基于强化学习的SCR脱硝系统优化控制

Optimal Control of SCR Denitrification System Based on Reinforcement Learning

为了解决选择性催化还原(SCR)脱硝系统中NO_(x)排放控制遇到的时效滞后和负荷波动引发的响应问题,采用强化学习来调控比例积分微分(PID)参数,基于深度确定性策略梯度(DDPG)算法,重新设计了Critic网络的损失函数,并引入了一个延迟队列来模拟系统的延迟性,并将所提控制策略应用到国内某660 MW超超临界火电机组中。结果表明:强化学习控制方法在调节时间、超调量、稳定性方面均优于传统的PID控制方法;所提控制策略克服了传统PID控制方法无法解决的时效滞后和负荷波动问题,具有重要的理论与实际应用价值。

To address the response issues caused by time lag and load fluctuations in NO_(x)emission control of SCR denitrification system,reinforcement learning was employed to adjust the proportional-integral-derivative(PID)parameters.The loss function of Critic network was redesigned according to the deep deterministic policy gradient(DDPG)algorithm,and a delay queue was introduced to simulate system latency.The proposed control strategy has been applied to a 660 MW ultra-supercritical coal-fired power unit in China.Results show that the reinforcement learning control method is superior to traditional PID control in terms of adjustment time,overshoot,and stability.The proposed strategy overcomes the time lag and load fluctuations that traditional PID control cannot resolve,demonstrating the significant theoretical and practical values.