Method for Locating Fault Points of Medium-Voltage Cables Based on Improved Wavelet Neural Network

Abstract

In view of the slow convergence and easy fall into local optimum of traditional WNN, this paper introduces wavelet basis adaptive selection and GA-PSO hybrid strategy to jointly tune the scaling factor, translation factor, weight, threshold and number of hidden nodes. First, the optimal wavelet basis is automatically selected by mutual information. Then, through GA global search and PSO local fine tuning, an improved WNN model is constructed and applied to medium voltage cable fault location. Simulation results show that under the condition of 100km cable, the improved WNN converges in 70 iterations, and the MAE and RMSE are reduced by 1.95km and 2.25km respectively compared with the traditional WNN, which significantly improves the positioning accuracy and convergence speed.