Considering the randomness and volatility of wind, a method based on B-spline neural network optimized by particle swarm (PSO-BSNN) is proposed to predict the short-term wind speed. The input space variable of BSNN can be determined by phase space reconstruction, BSNN can change the division of input space and the definition of basis function flexibly. For any input, only a few outputs of hidden layers are nonzero, the outputs are simple and the convergence speed is fast, but the traditional method is rough when it divides the input space evenly, and easy to fall into local minimum which will influence the final prediction accuracy. PSO is an intelligent search method, it has strong global search ability and is easy to be realized, which can be used to optimize the BSNN input space internal nodes and it can avoid the prediction result falling into local minimum. Simulation results show that PSO-BSNN has higher prediction accuracy than BSNN and BPNN.