Hybrid Artificial Intelligence-based Computational Fluid Dynamics model for optimizing Offshore Wind Farm aerodynamics under variable marine climate conditions

Abstract

Aerodynamic optimization of Offshore Wind Farms (OWF) is challenged by nonlinear wake interactions, turbulence transport, and stochastic Marine Climate Conditions (MCC). High-fidelity Computational Fluid Dynamics (CFD) models capture these dynamics accurately but impose prohibitive computational costs for large-scale optimization. Analytical wake models offer computational efficiency but oversimplify complex turbulence interactions. This study presents a hybrid Artificial Intelligence (AI)-CFD that integrates Deep Neural Network (DNN) surrogates with selective CFD validation to enable efficient, robust optimization under MCC. The model employs unsteady Reynolds-averaged Navier-Stokes simulations with actuator-line turbine representation to generate training data, which are used to train an ensemble surrogate model incorporating dimensionally reduced climate states. Uncertainty-driven adaptive sampling triggers CFD validation for high-uncertainty configurations, maintaining physical fidelity while accelerating optimization. A multi-objective evolutionary algorithm (NSGA-II) optimizes turbine layout, yaw angles, and pitch controls to balance power generation, wake losses, and structural loading. Validation on an 80-turbine, 320 MW wind farm in Dhanushkodi, Tamil Nadu, sea proves 7.8% power improvement, 23.5% wake-loss reduction, and 11.2% network load decrease compared to the baseline, with 7.2× computational speedup vs. CFD-only optimization. Sensitivity analyses confirm robustness across wind speeds (6–12 ms−1 ), turbulence intensities (5–15%), and inflow directions (0–30°). The model establishes a scalable methodology for optimizing OWF under realistic MCC.