Data-driven design and optimization of multi-chamber oscillating water column using CFD and machine learning

This study presents a comprehensive data-driven approach for the design and optimization of multi-chamber oscillating water column (OWC) wave energy converters by integrating high-fidelity computational fluid dynamics (CFD) simulations with machine learning (ML) techniques. The CFD model was rigorously validated against experimental data from literature results, with good agreement observed in both hydrodynamic efficiency and power output. Further, a large input data has been generated with distinct simulation cases, spanning single-, double-, and triple-chamber chamber configurations under various wave conditions with kh ranging from 2.0 s to 5.5 s, were conducted. The CFD-generated dataset was employed to train several ML models—polynomial regression, decision trees, random forest, XGBoost, support vector regression, and multilayer perceptron. XGBoost demonstrated better performance compared to the other machine learning models evaluated. Furthermore, to identify the optimal design configuration, Latin Hypercube Sampling was employed to randomly generate 1,000 distinct OWC configurations, which were then evaluated using the XGBoost model. The top ten configurations were identified, with the highest predicted power output of 36.40 W obtained from the dual-chamber OWC configuration. These findings confirm the potential of ML-driven models to significantly reduce computational cost and accelerate the design of efficient wave energy systems.