Application of Reinforcement Learning to Wind Farm Active Power Control Design

Wind power is becoming a key player of the world's energy landscape thanks to its cleanliness, abundance and huge potential. At the same time, it attracts an increasing attention when it comes to its efficient production and financial viability, which otherwise may restrict its development in the foreseeable future. Indeed, enhancing energy production efficiency to reduce costs and improve the ability of the system to resist faults are research areas of great interest. With the advancement of machine learning and artificial intelligence along with increasing computational power at hand, reinforcement learning enables achieving an optimal control solution in an application environment after continuous attempts and updates. In this paper, a novel solution based on reinforcement learning is applied to the control of wind farm. An intelligent agent is designed to explore the environment, and after training, it effectively maintains the necessary balance between power generation and load, which in turn regulates the wind farm grid frequency when enough wind is available. The trained agent is tested under different loads, realistic wind fields, as well as fault scenarios. All simulation results show that the agent accurately understands the environment and load requirements, mitigates the impact of faults, and thus, improves the stability of the grid frequency.