Interpretable Short-Term Wind Speed Prediction Based on LSTM and SHAP: A Case Study in Northwest China

Accurate wind speed forecasting is crucial for renewable energy development and wind farm operations. With the advancement of artificial intelligence, deep learning (DL) models based on Long Short-Term Memory (LSTM) networks have achieved significant progress in wind speed forecasting. However, ensuring the interpretability of why models make specific predictions is as important as prediction accuracy, as it guarantees the credibility and transparency of the forecasting process. To address the interpretability issues inherent in the "black-box" nature of DL models, this study constructs a short-term wind speed forecasting model based on LSTM, applied to four meteorological stations in Northwest China, and employs the SHapley Additive exPlanations (SHAP) method to quantitatively evaluate the contribution of each meteorological feature to prediction results. Through the analysis of summary plots and force plots, we provide a comprehensive explanation of the prediction process from both global and local perspectives, helping wind farm managers make more informed decisions based on key meteorological features and offering data-driven support for decision-making in response to extreme weather conditions. This study innovatively employs the SHAP method to interpret the LSTM-based wind speed forecasting model, filling a gap left by existing studies that mainly focus on improving prediction accuracy but overlook interpretability.