DISCRIMINANT ANALYSIS IN SECURITY ASSESSMENT OF POWER SYSTEMS BY STATISTICAL PATTERN RECOGNITION

Abstract The method of pattern recognition for analyzing power system transient stability is particularly attractive. It gives the opportunity to assess the transient stability of power systems more directly and effectively than the conventional approach based on simulation. The success of the method depends on the choice of a pattern vector which contains two types of information: - information which permits to determine whether a state is stable or unstable, - redundant information which must be eliminated to reduce on line computational time. In this paper, we present the discriminant analysis methods used for extracting discriminant information contained in two pattern vectors. The first pattern vector consists of transient energy gained during the fault and a variable correlated to the critical energy. Its dimension is not governed by the number of generators and the discriminant information is extracted using a method based on within-class and between-class scatter matrices. The second pattern vector is derived from the first by decomposition of the transient energy into individual energy machines. Because the previous method is not suitable for this case, the redundant information is eliminated by projecting pattern space onto a set of discriminant vectors proposed by Foley and Sammon which are optimal for the BAHADUR criterion. The results are applied to a six machine system for a three phase fault.