Decentralized yaw optimization for maximizing wind farm production based on deep reinforcement learning

This study describes a deep reinforcement learning (DRL) based decentralized yaw optimization method to maximize the power production of wind farms. Specifically, we apply the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm to design agents separately for each turbine in the wind farm to make yaw de-cisions independently. This design allows setting different yaw rewards for the agents to promote faster and better convergence of MADDPG. To test the control effect of the proposed method, a novel analytical dynamic wake model is derived first, which can dynamically reflect the wake propagation of the wind turbine after the inflow wind speed, axial induction factor, and yaw angle change. Then, the static and dynamic characteristics of the proposed wake model are verified. The wind farm simulation is realized through the dynamic wake model, which provides an interactive environment for the simulation experiments to test the effect of the proposed DRL-based decentralized yaw optimization method. The results show that the proposed method can significantly increase the power generation of the wind farm, and the set yaw reward helps to guide MADDPG to converge to the optimal strategy.