Abstract: Considering the traditional BP neural network prediction is sensitive to the initial weight and is easy to fall into the local optimal solution, an ultra short-term wind power output forecasting is proposed based on the particle swarm algorithm and the simulated annealing algorithm(SA-PSO) which is used to jump out of the local optimal trap and find the global optimal network parameter. The particle swarm algorithm is used to update the particles to the optimal solution during the neural network parameter training process. At the meantime, a variety of environmental factors are considered in order to improve the forecasting accuracy The BP, PSO-BP, and SA-PSO-BP neural network algorithms are used to predict the wind power output in a wind farm in Bayannur. The results show that the SA-PSO-BP algorithm has the smallest average prediction error. It verifies that the improved SA-PSO-BP neural network algorithm has higher prediction accuracy.