Bin Yuan

Accurate wind power forecasting is essential for the stable and economic operation of power systems with high renewable penetration. Although machine learning models have been widely adopted for this task, the assumption that greater model complexity invariably yields superior forecasting accuracy has received insufficient scrutiny. This paper presents a systematic experimental investigation that covers two complementary stages, i.e., wind speed correction and wind power forecasting. For wind speed correction, we compare 10 machine learning methods, including spanning linear, instance-based, and tree-based ensemble learners, under four newly proposed progressively enriched feature configurations. For wind power forecasting, we benchmark 20 methods spanning traditional machine learning, time-series deep learning, and Transformer-based architectures on two geographically distinct wind farms. Our results reveal a clear task-dependent pattern. In wind speed correction, tree-based ensemble methods, particularly gradient boosting variants, consistently dominate, and feature engineering contributes more to accuracy gains than model selection. In wind power forecasting, deep learning architectures substantially and consistently outperform traditional methods, with attention-based models generalizing the most robustly across regimes and recurrent networks proving to be the most sensitive to regime shifts. These findings provide actionable task-specific guidance for model selection in operational wind power forecasting systems.

This study conducts a comprehensive evaluation of four scenario experiments using the CMA_WSP, WRF, and WRF_FITCH models to enhance forecasts of hub-height wind speeds at multiple wind farms in Northern China, particularly under significant wind speed fluctuations during high wind conditions. The experiments apply various wind speed calculation methods, including the Monin–Obukhov similarity theory (ST) and wind farm parameterization (WFP), within a 9 km resolution framework. Data from four geographically distinct stations were analyzed to assess their forecast accuracy over a 72 h period, focusing on the transitional wind events characterized by substantial fluctuations. The CMA_WSP model with the ST method (CMOST) achieved the highest scores across the evaluation metrics. Meanwhile, the WRF_FITCH model with the WFP method (FETA) demonstrated superior performance to the other WRF models, achieving the lowest RMSE and a greater stability. Nevertheless, all models encountered difficulties in predicting the exact timing of extreme wind events. This study also explores the effects of these methods on the wind power density (WPD) distribution, emphasizing the boundary layer’s influence at the hub-heighthub-height of 85 m. This influence leads to significant variations in the central and coastal regions. In contrast to other methods that account for the comprehensive effects of the entire boundary layer, the ST method primarily relies on the near-surface 10 m wind speed to calculate the hub-height wind speed. These findings provide important insights for enhancing wind speed and WPD forecasts under transitional weather conditions.