Research on the Diagnosis of Distribution Network Fault Data Using a Fault Prediction Model

Abstract

To ensure the stable operation of the distribution network, it is necessary to timely predict and diagnose its faults. In this paper, distribution network faults were briefly analyzed. To address the problem of redundant and incomplete fault data, they were reduced by the rough set (RS), and then the reduced data were predicted based on the back-propagation neural network (BPNN) model. To further improve the prediction performance, an improved particle swarm optimization (IPSO) algorithm was designed to optimize the parameters of the BPNN model and establish the IPSO-BPNN model. Through example analysis, it was found that RS reduction retained the discriminative ability of the data and improved the prediction accuracy. The accuracy of the IPSO-BPNN algorithm was 79.36% when using non-reduced data; the accuracy of the BPNN algorithm and the IPSO-BPNN algorithm was 87.64% and 95.77%, respectively, when using the reduced data. The experimental results demonstrate the reliability of the IPSO-BPNN algorithm for fault data prediction and its applicability in the actual distribution network.