Short-Term Wind Power Interval Forecasting via VMD Decomposition and MKDE-Based Error Modeling

To address the limitations of point forecasts in wind power uncertainty quantification, this paper proposes a shortterm interval prediction method combining Variational Mode Decomposition (VMD), Grey Wolf Optimizer-Random Forest (GWO-RF), and Mixed Kernel Density Estimation (MKDE). The original power series is decomposed via VMD to reduce nonstationarity effects. Modal components and residuals are then fused for feature selection, and GWO-RF generates initial point predictions. Prediction errors are modeled using MKDE with Gaussian and Epanechnikov kernels to adaptively estimate error distributions and construct quantile intervals. Experimental results show that at a 95 % confidence level, the proposed method achieves a Prediction Interval Coverage Probability (PICP) of 97.9% and a Normalized Average Width (PINAW) of 0.135. This approach effectively balances coverage and precision, offering robust uncertainty quantification for grid scheduling.