Flashover voltage and time prediction of polluted silicone rubber insulator based on artificial neural networks

Abstract

Insulator flashover prediction is an important task that should be done before any hazards. In this paper, leakage current (LC) analysis using neural networks is utilized to predict the flashover voltage (FOV) and flashover time (FOT). Experiments are performed on silicone rubber (SiR) insulators in the salt fog test chamber under different conditions and levels of contamination for LC sampling. To predict flashover, sampled LC at different contamination levels are first clustered by the self-organizing map (SOM) artificial neural network (ANN). Clustering results and other factors are employed to level the situation of LC periods. Afterward, these levels are fed sequentially to another ANN to predict the FOV and FOT. Various sample data are tested and compared with Back Propagation (BP) and Multilayer Perceptron (MLP) neural networks to evaluate the proposed neural network and algorithm. The results confirm the acceptable performance of the proposed neural networks and their ability as an online monitoring system to raise alarms before potential flashover hazards.