Application of artificial neural networks for characterizing the hydrodynamic performance of oscillating water column devices

Abstract

This work investigates the application of artificial neural networks to improve the characterization and assist in the optimization process of oscillating water column wave energy converters, thus increasing the reliability and accuracy of performance predictions. More specifically, this work addresses the challenges associated with resource-intensive physical modelling by combining neural networks with specifically designed experimental campaigns, thereby reducing the dependency on exhaustive experimental testing while maintaining predictive accuracy. The methodology integrates multilayer perceptron networks with Levenberg-Marquardt training, selected for their robustness and suitability for regression tasks. Key parameters, including significant wave height, energy period, and turbine-induced damping, are used as inputs to predict accurate performance in terms of capture-width ratio. Among the tested configurations, a network with a single hidden layer of 20 neurons demonstrated the best balance between accuracy and generalization, achieving a root mean square error below 2.5% and a correlation coefficient very close to unity (R > 0.99) when validated against experimental data. The selected neural network model is subsequently applied to analyse the performance of an oscillating water column device operating at a case study site, accurately predicting its capture-width ratio matrix for untested turbine-induced damping values and enabling the identification of the optimum damping through a parametric search. The findings confirm that artificial neural network based frameworks are a valuable tool that can provide information which streamline the design and help the optimization process of oscillating water column devices.