Insulator leakage current prediction using surface spark discharge data and particle swarm optimization based neural network

Abstract The paper has completely built an online monitoring system to detect deteriorative insulator by predicting the leakage current which assist maintenance personnel in determining whether insulators need to be cleaned or replaced and improve maintenance efficiency in Taiwan. The real-time monitoring systems are installed on the 69 kV and 161 kV transmission towers. The local meteorological data such as temperature, humidity and dew point, and images of spark discharge phenomenon are collected to identify the correlation with the leakage current of deteriorative insulators. The forecasting model is enhanced by using the image features of surface spark discharge phenomenon (SSDP). The percentage of spark area and the change in the brightness of region of interest (ROI) can be utilized as exogenous inputs to the predicting model. Correspondently, a particle swarm optimization based back propagation neural network (PSO-BPNN) is applied to improve the accuracy in prediction. The proposed developing model combined enhancement inputs are compared with persistent models, such as back propagation neural network (BPNN), radial basis function neural network with K-means cluster (RBF-K-means) and support vector regression (SVR), to evaluate the performance based on common statistic metrics. The simulation experiments have proved the integrated surface spark discharge features combined PSO-BPNN to improve the accuracy and stability of predicting leakage current comparing with other models. Therefore, the surface spark discharge data have strong correlation with the leakage current. In addition, the PSO-BPNN also achieves higher accuracy, better effectiveness and faster convergence comparing with other persistent models.