Accurate short-term power load forecasting is an important guarantee for power production scheduling and safe and stable operation. It is also an important part in the construction of smart grid. In order to improve the prediction accuracy of the model, a new wavelet packet echo state network optimized by particle swarm optimization is proposed in this paper to predict the short-term power load. Firstly, the load data is decomposed and reconstructed by wavelet packet theory, and wavelet packet echo state network prediction model is established. Then, the prediction model parameters of dynamic neurons reservoir is optimized by particle swarm optimization algorithm. The results show that the forecasting accuracy, stability and generalization ability of PSO-WPESN have been significantly enhanced, compared with BP, Elman, traditional ESN, especially eases ESN disadvantages caused by excessive sick solution.