Power Quality Disturbance Identification Method Based on Improved Fully Convolutional Network

Abstract

Power quality disturbance(PQD) identification is the premise of power system fault analysis and processing. This study proposed a PQD identification method based on an improved fully convolutional network. It combines the representation ability of fully convolutional networks for data and the memory function of long short-term memory networks for time series data. First of all, the one-dimensional voltage signal is directly used as input into the FCN-LSTM serial model. After the FCN layer is processed by convolution, batch normalization, and global average pooling, it enters the LSTM layer to extract timing features, and finally determines the type of disturbance. Numerical experiments show that the method had high recognition rates for PQD and strong anti-noise interference ability, which is superior to a single convolutional neural network and LSTM network. The engineering data verify the feasibility of this method in the field of power grid fault diagnosis and analysis.