Online Power Quality Disturbance Classification with Recurrent Neural Network

Abstract

Increasing penetration of power electronic devices is resulting in power quality disturbances that are causing marked degradation in grid performance and efficiency. Although phasor measurement data is readily available at high granularity due to the advent of advanced grid monitoring capabilities, effective processing algorithms are necessary to glean in-depth insights into the disturbances that have transpired. In this paper, a novel power quality disturbance classifier is proposed for online application by leveraging on wavelet transforms and recurrent neural networks. The existing approaches requires fixed window size, and there is a fundamental trade-off between the accuracy and localization of the event. The recurrent neural network efficiently store and memorize the past information and overcomes this limitation. The proposed technique is tested on simulation data based on IEEE-1159 standards.