It is difficult for traditional wind power forecasting methods to meet the requirements of refined and dynamic modeling. It is also easy for them to fall into local optimality. Thus an intelligent wind power forecasting method based on cluster analysis and a Hybrid Adaptive Evolution Algorithm (KHEA) is proposed. First, the K-means clustering algorithm is used to cluster the annual wind speed and power data to eliminate unreasonable data. Then, Wavelet Transform (WT) is used to identify the behavioral characteristics of the power data, and the set of deconstructed sequences is obtained. Then a BP neural network model is established to predict the power deconstructed sequence in a future time period. In order to reduce the prediction error, the Evolutionary Particle Swarm Optimization (EPSO) is used to adjust and optimize the weights and thresholds of the model, so as to realize the function complementation of EPSO evolution characteristics and neural network self-learning capabilities. Finally, the inverse wavelet transform is used to reconstruct the prediction sequence to obtain the final power prediction value. Using the data of a wind farm in southern China to carry out simulation experiments, and comparing with other models, it shows that KHEA has higher wind power short-term prediction accuracy and reliability, and provides new technical solutions for improving wind power prediction accuracy and optimizing dispatch management. This work is National Natural Science Foundation of China (No. 61503348).