复杂工况下选择性催化还原脱硝系统的迁移强化学习控制
Transfer reinforcement learning control for a selective catalytic reduction denitration system under complex conditions
孙小明 1彭晨 1程传良1
作者信息
- 1. 上海大学机电工程与自动化学院,上海 200444
- 折叠
摘要
针对复杂工况下选择性催化还原(SCR)系统难以实现精确脱硝控制的问题,本文提出一种基于迁移强化学习的智能控制方法.首先根据机组负荷的变化将整体运行过程划分为不同阶段.然后训练了强化学习控制器以分别学习各个阶段的不同特征,从而实现了变工况下SCR脱硝系统的精确控制.此外,借鉴了迁移学习的思路以应对预料之外的未知工况,避免了因工况未知所造成的不利影响.最后将训练好的控制器用于实际SCR脱硝系统的控制中,实验结果表明所提方法可以有效地控制复杂工况下燃煤机组NOx的排放量,为复杂工况下SCR脱硝系统的智能控制提供了借鉴.
Abstract
Aiming at the problem that selective catalytic reduction(SCR)system is difficult to achieve precise denitration control performance under complex working conditions,an intelligent control method based on transfer reinforcement learning is proposed in this paper.The overall operation process is firstly divided into different stages according to the changes of unit load.Then a reinforcement learning controller is trained to learn different characteristics of each stage,respectively,so as to realize accurate control of the SCR denitration system under variable working conditions.In addition,the idea of transfer learning is used for reference to deal with unexpected unknown working conditions and avoid adverse effects caused by unknown working conditions.Finally,the trained controller is applied to the control of an actual SCR denitration system.Experimental results show that the proposed method can effectively control NOx emissions of a coal-fired power unit under complex working conditions,and provide an idea for intelligent control of the SCR denitration system under complex working conditions.
关键词
SCR脱硝系统/变工况/未知工况/强化学习/迁移学习Key words
SCR denitration system/variable working condition/unknown working condition/reinforcement learning/transfer learning引用本文复制引用
基金项目
国家自然科学基金重点项目(61833011)
出版年
2024
国家科技期刊平台
NSTL
万方数据