Wavelet feature-based modular neural network for detection and classification of power quality disturbances

Abstract

Listen

Translate article

Disturbances such as voltage sag, swell, interruption and harmonics are very typical in a power system. Power quality monitoring should be capable of identifying these disturbances to initiate mitigation action and protect sensitive loads. This paper presents wavelet-neural network-based detection and classification of power quality disturbances. Wavelet transform has the ability to analyse signals simultaneously in both time and frequency domains and is used to extract features of the disturbances by decomposing the signal using multi resolution analysis. These features, used to detect and localise the disturbances and are not easily separable, will reduce the performance of multilayer neural network. Improvement in the classification accuracy is suggested by employing modular neural network obtained by dividing a complex task into easier subtasks. The algorithm proposed is tested for classification of various power quality disturbances and it is found that a modular neural network has a higher classification accuracy over traditional multilayer neural network.