Power system overvoltage identification using feedforward neural network

The main causes of electrical accidents in power distribution networks are internal and external overvoltages. In this paper, the authors introduce an intelligent way of identifying internal and external overvoltages based on time domain studies of their waveforms. Seven parameters are extracted from the signal to build a sample vector as entry data for every sampled overvoltage to be identified. A neural network based on feedforward multilayer perceptron topology is used for the identification and classification. The overvoltage samples used in this work are obtained from ATP-EMTP software and the data extraction is performed with a script written in Matlab. A variety of power system overvoltages have been used to test the validity of the proposed method and we obtained an result of 98.82% of positive identification and classification. Results prove that, the feedforward multilayer perceptron is a suitable candidate in power system overvoltage identification and classification.