Improvement of estimation of surge arrester parameters by using Modified Particle Swarm Optimization M. Nafar a, * , G.B. Gharehpetian b , T. Niknam c a Department of Electrical Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran b Department of Electrical Engineering, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran c Department of Electrical and Electronics Engineering, Shiraz University of Technology, Shiraz, P.O. 71555-313, Iran article info Article history: Received 3 December 2010 Received in revised form 12 May 2011 Accepted 15 May 2011 Keywords: Surge arrester Models Particle Swarm Optimization (PSO) Ant Colony Optimization (ACO) Parameter Estimation EMTP abstract Metal Oxide Surge Arrester (MOSA) accurate modeling and its parameter identification are very important aspects for arrester allocation, system reliability determination and insulation coordination studies. In this paper, Modified Particle Swarm Optimization (MPSO) algorithm is used to estimate the parameters of surge arrester models. The convergence to the local optima is often a drawback of the Particle Swarm Optimization (PSO). To overcome this demerit and improve the global search capability, Ant Colony Optimization (ACO) algorithm is combined with PSO algorithm in the proposed algorithm. The suggested algorithm selects optimum parameters for the arrester model by minimizing the error among simulated peak residual voltage values given by the manufacturer. The proposed algorithm is applied to a 120 kV MOSA. The validity and the accuracy of estimated parameters are assessed by comparing the predicted residual voltage with experimental results. Ó 2011 Elsevier Ltd. All rights reserved. 1. Introduction Lightning and switching overvoltages in power systems are very common causes of interruptions [10e12]. The Metal Oxide Surge Arresters (MOSAs) are extensively used as protective devices against lightning and switching over voltages. Proper volta- geecurrent characteristics, ignorable power losses, high level reli- ability in the operation time, high speed response to overvoltages and long life time are some advantages of MOSAs. Accurate modeling and simulation of their dynamic characteristics are very important for arrester allocation, system reliability assessment and insulation coordination studies [1e 16]. For switching studies, MOSAs can be modeled by their nonlinear VeI characteristics [1,2]. However, such a presentation would not be appropriate for fast front transients and lightning surge studies; since dynamic char- acteristics obtained from MOSA show that voltage across the surge arrester increases as the time-to-crest of the arrester current decreases and the voltage of the arrester reaches a peak before arrester current reaches its maximum [1]. Typically, the residual voltage of an impulse current having a front time of 1 ms is 8e12% higher than that of predicted for an impulse current having a front time of 8ms. The residual voltage of longer time-to-crests between 45 and 60ms, is 2e4% lower than that of a 8ms current impulse [5e7]. This dynamic behavior requires a more sophisticated model for fast front waves. In order to reproduce the MOSA dynamic characteristics, several studies have been focused on the modeling and simulation of MOSAs [1e 10]. IEEE, Pinceti and Fernandez models are the main models of surge arresters that have been presented to simulate the dynamic behavior of surge arresters. These models have different parameter estimation procedures. It is very often difficult to identify the dynamic model parameters from the available data [1e 14]. In recent years, different procedures have been presented for estimating the parameters of all models [6e10]. In [7], a numerical method has been proposed for estimating the parameters of three mentioned models. New procedures, using heuristic algorithms, have been proposed to determine the best parameters of MOSAs models in Ref.[8 and 9]. These methods are based on the comparison of simulation results of residual voltages and the results of 8=20ms experimental measurements. Proposed methods of Ref. [7e9] are general and can be applied to all models. But, they need measurement results obtained under 8=20ms impulse test, which are not always available and reported in datasheets. To overcome this problem, a procedure has been pre- sented in [10], which matches the peak of the discharge voltage obtained from 8=20ms impulse current. In this paper, it is shown that the comparison between the simulated peak residual voltage * Corresponding author. Tel.: þ98 917 3018583; fax: þ98 728 3311172. E-mail address: m.nafar@srbiau.ac.ir (M. Nafar). Contents lists available at ScienceDirect Energy journal homepage: www.elsevier.com/locate/energy 0360-5442/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.energy.2011.05.021 Energy 36 (2011) 4848e4854