Abstract: In view of problems that it can only obtain local optimal solution by using random choice method and K-means clustering method to determine center and width of nodes of hidden layer of RBF neural network, and PSO algorithm is easy to premature convergence and has bad precision for some functions, the paper proposed an improved PSO algorithm which combines with inertia weight model and convergence factor model. In view of problems of difficult positioning for single-phase grounding fault in underground distribution network, bad reliability and low precision existed in traditional fault location methods, the paper proposed a single-phase grounding fault location method of underground distribution network by using the improved PSO algorithm to optimize RBF neural network. The simulation results show that location precision of RBF neural network optimized by the improved PSO algorithm is higher than RBF neural network, and can realize accurate and reliable location for fault point.