Deep Reinforcement Learning Applied to Wake Steering

Abstract Wake steering is a wind farm control strategy where the yaw angle of turbines is intentionally misaligned with the incoming wind direction to deflect the wake in such a way as to increase the power production from the downstream turbines in exchange for producing less power in the upstream turbines. This paper presents a Deep Reinforcement Learning approach for predicting the optimal turbine misalignment using steady‐state flow simulations and the Proximal Policy Optimization (PPO) algorithm. This approach leads to a 2.25–5.27% improvement over the greedy strategy, averaged over all incident wind directions, with a computing time of less than 30 s per configuration, although it does not outperform the state‐of‐the‐art optimizers.