Physically Informed and Machine-Learning Power Prediction for a Wave Energy Converter under Bayesian Uncertainty: A Comparative Carbon-Footprint Assessment against Fossil-Fuel Generation

Abstract

Abstract
Ocean wave energy is a large, largely untapped renewable resource, but quantitative, uncertainty-aware comparisons of its climate benefit against fossil-fuel generation remain scarce. This study develops a combined physical and machine-learning (ML) modelling framework, wrapped in a Bayesian uncertainty-quantification (UQ) layer, to estimate the annual energy production (AEP) and life-cycle carbon-footprint mitigation of a reference two-body point-absorber wave energy converter (WEC), using the PacWave South (Oregon, USA) test site as a case study. A deterministic wave-to-wire physical model (linear wave power flux, a resonance-shaped capture-width response, and rated-power saturation) is cross-validated against a data-driven Gaussian Process Regression (GPR) surrogate trained on a synthetic power-performance dataset. The physical model's hydrodynamic parameters are calibrated with a four-chain random-walk Metropolis-Hastings Markov Chain Monte Carlo (MCMC) sampler (Gelman-Rubin R-hat <= 1.01 for all parameters), while the GPR supplies an independent, non-parametric posterior-predictive uncertainty. The two approaches converge on a closely comparable AEP (physical-Bayesian: 500.8 MWh yr⁻¹; ML-Bayesian: 453.5 MWh yr⁻¹), which are propagated, together with literature-based life-cycle carbon-intensity distributions for the WEC and for coal- and gas-fired generation, through a 100,000-draw Monte Carlo simulation. The device is estimated to avoid 376 t CO₂-eq yr⁻¹ (95% credible interval, CrI: 287-479) relative to an equivalent coal-fired generator (93.3% reduction) and 220 t CO₂-eq yr⁻¹ relative to natural gas (89.1% reduction), equivalent to roughly 9.4 kt CO₂ over a 25-year design life relative to coal. These results demonstrate that combining physically grounded and data-driven models within a rigorous Bayesian framework yields credible, auditable uncertainty bounds on the decarbonisation potential of wave energy, supporting evidence-based marine-energy policy and investment appraisal.