Multifactor interpretability method for offshore wind power output prediction based on TPE-CatBoost-SHAP

Abstract

Accurate forecasting of offshore wind power output is crucial to the operation and scheduling of wind farms. However, predicting offshore wind power output is challenging because numerous factors influence wind energy. Current research focuses on achieving precise predictions, but the interpretability of models remains poor. In this context, this paper proposes a method that integrates Tree-structured Parzen estimator (TPE), categorical boosting (CatBoost), and SHapley Additive exPlanations (SHAP) to predict and interpret the offshore wind power output. PCA (principal component analysis) is used for meteorological feature preprocessing. The TPE algorithm iteratively searches for the optimal hyperparameters for the CatBoost prediction model. SHAP values are used to analyze the importance of various factors. By examining interactions, we identify the meteorological conditions conducive to efficient wind farm operation. The results indicate that the TPE-CatBoost model outperforms various other models in terms of evaluation metrics. SHAP explains the dependency between features and the seasonal variations in wind farms.