Model-free wind farm wake-steering optimization using the cross-entropy method

Abstract

Wake steering by yaw misalignment can increase wind farm energy capture by redirecting wakes away from downstream turbines, but the resulting optimization problem is nonconvex, constrained, and costly to evaluate when high-fidelity wake models or field measurements are used. This paper presents a model-free optimization approach based on the Cross-Entropy Method (CEM) to tune turbine yaw offsets for steady inflow conditions. CEM maintains a parametric sampling distribution over yaw settings, iteratively samples candidate control vectors, selects an elite set according to measured or simulated farm power, and updates the distribution to concentrate probability mass around high-performing regions. We detail constraint handling, variance control, and robust extensions for wind-direction uncertainty, and outline a reproducible simulation workflow in which a wake model acts as the evaluation oracle.