2010 IREP Symposium- Bulk Power System Dynamics and Control – VIII (IREP), August 1-6, 2010, Buzios, RJ, Brazil

Estimation of Voltage Stability Index in a Power System with Plug-in Electric Vehicles
K. Joseph Makasa Real-Time Power and Intelligent Systems Laboratory Missouri University of Science and Technology Rolla, MO 65409, USA jkm6g7@mst.edu Abstract
A Multilayer Perceptron (MLP) neural network based approach for estimation of the voltage stability L-index in a power system with Plug-in Electric Vehicles (PEVs) is presented. This technique overcomes the limitations of direct calculation of L-index from measurements at a load bus. The L-index calculation is dependent upon the noload voltage phasor for any given system topology and operating condition. In practice it is difficult to obtain noload voltage phasor at a bus each time the system topology or operating point changes. An MLP neural network based method capable of estimating voltage stability L-index at a load bus using direct measurements and that does is independent of the no-load voltage phasor of the bus is presented. Results show that the MLP accurately estimates the voltage stability L-index for different cases of system topologies and operating conditions.

Ganesh K. Venayagamoorthy Real-Time Power and Intelligent Systems Laboratory Missouri University of Science and Technology Rolla, MO 65409, USA gkumar@ieee.org
voltage phasor at the load bus for a given topology of the system. Since the no-load voltage is dependent upon the system topology and operating point, it varies as the system topology or operating point changes. In practice it is difficult to obtain no-load voltage at a bus. The proposed Multilayer Perceptron (MLP) approach capable of estimating the L-index without directly obtaining the no-load voltage overcomes this limitation of and facilitates on-line determination and use of L-index. The electric power grid is rapidly growing and demanding new technologies for efficient and rapid control in order to ensure reliable and secure power networks [6]. Reference [7] carries out a study of the impact of Plug-in Electric Vehicles (PEVs) parking lots (SmartParks) on the stability of a power system. In particular, the study shows voltage characteristics following changes in power demand of PEVs. When PEVs discharge into the power network, system voltage support is enhanced, while charging action is accompanied by voltage drop in the load area. The method for estimation of voltage stability L-index developed in this study is applied for monitoring voltage stability in a power system with SmartParks included.

Introduction
Power system voltage stability has become an issue of great concern for both power system planning and operation in recent years, as a result of a number of major black outs that have been experienced in many countries due to voltage stability problems [1,2]. This has been mainly due to power systems being operated closer to their stability limits because of increased demand for electricity [1]. Many studies have been carried out done to determine voltage stability indices in order to take necessary action to preclude eminent instability and thereby improving voltage stability in a power system. References [3, 4] present comparative studies and analysis of six different voltage stability indices, while [5] introduces the voltage stability L-index to be a simple but effective means of measuring the distance of a power system to its stability limit. The disadvantage of using the L-index, is that its calculation depends upon the no-load
This work was supported by the US National Science Foundation under the Grant EFRI #0836017.

Power System with Plug-in Electric Vehicles
Figure 1 shows the 11 bus voltage stability test system with plug-in electric vehicle parking lots models included. The power system consists of two generators, G1 and G2 supplying the load area through five long (200km) parallel transmission lines, and one local generator, G3 providing voltage support in the load area. Bus 11 is a voltage controlled bus using an on-load transformer tap changer. System parameters and loading conditions of the system used in this paper are those given in Appendix E of reference [1]. The original system with 10 buses was modified by adding two SmartPark buses 12 and 13.

978-1-4244-7465-3/10/$26.00 ©2010 IEEE

the system is voltage stable.0 at the referred bus [8]. L= 4 V0 VLcos(θ0 − θ L ) − VL cos2 (θ0 − θ L ) 2 V0 2 [ ] ……..(1) Voltage Stability Index Estimation Reference [8] presents the formulation of a voltage stability load index at a load bus using voltage equations. VSI Estimation using a Multilayer Perceptron A feedforward multilayer perceptron neural network can be used to estimate voltage stability L-index at a bus. load is 3000MW. The technique uses measurements of voltage phasors and no-load voltage at the bus to calculate the voltage stability L-index.8 kV 500 kV 5000MVA G1 1 5 500 kV 200 km transmission lines 500 kV 13. When the value of L at every load bus in the system is less than 1. 1800MVAr modeled as a constant power load. The index gives the distance of the bus to the voltage stability limit... 1. 1800MVAr 6 7 9 PL3 115kV 8 PL4 Industrial load SmartPark bus 13 1600MVA PL6 . The complete mathematical derivation of the Lindex is presented in the Appendix. while load (residential) at bus 11 is 3000MW modeled as constant power load. The nominal (industrial) load at bus 8. the system approaches its stability limit and becomes unstable when L is exceeds 1. Plug-in electric vehicle parking lots represent six SmartParks at bus 12 and 13 with capacity +/-180 MW each.0 at any bus.. Bus 8 is a load bus located in an industrial area and bus 11 is a load bus in a residential area.13. As the value of L approaches 1.0..8 kV 2200MVA 115 kV 10 11 Residential load 3000MW 13. Vo is the no load voltage at the node and VL is load voltage.8 kV Infinite Bus 13.8 kV PL5 3 G3 13. The voltage stability L-index is given by the equation: Where. Power System with plug-in electric vehicle parking lots (SmartParks). Equations (12) to (16) in the Appendix show that voltage PL1 PL2 G2 2 3000MW.8 kV PL7 PL8 PL9 PL10 PL11 PL12 SmartPark bus 12 Fig.

Evaluation of L-index was Fig.8 to 1. The first set of results are results for the development phase that show that the MLP was successfully trained for estimating voltage index to a high degree of accuracy. These quantities are selected as input variables in the estimation of L-index using the MLP neural network as shown in Fig. Particle swarm optimization training algorithm [10] was used to train the multilayer peceptron. 40 input patterns are used to test the accuracy of the neural network in estimating voltage stability L-index. The MLP consists of four neurons in the input layer. θ) measurements at the concerned load bus. the operation phase. Performance of the MLP is then validated using patterns of input output data sets to tests its performance. Values of real power and reactive power at non generator buses as well as voltages at the load buses were taken with the load at bus 8 and 11 varied simultaneously from 0. Activation functions in the input and output layers are linear activation functions while the hidden layer uses sigmoidal activation functions. reactive power (Q). The table shows that the MLP output values are very close to the target values at both bus 8 and 11. In the development phase. . . the trained MLP was applied in the estimation of voltage stability L-index in the system. .stability L-index is a function of real power (P). In the second phase. and voltage magnitude (V) and phase (θ) at the bus. Sixty input output patterns were selected at random and used as training data. and then the contingency of the outage of one transmission line was also evaluated. 2. . The output of the neural network is the estimated voltage stability L-index at the load bus. L-index Development and Training phase Training MLP using training data Test MLP performance Output layer Operation phase Application of trained MLP in estimating VSI Fig. training data was obtained. development phase and the operation phase. with no plug-in electric vehicles connected. Figures 4 and 5 show plots of bus voltage and L-index against real power and reactive power at bus 8. On-line application of the trained MLP is then carried out by applying the MLP to estimate the voltage stability L-index of the power system with SmartParks. 3. The inputs to the neural network are active and reactive power (P. After successfully training and validating the MLP neural network was used for estimation of voltage stability L-index of the test system with SmartParks included. 3. carried out with all five transmission lines in the system available. . In Fig. Voltage index target and MLP output values used in the testing phase are shown in Table I. the process of developing and implementing an MLP/ESN for estimation of voltage stability L-index involved two phases. Multilayer perceptron structure for L-index estimation.2 load factor in small steps to obtain one hundred sets of data. 4 increasing load demand at the load buses results in increasing voltage L-index and thus approaching the limit. First. Voltages at bus 8 and 11 were used to calculate voltage stability Lindices for bus 8 and 11 respectively used in the MLP training process as target values for corresponding sets of real and reactive power. Development and operation phases for VSI. Start Collect training data form smart grid P Q V θ Input layer Hidden layer . Training was carried out for the number of iterations that resulted in acceptable mean square error. 35 neurons in the hidden layer and a single neuron in the output layer. End The process followed in development of the MLP involved two phases shown in a flowchart in Fig. Results and Discussion As described in the preceding section. Q) and voltage magnitude and angle (V. In the development phase the MLP is trained for accurate estimation of voltage L-index. 2.

6823 0.9923 0.7492 1.0211 1.7292 0.9 120.6720 0.7658 0.68 -32.9829 0.8199 0.6221 0.41 -29.41 -35.7284 0.8 -930 -925.7279 0.14 -32.6531 0.3 -817 Q(MVar) L-index (Calculated) 0.8016 0.7 173.9769 0.82 -30.7374 0.9 -825.96 -37.8008 0.0313 1.67 2980 2991 3003 3013 3024 3035 3045 3055 3065 3074 3084 3093 3102 3111 3120 3128 3137 3145 3153 3161 3168 3176 3183 3190 3197 3204 3210 3217 3223 3229 3235 3241 3246 -963.16 93.0235 1.0184 1.42 -30.7605 0.7199 0. Testing results have validated the MLP in accurately estimating voltage stability L-index of the power system.2 -29.9871 0.0105 Voltage decreases with increasing load.6863 0.7457 0.0444 1.2 -147.5 134.6595 0.19 -38.0157 1.08 79.6786 0.86 2960 2971 2982 2993 3004 3014 3025 3035 3045 3055 3064 3074 3083 3092 3101 3110 3118 3126 3135 3143 3151 3158 3166 3173 3181 3188 3195 3202 -197.7941 0. In Figures 6 and 7.6774 0.21 -37.6 -934.76 -31.99 -29.9693 Plots of the voltage stability L-index output of the MLP for a 24 hour period at bus 8 and 11 are shown in Figs.6450 0.8 -897.7242 0.7675 0.7781 0.9903 0.6603 0.8 -131.4 -901.7804 0.9878 0.282 21.6519 0.68 -35.536 7.63 -39.6218 0.91 -37.49 -27.0343 1.9804 0.7513 0.7622 0.23 -41.6122 0.6 -99.7 -874.4 -180.0027 1.9979 0.2 -829.62 -30.9853 0. After 3200 MW bus 11 is the critical bus.0132 1.9846 0.7128 0.0391 1.6319 0.6124 0.6311 0.6411 0.5 -32.7034 0.0530 1.5921 0.2 -31.86 -40.0261 1.22 -30.1 147.9820 0.7549 0.94 -36.5810 0.7513 0.32 -40.63 -22.0396 (deg) -33.57 -28.8 -847.6851 0.7903 0.6750 1.6693 0. MSE obtained using an MLP: 8.55 -40.9 -883.95 -32.8141 0.9 -959 -954.7436 0.7568 0.Table I: Bus Load Flow and Calculated and Estimated Voltage Stability L-index Bus 8 Voltage P(MW) (pu) 1.1 106.9779 (deg) -27.7749 0.0470 1.6978 0.92 -68.0004 0.7359 0.6915 0.7883 0.68 -38.15 -35.5918 0.8312 0.9743 0.9794 1.44 -38.6896 0.7370 0.6686 0.7176 0.0450 1.93 -28.0290 1.6388 0.8261 0.35 -34.66 -83.61 -29. 8 and 10 respectively.0317 1.9975 0.7661 0.7 199.7041 0.7045 0.34 -52.8321 L-index (estimated) 0.7588 0.0183 1.0001 0.5 -887.6784 0.7946 1.8 186.7212 0.02 -30.7676 0.6417 0.45 -41.1 -906.7733 0.39 -31.85 65.0264 1.7431 1.78 -28.8200 0.2 -851.6021 0. The trained neural network is used to estimate the voltage stability L-index of the 11 bus test system with SmartParks included at bus 12 and 13.81 -34.7103 0.0503 1.6677 from the SmartPark to the grid.6959 0.7342 0.0477 1.6381 0. negative values of power represent charging of electric vehicles where real power flow is from the grid to the SmartPark and positive values of power represent discharging action where power flow is 0.0423 1.0081 1.95 -38.4 -939.6949 0.3 -842.09 -40.9949 0.7711 0.7452 0.48 50.6314 0.6320 0.6511 0.6 -115.6 -834.0209 1.7730 0.7108 0.7 -865.58 -31.7039 0.0053 0.7951 0.7164 0.92 -39.6 Q(MVar) L-index (Calculated) 0.08 -34.5819 0.5716 0.6022 0.0132 1.7833 0.7885 0.46 -37.9897 0.0055 1.7399 0.9718 0.1 -860.71 -27.9 -838.8136 0.2 -869.6873 0.7116 0.0365 1.7 -37.7224 0.81 -31 -31.5704 0.8255 0.5 -821.0339 1.4 160.6458 0.39 -39.8 -911.0079 0.0237 1.36 -28.1 -949.7536 × 10-5.7762 0.6587 0.9954 0.3 -920.0417 1.7857 0.7604 0.16 -39.9929 0. Two different operating conditions of the power system have been considered: the first case .8073 0.6504 0.7 -164.7817 0.5 -915.7933 L-index (estimated) 0.0158 1.14 -28.1 -892.78 -41 -41.6251 0.62 -34.19 -36.32 -32.95 36.51 -7.0030 1.6654 0.45 -36.3 -878.0287 1.3 -944. Table II: Bus 8 Load Flow and Calculated and Estimated Voltage Stability L-index Bus 11 Voltage P(MW) (pu) 1.8080 0.7 -856.89 -35.6967 0.7 -36.7811 0.0370 0.7528 0.6603 0.0106 1.7873 0.7308 0.

47 0.7 0.65 L-index 0. when the SmartParks are discharging.64 0.61 0. When the SmartPark is charging on the other hand. i. Smartpark power vaiation at bus 8. The MLP estimated L-index outputs for these two cases are shown in Figs. . 6.51 0.57 0.6 0. positive power. This is when the SmartPark is providing power to the grid. indicating the system is less stable.98 0. 7.e. 0. 9.45 0.06 0. 1.02 Voltage L-index 0. the voltage indices at the buses are lower.5 0.58 0.96 0.8 1. 1.02 0.59 0. Estimated voltage stability L-indices with SmartParks included. 4. 8.44 0 2 4 6 8 10 12 14 16 18 20 22 24 Fig.48 0.04 0.62 0. Smartpark power vaiation at bus 11.8 Voltage Stability L-index 1. Estimate voltage stability L-indices with Smartparks included.85 180 140 Bus 8 80 40 0 -40 -80 -120 0 2 4 6 8 10 12 14 16 18 20 22 24 0.56 0 2 5 lines in service 4 lines in service 4 6 8 10 12 14 16 18 20 22 24 Time [hours] Fig.75 1 Voltage L-index 0.98 0.85 Time [hours] Fig.66 Bus 8 L-index At both bus 8 and bus 11.75 Voltage Stability L-index 1.94 2950 Power [MW] 80 40 0 -40 -80 -120 0 2 4 6 8 10 12 14 16 18 20 22 24 3000 3050 Load [MW] 3100 3150 3200 3250 3300 Fig. voltage stability index values are higher.7 0. 0.Power [MW] considers the system fully operational with no fault.04 Bus Voltage [pu] 1 0.96 0. 0. Time [hours] Fig.49 Bus 11 5 lines in service 4 lines in service Bus Voltage [pu] 1. hence helping to increase the system stability margin.06 0.65 0.94 -200 -100 Reactive power [MVar] 0 100 200 300 400 0.46 0.65 Time [hours] Fig.63 0. 8 and 9. The parking lot is supplying additional power to the system (generating). In the second case the contingency with one of the five parallel transmission lines out of service is considered.67 0. 5. Plots of voltage stability L-index and voltage against reactive power for bus 8. 200 160 120 Bus 11 0. Plots of voltage stability L-index and voltage against real power for bus 8.68 0.

EMPD '95. EPRI Power System Engineering Series. Venayagamoorthy. Real power. Yixin. K. Atlanta. K. issues 2-3. S.de/hjaeger/pubs/ESNTutorial. R. USA. Iwamoto. “A Comparative Study of Voltage Stability Indices in a Power System”. Sinha. vol. Srivastava. P. Power System Voltage Stability. Power System Stability and Control. Q2 . GA. M. Vol. pp. reactive power and voltage at load bus are sufficient measurements to estimate the L-index. K. “Wide area Control for Improving Stability of a Power System with Plug-in Electric Vehicles”. Y.faculty. P. When one line is out of service. Q1 . 27.…………………. G. “Potentials and Promises of Computational Intelligence for Smart Grids”. A. 8 and 9. “Estimating of Loadability Margin of a Power System by Comparing Voltage Stability Indices”. D. EPRI Power System Engineering Series. Proceedings of the 2003 Power Engineering Conference. Mitra. 22. “A New Voltage Stability Index Considering Voltage Limits”. issue 5. pp. 45-351 S. International Conference on Energy Management and Power Delivery. International Conference on Control. pp. McGraw-Hill. Suganyadevia. Y. S. The presented neural network approach is independent of the no-load voltage at given bus. T. 567574. 24-26 April 2003. “A New Intelligent Algorithm for Online Voltage Stability Assessment and Monitoring”. Abidin. K. The following equations can be derived.. Calgary. 100-110 Y. 589-596. K.V1 θ 1 P 2 . Venayagamoorthy. “Development of Artificial Neural Network for Voltage Stability Monitoring”.117 R. Gudise. S. “Comparison of Particle Swarm Optimiszation and Backpropagation as Training Algorithms for Neural Networks”. G.: “An Improved Voltage Stability Index and its Application” . Consider a line connecting two buses 1 and 2 where P1 and Q1 are the power injected into the line as shown above.: “A New Index for Voltage Stability Monitoring and Enhancement”. October 2005. pp. McGraw-Hill. Results show that the MLP approach is able to estimate accurately the L-index even with changes in topology and operating conditions.…………………………………………. 71 4 – 719. pp. J. International Journal of Electrical Power & Energy Systems.K. D. Xiaodan. 4-6 June 2009. PSCE '06. Automation. 2009. ISBN 0-07035958-X. 20. November 2000. Communication and Energy Conservation. 1995. T. http://www. Kataoka. IEEE Power Society General Meeting. pp. of charging and discharging. C. Venayagamoorthy. International Journal of Electrical Power & Energy Systems. Watanabe. higher values of voltage stability L-index indicate that the system is less stable. Proceedings of 1995 P1 . The results also show that the system has lower values of voltage stability L-index. July 26 -30. K. 2. “A New Technique for Voltage Stability Analysis in a Power System and Improved Loadflow Algorithm for Distribution Network”. 1993. G. 110 . Kamalasadan. with all five transmission lines in service. Future work involves demonstration of the proposed approach with PMU measurements.iu-bremen. V. Bijwe. 10 Two bus network I1 = 2 P2 + Q 2 2 V2 2 2 [4] ……………………. PECon. G. Babulal. ISBN 0-07063164-0. P. and hence more stable. vol. Proceedings of the 2003 IEEE Swarm Intelligence Symposium. A. Tare. Thukaram. INCACEC. February-March 2009. 31. Kundur. issue 8. M. Taylor. V. Appendix The mathematical formulation of the Voltage Stability Lindex technique used in this paper is derived from voltage equations of a two bus network as shown in Fig. This can be seen in the graphs of Figs. Issue 8. Q2 Fig. P1 . V 2 θ 2 References [1] [2] [3] C. Rahman. Q1 P 2 . F. Sahari. Hongjie. Hazarika. A.……………………………(3) P2 = P1 − Ploss …………. 1994.pdf. K. Jasmon.…………. B. June 1998. where the graphs of L-index corresponding to the system with all five transmission lines in service as lower L-index values than that corresponding to the system with one of the five transmission line out of service. 1 – 4. W. CA. 10. pp. (2) [5] I1 = 2 P1 + Q1 2 V1 2 2 [6] [7] ………………………. G. [9] [10] [11] [12] [13] [14] Conclusion A multilayer perceptron feedfoward neural network based approach for estimation of the voltage stability L-index in a power system with SmartParks has been presented.Thus system stability margin varies according to the operation of the SmartPark operation. vol. Rahman. International Journal of Electrical Power & Energy Systems.(4 ) [8] . Proceedings of the Power Systems Conference and Exposition. IET Proceedings Generation Transmission and Distribution [In press]. vol. International Journal of Electrical Power & Energy Systems.

……. L= 4 V1 (P2 r1 + Q 2 r1 ) + (P2 x1 − Q 2 r1 ) 2 ..………..(16) V1 2 [ ] Q loss ⎛ Q22 + Q22 ⎞ ⎟ * x1 =⎜ ⎜ V2 ⎟ 2 ⎝ ⎠ …………….(11) Which can be simplified to 4 V1 (P2 r1 + Q2 r1 ) + (P2 x1 − Q2 r1 ) ≤1 2 V1 2 2 Applying equation (16) to the thevenin equivalent circuit gives.…(13) V1 2 2 [ ] Since V1V2 cos(θ1 − θ 2 ) − V2 = P2 r1 + Q 2 x1 2 ……. 11 Thevenin equivalent network 8P2Q2 r1x1 − 4V1 (P2 r1 + Q 2 x1 ) + V1 − 4 P2 r1 + Q2 x1 ≥ 0 2 4 2 2 2 2 ( ) .…….…(15) Substituting equations (14) and (15) into (13) we have: L= …………….(7) 2 2 The equation for the voltage stability L-index is applied to the thevenin equivalent circuit looking at the load bus as shown in fig.(6) 4 V1V2 cos(θ1 − θ 2 ) − V2 cos 2 (θ1 − θ 2 ) 2 .…. The circuit shows the system equivalent thevenin voltage and impedance. and has real roots when Fig...…….(14) and . The steps used in obtaining the Thevenin equivalent are as follows: a) Load flow solutions are used to determine the voltage profile of the system at a given load condition b) Thevenin voltage is obtained by load flow of the system with the load at the concerned bus removed I1 2 ⎡ 2 ⎛ P2 2 + Q 2 2 ⎞ ⎤ ⎡ 2 ⎛ P2 2 + Q 2 2 ⎞ ⎤ ⎟r1 ⎥ + ⎢Q 2 + ⎜ ⎟x1 ⎥ ⎢P2 + ⎜ ⎜ ⎜ ⎟ ⎟ V2 V2 ⎥ ⎢ ⎥ ⎢ ⎝ ⎝ ⎠ ⎦ ⎠ ⎦ ⎣ ⎣ = 2 V1 .……….……………….. [ ] ………… (12) L= 4 V0 VL cos(θ 0 − θ L ) − VL cos 2 (θ 0 − θ L ) 2 ……(17) V0 2 [ ] Therefore the voltage stability index is given by. (8) ⎛ P2 2 + Q 2 2 ⎞ 2 2 2 ⎟ r1 + x 12 V1 = V2 + 2(P2 r1 + Q 2 x 1 ) + ⎜ ⎜ ⎟ V 2 ⎝ ⎠ … (9) ( ) The voltage equation can be written as: V2 + V2 2(P2r1 + Q2 x1 ) − V1 + P2 + Q2 r1 + x1 = 0 4 2 2 2 2 2 2 [ ]( )( ) Zth= rth+jxth …(10) Vth θth VL θL VL L The expression is a quadratic equation in V22 ..Q 2 = Q1 − Qloss …………………….………………………………(5 ) ⎛ P2 2 + Q 2 2 ⎞ ⎟ Ploss = ⎜ ⎜ V 2 ⎟ * r1 2 ⎝ ⎠ V1V2sin (θ1 − θ 2 ) = P2 r1 − Q 2 x1 ………………. 11..

Sign up to vote on this title
UsefulNot useful