基于双向长短期记忆网络和最小二乘支持向量机模型的SCR脱硝反应器进口截面NO x 浓度预测

doi:10.12096/j.2096-4528.pgt.24008

发电技术 ›› 2025, Vol. 46 ›› Issue (5): 996-1004.DOI: 10.12096/j.2096-4528.pgt.24008

基于双向长短期记忆网络和最小二乘支持向量机模型的SCR脱硝反应器进口截面NO _x 浓度预测

曾稳根, 陈冬林, 唐明珠, 汪淑奇, 胡章茂

长沙理工大学能源与动力工程学院，湖南省长沙市 410114

收稿日期:2024-06-28 修回日期:2024-10-02 出版日期:2025-10-31 发布日期:2025-10-23
作者简介:曾稳根(1997)，男，硕士研究生，从事喷氨智能控制策略优化研究，zeng.wengen@foxmail.com；
陈冬林(1963)，男，博士，教授，从事锅炉及燃烧理论与技术研究应用、燃烧污染物减排技术等研究，本文通信作者，chendl_01@126.com；
唐明珠(1983)，男，博士，副教授，从事故障诊断与机器学习及应用研究，tmzhu@163.com；
汪淑奇(1966)，男，硕士，副教授，从事锅炉燃烧与优化控制等研究，wangsq_mail@163.com；
胡章茂(1985)，男，博士，副教授，从事高效低污染燃烧技术等研究，huzhangmao@163.com。
基金资助:
国家自然科学基金项目(62173050)

Prediction of NO _x Concentration at Inlet Section of SCR Denitrification Reactor Based on Bidirectional Long Short-Term Memory Network and Least Squares Support Vector Machine Model

Wengen ZENG, Donglin CHEN, Mingzhu TANG, Shuqi WANG, Zhangmao HU

School of Energy and Power Engineering, Changsha University of Science and Technology, Changsha 410114, Hunan Province, China

Received:2024-06-28 Revised:2024-10-02 Published:2025-10-31 Online:2025-10-23
Supported by:
National Natural Science Foundation of China(62173050)

摘要/Abstract

摘要：

目的为解决目前选择性催化还原(selective catalytic reduction，SCR)脱硝喷氨控制系统喷氨指令与SCR反应器进口截面NO _x 浓度动态适配迟延及喷氨精度不足问题，构建了某台660 WM燃煤机组SCR脱硝喷氨反应器进口截面NO _x 浓度预测模型。方法结合燃煤机组烟道结构对燃煤机组采样数据进行预处理，然后采用双向长短期记忆网络(bidirectional long short-term memory network，BiLSTM)对SCR脱硝喷氨反应器进口截面NO _x 浓度进行初步预测，并使用最小二乘支持向量机(least squares support vector machine，LSSVM)对预测残差进行修正，从而最大程度减小模型的预测误差。最后，为评估所建模型的预测性能，进行了仿真与现场同步验证。结果该模型在机组各负荷段及变煤质工况下对SCR反应器入口截面NO _x 浓度的预测具有良好的效果，同步验证预测误差仅为5.772 mg/m³，并且具备8.5 s的提前预测能力。结论该预测模型可为喷氨控制系统提供精确的前馈信息。

关键词: 燃煤机组, 数据预处理, 神经网络预测, 脱硝, 双向长短期记忆网络(BiLSTM), 最小二乘支持向量机(LSSVM), 选择性催化还原(SCR)反应器, NO _x 浓度预测

Abstract:

Objectives To address the delayed dynamic adaptation between the ammonia injection command and the NO _x concentration at the inlet section of the selective catalytic reduction (SCR) denitrification reactor, as well as the insufficient ammonia injection accuracy in current SCR denitrification ammonia injection control systems, a prediction model for the NO _x concentration at the inlet section of the SCR denitrification reactor of a 660 MW coal-fired unit is developed. Methods The sampled data from the coal-fired unit are preprocessed in combination with the flue duct structure of the unit. Subsequently, a bidirectional long short-term memory network (BiLSTM) is applied to generate a preliminary prediction of the NO _x concentration at the inlet section of the SCR denitrification ammonia injection reactor. Next, a least squares support vector machine (LSSVM) is utilized to correct the prediction residuals, thereby minimizing the prediction error of the model to the greatest extent. Finally, in order to evaluate the prediction performance of the proposed model, simulation and real-time validation are carried out. Results This model exhibits excellent performance in predicting the NO _x concentration at the inlet section of the SCR reactor under various unit load conditions and variable coal quality conditions. The prediction error in real-time validation is only 5.772 mg/m³, with an advance prediction capability of 8.5 s. Conclusions This prediction model can provide accurate feedforward information for the ammonia injection control system.

Key words: coal-fired unit, data preprocessing, neural network prediction, denitrification, bidirectional long short-term memory network (BiLSTM), least squares support vector machine (LSSVM), selective catalytic reduction (SCR) reactor, NO _x concentration prediction

中图分类号:

TK 323

曾稳根, 陈冬林, 唐明珠, 汪淑奇, 胡章茂. 基于双向长短期记忆网络和最小二乘支持向量机模型的SCR脱硝反应器进口截面NO _x 浓度预测[J]. 发电技术, 2025, 46(5): 996-1004.

Wengen ZENG, Donglin CHEN, Mingzhu TANG, Shuqi WANG, Zhangmao HU. Prediction of NO _x Concentration at Inlet Section of SCR Denitrification Reactor Based on Bidirectional Long Short-Term Memory Network and Least Squares Support Vector Machine Model[J]. Power Generation Technology, 2025, 46(5): 996-1004.

图/表 18

表1 NO x 浓度预测模型初步采集特征

Tab. 1 Preliminary collected features for NO x concentration prediction model

序号	特征
1	机组负荷(有功功率)
2	一次风机出口流量
3	空预器出口二次风量
4	锅炉总风量
5	锅炉含氧量
6	锅炉水煤比
7	给煤机A瞬时煤量
8	给煤机B瞬时煤量
9	给煤机C瞬时煤量
10	给煤机D瞬时煤量
11	磨煤机A分离器出口风粉温度
12	磨煤机B分离器出口风粉温度
13	磨煤机C分离器出口风粉温度
14	磨煤机D分离器出口风粉温度
15	SCR反应器进口NO _x 浓度

图1 数据清洗流程图

Fig. 1 Data cleaning flowchart

图2 机组负荷数据滤波图

Fig. 2 Filtering diagram of unit load data

图3 锅炉煤质数据箱型图Mt—全水分质量分数；Aar—收到基灰分质量分数；Ad—干燥基灰分质量分数；Aad—空气干燥基灰分质量分数；Vad—空气干燥基挥发分质量分数；FCad—空气干燥基固定碳质量分数；Vdaf—干燥无灰基挥发分质量分数。

Fig. 3 Box plot of boiler coal quality data

图4 采集特征方差解释比例直方图

Fig. 4 Histogram of variance explanation ratio of collected features

表2 NO x 浓度预测模型预测变量

Tab. 2 Predictor variables of NO x concentration prediction model

序号	变量
1	机组负荷(有功功率)
2	一次风机出口流量
3	锅炉水煤比
4	给煤机A瞬时煤量
5	给煤机B瞬时煤量
6	给煤机C瞬时煤量
7	给煤机D瞬时煤量
8	磨煤机A分离器出口风粉温度
9	磨煤机B分离器出口风粉温度
10	磨煤机C分离器出口风粉温度
11	磨煤机D分离器出口风粉温度

图5 SCR脱硝系统烟道测点布置及CEMS采样示意图

Fig. 5 Schematic diagram of flue measurement points layout in SCR denitrification system and CEMS sampling

图6 机组负荷与烟气流速(ESP入口前)关系图

Fig. 6 Relationship between unit load and flue gas velocity (in front of ESP entrance)

图7 DCS采样数据时间对齐示意图

Fig. 7 Schematic diagram of sampling data time alignment in DCS

图8 BiLSTM-LSSVM模型结构及训练流程

Fig. 8 BiLSTM-LSSVM model structure and training process

图9 BiLSTM和BiLSTM-LSSVM模型训练效果图

Fig. 9 Training performance of BiLSTM and BiLSTM-LSSVM models

表3 预测模型训练指标对照表

Tab. 3 Comparison table of training metrics for predictive models

模型	R_E/(mg/m³)	M_E/%	R
BiLSTM	18.401	6.021	0.910
BiLSTM-LSSVM	10.532	3.755	0.941

图10 BiLSTM和BiLSTM-LSSVM模型预测效果图

Fig. 10 Prediction performance of BiLSTM and BiLSTM-LSSVM models

图11 BiLSTM和BiLSTM-LSSVM模型预测误差分布图

Fig. 11 Prediction error distribution of BiLSTM and BiLSTM-LSSVM models

图12 BiLSTM和BiLSTM-LSSVM模型预测误差堆积图

Fig. 12 Stacked prediction error diagram of BiLSTM and BiLSTM-LSSVM models

表4 各预测模型对新数据的预测性能指标对比

Tab. 4 Comparison of prediction performance metrics of each prediction model for new data

模型	R_E/(mg/m³)	M_E/%	R
LSSVM	21.533	8.353	0.884
LSTM	17.804	6.103	0.915
BiLSTM	16.032	5.592	0.924
BiLSTM-LSSVM	10.256	3.926	0.964

图13 BiLSTM-LSSVM预测模型同步验证曲线图

Fig. 13 Real-time validation curve of BiLSTM-LSSVM prediction model

表5 SCR脱硝系统(A侧)脱硝喷氨参数

Tab. 5 Parameters for ammonia injection in SCR denitrification system (side A)

参数	最大值	最小值	平均值	标准差
脱硝效率/%	92.525	76.886	85.771	3.520
氨逃逸率/(×10^-6)	0.599	0.598	0.599	0.000 265
出口NO _x 质量浓度/(mg/m³)	47.433	30.297	38.650	3.236

参考文献 19

[1]	罗志刚，何成兵，孟浩然，等．燃煤电厂SCR脱硝系统精准控氨优化方法研究[J]．发电技术，2023，44(4)：525-533．
	LUO Z G， HE C B， MENG H R，et al ．Research on optimization method of precise ammonia injection in SCR de-NO _x system of coal-fired power plant[J]．Power Generation Technology，2023，44(4)：525-533．
[2]	袁晓君，马青，李健，等．基于数据驱动的SCR脱硝系统控制优化[J]．电力科技与环保，2023，39(3)：227-234．
	YUAN X J， MA Q， LI J，et al ．Data driven optimal control of SCR denitration system[J]．Electric Power Technology and Environmental Protection，2023，39(3)：227-234．
[3]	董威，林子杰，王雅昀．基于CIFE-FOA-DELM的SCR脱硝入口NO _x 浓度预测方法研究[J]．电力科技与环保，2024，40(3)：313-320．
	DONG W， LIN Z J， WANG Y Y ．A CIFE-FOA-DELM method for predicting NO _x concentration at the inlet of SCR denitration system[J]．Electric Power Technology and Environmental Protection，2024，40(3)：313-320．
[4]	ADAMS D， OH D H， KIM D W，et al ．Prediction of SO _x -NO _x emission from a coal-fired CFB power plant with machine learning：plant data learned by deep neural network and least square support vector machine[J]．Journal of Cleaner Production，2020，270：122310． doi:10.1016/j.jclepro.2020.122310
[5]	AHMED F， CHO H J， KIM J K，et al ．A real-time model based on least squares support vector machines and output bias update for the prediction of NO _x emission from coal-fired power plant[J]．Korean Journal of Chemical Engineering，2015，32(6)：1029-1036． doi:10.1007/s11814-014-0301-2
[6]	LV Y， LIU J Z， YANG T T，et al ．A novel least squares support vector machine ensemble model for NO _x emission prediction of a coal-fired boiler[J]．Energy，2013，55：319-329． doi:10.1016/j.energy.2013.02.062
[7]	YANG T T， MA K F， LV Y，et al ．Real-time dynamic prediction model of NO _x emission of coal-fired boilers under variable load conditions[J]．Fuel，2020，274：117811． doi:10.1016/j.fuel.2020.117811
[8]	ZHAI Y J， DING X D， JIN X Z，et al ．Adaptive LSSVM based iterative prediction method for NO _x concentration prediction in coal-fired power plant considering system delay[J]．Applied Soft Computing，2020，89：106070． doi:10.1016/j.asoc.2020.106070
[9]	KANG J J， NIU Y G， HU B，et al ．Dynamic modeling of SCR denitration systems in coal-fired power plants based on a bi-directional long short-term memory method[J]．Process Safety and Environmental Protection，2021，148：867-878． doi:10.1016/j.psep.2021.02.009
[10]	QIAO J F ．A novel online modeling for NO _x generation prediction in coal-fired boiler[J]．Science of the Total Environment，2022，847：157542． doi:10.1016/j.scitotenv.2022.157542
[11]	张文祥，徐文韬，黄亚继，等．基于数据驱动660 MW循环流化床锅炉多目标燃烧优化[J]．电力科技与环保，2024，40(2)：97-107．
	ZHANG W X， XU W T， HUANG Y J，et al ．Multi-objective combustion optimization for 660 MW circulating fluidized bed boiler based on data-driven approach[J]．Electric Power Technology and Environmental Protection，2024，40(2)：97-107．
[12]	YANG G T， WANG Y N， LI X L ．Prediction of the NO _x emissions from thermal power plant using long-short term memory neural network[J]．Energy，2020，192：116597． doi:10.1016/j.energy.2019.116597
[13]	黄友华，马善为，刘吉，等．燃气轮机烟气SCR脱硝系统优化设计与工程应用[J]．发电技术，2021，42(3)：350-356． doi:10.12096/j.2096-4528.pgt.21029
	HUANG Y H， MA S W， LIU J，et al ．Optimization design and engineering application of gas turbine SCR denitrification system[J]．Power Generation Technology，2021，42(3)：350-356． doi:10.12096/j.2096-4528.pgt.21029
[14]	李存文，王在华，陈涛，等．基于燃煤电站运行数据的烟气脱硫系统性能预测研究[J]．发电技术，2022，43(4)：673-678． doi:10.12096/j.2096-4528.pgt.20009
	LI C W， WANG Z H， CHEN T，et al ．Research on performance prediction of flue gas desulfurization system based on operation data of coal-fired power station[J]．Power Generation Technology，2022，43(4)：673-678． doi:10.12096/j.2096-4528.pgt.20009
[15]	唐振浩，王世魁，曹生现，等．基于混合数据驱动算法的SCR氮氧化物排放量动态预测模型[J]．中国电机工程学报，2022，42(9)：3295-3307．
	TANG Z H， WANG S K， CAO S X，et al ．Dynamic prediction model for NO _x emission of SCR system based on hybrid data-driven algorithms[J]．Proceedings of the CSEE，2022，42(9)：3295-3307．
[16]	谭增强，李元昊，牛拥军，等．火电厂NO _x 的均布优化及预测研究[J]．环境工程，2023，41(S1)：349-353．
	TAN Z Q， LI Y H， NIU Y J，et al ．Research on uniform distribution optimization and prediction of NO _x in thermal power plants[J]．Environmental Engineering，2023，41(S1)：349-353．
[17]	石翠翠，刘媛华．基于组合模型的燃气轮机NO _x 排放影响因素研究[J]．计算机系统应用，2022，31(6)：354-360．
	SHI C C， LIU Y H ．Research on influencing factors of NO _x emission from gas turbine based on combined model[J]．Computer Systems and Applications，2022，31(6)：354-360．
[18]	王祖林，韩硕，康俊杰，等．基于双向深度学习的电站锅炉SCR脱硝系统入口NO _x 浓度预测[J]．自动化与仪表，2021，36(1)：82-87．
	WANG Z L， HAN S， KANG J J，et al ．Pridiction of NO _x concentration at the inlet of SCR denitration system of utility boiler based on bidirectional deep learning[J]．Automation & Instrumentation，2021，36(1)：82-87．
[19]	温鑫，钱玉良，彭道刚，等．基于深度双向LSTM的SCR系统NO _x 排放预测模型研究[J]．热能动力工程，2020，35(10)：57-64．
	WEN X， QIAN Y L， PENG D G，et al ．NO _x emission prediction model of SCR system based on deep bidirectional LSTM[J]．Journal of Engineering for Thermal Energy and Power，2020，35(10)：57-64．

基于双向长短期记忆网络和最小二乘支持向量机模型的SCR脱硝反应器进口截面NO _x 浓度预测

Prediction of NO _x Concentration at Inlet Section of SCR Denitrification Reactor Based on Bidirectional Long Short-Term Memory Network and Least Squares Support Vector Machine Model

RichHTML

PDF (PC)