Abstract: In allusion to shortcomings of particle swarm optimization (PSO) algorithm being likely to run into partial optimization and generalization of Elman neural network being insufficient, this paper presents a kind of method for short-term wind power forecasting based on wavelet packet decomposition (WPD) and catalytic particle swarm optimization (CPSO) algorithm for optimizing Elman neural network (ENN). This method uses wavelet transform to decompose wind power samples into multi-level sequences, adopts CPSO-ENN optimized by CPSO algorithm for forecasting the sub-sequence of wind power obtained from reconstruction, and finally overlays forecasting value of each sub-sequence to obtain actual forecasting results. Simulating verification for actual operational data of one wind power field indicates that this new model has higher forecasting precision.