Combined genetic algorithms and neural-network approach for power-system transient stability evaluation

As the electric power system grows in size and complexity with a large number of interconnections, the assessment of the transient stability of power systems became an extremely intricate and highly non-linear problem. Its solution needs either numerical methods involving bulk computations or specific dedicated methods to analyse dynamic non-linear systems. Either method mostly assesses, particularly in the post-fault condition, the critical clearing time (CCT). This parameter constitutes very complex functional relationships between the pre-fault condition, type, and location of fault beside the clearance sequence. The available methods for evaluating such parameter had been previously reviewed. New approaches using the locally-tuned radial basis function (RBF) network, an artificial neural network (ANN) paradigm have been recently proposed. The goal of this paper is to develop methods that can combine both neural networks and genetic algorithms (GA) into a common framework, and apply them to prediction problems. In the paper, the application of genetic algorithms in selecting the input patterns for the RBF network is proposed. Description of this combined approach and the results of its application to two power systems, one for four-machine six-bus system and the other for an existing system of North Sumatra, Indonesia, are also given in the paper. The attainable results show that the performance of the RBF network can be maintained and improved in spite of less features in the input patterns.