Power data are easily affected by climate, seasons, holidays and other factors, with different fluctuation characteristics. Aiming at the problems of low prediction accuracy of power data with different characteristics and weak generalization ability of prediction methods, a power data prediction method based on adaptive hybrid optimization is proposed. By using wavelet transform and stability analysis, the power data is adaptively decomposed into non-stationary series and multiple stationary series containing trend, season and periodic information. The state transition algorithm is used to optimize the long and short term memory deep learning network and autoregressive moving average model, respectively, to fit and predict the non-stationary and stationary series. Finally, the predicted sequences are reconstructed to obtain the final prediction results. Compared with other methods, the proposed method not only has higher prediction accuracy, but also has strong generalization ability.