2.00% 1.60% 5000 runs 78.98% 17.52% 0% 2.48% 1.02% As expected, the accuracy is increasing with the number of runs, although the percentage of some fault types is actually closer to the given distribution at 1000 runs. Figure 4 depicts the cumulative voltage dip density function which results at phase “A” of Node 4, see Figure 1, after 1000 and 5000 runs. Frequency (%) 10 8 6 4 2 0 11.4 22.8 34.2 45.6 57 68.4 79.8 91.2 102.6 114 Distance (km) a) Fault location Frequency (%) 25 20 15 10 5 0 6 7 8 9 10 11 12 13 14 Resistance (ohms) b) Fault resistance 14 Frequency (%) 12 10 8 6 4 2 0 0.04 0.0475 0.055 0.0625 Time (s) c) Initial fault time 9 8 Frequency (%) • Fault location : The whole distribution line length was divided into intervals of 11.4 km, that is each interval is the 10% of the whole length After 1000 runs : Standard deviation = 1.0154 Confidence interval = 0.5663 After 5000 runs : Standard deviation = 0.3562 Confidence interval = 0.2229 12 7 6 5 4 3 2 1 0 0.03 0.04 0.05 0.06 0.07 0.08 0.09 Duration (s) d) Fault duration 90 80 Frequency (%) 5.2 Convergence of the Monte Carlo method The study was performed using a single input file and taking advantage of the new ATP multiple run option. The following information was recorded at every run • the characteristics of the fault (location; initiation time, duration, resistance, faulted phases, and fault type) • the characteristics of the dips (remaining voltage, duration) at every node and every phase. Several criteria can be used to decide when the convergence of the Monte Carlo simulation is achieved. For instance, by checking the error between a few points on the cumulative probability curve of every random variable obtained by simulation, against the theoretical values, or by checking the confidence level, see Chapter 2. The test system was simulated up to 5000 times. If it is assumed that 12 short-circuits are generated by year and 100 km of overhead lines, as the network has 114 km, then it equals the performance of the network during 366 years. Figure 3 shows the probability distribution of all random variables after the 5000 runs. The following paragraphs present some of these results, deduced with a confidence level of 95%. 70 60 50 40 30 20 10 0 L-G 2L-G 3L-G L-L 3L e) Fault type Figure 3: Distribution of random variables. 14th PSCC, Sevilla, 24-28 June 2002 Session 24, Paper 4, Page 5 100 90 80 70 60 50 40 30 20 10 t> 0ms t >2 0ms t> 40m s t >6 0ms t >8 0ms 0 V V< V< 20 10 % % <3 0 V< % V V <9 0% V < V < 7 < 80 V< % 0% 60 50 % 40 % % a) Dip density function after 1000 runs 100 90 80 70 60 50 40 30 20 10 t> 0ms t >2 0ms t> 40m s t >6 0ms t >8 0ms 0 V V< <3 V< V< 0% 20 10 % % V< 40 % V V < V < 8 < 90 V % 0% 7 <6 0% 0 % 50 % b) Dip density function after 5000 runs detected, and close contacts after a certain reclosing interval. Figure 5 shows the dip density function at Node 4 derived with reclosers installed at the head of both feeders and from different recloser intervals. The charts are based on two different reclosing intervals : 20 and 60 ms. As expected, one can observe that the reclosing interval will significantly affect the voltage dip density on those dip durations shorter than the reclosing intervals. Since Figure 5 charts show the dip density function in percentage, it is not easy to deduce how many voltage dips are produced. Figure 6 compares the effect of a recloser on the average number of voltage dips per year, with and without reclosers. The results presented in Figures 5 and 6 should be interpreted as follows. A recloser needs some time to operate after a fault current has been detected, but once their terminals are open, the dip duration is increased at nodes located at the feeder where the fault is originated, and decreased at nodes located at the other feeder. As for the remaining voltage, it is decreased to zero at nodes located at the faulted feeder, and recovered to the pre-fault voltage at the other feeder nodes. This performance is obvious from Figures 5 and 6. On one hand, the number of dips with long duration and small remaining voltage has increased; on the other hand, there has been an increment in the number of dips with short duration. Figure 4: Dip density function at Node 4 – Phase A. 100 90 5.3 Influence of the protective devices The previous study has been made without including any feeder protection in the distribution network model. However, even assuming only short-duration faults, the protection equipment can significantly modify the voltage dip density function. A complete model of a distribution system should include most protection equipment : breakers, reclosers, and fuses. Even some protective relays should be included in some cases if an accurate enough simulation is needed. The results presented in this document has been deduced by assuming that only reclosers are installed and they have only one operation after the contacts open, so if the fault condition is still present, the contacts will remain open. The recloser model is not based on a time-current characteristic, instead the recloser will operate 10 ms after the fault current is 80 70 60 50 40 30 20 10 t> 0ms t >2 0ms t> 40m s t >6 0ms t >8 0ms 0 V< V< V< 20 10 % % 30 V< % V< V 90 V < V < 7 < 80 % V< % 0% 60 50 % 40 % % a) Reclosing interval of 20 ms 100 90 80 70 60 50 40 30 20 10 0 t> 0ms t >2 0ms t> 40m s t >6 0ms t >8 0ms Both charts give the probability of every characteristic interval. They show no significant differences between them. Similar conclusion was derived for the dip density functions at other phases and other nodes. So one can conclude that 1000 runs are enough to obtain accurate enough dip density functions for the present system. In fact, it is the income information (fault statistics) used for this study where error can be more significant. Therefore, the subsequent studies will be based on 1000 runs, which equal the performance of the network for 73 years. V< V 90 V < V < 7 < 80 % V< % 0% 60 V< 50 V< % 4 % V< 0% 3 V< 0 20 % 10 % % b) Reclosing interval of 60 ms Figure 5: Dip density function with recloser –Node 4 14th PSCC, Sevilla, 24-28 June 2002 Session 24, Paper 4, Page 6 3 2.5 2 1.5 1 0-2 0ms 20-4 0ms 40-6 0m s 6080m s 80100 ms 0.5 0 80 70 %60 %90 %8 % % 0 70 % 30 -6 0 %% 20 %50 % 10 %40 % 0% %30 % 2 % -1 0 0% % 40 50 a) Without recloser 4 3.5 3 2.5 2 1.5 1 0.5 0-2 0ms 20-4 0ms 40-6 0m s 6080m s 80100 ms 0 80 70 %60 %90 %80 % %70 % 30 %6 % 2 % 0 50 0% % 1 4 0 % 0 0% -3 0 %% 20 % -1 0 % % 40 50 b) Reclosing interval of 60 ms Figure 6: Number of dips per year – Node 4. 5.4 Calculation of voltage dip indices Several voltage dip indices have been proposed up to date to reflect the behaviour of a system and to assess the effect of mitigation techniques [8], [11], [12]. They can characterize either a site or a full system, using single event or a group of events. The current version of the module library allows users to obtain site and system indices. Some of the indices proposed are based on the concept of voltage dip energy, which can be calculated according to [11], [12] 2 W = {1 − V pu } T only those cases for which the voltage drops below 90% of the rated voltage. Voltage dip index values will depend on the fault statistics and the performance of the protection system. Based on the same studies described above, several calculations were made to analyze these influences. To compare the performance of the system and calculate the voltage dip index, the same sequence of events should be used with both configurations. A feature was implemented in order to record randomly generated events and use them every time they were needed, as for this example. Figure 7 shows the AVSEI values at all load nodes, see Figure 1, after 500 runs. Initially, the cases were run without installing any feeder protection; the subsequent simulations were run with reclosers installed at both feeders. All simulations were performed without permanent faults. It is obvious from Figure 7 that the voltage dip energy is significantly increasing when the reclosers do act. An explanation for this performance is shown in Figure 8. This chart compares the voltage dip energy values that correspond to 10 runs obtained at Node 4, which is located at Feeder 1. It is easy to deduce from this chart on which feeder the fault was originated. For instance, events 15, 16, 17, 21 and 23 were originated at Feeder 1, and the rest were originated at the other feeder. For those dips originated at Feeder 1, the actuation of the recloser increased the index value since the dip became a short interruption. For those dips originated at the other feeder, the index value decreased when reclosers did act, but the reduction was less significant. 0.10 0.09 0.08 AVSEI 4 3.5 0.07 0.06 Without Recloser With Recloser 20 ms With Recloser 60 ms 0.05 0.04 0.03 0.02 1 2 3 4 5 6 (3) being V the dip magnitude and T the dip duration. For three-phase events the dip energy is added for the three phases. One of these indices is the so-called Average Voltage Sag Energy Index (AVSEI) 7 8 9 10 11 12 13 14 Load Figure 7: Voltage Dip Energy at load nodes. 0.2 Without Recloser With Recloser 20 ms With Recloser 60 ms 0.15 0.1 N AVSEI = 1 ⋅ Wk N k =1 ∑ (4) being Wk the lost energy during the dip event “k”, and N the number of events. Although this expression does not differentiate dips from swells, the calculation of this index will be made taking into account only those phases in which dips are originated, and considering 0.05 0 15 16 17 18 19 20 21 22 23 24 Case Figure 8: Voltage Dip Energy at Node 4 25 14th PSCC, Sevilla, 24-28 June 2002 6 CONCLUSIONS This paper has presented the application of the ATP package in voltage dip analysis. The capabilities of this tool have been used for developing new custom-made modules adapted for this type of studies. Digital simulation is a very efficient alternative for predicting the performance of a network and for testing devices and techniques which could mitigate voltage dip effects. A tool based on a time-domain has many advantages, but the limitations of such tools and those of the procedure developed for this work are still important. The procedure can be used to analyze the performance of a system by assuming that dips are originated only by faults of short duration. Although it can also be used to compare its performance after installing some mitigation devices, in reality this second possibility is not yet realistic if custom power controllers are used. Since a very small time step size is needed when variable topology converters are included, simulating a big network several thousands times would be extremely long. Both new custom-made modules and ATP capabilities are still needed to obtain more realistic results. The aim of a stochastic prediction is not only to deduce the number of voltage dips but also the number of trips of sensitive equipment. Therefore, the representation of equipment sensitivity is also required. Several approaches can be considered; for instance, ATP capabilities can be used to include a very detailed ASD model (rotating machines plus static converters), when this is the type of affected equipment. However, for some studies, i.e. voltage dip index calculation, the model of a big industrial plant can be reduced to a few linear loads, being represented each of them by a module that incorporates voltage dip tolerances curves. An important aspect of any voltage dip analysis is the influence of protecive equipment. All cases analyzed with protective devices (reclosers) included in the system representation were very simple; however, they have been very useful to prove the influence that these devices can have on voltage dip characteristics and indices. The calculation of voltage dip indices is a rather recent approach for qualifying the performance of a power system [8], and no consensus has been reached about the most appropriate indices. The document has presented the calculation of only one index, and some of the influences that protective devices could have on its values. However, there would be no problem to improve present custom-made modules that represent measuring devices in order to calculate some other voltage dip indices. Load representation is an important subject in which capabilities of a tool like ATP have several advantages. All calculations presented in this work were made Session 24, Paper 4, Page 7 assuming that loads could be represented as constant impedances, and peak loads were simultaneous at all nodes. More realistic representations can be made by introducing any voltage dependence and assuming probabilistic models. As mentioned above, voltage tolerance is another important feature to be considered. REFERENCES [1] M.H.J. Bollen, “Understanding Power Quality Problems. Voltage Sags and Interruptions,” IEEE Press, 2000, New York. [2] H. Sarmiento and E. Estrada, “A voltage sag study in an industry with adjustable speed drives”, IEEE Industry Applications Magazine, vol. 2, no. 1, pp. 16-19, January/February 1996. [3] N.G. Hingorani, “Introducing Custom Power”, IEEE Spectrum, vol. 32, no. 6, pp. 41-48, June 1995. [4] N. Woodley, L. Morgan and A. Sundaram, “Experience with an inverter based dynamic voltage restorer”, IEEE Trans. on Power Delivery, vol. 14, no. 3, pp. 1181-1186, July 1999. [5] M.H.J. Bollen, “Fast assessment methods for voltage sags in distribution systems”, IEEE Trans. on Industry Applications, vol. 32, no. 6, pp. 14141423, November/December 1996. [6] M.H. Qader, M.H.J. Bollen and R.N. Allan, “Stochastic prediction of voltage sags in a large transmission system”, IEEE Trans. on Industry Applications, vol. 35, no. 1, pp. 152-162, January/February 1999. [7] H.W. Dommel, “ElectroMagnetic Transients Program. Reference Manual (EMTP Theory Book)”, Bonneville Power Administration, Portland, 1986. [8] IEEE Voltage Quality Working Group, “Recommended practice for the establishment of voltage sag indices”, IEEE P1564, Draft, March 2001. [9] A. Dubi, “Monte Carlo Applications in Systems Engineering”, John Wiley, 1999. [10] J.A. Martinez, “The ATP package. An environment for power quality analysis”, 9th International Conference on Harmonics and Quality of Power, October 14-16, 2000, Orlando. [11] D.L. Brooks, R.C. Dugan, M. Waclawiak and A. Sundaram, “Indices for assessing utility distribution system RMS variation performance”, IEEE Trans. on Power Delivery, vol. 13, no. 1, pp. 254-259, January 1998. [12] R.S. Thallam and G.T. Heydt, “Power acceptability and voltage sag indices in the three phase sense”, 2000 IEEE PES Summer Meeting, July 16-20, 2000, Seattle.