First this paper briefly analyzes the electric power transient signal of single phase grounding fault, and then the fault characteristics of zero sequence current and bus voltage is obtained by wavelet transformation. Finally by using genetic algorithm to optimize the original weights of back-propagation neural network, as well as taking the fault characteristics as the input characteristic vector of optimized back propagation neural network, the network is trained and tested. The simulation result shows that the prediction effects and convergence rate of optimized back-propagation network have better performance than traditional back-propagation network. It's relatively error were less than 3%, independent of fault distance, power supply phase angle and the impact of transient resistance.