Wind farm optimization with computationally efficient AEP models.
Abstract
As wind farms expand to meet energy transition goals, wake effects create significant challenges, giving rise to the Wind Farm Layout Optimization (WFLO) problem. Current methods to maximize Annual Energy Production (AEP) by minimizing wakes are computationally intensive, requiring hundreds of iterations and multiple optimizations per project. Several AEP models with a variety of simplifications, assumptions and approaches, were tested as objective functions in a gradient based optimization on a complex 167 turbine case study, both with and without grid constraints, alongside other constraints such as minimum turbine distance, boundaries, and exclusion zones. The FLOW Estimation and Rose Superposition (FLOWERS) method emerged as the most effective, balancing efficiency and efficacy with the lowest wake losses and optimization times, thanks to its fast evaluations, smooth design space – coming from the continuous form using Fourier – and single CPU configuration. Stochastic Gradient Descent (SGD), thanks to its algorithm and stochastic nature, performed greatly in terms of AEP, with a performance close to FLOWERS, accomplishing fair efficiency only when using a reduced number of iterations, despite its loss in efficacy. Although complex models with many speed and direction bins delivered consistent results, they required 5-6 times more computational time despite using 8 CPUs. On the other hand, simpler models, such as those using average wind speed or a uniform thrust coefficient, proved highly efficient but poorly reflected the physics, leading to suboptimal layouts. In addition, it was concluded that models with fewer wind directions and Gaussian-shaped wake models effectively captured layout changes, whereas those averaging wind speed per direction missed critical physical dynamics. Overall, FLOWERS outperformed all other models and demonstrated strong potential for other optimization problems applications, including nested and multi-stage optimizations.
Cite this
Bueno, Javier Andueza. (2026). Wind farm optimization with computationally efficient AEP models.
Link
Download here.