基于深度强化学习的SCR脱硝系统协同控制策略研究

Research on Cooperative Control Strategies for SCR Denitrification Systems Based on Deep Reinforcement Learning

针对选择性催化还原(SCR)脱硝系统大惯性、多扰动等特点,提出了一种基于多维状态信息和分段奖励函数优化的深度确定性策略梯度(DDPG)协同比例积分微分(PID)控制器的控制策略。针对SCR脱硝系统中存在部分可观测马尔可夫决策过程(POMDP),导致DDPG算法策略学习效率较低的问题,首先设计SCR脱硝系统的多维状态信息;其次,设计SCR脱硝系统的分段奖励函数;最后,设计DDPG-PID协同控制策略,以实现SCR脱硝系统的控制。结果表明:所设计的DDPG-PID协同控制策略提高了DDPG算法的策略学习效率,改善了PID的控制效果,同时具有较强的设定值跟踪能力、抗干扰能力和鲁棒性。

A cooperative control strategy of deep deterministic policy gradient(DDPG)and proportion integration differentiation(PID)based on multidimensional state information and segmental reward function optimization was proposed for the selective catalytic reduction(SCR)denitrification system with large inertia and multi-disturbance.Addressing the problem of low strategy learning efficiency of the DDPG algorithm caused by the partially observable Markov decision process(POMDP)in the SCR denitrification system,the multidimensional state information of the SCR denitrification system was designed firstly.Secondly,the segmented reward function of the SCR denitrification system was designed.Finally,a DDPG-PID cooperative control strategy was designed to achieve the control of SCR denitrification system.Results show that the designed DDPG-PID cooperative control strategy improves the strategy learning efficiency of the DDPG algorithm and the control effect of PID.Meanwhile,the designed cooperative control strategy has strong set value tracking capability,anti-interference capability and robustness.