Interpretable wind power forecasting with residual learning-based model

Accurate wind power forecasting is crucial for ensuring grid stability and enhancing energy output in renewable energy systems. Existing studies have prioritized maximizing accuracy through data preprocessing and model optimization, often overlooking the significant aspect of interpretability. This study proposes a novel ensemble approach to wind power forecasting that combines the strengths of two robust machine learning algorithms to enhance predictive accuracy while providing transparent and explainable results. The proposed model sequentially integrates CatBoost for initial predictions and XGBoost for modeling residuals. The proposed ensemble's sequential architecture is effective in capturing complex non-linear relationships and thereby addressing model biases. Additionally, integrating explainable AI methods ensures the interpretability of the factors affecting forecasts, thereby confirming the model's transparency and reliability. This clarity enriches the understanding of the model's decision-making process, thereby validating the results and enhancing their applicability for implementation in renewable energy systems. The dataset used in this study integrates several meteorological, turbine, and rotor parameters, and the model's performance is assessed using standard evaluation metrics, such as MSE, MAE, R² score, and MAPE. The results reveal that the ensemble technique outperforms individual models, emphasizing its potential to enhance accuracy in wind power prediction.