Fault Detection and Identification Method of Power System Based on Deep Residual Network with Spectral Normalization

Abstract

The intelligent technology of power system gradually becomes mature. Fault detection and identification, as an important part of intelligent technology, has been widely concerned. In this paper, a fault detection and identification method based on deep residual network with spectral normalization is proposed. Firstly, this paper studies how to conFigure phasor measurement units in power system in the most efficient way. Then, the matching pursuit decomposition algorithm based on Gauss atom library is used to extract features from synchronous frequency signals and voltage signals collected by PMU. Finally, the extracted features are classified by the deep residual network with spectral normalization. The simulation results show that our method is efficient, effective and robust at the same time.