The influence degree of wind power large-scale grid-connected on power system planning and operation is deepening day by day. Aiming at the randomness and uncertainty characteristics of the output fluctuation of wind turbines, in order to improve prediction accuracy of short-term wind power, a short-term wind power prediction method combining bat optimization algorithm (AMBA) and BP Neural network based on the adaptive variation of population fitness variance is proposed. According to variance of population fitness and the size of the current optimal solution, the model determines the mutation probability of the current optimal individual and the T-distribution variation of the global optimal individual, and two optimization of the mutated bat individuals. Then network parameters of BP neural network are optimized by AMBA, and then prediction accuracy of BP neural network is improved. By analyzing the example, prediction effect of AMBA-BP model is compared with other model prediction results. The results show that the model can effectively improve prediction accuracy of short-term wind power.