A method based on Improvement Artificial Fish Swarm Algorithm (IAFSA)-BP neural network algorithm is presented for improving the accuracy of short-term wind power forecasting. It optimizes the weights and thresholds of BPNN and improves the BPNN generalization capacity and the rate of convergence. By using the historical data of a wind farm of Shanghai in 2014, IAFSA is proposed to overcome the defects of traditional Artificial Fish Swarm Algorithm such as the blindness of searching, slow convergence speed and low searching precision at the later stage. The simulation result compared with BP neural network and AFSA-BPNN algorithm shows that the IAFSA-BPNN algorithm can not only improve the prediction accuracy and stability, but also shorten the model’s rate of convergence, and improve the precision and stability in short-term wind power forecasting. This work is supported by National Natural Science Foundation of China (No. 41675156), and Prior Discipline Construction Project of Jiangsu Universities (No. PAPD), and Jiangsu Six Major Talents Peaks Project (No. WLW-021).