Data-driven aerodynamic optimization for enhancing unsteady performance in wind energy systems

Wind turbine blades operating under unsteady inflow conditions often suffer from adverse aerodynamic effects induced by dynamic stall, which can significantly compromise energy capture efficiency and increase structural loading. While computational fluid dynamics (CFD) combined with optimization algorithms has been widely adopted for aerodynamic performance improvement, the high computational cost remains a major barrier for practical deployment. This study proposes a data-efficient optimization framework based on the Gaussian process regression (GPR) model to enhance unsteady aerodynamic performance in wind energy system. A machine learning-based predictive model is integrated with an evolutionary optimizer to reduce the reliance on high-fidelity CFD simulations during iterative design. The model accuracy is progressively improved by sequentially infilling sample points based on the expectation improvement (EI) criterion. The approach is validated through the optimization of a standard wind turbine blade profile under oscillating flow conditions. Results indicate that the optimized airfoil exhibits an increased thickness and a rounded leading-edge, efficiently suppressing the formation and upstream propagation of trailing-edge vortices. Compared to the baseline design, the optimized design achieves reductions of 32.87 % and 23.74 % in averaged drag and moment coefficients, respectively, over one oscillation cycle. When applied to the NREL Phase VI rotor, the optimized profile exhibited improved aerodynamic performance, supporting its potential for broader application in wind turbine systems operating under unsteady inflow.