The XGBoost wind speed prediction model based on VMD-LSTM error correction

Wind speed prediction is crucial for the development and utilization of renewable energy, particularly for the efficient use of wind power, which depends on accurate forecasting. Wind speed is influenced by multiple factors, making it essential to model its complex temporal dependencies. This study proposes an enhanced deterministic wind speed prediction model for a short-term forecasting horizon of 10 h ahead. The model first employs the Whale Optimization Algorithm (WOA) to optimize the hyperparameters of an XGBoost model, improving its initial predictive performance. To further enhance accuracy, Variational Mode Decomposition (VMD) is applied to decompose the prediction residuals, extracting key temporal components. Long Short-Term Memory (LSTM) networks are then used to model and correct these residual components. The corrected values are added back to the initial WOA-XGBoost predictions, forming the final deterministic output. Experimental results demonstrate that the proposed model reduces the mean absolute error (MAE) from 0.2935 m/s to 0.1523 m/s, corresponding to a reduction of 48.1%, and the root mean square error (RMSE) from 0.4215 m/s to 0.2049 m/s, representing a reduction of 51.4%. The R 2 score improves from 0.8242 to 0.9584, reflecting a 16.3% enhancement. The key advantage of this hybrid model lies in its ability to effectively capture and correct multi-scale residual errors through the combination of WOA-optimized XGBoost and VMD-LSTM, resulting in significantly improved prediction accuracy, generalization capability, and robustness compared to traditional single models. This provides an effective approach for handling complex time series in wind speed forecasting. • Proposes a solution to the challenge of modeling complex temporal dependencies in wind speed forecasting. • Employs WOA to automatically optimize XGBoost hyperparameters. • Conducts feature importance analysis to identify key influencing factors of wind speed and improve training efficiency. • Applies VMD to prediction residuals to extract key temporal features. • Utilizes LSTM to model and correct residual components, enhancing prediction accuracy.