Power loss minimization-oriented reactive power control for wind farm equipped with distributed energy storages using clustering-based data-driven method

This paper presents a data-driven based reactive power control method for the wind farm, in which every wind turbine is equipped with a standalone distributed energy storage unit. Firstly, the hierarchical control framework is adopted through dividing the wind turbines and standalone distributed energy storage into groups. In the process of grouping, multi-view clustering is applied to help the controller get valid results in a wide range of wind conditions. Then, the active power loss models within the wind farm are constructed, and a reactive power control framework is designed through the upper-level coordination of multiple deep reinforcement learning agents and the lower-level power distribution. Especially, a similarity-based discount factor is proposed in the reward function of each agent to reduce the impacts from the agents with wide differences and improve the control convergence. A modified wind power system is utilized for case study, and several frameworks using different grouping schemes and deep reinforcement learning algorithms are defined for comparison. The results indicate that the participation of the standalone distributed energy storage in the reactive power control can effectively reduce the active power losses within the wind farm; the proposed method can achieve stable, fast and accurate control under variable operation conditions.