Physics-guided multiscale fusion network for ultra-short-term wind power forecasting

Artikel i vetenskaplig tidskrift, 2026

With the increasing penetration of wind power in modern power systems, ultra-short-term wind power forecasting plays an important role in maintaining grid security and operational stability. However, wind power sequences are characterized by strong nonlinearity and nonstationarity, and are further affected by turbine wake coupling, making it difficult for purely data-driven methods to simultaneously achieve high forecasting accuracy and physical consistency. To address this issue, this paper proposes a Physics-Guided Multiscale Fusion Network (PGMFNet) for ultra-short-term wind power forecasting. Specifically, the proposed model employs a State-Space Model (SSM) to capture long-term dependencies and periodic evolutionary patterns in wind power sequences, thereby enhancing the modeling capability for global temporal dynamics. In addition, a Multi-Scale Fourier Transform Network (MSFNet) is constructed to improve the representation of local dynamics, short-term variation patterns, and abrupt fluctuations through frequency-domain encoding under multiple temporal resolutions. A Channel-Spatial Attention Module (CBAM) is further introduced to adaptively emphasize critical variables and informative time steps. Finally, a physics-constrained loss derived from Jensen's wake theory is incorporated to enforce consistency between the forecasting results and turbine aerodynamic behavior. Experimental results on real-world inland and offshore wind farm datasets in the United States show that the proposed model achieves the best overall performance in terms of the coefficient of determination (R2), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute scaled error (MASE) among the compared models. The results further indicate that the proposed framework can improve forecasting robustness and encourage wake-consistent prediction behavior under different wind farm conditions. Overall, PGMFNet provides an effective framework for accurate and physically consistent real-time wind power forecasting, offering practical support for intelligent wind farm operation and power grid dispatch.