Dynamic security assessment of a power system based on Probabilistic Neural Networks

Abstract

In this paper, a method of utilizing Probabilistic Neural Networks (PNNs) in the dynamic security assessment of power systems is proposed. The method involves an approach of a proper training data selection for a PNN which classifies the operating conditions of a power system with high accuracy. The classification is based on the power system security against critical contingencies that may cause transient instabilities. By the proposed method, high classification performances are attained without requiring large training sets. This work also includes an application of multi-spread PNN structures which provide more flexibility in enhancing the security assessment performance. A simple genetic algorithm (GA) is applied to calculate proper spread parameters of multi-spread PNN structure. The proposed methods are implemented on the Iowa power system model and the results regarding dynamic security assessment performances are discussed.