Optimization control for CCS of coal-fired power unit based on reinforcement learning using process data

Due to the large-scale grid-connection of new energy power generation, the requirements for the control system of coal-fired power units have become increasingly stringent. This paper investigates the optimization control of the coal-fired unit and proposes a reinforcement learning control framework using process data of coal-fired unit. The core of the proposed method is to employ the historical operation data to train the policy network by the supervised learning algorithm, and then the pretrained policy is further improved by the reinforcement learning via interacting with the power unit. The results of simulation experiment show that the proposed method can accelerate the training process, alleviate the computational burden compared with the standard reinforcement learning without using process data.