In today’s power system, the installed capacity of hydro-generators is increasing, resulting in an increase in its loss, which in turn leads to an increase in the temperature of hydro-generators and affects the performance of generators. In the temperature fault of hydro-generator, the stator part accounts for 38%, so the research on the stator is also increasingly important. Based on the hydro-generator of Zhanghewan Pumped Storage Power Station, this paper analyzes the electromagnetic field and stator loss of the hydro-generator under working conditions from the electromagnetic field theory, and then establishes its three-dimensional finite element model. The temperature field distribution of the stator of the hydro-generator is calculated by magneto-thermal coupling. Then the artificial fish swarm algorithm and BP neural network algorithm are combined to construct the temperature prediction model of stator winding and stator top, and the simulation results are compared with the measured data after monitoring and transformation. The results show that the optimization of BP neural network by artificial fish swarm algorithm(AFSA) improves the accuracy of the stator temperature prediction model, It provides a way of thinking for motor stator temperature fault analysis and monitoring determination.