A Short-Term Wind Power Forecasting Method Based on Multi-Source Numerical Weather Prediction Fusion

Short-term wind power forecasting plays a crucial role in day-ahead grid scheduling and electricity market operations. Traditional forecasting methods typically rely on numerical weather prediction (NWP) data from a single source and utilize deep learning models to establish a mapping between meteorological inputs and wind farm power generation. However, single-source NWP data often exhibit high uncertainty, particularly under sudden weather changes or complex meteorological conditions, leading to unstable performance and limited forecasting accuracy. To address these challenges, this study integrates multi-source and multi-location NWP data into the wind power forecasting model. Furthermore, to dynamically identify the optimal NWP sources, we propose a short-term wind power forecasting model that combines the AutoCorrelation mechanism with the ProbAttention mechanism. This model adaptively captures the time-varying relationships among multiple NWP sources, enhancing forecasting robustness. Experimental results from a wind farm in western Inner Mongolia demonstrate that the proposed approach significantly improves forecasting accuracy compared to baseline methods.