The inherent intermittency and uncertainty of wind power generation pose significant challenges to grid security and the integration of renewable energy. Accurate and reliable short-term wind power forecasting is crucial for enhancing wind energy usage and ensuring the safe operation of power systems. Current mainstream forecasting methods inadequately model spatial correlations among regional wind farms. Additionally, wind power generation is susceptible to sudden changes in weather conditions and environmental factors, limiting the robustness of existing forecasting methods when confronting dynamically changing prediction environments. This poses major challenges for accurate and reliable regional wind power forecasting. This paper employs Graph Convolutional Networks (GCN) to model spatial connections between wind farms while introducing a combined TCN-Transformer model for temporal feature extraction and dependency modeling. Furthermore, to enhance prediction accuracy and reliability, Deep Q-Network (DQN) is incorporated to dynamically correct model prediction errors. Experimental results demonstrate that the proposed short-term wind power forecasting method achieves an RMSE of 60.14 and an MAE of 45.98, showing significant improvement over predictions from models without DQN error correction and other comparative models. Future work may extend the forecasting horizon to provide more information support for grid supply security decisions.
Xu et al. (Thu,) studied this question.