In the small current grounding system, the most is single-phase grounding fault. how to quickly and accurately find the fault line has been a key research topic, and this didn"t get effective solution. This paper presents a fault line selection method of power distribution network based on improved genetic algorithm (GA) to optimize T-S fuzzy neural network. By improving the fitness function of GA, initial parameters and weights are optimized firstly, using the gradient descent method to optimize the second time. the influence of T-S fuzzy neural network, the traditional GA optimization of T-S fuzzy neural network and different network structures to network performance are discussed. The results of the study illustrate the improved GA to optimize T-S fuzzy neural network is better than T-S fuzzy neural network and traditional GA to optimize T-S fuzzy neural network in the term of line selection effect, can accurately, effectively, reliablely find fault line.