Data-Driven Wind Farm Layout Optimization Using Bayesian Adaptive Direct Search

With a burgeoning global appetite for clean energy, wind energy has swiftly risen to become the second most burgeoning renewable energy source, characterized by a notable annual growth rate, and has secured a position of about 6.3% in the global energy landscape. Optimizing the layout of wind farms is critical to maximizing power generation and reducing costs, a process that involves challenges such as complex multivortex wake effects, increased design variables, and non-convex objective function spaces. To overcome these challenges, the authors propose a new optimization framework. The framework includes the analysis of the vortex wake model, the eddy deflection model and the partial eddy interference effect considering the vortex superposition effect, and transforms the wind farm layout optimization into a constrained black-box optimization problem. By merging Bayesian optimization techniques with the Mesh Adaptive Direct Search (MADS) methodology, a novel algorithm known as Bayesian Adaptive Direct Search (BADS) has been developed. As a model-independent, gradient-free, efficient sample utilization and fast convergence optimization tool, BADS can effectively solve the non-smoothness in layout optimization and avoid the local optimal trap. The results not only confirm the important role of wind farm layout in improving efficiency, but also highlight the efficiency and potential of the BADS framework in dealing with actual complex models, and provide compatibility for further integration of advanced vortex wake models and high-fidelity simulations of topographic wind field interaction, indicating that wind energy utilization will be further improved in the future.