Abstract
Wave energy offers immense potential as a renewable energy source. However, accurately estimating the Total Absorbed Power (TAP) at various sites remains a significant challenge, requiring resource-intensive physical modelling and numerical simulations to capture the complex hydrodynamic behaviour of Wave Energy Converters (WECs) across different designs and wave conditions. To address this, we propose a novel, computationally efficient Machine Learning-Transfer Function (ML-TF) approach to estimate the TAP of Multi-Body Floating WECs (MBFWEC). The methodology integrates frequency-domain and time-domain analyses to generate a sparse dataset of MBFWEC responses under regular waves, which is used to train Machine Learning (ML) models. Wave height, wave period, and Power Take-Off (PTO) damping are the key inputs for predicting the Capture Width Ratio (CWR). Among the models tested, Multi-Layer Perceptron (MLP) model performed best (R2 = 0.995). This model was then used to derive a high-resolution CWR dataset, with error margins within ±6.11 %, proving its reliability for out-of-range CWR predictions. To extend the model's applicability to irregular wave conditions, a Transfer Function (TF) was developed from the CWR dataset across a desired frequency range. The TAP was subsequently estimated based on the TF, site-specific wave power spectra, and the converter's effective length. Validation using time-history simulations in uni-modal and bi-modal sea states showed excellent accuracy (4 % maximum error), while achieving an 80 % reduction in computational cost. The methodology was further applied in a real-world case study using wave data from three locations in the northern Oman Sea, to evaluate the region's year-round power potential.