A particle swarm optimization‐based support vector machine model to detect charging pile arc faults by using the pre‐treatment generalized S transform

Abstract

Abstract An arc fault is the most common cause of charging pile fire. The series arc fault current is usually lower than the short‐circuit fault current and is challenging to detect, resulting in the inaction of the circuit breaker. Therefore, a method for detecting series arc faults of charging piles based on generalized Stockwell transform (GST) is proposed in this paper. First, an real time digital simulation system (RTDS)‐based charging pile arc experimental platform is constructed, and the point contact and the carbonization path arc fault current are measured. Then, the fault information is extracted using GST, and the 250‐Hz component is selected as the arc fault identification feature. On this basis, an improved sampling point selection method is proposed. Finally, multiple feature quantities are constructed as feature vectors, and the trained particle swarm optimization _support vector machine (PSO_SVM) is used for arc fault detection. The experimental results show that the accuracy rate reaches 98.75%; thus, the scheme proposed by this paper could detect the series arc fault current accurately and avoid the incorrect action of the circuit breaker.