基于强化学习的再热汽温系统优化控制

Reheat Steam Temperature System Optimization Control Based on Reinforcement Learning

摘要: 针对具有传统控制策略面对再热汽温这种大惯性、大迟延系统时控制性能不足的问题，提出了一种基于DQN前馈与PID复合控制的策略，以提高系统的控制精度。在控制过程中，根据系统状态选择最优动作对PID控制器输出进行补偿。并且针对强化学习面对迟延对象训练难以收敛的问题，设计多维奖励函数和迟延缓冲区。结果表明：所设计的控制方法在超调量、调节时间、鲁棒性方面均优于传统PID控制，为再热汽温的高效稳定控制提供了新方案。

Abstract: Aiming at the problem of insufficient control performance of traditional control strategies when dealing with the large inertia and large delay system of reheated steam temperature, a strategy based on DQN (Deep Q Network) feedforward and PID compound control is proposed to improve the control accuracy of the system. During the control process, the optimal action is selected according to the system state to compensate the output of the PID controller. Moreover, to address the issue of difficulty in convergence during the training of reinforcement learning for delay objects, a multi-dimensional reward function and delay buffer are designed. The results show that the designed control method outperforms traditional PID control in terms of overshoot, regulation time, and robustness, providing a new solution for the efficient and stable control of reheated steam temperature.