Drainage system real-time control method and device

A drainage system real-time control method introduces a reinforcement learning method, and is constructed according to a model structure and an operation mode of reinforcement learning RL, wherein a drainage system model is used as an environment, a deep neural network is used as an Agent, and a large amount of State, evaluation score Reward and operation strategy Action data are obtained throughinteractive operation between the Agent and the environment for repeated training. The agent is continuously optimized, and an operation strategy Action is generated through the Agent in practical application so that the purpose of improving the operation effect of the drainage system is achieved. Optimal control over the drainage system is achieved through reinforcement learning, and compared with existing heuristic real-time control, a global optimal strategy can be searched for, and operation of the drainage system is better optimized; and compared with model prediction control, the problems caused by prediction er