Dioxin Emission Concentration Modeling Based on Simulation Mechanism and Improved Linear Regression Decision Tree
-
摘要: 城市固废焚烧(Municipal solid waste incineration, MSWI)过程是“世纪之毒”二噁英(Dioxin, DXN)的重要排放源之一. 截止目前为止, DXN的演化机理和实时检测仍是尚未解决的难题. 现有研究主要基于离线化验数据构建数据驱动模型, DXN的检测未有效结合燃烧过程机理. 针对该问题, 本文提出基于仿真机理和改进线性回归决策树(Linear regression decision tree, LRDT)的DXN排放建模. 首先, 采用基于床层固废燃烧模拟软件FLIC (Fluid dynamic incinerator code)和过程工程先进系统软件(Advanced system for process engineering Plus, Aspen Plus)耦合的数值仿真模型, 获取蕴含多运行工况的虚拟机理数据; 接着, 利用虚拟机理数据构建基于改进LRDT的CO2、CO和O2燃烧状态表征变量模型; 然后, 以真实CO2、CO、O2作为输入和以DXN真值作为输出, 构建多入单出LRDT的过程映射模型(Process mapping model, PMM), 再利用该模型进行半监督学习和结构迁移得到机理映射模型1 (Mechanism mapping models1, MMM1); 最后, 通过结构增量学习获得基于半监督迁移学习的MMM2模型. 在实验室的半实物平台和北京某MSWI厂的边侧验证平台对所提方法进行了工业应用验证. 实验结果证明了所提方法与研发的软测量系统可有效实现二噁英排放浓度在线检测.Abstract: Municipal solid waste incineration (MSWI) process has been one of the important emission sources of dioxin (DXN) in terms of century posion. Untill now, the evolution mechanism and real-time detection of DXN emission concentration are still unsolved challenges. Existing studies mainly rely on available data to build data-driven modeling, and how to effectively combine the mechanism of combustion process for DXN detection is a problem that is not considered. To solve this problem, this article proposes DXN emission modelling method based on simulation mechanism and improved linear regression decision tree (LRDT). First, a numerical simulation model based on coupling fluid dynamic incinerator code (FLIC) and advanced system for process engineering Plus (Aspen Plus) software is used to obtain virtual mechanism data with multiple operating conditions. Then, virtual mechanism data is used to construct an improved LRDT combustion state representation variable CO2, CO, and O2 model. Next, a process mapping model (PMM) based on multiple input single output LRDT is constructed using real CO2, CO, and O2 as input and DXN as output. Semi-supervised learning and structural transfer learning based on PMM are used to obtain the mechanism mapping models1 (MMM1). Finally, the final MMM2 based on semi-supervised transfer learning is obtained by the structural growth learning of the MMM1. The proposed method was validated for industrial application on a hardware-in-loop simulation platform in the laboratory and an edge verification platform at an MSWI plant in Beijing. The experimental results show that the proposed method and the developed soft measurement system can effectively realize the on-line detection of DXN emission concentration.
-
图 6 基于基准工况的固相燃烧结果图
Fig. 6 Solid phase combustion results based on benchmark conditions
图 8 虚拟机理数据异常值去除前后的结果
Fig. 8 Results of before and after removal of outliers from virtual mechanism data
图 11 基于伪标记机理数据的模型测试曲线
Fig. 11 Model testing curves based on pseudo-labeled mechanical data
图 13 MSWI过程半实物仿真平台DXN检测软件界面
Fig. 13 DXN testing software interface of hardware-in-loop simulation platform of MSWI process
图 14 北京某MSWI 厂的基于安全隔离采集设备的边侧验证平台
Fig. 14 Edge verification platform with secure isolation acquisition equipment at an MSWI factory in Beijing
表 1 MSW成分分析
Table 1 Analysis of MSW components
分析项 值 单位 工业分析 水分 (ar) 38.48 wt% 挥发性 (ar) 41.80 wt% 固定碳 (ar) 6.56 wt% 灰烬 (ar) 13.16 wt% 元素分析 C (daf) 64.31 wt% H (daf) 9.91 wt% N (daf) 24.93 wt% S (daf) 0.51 wt% O (daf) 0.34 wt% 
下载: 导出CSV
表 2 焚烧炉基本情况
Table 2 Basic information about incinerators
参数 值 单位 额定产能 800 t/d 实际产能 624 t/d 炉排 往复式顺推 / 长 × 宽 11 × 12.9 m 速度 8 m/h 一次风量 65400 ${\rm{m}}^3$/h 二次风量 7500 ${\rm{m}}^3$/h 一次风温度 200 ℃ 一次风在干燥段的风量分布比例 24.31 % 一次风在燃烧一段的风量分布比例 43.35 % 一次风在燃烧二段的风量分布比例 19.27 % 一次风在燃烬段的风量分布比例 13.07 %
下载: 导出CSV
表 3 正交实验参数信息
Table 3 Orthogonal experimental parameter information
10因素5水平 5因素5水平 参数 因素 单位 水平-1 水平-2 操作参数 炉排速度 m/h 7, 7.5, 8, 8.5, 9 −0.1 给料量 t/h 24.2, 24.7, 25.2, 25.7, 26.2 +0.1 第1区域进风 ${\rm{m}}^3$/h 16080 ,16440 ,16800 ,17160 ,17520 +1.8 第2区域进风 ${\rm{m}}^3$/h 28620 ,29280 ,29940 ,30600 ,31260 +3.2 第3区域进风 ${\rm{m}}^3$/h 12660 ,12960 ,13260 ,13560 ,13860 +1.4 第4区域进风 ${\rm{m}}^3$/h 8640 ,8820 ,9000 ,9180 ,9360 +1 微观参数 颗粒大小 mm 15, 20, 25, 30, 35 / 颗粒混合系数 / $2{\rm{e}}{-6}$, $3{\rm{e}}{-6}$, $4{\rm{e}}{-6}$, $5{\rm{e}}{-6}$, $6{\rm{e}}{-6}$ / 组分参数 水分含量 % 48, 49.75, 51.5, 53.25, 55 / C : H : O 比率 % (58 : 7.5 : 33), (59 : 7.5 : 32), (60 : 7.5 : 31), (61 : 7.5 : 30), (62 : 7.5 : 29) /
下载: 导出CSV
表 4 机理数据的不同方法性能比较结果
Table 4 Results of performance comparison between different methods of mechanism data
方法 目标值 训练集 测试集 RMSE ${\rm{R}}^2$ RMSE ${\rm{R}}^2$ DT ${\rm{CO}}_2$ 0.2688 0.9702 0.6153 0.8457 CO 0.9519 0.9659 1.8184 0.8751 ${{\rm{O}}_2}$ 0.2811 0.9710 0.6536 0.8446 RDT ${\rm{CO}}_2$ 0.5400 0.8798 0.6237 0.8414 CO 2.2356 0.8117 2.5335 0.7575 ${{\rm{O}}_2}$ 0.6752 0.8325 0.7862 0.7751 RR ${\rm{CO}}_2$ 1.40025 0.1918 1.3945 0.2072 CO 4.9986 0.0586 4.9738 0.0652 ${{\rm{O}}_2}$ 1.4306 0.2481 1.4233 0.2630 MISO LRDT ${\rm{CO}}_2$ 0.4138 0.9294 0.5894 0.8584 CO 1.3046 0.9359 1.7057 0.8901 ${{\rm{O}}_2}$ 0.4282 0.9326 0.5487 0.8905 MIMO LRDT ${\rm{CO}}_2$ 0.1645 0.9357 0.3089 0.8869 CO 0.2220 0.9357 0.2991 0.8867 ${{\rm{O}}_2}$ 0.4056 0.9812 0.5558 0.9747
下载: 导出CSV
表 5 基于伪标记机理数据的模型统计结果
Table 5 Model statistical results based on Pseudo-labeling mechanism data
方法 训练集 测试集 RMSE ${\rm{R}}^2$ RMSE ${\rm{R}}^2$ PMM 0.0020 0.8021 0.0020 0.8072 MMM1 0.0015 0.8909 0.0015 0.8918 MMM2 0.0014 0.8960 0.0015 0.8965
下载: 导出CSV
表 6 缩写词说明
Table 6 Abbreviation description
缩写词 英文全称 中文全称 MSWI Municipal solid waste incineration 城市固废焚烧 DXN Dioxin 二噁英 SNCR Selective non-catalytic reduction 选择性非催化还原 FLIC Fluid dynamic incinerator code 床层固废燃烧模拟软件 Aspen Plus Advanced system for process engineering Plus 过程工程先进系统 LRDT Linear regression decision tree 线性回归决策树 PMM Process mapping model 过程映射模型 MMM1 Mixed-driven models1 混合驱动模型1 DD Data-driven 数据驱动 MD Mechanism-driven 机理驱动 MIMO Multiple-in multiple-out 多入多出 PICs Products of incomplete combustion 不完全燃烧产物 LHV Lower heat value 低热值 PA Primary airflow 一次风量 FC Feeding capacity 给料量 QAF Quartile abnormal filter 四分位异常滤波 DCS Distributed control system 集散控制系统 CmHn Hydrocarbons 碳氢化合物 MISO Multiple-input and single-output 多输入单输出 CART Classification and regression tree 分类回归树 MSE Mean squared error 均方误差 MMM Mechanism mapping model 机理映射模型 DT Decision tree 决策树 RDT Random decision tree 随机决策树 RR Ridge regression 岭回归 RF Random forest 随机森林 RMSE Root mean square error 均方根误差 ${\rm{R}}^2$ Coefficient of determination 决定系数
下载: 导出CSV
-
[1] Xia H, Tang J, Aljerf L, Wang T Z, Gao B, Xu Q D, et al. Assessment of PCDD/Fs formation and emission characteristics at a municipal solid waste incinerator for one year. The Science of the Total Environment, 2023, 883 : Article No. 163705 [2] Gómez-Sanabria A, Kiesewetter G, Klimont Z, Schoepp W, Haberl H. Potential for future reductions of global GHG and air pollutants from circular waste management systems. Nature Communications, 2022, 13 : Article No. 106 [3] 乔俊飞, 郭子豪, 汤健. 面向城市固废焚烧过程的二噁英排放浓度检测方法综述. 自动化学报, 2020, 46(6): 1063−1089Qiao Jun-Fei, Guo Zi-Hao, Tang Jian. Dioxin emission concentration measurement approaches for municipal solid wastes incineration process: A survey. Acta Automatica Sinica, 2020, 46(6): 1063−1089 [4] 汤健, 夏恒, 余文, 乔俊飞. 城市固废焚烧过程智能优化控制研究现状与展望. 自动化学报, 2023, 49(10): 2019−2059Tang Jian, Xia Heng, Yu Wen, Qiao Jun-Fei. Research status and prospects of intelligent optimization control for municipal solid waste incineration process. Acta Automatica Sinica, 2023, 49(10): 2019−2059 [5] Yang Z Y, Ge Z Q. On paradigm of industrial big data analytics: From evolution to revolution. IEEE Transations on Industrial Informatics, 2022, 18(12): 8373−8388 doi: 10.1109/TII.2022.3190394 [6] Lu S W, Chai T Y. Mesoscale particle size predictive model for operational optimal control of bauxite ore grinding process. IEEE Transations on Industrial Informatics, 2020, 16(12): 7714−7721 doi: 10.1109/TII.2020.2967067 [7] Chang N B, Huang S H. Statistical modelling for the prediction and control of PCDDs and PCDFs emissions from municipal solid waste incinerators. Waste Management & Research, 1995, 13(4): 379−400 [8] Bunsan S, Chen W Y, Chen H W, Chuang Y H, Grisdanurak N. Modeling the dioxin emission of a municipal solid waste incinerator using neural networks. Chemosphere, 2013, 92(3): 258−264 doi: 10.1016/j.chemosphere.2013.01.083 [9] Tang J, Xia H, Zhang J, Qiao J F, Yu W. Deep forest regression based on cross-layer full connection. Neural Computing and Applications, 2021, 33: 9307−9328 doi: 10.1007/s00521-021-05691-7 [10] Xia H, Tang J, Yu W, Qiao J F. Online measurement of dioxin emission in solid waste incineration using fuzzy broad learning. IEEE Transations on Industrial Informatics, 2024, 20(1): 358−368 doi: 10.1109/TⅡ.2023.3259419 [11] 夏恒, 汤健, 崔璨麟, 乔俊飞. 基于宽度混合森林回归的城市固废焚烧过程二噁英排放软测量. 自动化学报, 2023, 49(2): 343−365Xia Heng, Tang Jian, Cui Can-Lin, Qiao Jun-Fei. Soft sensing method of dioxin emission in municipal solid waste incineration process based on broad hybrid forest regression. Acta Automatica Sinica, 2023, 49(2): 343−365 [12] Xia H, Tang J, Qiao J F, Zhang J, Yu W. DF classification algorithm for constructing a small sample size of data-oriented DF regression model. Neural Computing and Applications, 2022, 34: 2785−2810 doi: 10.1007/s00521-021-06809-7 [13] Wyper P, Antiochos S, DeVore C R. A universal model for solar eruptions. Nature, 2017, 544: 452−455 doi: 10.1038/nature22050 [14] 阳春华, 孙备, 李勇刚, 黄科科, 桂卫华. 复杂生产流程协同优化与智能控制. 自动化学报, 2023, 49(3): 528−539Yang Chun-Hua, Sun Bei, Li Yong-Gang, Huang Ke-Ke, Gui Wei-Hua. Cooperative optimization and intelligent control of complex production processes. Acta Automatica Sinica, 2023, 49(3): 528−539 [15] Sun Q Q, Ge Z Q. A survey on deep learning for data-driven soft sensors. IEEE Transations on Industrial Informatics, 2021, 17(9): 5853−5866 doi: 10.1109/TII.2021.3053128 [16] Ren J C, Liu D, Wan Y. Modeling and application of Czochralski silicon single crystal growth process using hybrid model of data-driven and mechanism-based methodologies. Journal of Process Control, 2021, 104: 74−85 doi: 10.1016/j.jprocont.2021.06.002 [17] 吴志伟, 柴天佑, 吴永建. 电熔镁砂产品单吨能耗混合预报模型. 自动化学报, 2013, 39 (12): 2002−2011Wu Zhi-Wei, Chai Tian-You, Wu Yong-Jian. A hybrid prediction model of energy consumption per ton for fused magnesia. Acta Automatica Sinica, 2013, 39(12): 2002−2011 [18] Ni Y L, Xu J N, Zhu C Y, Pei L. Accurate residual capacity estimation of retired LiFePO4 batteries based on mechanism and data-driven model. Applied Energy, 2022, 305 : Article No. 117922 [19] Meng Y M, Yu S S, Zhang J L, Qin J, Dong Z, Lu G C, et al. Hybrid modeling based on mechanistic and data-driven approaches for cane sugar crystallization. Journal of Food Engineering, 2019, 257: 44−55 doi: 10.1016/j.jfoodeng.2019.03.026 [20] Xiao D, Xie H, Jiang L, Le B T, Wang J, Liu C, et al. Research on a method for predicting the underflow concentration of a thickener based on the hybrid model. Engineering Applications of Computational Fluid Mechanics, 2020, 14(1): 13−26 doi: 10.1080/19942060.2019.1658228 [21] Dong X, Yan X, Qu H. Advanced process control for salvianolic acid a conversion reaction based on data-driven and mechanism-driven model. Process Biochemistry, 2022, 118: 1−10 doi: 10.1016/j.procbio.2022.04.001 [22] Ren J C, Liu D, Wan Y. Data-driven and mechanism-based hybrid model for semiconductor silicon monocrystalline quality prediction in the czochralski process. IEEE Transations on Semiconductor Manufacturing, 2022, 35(4): 658−669 doi: 10.1109/TSM.2022.3202610 [23] 朱鹏飞, 夏陆岳, 潘海天. 基于改进Kalman滤波算法的多模型融合建模方法. 化工学报, 2015, 66(4): 1388−1394Zhu Peng-Fei, Xia Lu-Yue, Pan Hai-Tian. Multi-model fusion modeling method based on improved Kalman filtering algorithm. CIESC Journal, 2015, 66(4): 1388−1394 [24] Zhuang Y L, Liu Y X, Ahmed A, Zhong Z G, Rio Chanona E A, Hale C P, et al. A hybrid data-driven and mechanistic model soft sensor for estimating CO2 concentrations for a carbon capture pilot plant. Computers in Industry, 2022, 143 : Article No. 103747 [25] 张梦轩, 刘洪辰, 王敏, 蓝兴英, 石孝刚, 高金森. 化工过程的智能混合建模方法及应用. 化工进展, 2021, 40(4): 1765−1776Zhang Meng-Xuan, Liu Hong-Chen, Wang Min, Lan Xing-Ying, Shi Xiao-Gang, Gao Jin-Sen. Intelligence hybrid modeling method and applications in chemical process. Chemical Industry and Engineering Progress, 2021, 40(4): 1765−1776 [26] Altarawneh M, Dlugogorski B, Kennedy E, Mackie J. Mechanisms for formation, chlorination, dechlorination and destruction of polychlorinated dibenzo-p-dioxins and dibenzofurans (PCDD/Fs). Progress in Energy and Combustion Science, 2009, 35(3): 245−274 doi: 10.1016/j.pecs.2008.12.001 [27] Hunsinger H, Jay K, Vehlow J. Formation and destruction of PCDD/F inside a grate furnace. Chemosphere, 2002, 46(9−10): 1263−1272 doi: 10.1016/S0045-6535(01)00256-9 [28] Peng Y Q, Lu S Y, Li X D, Yan J H, Cen K F. Formation, measurement, and control of dioxins from the incineration of municipal solid wastes: Recent advances and perspectives. Energy & Fuels, 2020, 34(11): 13247−13267 [29] Hasselriis F. Minimizing trace organic emissions from combustion of municipal solid waste by the use of carbon monoxide monitors. In: Proceedings of National Waste Processing Conference. Denver, 1986: 129−144 [30] Tillman D, Rossi A, Vick K. Controlling products of combustion. Incineration of Municipal and Hazardous Solid Wastes. London: Elsevier, 1989: 283−337 [31] Yang Y B, Yamauchi H, Nasserzadeh V, Swithenbank J. Effects of fuel devolatilisation on the combustion of wood chips and incineration of simulated municipal solid wastes in a packed bed. Fuel, 2003, 82(18): 2205−2221 doi: 10.1016/S0016-2361(03)00145-5 [32] Atsonios K, Zeneli M, Nikolopoulos A, Nikolopoulos N, Grammelis P, Kakaras E K. Calcium looping process simulation based on an advanced thermodynamic model combined with CFD analysis. Fuel, 2015, 153: 370−381 doi: 10.1016/j.fuel.2015.03.014 [33] Yang Y B, Sharifi V, Swithenbank J. Converting moving-grate incineration from combustion to gasification-numerical simulation of the burning characteristics. Waste Management, 2007, 27(5): 645−655 doi: 10.1016/j.wasman.2006.03.014 [34] 王天峥, 汤健, 夏恒, 乔俊飞. 城市固废焚烧过程的回路控制半实物仿真平台. 系统仿真学报, 2023, 35(2): 241−253Wang Tian-Zheng, Tang Jian, Xia Heng, Qiao Jun-Fei. Hardware-in-the-loop simulation platform of loop control for municipal solid waste incineration process. Journal of System Simulation, 2023, 35(2): 241−253 [35] 王天峥, 汤健, 夏恒, 潘晓彤, 乔俊飞, 刘溪芷. 多模态数据驱动的城市固废焚烧过程验证平台设计与实现. 中国电机工程学报, 2023, 43(12): 4697−4708Wang Tian-Zheng, Tang Jian, Xia, Heng, Pan Xiao-Tong, Qiao Jun-Fei, Liu Xi-Zhi. Design and implementation of multi-modal data-driven verification platform for municipal solid waste incineration process. Proceedings of the CSEE, 2023, 43(12): 4697−4708 [36] 汤健, 王天峥, 夏恒, 崔璨麟, 潘晓彤, 郭海涛, 等. 城市固废焚烧智能算法测试与验证模块化半实物平台. 自动化学报, DOI: 10.16383/j.aas.c230762Tang Jian, Wang Tian-Zheng, Xia Heng, Cui Can-Lin, Pan Xiao-Tong, Guo Hai-Tong, et al. Research on modular hard-in-loop platform of intelligent algorithm testing and verification for municipal solid waste incineration. Acta Automatica Sinica, DOI: 10.16383/j.aas.c230762 -
图(14) / 表(6)
计量
- 文章访问数: 320
- HTML全文浏览量: 57
- PDF下载量: 78
- 被引次数: 0
