Distributed Multivariable Control for Wind Farm Power Generation Based on Multi-Agent Deep Reinforcement Learning

To address the reduction in power output due to wake effects in wind farms, a distributed control strategy based on multi-agent deep reinforcement learning is proposed to maximize the power generation efficiency of wind farms. The multivariate power model of wind farms is introduced to analyze the impact of wake effects on power output. This model considers controllable variables such as yaw angle, tilt angle, pitch angle, and tip speed ratio of wind turbines. To effectively control multiple variables of wind turbines, each wind turbine is modeled as an agent. The wind turbine control problem is then modeled as a partially observable Markov decision process, with each agent using the proximal policy optimization algorithm to solve for real-time optimized control strategies. Simulation results show that the proposed distributed control strategy for wind farms significantly improves power output compared to existing wind farm control algorithms.