基于混合建模方法循环流化床锅炉深度调峰NO_(x)排放预测  

Prediction of NO_(x)emissions from deep peaking circulating fluidizedbed boilers based on a hybrid modelling approach

作　　者：张鹏新 高明明[1] 郭炯楠 于浩洋 黄中[2] 周托 ZHANG Pengxin;GAO Mingming;GUO Jiongnan;YU Haoyang;HUANG Zhong;ZHOU Tuo(State Key Laboratory of Alternate Electrical Power System With Renewable Energy Sources,North China Electric Power University,Beijing 102206,China;Department of Energy and Power Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]华北电力大学新能源电力系统国家重点实验室,北京102206 [2]清华大学能源与动力工程系,北京100084

出　　处：《洁净煤技术》2024年第9期85-94,共10页Clean Coal Technology

基　　金：国家重点研发计划资助项目(2022YFB4100304)。

摘　　要：为响应碳达峰,碳中和目标,我国循环流化床锅炉大规模参与深度调峰运行,导致锅炉NO_(x)排放浓度波动范围大,控制效果不佳,难以满足污染物超低排放需求,因此对深度调峰NO_(x)排放浓度进行精准建模预测有重要意义。以即燃碳模型为基础,深度剖析炉内NO_(x)生成和还原机理,建立炉内即燃碳燃烧模型、O2动态平衡模型、CO软测量模型、NO_(x)生成与还原模型,完成SNCR入口NO_(x)浓度机理计算;选取给煤量、床温、烟气温度及含氧量、一二次风量、尿素溶液流量作为NO_(x)排放浓度的输入变量,将SNCR入口NO_(x)浓度计算值作为拓展输入变量,对所有输入变量与NO_(x)排放浓度进行相关性分析和迟延补偿,完成数据集重构;采用长短期记忆神经网络对重构数据集进行训练和预测,并将鲸鱼优化算法用于长短期记忆神经网络的参数优化,建立循环流化床锅炉深度调峰NO_(x)排放浓度机理——数据混合预测模型。仿真验证表明混合预测模型不同工况下预测性能和泛化能力好,能够实现循环流化床锅炉变负荷时NO_(x)排放浓度的实时预测,相较其他预测模型的各项误差性能指标均显著提升,平均绝对误差δMAE达2.14 mg/m^(3),平均相对百分误差δMAPE达5.68%,决定系数R^(2)达0.902 1。混合预测模型能精准预测循环流化床锅炉深度调峰下NO_(x)排放浓度,为循环流化床锅炉超低排放智能控制系统的设计提供参考。In response to the goal of Carbon peak Carbon neutral,China′s circulating fluidized bed boilers participate in deep peaking operation on a large scale,resulting in large fluctuation ranges of NO_(x)emission concentration in boilers,poor control effect,and difficulty in meeting the demand for ultra-low emission of pollutants,so it is important to accurately model and predict the NO_(x)emission concentration in deep peaking.Based on the instantaneous carbon model,the NO_(x)generation and reduction mechanism in the furnace was deeply analyzed,and the instantaneous carbon combustion model,O_(2)dynamic balance model,CO soft measurement model,NO_(x)generation and reduction model were established to complete the calculation of the mechanism of the NO_(x)concentration at the entrance of the SNCR.The amount of coal feed,bed temperature,flue gas temperature and oxygen content,the first and second airflow,and the flow rate of the urea solution were selected as the input variables for the NO_(x)emission concentration,and the NO_(x)emission concentration was predicted by the SNCR inlet model.The SNCR inlet NO_(x)concentration was used as an extended input variable,and the data set was reconstructed by correlation analysis and delay compensation between all input variables and NO_(x)emission concentration.The reconstructed data set was trained and predicted by using long and short-term memory neural network,and whale optimization algorithm was used for the optimization of parameters of the long and short-term memory neural network to establish a NO_(x)emission concentration model,the mechanism-data hybrid prediction model,for deep peaking of circulating fluidized bed boilers.The simulation validation shows that the hybrid prediction model has good prediction performance and generalization ability under different working conditions,and is able to realize real-time prediction of NO_(x)emission concentration in circulating fluidized bed boilers at variable loads,and significantly improves all the error performance indexes compa

关 键 词：循环流化床锅炉 深度调峰 NO_(x)排放浓度 迟延补偿 混合预测模型 

分 类 号：TM621.2[电气工程—电力系统及自动化] TK229.6[动力工程及工程热物理—动力机械及工程]