A hybrid wind speed forecasting model with rolling mapping decomposition and temporal convolutional networks

Accurate wind speed (WS) forecasting is essential for the effective integration of wind energy into power grids. However, the non-stationary and highly volatile nature of WS data presents significant challenges. Decomposition-based hybrid models have proven effective in addressing these issues, but many suffer from future information leakage during data preprocessing , which limits their real-world applicability. To address this limitation, this paper proposes a novel hybrid forecasting model that eliminates future information leakage while enhancing prediction accuracy. The model incorporates three key innovations: (1) a rolling mapping neural network (MNN) for real-time decomposition, designed to minimize boundary effects; (2) an improved artemisinin optimization (IAO) algorithm to enhance the performance of the MNN; and (3) temporal convolutional networks (TCNs) for precise subseries forecasting. The rolling framework ensures that only current and past data are used, thereby preventing future data from influencing the model during training. Experiments using hourly WS data from multiple locations in China demonstrate that the proposed IAO-MNN-TCN model outperforms traditional and existing hybrid models in terms of both accuracy and robustness. Comparative studies further confirm its superiority for short-term WS forecasting. • A hybrid wind speed forecasting model integrating IAO, MNN, and TCN is proposed. • Rolling MNN-based decomposition reduces boundary effects and prevents information leakage. • The enhanced IAO algorithm improves optimization efficiency for better decomposition. • TCN captures long-term dependencies for more accurate forecasting. • The model outperforms six individual and eight hybrid models in multiple datasets.