中国电机工程学报2024，Vol.44Issue(12)：4815-4825,中插17,12.DOI:10.13334/j.0258-8013.pcsee.221335
水泥分解炉SNCR脱硝系统的深度强化学习多目标优化控制研究
Research on Deep Reinforcement Learning Multi-objective Optimization Control of SNCR Denitration System of Cement Calciner
刘定平 1吴泽豪1
作者信息
- 1. 广东省能源高效低污染转化工程技术研究中心(华南理工大学),广东省 广州市 510640
- 折叠
摘要
Abstract
The process parameter optimization of selective non-catalytic reduction(SNCR)denitration process can effectively reduce NOx emission and denitration cost of cement calciner.Taking a cement calciner as the research object,a NOx concentration prediction model based on LightGBM is established.Taking the minimization of denitration cost and NOx concentration as the optimization objective,the deep deterministic policy gradient(DDPG)algorithm is used to optimize and control the relevant process parameters of SNCR denitration process of cement calciner mixed with sludge.The results show that the RMSE of NOx concentration prediction model is 6.8 and MAPE is 3.48%.The DDPG algorithm can effectively optimize the relevant process parameters:when the ammonia injection amount and sludge combustion amount are 427.87 L/h and 9.78 t/h,respectively,the NOx emission concentration is 225.99 mg/(Nm3)and the denitration operation cost is 1 747.8 yuan/h.Compared with the results of other optimization algorithms and conventional working conditions,the NOx emission concentration and denitration cost of DDPG optimal control model show different degrees of decline.The simulation and effect verification of the model show that the model can output a reasonable combination of ammonia injection and sludge combustion,reduce the fluctuation of NOx concentration at the outlet of SNCR,effectively reduce NOx emission concentration and denitration cost,and realize the multi-objective optimal control of SNCR denitration system.This study can provide a reference for the multi-objective optimal control design of SNCR denitration of cement calciner based on intelligent algorithm.关键词
喷氨/污泥掺烧/选择性非催化还原优化控制/LightGBM/强化学习/深度确定性策略梯度Key words
ammonia injection/sludge mixed combustion/SNCR optimal control/LightGBM/reinforcement learning/deep deterministic policy gradient分类
信息技术与安全科学引用本文复制引用
刘定平,吴泽豪..水泥分解炉SNCR脱硝系统的深度强化学习多目标优化控制研究[J].中国电机工程学报,2024,44(12):4815-4825,中插17,12.
