Effective and accurate wave farm power prediction using Cooperative Multi-objective Covariance Adaptation Gradient Ensemble Model with comprehensive comparative experiments

Wave-farm power prediction is central to large-scale deployment but remains challenging due to hydrodynamic coupling, inter-WEC interactions, and the highly nonlinear layout–control design space. Existing surrogate models are typically calibrated with a single accuracy objective, which often fails to ensure robustness, low bias, and stable performance in high-power regimes. To address this limitation, this study proposes a Cooperative Multi-objective Covariance Matrix Adaptation Gradient Ensemble model, CMA-SXSCE, for high-fidelity prediction of total absorbed power in 16-WEC wave farms. The framework combines heterogeneous cooperative learning with graph-based representations of spatial and hydrodynamic interactions, while hyperparameter tuning is formulated as a constrained multi-objective optimisation problem to jointly balance accuracy, relative error, high-power robustness, and distributional consistency.