Power quality disturbances classification based on S-transform and probabilistic neural network

Abstract

Classifying power quality (PQ) disturbances is one of the most important issues for power quality control. A novel high-performance classification system based on the S-transform and a probabilistic neural network (PNN) is proposed. The original power quality signals are analysed by the S-transform and processed into a complex matrix named the S-matrix. Eighteen types of time–frequency features are extracted from the S-matrix. Then, after comparing the classification abilities of different feature combinations, a selected subset with 2 features is used as the input vector of the PNN. Finally, the PNN is trained and tested with simulated samples. By reducing the number of features in the PNN's input vector, the new classification system reduces the time required for learning and the computational costs associated with the features and the PNN's memory space. The simulation results show that 8 types of PQ disturbance signals with 2 types of complex disturbances were classified precisely and that the new PNN-based approach more accurately classified PQ disturbances compared to back propagation neural network (BPNN) and radial basis function neural network (RBFNN) approaches.