Voltage sag analysis on three phase systems using wavelet transform and probabilistic neural network

The ever growing diffusion of susceptible loads in power system result in increasing susceptibility to power quality problems. For this reason power quality data are recorded by utilities with the aim of identifying the origins of these disturbances in order to improve the reliability of the electrical system and thus to reduce their impact on customers. In this paper a classification algorithm for automatic analysis of recorded power quality event data is presented. The approach is based on wavelet analysis and a probabilistic neural network classificator. The discrete wavelet transform (DWT) is used to detect fast changes in the voltage signals. It is possible to detect carefully the starting and the recovery times of the disturbance and other features, such as the start slew rate and intermediate rate voltage and the number of phases involved for each fault event. These features are the input patterns of the second step of the classification algorithm based on a probabilistic neural network (PNN) in order to classify voltage sag events into four classes (network faults, motor starting, motor re-acceleration after fault, transformer energising).