Machine learning for predicting flow and pressure in an oscillating water column wave energy converter

The use of wave energy presents itself as a renewable alternative, diversifying the energy matrix and reducing the consumption of fossil fuels. One of the prominent devices for ocean wave harvesting is the Oscillating Water Column (OWC), which consists of a hydropneumatic chamber where the oscillation of the water surface compresses the confined air, driving a turbine. This work combines numerical simulation and machine learning algorithms, aiming to develop predictive models for mass flow rate and pressure through the OWC chamber. The database for the models was created using Design of Experiments (DOE) techniques, where different combinations of wave height and wavelength were simulated via Computational Fluid Dynamics (CFD) using the Finite Volume Method (FVM). Subsequently, the wave elevation height data is fed into a fully connected deep neural network, which uses the simulated data for learning and returns predictions based on the provided information. The model's performance is evaluated in terms of a cost function (Mean Squared Error). The employed model allows for the identification of flow rate and pressure peaks for the wave period provided at the input, and the results show differences of up to 5% in RMS (Root Mean Square) between the predictions and the CFD data. The main contribution of this work lies in the use of machine learning in predicting traditional operating conditions of an OWC device, introducing a concept that can be applied in the early stages of the design of such devices.