Assessing the efficacy of a sequential general variational mode decomposition-based combination model for United States wind power forecasting

Accurate forecasting of wind energy is essential for maintaining grid stability and ensuring efficient energy management. Hybrid modeling approaches offer enhanced predictive accuracy and reliability, supporting wind farm performance, seam- less grid integration, and informed market operations. This study presents an adaptive forecasting system that compares hybrid and standalone models using a univariate framework to preserve simplicity and reduce computational overhead. Historical U.S. wind energy data spanning 2001–2025 were utilized, with optimal lag selection based on correlation coefficients exceeding 0.899. The dataset was partitioned into training and testing subsets in a 70:30 ratio. Two hybrid frameworks were implemented: one combining Convolutional Neural Networks with Long Short-Term Memory (CNN- LSTM), and another integrating Artificial Neural Networks with Sequential Variable Mode Decomposition (SVMD-ANN). Two standalone models were implemented: Support Vector Regression (SVR) and ANN. The Combination Model Based on Sequential General Variational Mode Decomposition is important for wind power prediction because it significantly enhances the accuracy and robustness of forecasting wind power output by effectively decomposing complex, non-sta- tionary wind power time series into intrinsic mode components. This decomposition allows more accurate modeling of different wind power signal characteristics and improved handling of fluctuations and abrupt changes commonly seen in wind power data. Performance was assessed using graphical visualizations and quantitative metrics, including Root Mean Square Error (RMSE), Coefficient of Determination (R²), symmetric Mean Absolute Percentage Error (sMAPE), and Mean Absolute Percentage Error (MAPE). The SVMD-ANN model achieved the highest accuracy, with a total error of 4.16 × 10⁵ MWh, R² of 0.964, sMAPE of 10.43%, and MAPE of 20.12%. CNN-LSTM showed competitive results (RMSE: 5.75 × 10⁵ MWh, R²: 0.952, sMAPE:16.26%, MAPE: 31.34%), outperforming ANN (RMSE: 5.79 × 10⁵ MWh, R²: 0.949, sMAPE:16.67%, MAPE: 34.76%) and SVR (RMSE: 5.91 × 10⁵ MWh, R²: 0.947, sMAPE:18.43%, MAPE: 45.44%). These findings validate the robustness of the proposed SVMD-ANN model and highlight its potential for modeling monthly wind energy production.