Contribution to automatic detection and diagnosis of wide variety range of power quality disturbances using combined wavelet transform and neural network methods

A new approach for the detection and classification of a wide range (15 types) of power quality violations, based on the IEEE 1159 standard, is presented. It involves a broad range of disturbances, from low frequency dc offsets to high frequency transients or low duration impulse to steady state events. Wavelet multiresolution signal analysis is used to denoise, and then decompose, the signal of a power quality event to extract its useful information. An optimal vector (with 8 elements) of computed features is then selected and adopted in training a neural network classifier. This vector, which consists of a statistical parameter of frequency related details and approximation wavelet coefficients, represents a distinctive property of the studied power quality events. For the neural network structure, a multilayer perceptron (MLP) and a radial basis function (RBF) are used and compared. The proposed classifier can significantly improve the efficiency of the automatic diagnosis of power quality disturbances. Simulation results with low error rate confirm the capability of the proposed method.