A data-physics hybrid-driven layout optimization framework for large-scale wind farms

The global trend of wind energy utilization moves towards building large-scale and remotely-located wind farms, while strategic layout optimization is crucial to improving the power generation of wind farms. However, large-scale wind farm layout optimization (WFLO) faces challenges due to complicated calculations involving the high-dimensional decision variables and the need to trade-off between wake model accuracy and efficiency. To address these issues, this paper proposes a novel data-physics hybrid-driven framework for layout optimization of large-scale wind farms. This framework attempts to integrate physical equations with variable parameters to guide the modeling of wake effects and further facilitate the layout optimization process. Specifically, the physics-informed dual neural network (PIDNN) model is proposed to estimate the wind velocity. This model incorporates a variable thrust coefficient into the Navier–Stokes equations through dual neural networks. Moreover, the gene-targeted differential evolution (GTDE) algorithm is employed to optimize the wind farm layout, which is particularly designed for large-scale optimization problems. Simulation results demonstrate that the proposed PIDNN can estimate wake velocity effectively. Furthermore, the proposed optimization framework outperforms competing methods, achieving the highest power generation.