Active power regulation of wind turbine based on deep reinforcement learning guided active disturbance rejection control

Enhancing the active power regulation ability of wind turbine is a crucial approach to improving the flexibility of new power systems. This paper proposes a coordinated optimization strategy for reinforcement learning (RL) guided error-based active disturbance rejection controller (EADRC). Firstly, a well-established EADRC framework is employed, in which an extended state observer estimates and compensates for system disturbances in real time, thereby enhancing the system's disturbance rejection capability and adaptability. Secondly, a Markovian decision process is integrated into the EADRC, allowing RL to adaptively tuning EADRC parameters in real time. This innovative integration of data-driven learning with disturbance rejection control enables the system to achieve optimal active power responses across diverse operating scenarios. Finally, simulations on a 5 MW wind turbine show that this proposed strategy outperforms conventional control methods in both active power tracking accuracy and system dynamic response, significantly enhancing the robustness of wind turbine control. This provides effective technical support for large-scale wind power integration in new power systems.