NO x emission prediction method of CFB unit based on 1D mechanism model dynamicly corrected with LSTM

doi:10.11949/0438-1157.20241379

Abstract

Abstract:

The current monitoring methods for NO _x emission concentration of CFB unit still have some shortcomings. Mechanism based NO _x emission concentration prediction methods may have large prediction errors under certain operating conditions different from the modeling conditions. Machine learning based methods have high prediction accuracy, but lack physical significance and have poor interpretability. To this end, a fusion model for predicting NO _x emission concentration of CFB unit is proposed. Firstly, a one-dimensional circulating fluidized bed overall semi-empirical model is constructed to simulate the combustion in the furnace to give the initial prediction value of NO _x emission concentration. Secondly, an error correction model is constructed based on the long short-term memory artificial neural network to dynamically correct the initial prediction value. Taking two CFB units as research objects, the results show that the proposed model is superior to the single one-dimensional semi empirical model, as well as models such as long short-term memory network and BP neural network. The combination of mechanism and machine learning methods enables the fusion model to have both high prediction accuracy and good interpretability.

Key words: circulating fluidized bed, dynamic modeling, measurement, NO _x emission concentration, LSTM neural network

摘要：

CFB机组NO _x 排放浓度实时监测方法的不足主要表现在：基于机理模型的NO _x 排放浓度预测方法在某些不同于其建模条件的工况下，预测误差较大；基于机器学习模型的方法预测精度高，但缺乏物理意义，可解释性差。为此，提出一种CFB机组NO _x 排放浓度预测融合模型，首先构建一维循环流化床整体半经验模型模拟炉内燃烧给出NO _x 排放浓度初始预测值，其次基于长短期记忆人工神经网络构建误差校正模型，对初始预测值进行动态修正。以两台CFB机组为研究对象，结果表明，提出的模型优于单一的一维半经验模型以及长短期记忆网络、BP神经网络等模型。机理与机器学习方法的结合使得融合模型既具备高的预测精度，又具有良好的可解释性。

关键词: 循环流化床, 动态建模, 测量, NO _x 排放浓度, LSTM神经网络

CLC Number:

TK 39

Fang WANG, Suxia MA, Ying TIAN, Zhongyuan LIU. NO _x emission prediction method of CFB unit based on 1D mechanism model dynamicly corrected with LSTM[J]. CIESC Journal, 2025, 76(7): 3416-3425.

王芳, 马素霞, 田营, 刘众元. 基于LSTM动态修正一维机理模型的CFB机组NO _x 排放浓度预测方法[J]. 化工学报, 2025, 76(7): 3416-3425.

Figures/Tables 14

Fig.1 Cell division in CFB model

Table 1 Chemical reactions and reaction rates

化学反应	反应速率	系数
$C O + 12 O 2 → C O 2$	$R C O = k C C O C O 2 0.5$	$k = 1.0 × 1012 e x p - 16000 T$
$H C N + 12 O 2 → N C O + 12 H 2$	$R H C N = k C O 2 C H C N$	$k = 2.14 × 105 e x p - 10000 T$
$N C O + 12 O 2 → N O + C O$	$R N C O = k C O 2 C H C N k 1 k 1 + k 2 C N O$	$k 2 k 1 = 1.02 × 109 e x p - 25460 T$
$N H 3 + 54 O 2 → N O + 32 H 2 O$	$R N H 3, 1 = k C O 2 C N H 3$	$k = 2.73 × 1014 e x p - 38160 T$
$N H 3 + 34 O 2 → 12 N 2 + 32 H 2 O$	$R N H 3, 2 = k C O 2 C N H 3 C O 2 + 0.054$	$k = 3.38 × 107 e x p - 10000 T$
$N O + C O → 12 N 2 + C O 2$	$R N O - C O = K T k 1 C N O k 2 C C O + k 3 k 1 C N O + k 2 C C O + k 3$	$K T = 1.952 × 1010 e x p - 19000 T$
$2 N O + C → N 2 + C O 2$	$R N O - C = π N d c 2 k C N O$	$k = 1.3 × 105 e x p - 17111 T$

Table 1 Chemical reactions and reaction rates

化学反应	反应速率	系数
$C O + 12 O 2 → C O 2$	$R C O = k C C O C O 2 0.5$	$k = 1.0 × 1012 e x p - 16000 T$
$H C N + 12 O 2 → N C O + 12 H 2$	$R H C N = k C O 2 C H C N$	$k = 2.14 × 105 e x p - 10000 T$
$N C O + 12 O 2 → N O + C O$	$R N C O = k C O 2 C H C N k 1 k 1 + k 2 C N O$	$k 2 k 1 = 1.02 × 109 e x p - 25460 T$
$N H 3 + 54 O 2 → N O + 32 H 2 O$	$R N H 3, 1 = k C O 2 C N H 3$	$k = 2.73 × 1014 e x p - 38160 T$
$N H 3 + 34 O 2 → 12 N 2 + 32 H 2 O$	$R N H 3, 2 = k C O 2 C N H 3 C O 2 + 0.054$	$k = 3.38 × 107 e x p - 10000 T$
$N O + C O → 12 N 2 + C O 2$	$R N O - C O = K T k 1 C N O k 2 C C O + k 3 k 1 C N O + k 2 C C O + k 3$	$K T = 1.952 × 1010 e x p - 19000 T$
$2 N O + C → N 2 + C O 2$	$R N O - C = π N d c 2 k C N O$	$k = 1.3 × 105 e x p - 17111 T$

Fig.2 Flow chart of the GA-1D model

Fig.3 Boiler structure diagram

Fig.4 Sensitivity analysis results

Fig.5 RMSE results with different numbers of neurons in hidden layer

Fig.6 RMSE results with different look-back timesteps

Fig.7 Fusion model flowchart

Fig.8 Comparison between simulated and actual bed pressure values

Fig.9 Distribution of NO x, NH3, HCN along the height of the furnace

Fig.10 Comparison between predicted values and actual values of three models

Table 2 Prediction performance comparison of three models

模型	RMSE	R²	MAE
1D model	15.70	0.94	12.24
LSTM model	12.80	0.96	9.88
1D-LSTM model	10.76	0.97	8.47

Table 3 Prediction performance comparison of five models

模型	RMSE	R²	MAE
BP	13.96	0.95	11.74
ELM	13.17	0.95	11.05
1D-BP	13.25	0.95	10.95
1D-ELM	12.12	0.96	9.88
1D-LSTM	10.76	0.97	8.47

Fig.11 Comparison between predicted values and actual values of five models

References 31

[1]	Wang X W, Liu W J, Wang Y N, et al. A hybrid NO _x emission prediction model based on CEEMDAN and AM-LSTM[J]. Fuel, 2022, 310: 122486.
[2]	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.
[3]	Tan P, He B, Zhang C, et al. Dynamic modeling of NO _x emission in a 660 MW coal-fired boiler with long short-term memory[J]. Energy, 2019, 176: 429-436.
[4]	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.
[5]	任少君, 朱保宇, 翁琪航, 等. 基于物理信息神经网络的燃煤锅炉NO _x 排放浓度预测方法[J]. 中国电机工程学报, 2024, 44(20): 8157-8166.
	Ren S J, Zhu B Y, Weng Q H, et al. Forecasting method for NO _x emission in coal fired boiler based on physics-informed neural network[J]. Proceedings of the CSEE, 2024, 44(20): 8157-8165.
[6]	Azodi C B, Tang J L, Shiu S H. Opening the black box: interpretable machine learning for geneticists[J]. Trends in Genetics, 2020, 36(6): 442-455.
[7]	赵小军, 王学斌, 孙锦余, 等. 300MW电站煤粉锅炉耦合掺烧生物质的CFD数值模拟[J]. 洁净煤技术, 2022, 28(3): 56-64.
	Zhao X J, Wang X B, Sun J Y, et al. Numerical simulation of biomass coupled by pulverized coal boiler in 300MW power station[J]. Clean Coal Technology, 2022, 28(3): 56-64.
[8]	李佳伟, 陈智超, 张旭阳, 等. 预燃室燃烧器煤气化细灰混烧特性和数值模拟研究[J]. 工程热物理学报, 2022, 43(6): 1675-1683.
	Li J W, Chen Z C, Zhang X Y, et al. Study on co-combustion characteristics and numerical simulation of coal gasification fine ash in precombustion chamber burner[J]. Journal of Engineering Thermophysics, 2022, 43(6): 1675-1683.
[9]	Wang S, Luo K, Yang S L, et al. LES-DEM investigation of the time-related solid phase properties and improvements of flow uniformity in a dual-side refeed CFB[J]. Chemical Engineering Journal, 2017, 313: 858-872.
[10]	Tu Q Y, Wang H G, Ocone R. Application of three-dimensional full-loop CFD simulation in circulating fluidized bed combustion reactors—a review[J]. Powder Technology, 2022, 399: 117181.
[11]	Muhammad A, Zhang N, Wang W. CFD simulations of a full-loop CFB reactor using coarse-grained Eulerian-Lagrangian dense discrete phase model: effects of modeling parameters[J]. Powder Technology, 2019, 354: 615-629.
[12]	Bandara J C, Jayarathna C, Thapa R, et al. Loop seals in circulating fluidized beds—review and parametric studies using CPFD simulation[J]. Chemical Engineering Science, 2020, 227: 115917.
[13]	Li S Y, Shen Y S. Multi-fluid modelling of hydrodynamics in a dual circulating fluidized bed[J]. Advanced Powder Technology, 2020, 31(7): 2778-2791.
[14]	Ke X W, Engblom M, Yang H R, et al. Prediction and minimization of NO _x emission in a circulating fluidized bed combustor: a comprehensive mathematical model for CFB combustion[J]. Fuel, 2022, 309: 122133.
[15]	马达夫, 何翔, 吕为智, 等. 660 MW超临界W火焰锅炉低负荷稳燃特性研究[J]. 工程热物理学报, 2022, 43(1): 259-266.
	Ma D F, He X, Lyu W Z, et al. Investigations of combustion stability in a 660 MW supercritical W-flame boiler under low load[J]. Journal of Engineering Thermophysics, 2022, 43(1): 259-266.
[16]	李政. 循环流化床锅炉通用整体数学模型、仿真与性能预测[D]. 北京: 清华大学, 1994.
	Li Z. Modeling, simulation and performance prediction of a complete CFB boiler[D]. Beijing: Tsinghua University, 1994.
[17]	高明明, 岳光溪, 雷秀坚, 等. 循环流化床锅炉石灰石控制研究[J]. 动力工程学报, 2014, 34(10): 759-764, 777.
	Gao M M, Yue G X, Lei X J, et al. Research on limestone control of circulating fluidized bed boiler[J]. Journal of Chinese Society of Power Engineering, 2014, 34(10): 759-764, 777.
[18]	单露, 张缦, 张翼, 等. 循环流化床全回路气固流动动态模型及分析[J]. 中国电机工程学报, 2017, 37(S1): 98-104.
	Shan L, Zhang M, Zhang Y, et al. Dynamic model establishment and analysis on gas-solid flow in CFB whole loop[J]. Proceedings of the CSEE, 2017, 37(S1): 98-104.
[19]	Yan J, Lu X F, Wang Q H, et al. Study on the influence of secondary air on the distributions of flue gas composition at the lower part of a 600 MW supercritical CFB boiler[J]. Fuel Processing Technology, 2019, 196: 106035.
[20]	Kunii D. Flow modeling of fast fluidized beds[M]// Circulating Fluidized Bed Technology. 3rd ed. New York: Pergamon Press, 1991: 91-98.
[21]	Pallarès D, Johnsson F. Macroscopic modelling of fluid dynamics in large-scale circulating fluidized beds[J]. Progress in Energy and Combustion Science, 2006, 32(5/6): 539-569.
[22]	Yang H R, Yue G X, Xiao X B, et al. 1D modeling on the material balance in CFB boiler[J]. Chemical Engineering Science, 2005, 60(20): 5603-5611.
[23]	Nikolopoulos A, Malgarinos I, Nikolopoulos N, et al. A decoupled approach for NO _x -N₂O 3-D CFD modeling in CFB plants[J]. Fuel, 2014, 115: 401-415.
[24]	Desroches-Ducarne E, Dolignier J C, Marty E, et al. Modelling of gaseous pollutants emissions in circulating fluidized bed combustion of municipal refuse[J]. Fuel, 1998, 77(13): 1399-1410.
[25]	Kilpinen P, Kallio S, Konttinen J, et al. Char-nitrogen oxidation under fluidised bed combustion conditions: single particle studies[J]. Fuel, 2002, 81(18): 2349-2362.
[26]	Pan Q Q, Zheng S C, Liu X J. Deep-coupling neural network and genetic algorithm based on Sobol-PR for reactor lightweight optimization[J]. Applied Soft Computing, 2024, 167: 112458.
[27]	闫琳琦, 王振雷. 基于STA-BiLSTM-LightGBM组合模型的多步预测软测量建模[J]. 化工学报, 2023, 74(8): 3407-3418.
	Yan L Q, Wang Z L. Multi-step predictive soft sensor modeling based on STA-BiLSTM-LightGBM combined model[J]. CIESC Journal, 2023, 74(8): 3407-3418.
[28]	张中秋, 李宏光, 石逸林. 基于人工预测调控策略的复杂化工过程多任务学习方法[J]. 化工学报, 2023, 74(3): 1195-1204.
	Zhang Z Q, Li H G, Shi Y L. A multi-task learning approach for complex chemical processes based on manual predictive manipulating strategies[J]. CIESC Journal, 2023, 74(3): 1195-1204.
[29]	Knöbig T, Werther J, L-E Å, et al. Comparison of large- and small-scale circulating fluidized bed combustors with respect to pollutant formation and reduction for different fuels[J]. Fuel, 1998, 77(14): 1635-1642.
[30]	颜建国, 郑书闽, 郭鹏程, 等. 基于GA-BP神经网络的超临界CO₂传热特性预测研究[J]. 化工学报, 2021, 72(9): 4649-4657.
	Yan J G, Zheng S M, Guo P C, et al. Prediction of heat transfer characteristics for supercritical CO₂ based on GA-BP neural network[J]. CIESC Journal, 2021, 72(9): 4649-4657.
[31]	代学志, 熊伟丽. 基于核极限学习机的快速主动学习方法及其软测量应用[J]. 化工学报, 2020, 71(11): 5226-5236.
	Dai X Z, Xiong W L. A fast active learning method based on kernel extreme learning machine and its application for soft sensing[J]. CIESC Journal, 2020, 71(11): 5226-5236.