Due to its high parameters and high efficiency, ultra supercritical units have become a powerful support for deep frequency regulation, peak shaving, and suppression of power grid fluctuations. The optimization and transformation of control strategies for ultra supercritical units are of great significance for the safe and stable operation of the power grid. Aiming at the optimization problem of coordinated control system for ultra supercritical units, an intelligent control strategy based on error self-disturbance rejection control strategy and reinforcement learning algorithm is proposed. Firstly, in framework of the error-based self-disturbance rejection control strategy, the controlled object model of the machine furnace coupling process is simplified according to operating characteristics of the unit’s turbine-boiler coupled process, and an extended state observer is designed to estimate and compensate for the unmodeled dynamic characteristics and external disturbances of the unit in real time. Secondly, a reward function is constructed and the flexible actor-critic algorithm is used to achieve self-adaptive adjustment of controller parameters. Finally, the effectiveness of the proposed control strategy is verified through simulation based on actual historical operating data of a certain ultra supercritical 1 000 MW secondary reheating unit.