Due to the real-time characteristics and nonlinear feature of short-term wind power prediction, a wind power prediction model based on Plant Growth algorithm and neural network (PGNN) is proposed. PGANN model combines the ability of BP neural network for solving nonlinear problem and the ability of Plant Growth algorithm (PG) for global optimization. And in this model, the PG algorithm is utilized to optimize the weights of BPNN. In order to enhance the effective of PG, we add the Metropolis criterion in it to avoid the local minima value problem. The simulation results show that based on the actual data of a wind farm, the forecasting results predicted by the improved PGNN is more precise than those by BP neural network model. Finally, we added the turbulence intensity in our prediction model and it really enhanced the predict precision. Thus, we provide an effective way to forecast short-term wind farm generation output.