Fusing spatial–temporal information into deep learning via wind propagation theory to enhance wind power prediction

Accurately predicting wind power poses significant challenges because of the inherent randomness and intermittency of wind speed, thereby impeding effective wind power scheduling. This study proposes an improved deep learning model which leverages wind propagation theory to uncover spatial–temporal relationships among wind turbines to enhance the performance of wind power prediction. In addition, comprehensive theoretical and empirical analyses are conducted to justify the effectiveness of leveraging wind propagation theory for capturing spatio-temporal relationships among wind turbines. Moreover, spatio-temporal dependencies are modeled through a dual mechanism: multi-channel independent modeling for per-turbine temporal dynamics and wind propagation-based matrix computations for inter-turbine spatial relationships, which together significantly reduce computational complexity while preserving predictive performance. Data from 134 wind turbines and six comparison models were employed to validate the robustness and effectiveness of the proposed model. Empirical results indicate that the proposed model outperforms the baseline models, achieving an average improvement of 6.19% in Root Mean Square Error and 7.05% in Mean Absolute Error.