Optimization and machine learning analysis of a small-scale oscillating water column (OWC) in regular waves: A computational study

Abstract

Addressing the global challenge of energy scarcity necessitates innovative solutions like oscillating water columns (OWC), which offer significant potential in renewable energy. This study introduces a conceptual design and optimization of a small-scale OWC. A finite volume method (FVM) based wave modelling approach integrated with a volume of fluid (VOF) method is proposed to model and simulate the two-phase, viscous, time dependent, turbulent flow in a numerical wave flume (NWF) for realistic representation of wave propagation around the OWC model. Once validated against theoretical and experimental data with an error of 0.74 %, the present numerical methodology is extended to comprehensively optimize the OWC model by sampling varying geometric dimensions under different wave flow conditions using Latin Hypercube Sampling (LHS). This approach aims to not only improve efficiency but also to enhance the understanding of how these parameters affect overall performance. This is supported by machine learning analyses, such as feature importance and SHapley Additive exPlanations (SHAP), which facilitate to understand the effect of each input parameter. Key findings include the ratio of chamber height to chamber length (H₁/L) exhibiting the greatest impact on OWC efficiency, while the ratio of channel height to channel length (H₂/l) showing the least significance. Additionally, the response surface analysis reveals the optimum ranges of the parameters and highlights the necessity of multi-variable optimization utilized in this study. Optimum dimensions result in a primary efficiency of 45 %, while the least efficient is found to be 2 %, emphasizing the critical importance of optimization in increasing OWC efficiency.