基于TCN-BiLSTM的脱硫系统出口SO<sub>2</sub>浓度动态预测模型

王峰; 周倜; 于威振; 邹鹏; 李柏毅; 刘少杰; 赵征

doi:10.13205/j.hjgc.202609012

基于TCN-BiLSTM的脱硫系统出口SO₂浓度动态预测模型

doi: 10.13205/j.hjgc.202609012

王峰¹,
周倜¹,
于威振¹,
邹鹏¹,
李柏毅²,
刘少杰²,
赵征^3, ,

1. 中电华创(苏州)电力技术研究有限公司, 江苏苏州 215000;
2. 山西中电神头第二发电有限责任公司, 山西朔州 036000;
3. 华北电力大学自动化系, 河北保定 071003

基金项目:

中国电力国际发展有限公司科技项目（ZGDL-KJ-2024-C022）；深圳市科技计划项目专2021N041 “生活垃圾焚烧智能优化控制及污染物超低排放技术研发”（KCXFZ20201221173402007）

详细信息

作者简介:
王峰(1993—),男,硕士,主要研究方向为电力环境保护。wangfeng1227@foxmail.com

通讯作者:
赵征(1978—),女,副教授,主要研究方向为热工过程建模、新能源发电。zheng_zhao@ncepu.edu.cn

计量
- 文章访问数: 4
- HTML全文浏览量: 0
- PDF下载量: 0
- 被引次数: 0
出版历程
- 收稿日期: 2025-09-08
- 网络出版日期: 2026-11-05

A dynamic prediction model of SO₂ concentration at the exit of desulfurization system based on TCN-BiLSTM

1. China Power Huachuang (Suzhou) Electric Power Technology Research Co., Ltd, Suzhou 215000, China;
2. Shanxi Zhongdian Shentou Second Power Generation Co., Ltd, Shuozhou 036000, China;
3. Department of Automation, North China Electric Power University, Baoding 071003, China

摘要

摘要: 火电厂脱硫系统出口SO₂浓度的精准预测对运行优化和污染控制至关重要。但系统具有非线性、滞后性及多干扰因素，导致出口SO₂浓度难以准确预测。针对该问题，提出了一种基于时域卷积网络（TCN）和双向长短期记忆网络（BiLSTM）的脱硫系统出口SO₂浓度动态预测模型。首先，通过脱硫机理与影响因素分析，初步筛选特征变量；接着利用组合特征提取算法筛选与出口SO₂浓度高度相关的变量，去除冗余项；随后，通过黑翅鸢算法（BKA）智能优化输入与输出变量的阶次组合，确定模型最佳输入变量集；最后，基于TCN-BiLSTM深度学习算法构建脱硫系统出口SO₂浓度动态预测模型。基于电厂实际运行数据的实验结果表明：经过变量筛选与阶次优化后的模型在精度上明显提高，所提出的TCN-BiLSTM模型在性能上表现优异，其最佳平均绝对百分比误差为0.92%，平均绝对误差为0.15，均方根误差为0.24，与其他模型相比，在各项误差性能指标和预测效果上均有显著提升，展现出更高的预测精度和更强的拟合能力，可有效满足实际现场需求，实现脱硫系统出口SO₂浓度实时预测，从而为SO₂排放优化控制提供参考。
- 组合特征提取 /
- 时域卷积网络 /
- 双向长短期记忆网络 /
- 黑翅鸢算法 /
- SO₂浓度 /
- 动态预测
Abstract: The accurate prediction of SO₂ concentration at the outlet of the desulfurization system in thermal power plants is crucial for operation optimization and pollution control. However， due to its nonlinearity， lag and multiple interfering factors， it is difficult to accurately predict the SO₂ concentration at the outlet. A dynamic prediction model of SO₂ concentration at the exit of the desulfurization system based on a temporal convolutional network （TCN） and bidirectional long short-term memory network （BiLSTM） was proposed. Firstly， the characteristic variables were selected by analyzing the mechanism and influencing factors of desulphurization. Then， the combined feature extraction algorithm was used to screen the variables highly related to the outlet SO₂ concentration， and the redundant items were removed. Then， the optimal set of input variables was determined by using the black-winged kite algorithm （BKA） to intelligently optimize the order combination of input and output variables. Finally， based on the TCN-BiLSTM deep learning algorithm， the dynamic prediction model of SO₂ concentration at the exit of the desulfurization system was constructed. Based on the actual operation data of the power plant， the experimental results showed that： after variable screening and order optimization， the accuracy of the model was significantly improved. The proposed TCN-BiLSTM model had excellent performance， with the best average absolute percentage error of 0.94%， the average absolute error of 0.15， and the root-mean-square error of 0.24. Compared with other models， the error performance index and prediction effect had been significantly improved， showing higher prediction accuracy and stronger fitting ability， which can effectively meet the actual field needs， to achieve real-time prediction of SO₂ concentration at the exit of the desulfurization system， and to provide a reference basis for the optimization control of SO₂ emission.
- combined feature extraction /
- temporal convolutional network /
- bidirectional long short-term memory network /
- black-winged kite algorithm /
- SO₂ concentration /
- dynamic prediction

HTML全文

参考文献(23)

[1]	JIN X Z，SHI D J，QIAO P. Dynamic soft measurement model for SO₂ generation in coal-fired units based on equilibrium optimizer algorithm-radial basis function neural network［J］. Journal of North China Electric Power University（Natural Science Edition），2025，52（6）:123-132. 金秀章，史德金，乔鹏. 基于平衡优化算法-径向基神经网络的燃煤机组SO₂浓度动态软测量模型［J］. 华北电力大学学报（自然科学版），2025，52（6）:123-132.
[2]	JIN X Z，CHEN J Z，ZHANG J. Prediction of SO₂ concentration at outlet based on variable screening and OS-KELM［J/OL］. Journal of North China Electric Power University（Natural Science Edition）:1-10［ 2025-8-11］. DOI:10.3969/j.ISSN.1007-2691.2026.01.14. 金秀章，陈佳政，张瑾. 基于变量筛选和OS-KELM的出口SO₂浓度预测［J/OL］. 华北电力大学学报（自然科学版）:1-10［ 2025-8-11］. DOI:10.3969/j.ISSN.1007-2691.2026.01.14.
[3]	WANG Y，CHEN X，ZHAO C. A data-driven soft sensor model for coal-fired boiler SO₂ concentration prediction with non-stationary characteristic［J］. Energy，2024，300:131522.
[4]	JIN X Z，LIU Y，YU J，et al. Prediction of outlet SO₂ concentration based on variable selection and EMD-LSTM network［J］. Proceedings of the CSEE，2021，41（24）:8475-8484. 金秀章，刘岳，于静，等. 基于变量选择和EMD-LSTM网络的出口SO₂浓度预测［J］. 中国电机工程学报，2021，41（24）:8475-8484.
[5]	WEI G Y，GUO T F. Prediction of SO₂ concentration at absorber outlet based on variable and secondary decomposition［J］. Journal of North China Electric Power University（Natural Science Edition），2026，53（3）:147-158. 韦根原，郭腾飞. 基于变量选择和二次分解的吸收塔出口SO2浓度预测［J］. 华北电力大学学报（自然科学版），2026，53（3）:147-158.
[6]	WANG Q，CHAI Y H，WANG P C，et al. Prediction of SO₂ emission mass concentration based on improved INFO-Bi-LSTM algorithm［J］. Journal of the Chinese Society of Power Engineering，2024，44（4）:641-649. 王琦，柴宇唤，王鹏程，等. 基于改进INFO-Bi-LSTM模型的SO₂排放质量浓度预测［J］. 动力工程学报，2024，44（4）:641-649.
[7]	WANG Q，ZHAO H，ZHAO Q，et al. Prediction of SO₂ emission concentration in industrial flue gas based on deep learning:The ammonia desulfurization system of the Yunnan Aluminum Carbon Plant as the research object［J］. Process Safety and Environmental Protection，2024，185:340-349.
[8]	ZHAO Z，ZHOU Z Y，LU Y，et al. Dynamic modeling method for the combustion process of grate-type waste incinerators［J］. Proceedings of the CSEE，2025，45（11）:4360-4372. 赵征，周孜钰，卢叶，等. 炉排式垃圾焚烧炉燃烧过程动态建模方法研究［J］. 中国电机工程学报，2025，45（11）:4360-4372.
[9]	YAN H S，ZHAO W J. Prediction of desulfurization outlet SO₂ concentration based on MIC and MPA-KELM［J］. Acta Metrologica Sinica，2023，44（2）:271-278. 闫浩思，赵文杰. 基于MIC和MPA-KELM的脱硫出口SO₂浓度预测［J］. 计量学报，2023，44（2）:271-278.
[10]	YAO N，JIN X Z，LI Y F. NO_x modeling of denitrification system based on Bi-LSTM optimized by improved whale optimization algorithm［J］. Journal of North China Electric Power University（Natural Science Edition），2022，49（6）:76-83. 姚宁，金秀章，李阳峰. 基于改进鲸鱼算法优化Bi-LSTM的脱硝系统NO_x 建模［J］. 华北电力大学学报（自然科学版），2022，49（6）:76-83.
[11]	LU W P，WANG Y W，ZONG S L. Analysis of influencing factors of limestone-gypsum wet flue gas desulfurization process［J］. Energy Conservation，2024，43（4）:61-63. 陆万鹏，王雅雯，宗绍梁. 石灰石-石膏湿法脱硫工艺影响因素分析［J］. 节能，2024，43（4）:61-63.
[12]	GAO J，TIAN X F，PENG X Y，et al. Prediction model of desulfurization system pH based on bidirectional gated recurrent unit［J］. Science Technology and Engineering，2025，25（13）:5535-5543. 高钾，田雪峰，彭献永，等. 基于双向门控循环单元的脱硫系统pH预测模型［J］. 科学技术与工程，2025，25（13）:5535-5543.
[13]	LI R L，ZENG D L，CHEN Z Y，et al. Modeling and optimization control of single-tower double-circulation wet flue gas desulfurization system［J］. Journal of Chinese Society of Power Engineering，2025，45（3）:399-409. 李瑞连，曾德良，陈振宇，等. 单塔双循环湿式烟气脱硫系统建模与优化控制［J］. 动力工程学报，2025，45（3）:399-409.
[14]	WANG J，WANG W C，HU X X，et al. Black-winged kite algorithm:a nature-inspired meta-heuristic for solving benchmark functions and engineering problems［J］. Artificial Intelligence Review，2024，57（4）:1-53.
[15]	ZHANG M，HAN Y，WANG C，et al. Ultra-short-term photovoltaic power prediction based on similar day clustering and temporal convolutional network with bidirectional long short-term memory model:A case study using DKASC data［J］. Applied Energy，2024，375:124085.
[16]	SHI P M，GUO X Y，DU Q C，et al. Photovoltaic power prediction based on TCN-BiLSTM-Attention-ESN［J］. Acta Energiae Solaris Sinica，2024，45（9）:304-316. 时培明，郭轩宇，杜清灿，等. 基于TCN-BiLSTM-Attention-ESN的光伏功率预测［J］. 太阳能学报，2024，45（9）:304-316.
[17]	LI Z Y，ZHANG X W，GAO W. State of health estimation of lithium-ion battery during fast charging process based on BiLSTM-Transformer［J］. Energy，2024，311:133418.
[18]	HUANG X B，WANG Y K. Prediction of power plant NOx emission based on improved dung beetle optimization algorithm［J］. Science Technology and Engineering，2024，24（23）:9929-9936. 黄孝彬，王永凯. 基于改进蜣螂优化算法的电站NOx排放预测［J］. 科学技术与工程，2024，24（23）:9929-9936.
[19]	LIANG J，SONG J G，CHEN J M，et al. Modeling and analysis of operation performance evaluation for measurement and control equipment［J］. Chinese Journal of Engineering Mathematics，2023，40（4）:523-544. 梁军，宋建国，陈嘉铭，等. 测控设备运行效能评估建模及分析［J］. 工程数学学报，2023，40（4）:523-544.
[20]	HANG S，DONG Z，ASGHAR A H，et al. RIME:A physics-based optimization［J］. Neurocomputing，2023，532:183-214.
[21]	XUE J K，SHEN B. Dung beetle optimizer:a new meta-heuristic algorithm for global optimization［J］. The Journal of Supercomputing，2022，79（7）:7305-7336.
[22]	WU Q，WANG X H，ZENG Y F，et al. Method for borehole elastic wave processing and coal-rock identification based on EWOA-VMD［J］. Journal of China University of Mining& Technology，2024，53（5）:857-871. 武强，王显辉，曾一凡，等. 基于EWOA-VMD的钻孔弹性波处理与煤岩识别方法［J］. 中国矿业大学学报，2024，53（5）:857-871.
[23]	PAVEL T，MOHAMMAD D. Pelican optimization algorithm:a novel nature-inspired algorithm for engineering applications［J］. Sensors，2022，22（3）:855-855.