A Hybrid Framework for Short-Term Wind Power Forecasting Incorporating VMD and an Improved Sparrow Search Algorithm

Accurate short-term wind power forecasting is crucial for maintaining the stability of modern power grid dispatching systems; however, the high nonstationarity and volatility of wind power data pose challenges for traditional forecasting models. To address these issues, a hybrid forecasting framework, variational mode decomposition (VMD)–improved sparrow search algorithm (ISSA)–support vector regression (SVR), is proposed herein. First, VMD is employed to decompose raw wind power data into multiple stable mode components. Then, the ISSA is introduced to optimize the hyperparameters of SVR, thereby alleviating the tendency of conventional SVR hyperparameter tuning methods to become trapped in local optima. Independent SVR forecasting models are then established for each decomposed mode component, and the final forecasting output is obtained via signal reconstruction. Experiments conducted on real-world datasets collected from five wind turbines demonstrate that the proposed framework consistently outperforms several baseline models. Compared with the VMD–SSA–SVR model, the proposed VMD–ISSA–SVR framework reduces the average root mean square error by 8.03% (from 124.28 to 114.19 kW) and improves the average coefficient of determination to 0.9024, with a maximum value of 0.9115. These results verify the effectiveness of the proposed framework in modeling complex nonlinear wind power data and highlight the superior optimization capability of the proposed ISSA.