Article Preview
Top1. Introduction
The static security of the power system is defined as the ability of the system, following a contingency, to reach an operating state within the specified safety and supply quality (Kirschen, 2002). The present trend towards deregulation has forced electric utilities to operate their systems under stressed operating condition closer to their security limits. Important criteria for security assessment are violation of constraints for voltages, power flows etc. The load flow study is the basic requirement for planning and operation of a power system (Wood & Wollenberg, 1996). It gives the voltage profile and line flows in a system. The fast and efficient methods available at present solve with ease load flow problem of a large system, provided required data are available. However, to study the operational reliability, i.e. to see whether the system is secured or not, the detailed load flow results are not necessary.
In recent years soft computing techniques are used to solve power system problems (Arya, Mathur, & Gupta, 2012; Bhaskar, Srinivas & Maheswarapu, 2011; Kurbatsky, Sidorov, Tomin & Spiryaev, 2014; Milani, & Mozafari, 2011; Sadeghi & Hosseini, 2013).
The application of machine learning methods for security assessment has been proposed by many researchers. For static security assessment Artificial Neural Network (ANN) (Neibur & Germond, 1991; Ray, Chakravorti, & Mukherjee, 1994; Shanti, 2008; Zhou, Davidson & Fouad, 1994; Wehenkel, 1998; Kim and Singh, 2005) have been suggested. ANN has the ability to classify patterns and its good accuracy in comparison with other machine learning methods, so Artificial Neural Network is the most popular method suggested. Disadvantages of ANN are:
ANN requires selecting of a number of tunable parameters like the number of neurons in the hidden layer, momentum constant, learning rate parameter, and number of iterations to converge a given set of data in a feature space within a specified error limit.
Recently, Support Vector Network (SVN), based on statistical learning theory, have been used in different areas of machine learning, computer vision, pattern recognition and other practical applications (Banerjee, Lahiri & Bhattacharya, 2007; Burges, 1998; Guo, Niu & Chen, 2006; Moulin, Alves da Silva & El-Sharkawi, 2004; Platt, 1998; Sansom, Downs & Saha, 2002; Turkay & Demren, 2011; Vapnik, 1999). There are several causes for the superior performance of the SVN models to the artificial neural networks (ANN) models. First the SVN model has nonlinear mapping capabilities and, thus, can more easily capture electricity load data patterns than can the ANN models. Second, the SVN model performs structural risk minimization rather than minimizing the training errors. Compared with the ANN models, minimizing the upper bound on the generalization error improves the generalization performance.
In addition, the SVN regression is to map nonlinearly the original data into a higher dimensional feature space, it will be equivalent to solving a linear constrained quadratic programming problem so that the solution of SVN is always unique and globally optimal. The over-fitting problem can be easily controlled by the choice of a suitable data separation margin.
To see whether the system is statically secured or not, the detailed load flow study is being done at present to find out all bus voltages and line flows. Further, for evaluation of contingency due to loss of generating units, outage of line or transformer etc. load flow solution is necessary. For both the above cases, usually fast load flow algorithm, such as d.c. load flow or fast decoupled load flow is employed.