Model-Free Optimization Scheme for Efficiency Improvement of Wind Farm Using Decentralized Reinforcement Learning

Abstract Wake interactions caused by the complex wakes between the turbines within a wind farm have significant adverse effect on the total power generation of the wind farm. To mitigate the effect of wake interactions and optimize the total power output of wind farm, this paper proposes a model-free control scheme using reinforcement learning by developing a decentralized Q learning method. The proposed approach guarantees that the output power of wind farm converges to the optimal total power under different wind conditions, and further ensures the gradual changes of control variables of wind turbines and thus avoids the unexpected sharp drop of the power generation performance of wind farm. Simulation results are provided to demonstrate the effectiveness of the proposed method.