Modeling, prediction, and optimization of hybrid oscillating water column-Piezoelectric wave energy converter device using machine learning models

This study focuses on optimizing the geometric parameters of a hybrid wave energy converter (WEC) that integrates an oscillating water column (OWC) device with a piezoelectric plate wave energy converter (PWEC). The hydrodynamic response and corresponding power output are computed using a constant-panel boundary element method, and a high-fidelity database is generated through Latin hypercube sampling. Five data-driven models–Random Forest, XGBoost, LightGBM, CatBoost, and a multilayer perceptron–are evaluated, with XGBoost exhibiting the highest predictive accuracy (R2=0.9959, MAPE =0.02%, RMSE =61.86). Feature ranking using Spearman correlation and SHAP values identifies l1/h, r/h, and d1/h as the dominant contributors to total generated power. Interpretable machine-learning tools, particularly accumulated local effects (ALE), are used to locate high-performance regions in the input space, which subsequently guide a response surface methodology (RSM) search. The combined ML-RSM framework yields an optimal configuration of l1/h=0.21, d1/h=−0.08, gp/h=0.06, d2/h=−0.19, and r/h=0.63, resulting in a maximum power output of 7912Wm−1. This represents an improvement of approximately 23% over the maximum value present in the training dataset. This study represents a significant advancement in optimizing the total power output of hybrid WEC systems. The results demonstrate that machine-learning-assisted optimization offers an efficient and accurate pathway for enhancing overall device performance.