Dynamic Control of Fast Load Regulation for Thermal Power Unit Based on Reinforcement Learning and Edge Computing

In the current rapid load regulation dynamic control process of thermal power units, differential evolution algorithm is usually used to solve the optimal control decision. However, this algorithm is prone to getting stuck in local optimal solutions during the convergence process, resulting in deviations in the final control results. Therefore, a dynamic control method for fast load regulation of thermal power units based on reinforcement learning and edge computing is proposed. The edge computing technology is used to collect the operating data of thermal power units in real time, and input it into the stack self encoder neural network for autonomous learning, so as to quickly predict the load value of thermal power units. Based on the total load demand of thermal power units and the fast load prediction value, determine the load regulation control objectives, rely on reinforcement learning algorithms to complete intelligent analysis, and generate the best control decision that meets the control objectives. Establishing a fuzzy decoupling PID control scheme and executing dynamic control decisions can achieve dynamic load regulation control of thermal power units. The experimental results show that this method maintains the absolute error integral (IAE) below 40 MWh, improving the accuracy of load regulation control for the unit.