Ultra-Short-Term Wind Power Forecasting Based on Ensemble Multiscale Lag Selection and Adaptive Multi-Model Fusion

Ultra-short-term wind power forecasting is of great significance for renewable energy grid dispatching, reserve capacity allocation, and the secure operation of power systems. However, due to the influence of multiple environmental factors, the nonlinear and non-stationary characteristics of wind power pose considerable challenges to time-series modeling. To address this issue, an ultra-short-term wind power forecasting framework based on ensemble multiscale lag selection and adaptive multi-model fusion is proposed in this study. First, phase space reconstruction (PSR) is employed to map the original wind power series into a high-dimensional state space, where multi-order lag state information is introduced to enhance the model’s capability to characterize the dynamic evolution patterns of wind power sequences. On this basis, an integrated screening strategy combining the partial autocorrelation function (PACF) and maximum relevance minimum redundancy (mRMR) is designed to comprehensively evaluate the relevance and redundancy of lagged features, thereby extracting key lagged features with high relevance and low redundancy. Subsequently, long short-term memory (LSTM), gated recurrent unit (GRU), and extreme gradient boosting (XGBoost) are selected as prediction sub-models to achieve collaborative multi-model forecasting. Finally, improved dream optimization algorithm (IDOA) is proposed to adaptively weight and fuse the outputs of individual sub-models to obtain the final forecasting results. Experimental results across multiple representative seasons demonstrate that the proposed framework achieves RMSE, MAE, NRMSE, and NMAPE values of 0.7774, 0.4586, 0.0509, and 0.0300, respectively, outperforming other comparative models and validating its effectiveness and stability in ultra-short-term wind power forecasting tasks.