Hierarchical Reinforcement Learning EADRC Strategy for CHP Unit Integrated With Absorption Heat Pump Toward Fast Variable Load Demands

To smooth out the power fluctuations caused by the large-scale integration of new energy into the power grid, it is necessary to enhance the rapid load variation capacity of the combined heat and power (CHP) unit. Based on this, an advanced strategy combining deep deterministic policy gradient (DDPG) and error-based active disturbance rejection control (EADRC) is proposed to achieve flexible operation of the CHP unit integrated with absorption heat pump (AHP). First, a new coordinated control system (CCS) is designed to achieve heat-power decoupling by flexibly adjusting the heating valve and the heat pump regulating valve. The decoupled system can balance stable heating and power peak shaving, significantly improving the variable load rate of the unit. Second, a hierarchical control framework is proposed to achieve the optimal control of the CHP unit integrated with AHP. The EADRC estimates and compensates for the total system disturbance through an extended state observer to obtain the initial control law. The DDPG forms a compensation control law via online weight learning, enhancing the robustness of EADRC against uncertainty and disturbances. Finally, the hierarchical DDPG–EADRC strategy is applied to the 350 MW CHP unit integrated with AHP, and the effectiveness of the proposed scheme is evaluated in set value tracking, disturbance rejection, and model mismatch tests. The experimental results show that in the set value tracking test scenario, the settling time of the proposed strategy is reduced by 862 s compared to the traditional CCS, and the maximum overshoot is reduced by 3.74 MW.