Rapid assessment of power system security using pattern recognition concepts.

摘要

This research investigates the feasibility of on-line assessment of power system security using pattern recognition techniques. Three feature extraction techniques, the minimum entropy method, the Karhunen-Loeve expansion method and the differential entropy method, are studied to examine intraset clustering and interset class dispersion. A "branch-and-bound" method for feature selection is applied to provide an optimal set of features required for pattern classification. A pattern classifier system is developed to illustrate the feasibility of classifying any given operating condition into either the secure or insecure class. A performance index algorithm is developed and used as a weighting factor to enhance the separability between the secure and insecure classes. A biasing technique to reduce the possibility of misclassifications is discussed.; The proposed algorithm is applied to three power systems, the IEEE 14-bus, 30-bus, and 118-bus test systems. Classifier performance results indicate that transformation of pattern vectors to the feature space leads to substantial improvement in class separability. For small power systems, where direct application of conventional pattern recognition methods results in suboptimal solutions, the branch-and-bound method is proven to be an effective method for optimal feature selection, yielding, at times, better results than the full feature set. As the size of the power system increases, classifier performance improves significantly, yielding 100 percent classification for the 118-bus system. Better separability between the secure and insecure classes is obtained when a performance index is used to form weighted mean vectors and weighted covariance matrices of the secure and insecure sets. The study indicates that using statistical pattern recognition techniques to assess the security of power systems is feasible.