图神经网络预测烃类工质的热力学性质

doi:10.11949/0438-1157.20230942

化工学报 ›› 2023, Vol. 74 ›› Issue (11): 4466-4474.DOI: 10.11949/0438-1157.20230942

图神经网络预测烃类工质的热力学性质

洪小东¹(), 董轩², 林美金², 廖祖维²(), 任聪静³, 杨遥², 蒋斌波², 王靖岱², 阳永荣²

^1.浙江大学杭州国际科创中心，化工功能材料智能设计与制造浙江省工程研究中心，浙江杭州 311215
^2.浙江大学化学工程联合国家重点实验室，浙江杭州 310027
^3.浙江大学宁波科创中心，浙江宁波 315100

收稿日期:2023-09-11 修回日期:2023-11-14 出版日期:2023-11-25 发布日期:2024-01-22
通讯作者: 廖祖维
作者简介:洪小东（1991—），男，博士，研究员，hongxiaodong@zju.edu.cn
基金资助:
国家自然科学基金项目(U22A20415);浙江省尖兵领雁计划项目(2022C01SA442617)

Prediction of thermodynamic properties of hydrocarbon working fluids by graph neural network models

Xiaodong HONG¹(), Xuan DONG², Meijin LIN², Zuwei LIAO²(), Congjing REN³, Yao YANG², Binbo JIANG², Jingdai WANG², Yongrong YANG²

^1.Engineering Research Center of Functional Materials Intelligent Manufacturing of Zhejiang Province, ZJU-Hangzhou Global Scientific and Technological Innovation Center, Zhejiang University, Hangzhou 311215, Zhejiang, China
^2.State Key Laboratory of Chemical Engineering, Zhejiang University, Hangzhou 310027, Zhejiang, China
^3.Ningbo Innovation Center, Zhejiang University, Ningbo 315100, Zhejiang, China

Received:2023-09-11 Revised:2023-11-14 Online:2023-11-25 Published:2024-01-22
Contact: Zuwei LIAO

摘要/Abstract

摘要：

有机朗肯循环(ORC)因其低温热电转换的能力而备受关注，寻找高效环保工质，以代替具有较高全球变暖潜能值(GWP)的氢氯氟烃(HCFC)和氢氟烃(HFC)，是推动ORC应用的重要任务之一。构建一个基于图神经网络(GNN)的ORC烃类工质热力学性质预测模型，通过图神经网络学习分子结构的特征，并将分子结构信息与温度结合，利用多层感知机(MLP)构建热力学性质预测模型。模型基于2508种C₂~C₁₀的链状烃、环烃和芳香烃分子构建训练集，所得模型在预测临界温度、蒸发焓、气相摩尔热容和液相摩尔热容上均取得良好效果，优于文献的预测效果。此外，应用所得模型预测了超43万个氢氟烯烃(HFO)的热力学性质。

关键词: 热力学性质, 预测, 神经网络, ORC工质, 氢氟烯烃

Abstract:

Organic Rankine cycle (ORC) has attracted much attention due to its ability to convert low-grade heat to electricity. One of the important tasks to promote the application of ORC is to find efficient and environmentally friendly working fluids to replace high-GWP (global warming potential) hydrochlorofluorocarbon (HCFC) and hydrofluorocarbon (HFC). In this article, a prediction model for the thermodynamic properties of ORC hydrocarbon working fluids based on graph neural networks (GNN) is constructed. GNN is used to learn the characteristics of molecular structure, and the combination of molecular structure characteristics and temperature is used to build a prediction model of molecular structure and properties using multilayer perceptron (MLP). The model is based on a training set of 2508 linear, cyclic, and aromatic hydrocarbons with carbon chain lengths ranging from 2 to 10. The obtained model achieves better prediction results than previous literature on predicting critical temperature, evaporation enthalpy and gas-phase and liquid-phase molar heat capacity. In addition, the resulting model was applied to predict the thermodynamic properties of over 430000 hydrofluoroolefins.

Key words: thermodynamic properties, prediction, neural network, ORC working fluids, hydrofluoroolefin

中图分类号:

TQ 021.8

洪小东, 董轩, 林美金, 廖祖维, 任聪静, 杨遥, 蒋斌波, 王靖岱, 阳永荣. 图神经网络预测烃类工质的热力学性质[J]. 化工学报, 2023, 74(11): 4466-4474.

Xiaodong HONG, Xuan DONG, Meijin LIN, Zuwei LIAO, Congjing REN, Yao YANG, Binbo JIANG, Jingdai WANG, Yongrong YANG. Prediction of thermodynamic properties of hydrocarbon working fluids by graph neural network models[J]. CIESC Journal, 2023, 74(11): 4466-4474.

图/表 10

参考文献 30

1	Zhang S, Chen W Y. China's energy transition pathway in a carbon neutral vision[J]. Engineering, 2022, 14: 64-76.
2	Lin Q Y, Zhang X, Wang T, et al. Technical perspective of carbon capture, utilization, and storage[J]. Engineering, 2022, 14: 27-32.
3	Geffroy C, Lilley D, Parez P S, et al. Techno-economic analysis of waste-heat conversion[J]. Joule, 2021, 5(12): 3080-3096.
4	Imran M, Pili R, Usman M, et al. Dynamic modeling and control strategies of organic Rankine cycle systems: methods and challenges[J]. Applied Energy, 2020, 276: 115537.
5	Bellos E, Tzivanidis C. Investigation of a hybrid ORC driven by waste heat and solar energy[J]. Energy Conversion and Management, 2018, 156: 427-439.
6	da Silva Morais P H, Lodi A, Aoki A C, et al. Energy, exergetic and economic analyses of a combined solar-biomass-ORC cooling cogeneration systems for a Brazilian small plant[J]. Renewable Energy, 2020, 157: 1131-1147.
7	Chitgar N, Hemmati A, Sadrzadeh M. A comparative performance analysis, working fluid selection, and machine learning optimization of ORC systems driven by geothermal energy[J]. Energy Conversion and Management, 2023, 286: 117072.
8	Nondy J, Gogoi T K. Exergoeconomic investigation and multi-objective optimization of different ORC configurations for waste heat recovery: a comparative study[J]. Energy Conversion and Management, 2021, 245: 114593.
9	Schwöbel J A H, Preißinger M, Brüggemann D, et al. High-throughput screening of working fluids for the organic Rankine cycle (ORC) based on conductor-like screening model for realistic solvation (COSMO-RS) and thermodynamic process simulations[J]. Industrial & Engineering Chemistry Research, 2017, 56(3): 788-798.
10	Dong X A, Liao Z W, Sun J Y, et al. Simultaneous optimization for organic Rankine cycle design and heat integration[J]. Industrial & Engineering Chemistry Research, 2020, 59(46): 20455-20471.
11	Dong X A, Liao Z W, Sun J Y, et al. Simultaneous optimization of a heat exchanger network and operating conditions of organic Rankine cycle[J]. Industrial & Engineering Chemistry Research, 2020, 59(25): 11596-11609.
12	王羽鹏, 梁俊伟, 罗向龙, 等. 基于神经网络的有机朗肯循环过程及循环性能计算方法[J]. 化工学报, 2019, 70(9): 3256-3266.
	Wang Y P, Liang J W, Luo X L, et al. Novel prediction method of process and system performance for organic Rankine cycle based on neural network[J]. CIESC Journal, 2019, 70(9): 3256-3266.
13	李子航, 王占博, 苗政, 等. 亚临界有机朗肯循环系统工质筛选及热经济性分析[J]. 化工学报, 2021, 72(9): 4487-4495.
	Li Z H, Wang Z B, Miao Z, et al. Working fluid selection and thermo-economic analysis of sub-critical organic Rankine cycle[J]. CIESC Journal, 2021, 72(9): 4487-4495.
14	Fouad W A, Vega L F. Next generation of low global warming potential refrigerants: thermodynamic properties molecular modeling[J]. AIChE Journal, 2018, 64(1): 250-262.
15	Nair V. HFO refrigerants: a review of present status and future prospects[J]. International Journal of Refrigeration, 2021, 122: 156-170.
16	龚正. 分子动力学模拟预测工业流体的热力学性质的方法研究[D]. 上海: 上海交通大学, 2019.
	Gong Z. Study on the method of molecular dynamics simulation to predict the thermodynamic properties of industrial fluids[D]. Shanghai: Shanghai Jiao Tong University, 2019.
17	Xu Y H, Huang X, Li C P, et al. Predicting structure-dependent properties directly from the three dimensional molecular images via convolutional neural networks[J]. AIChE Journal, 2022, 68(8): e17721.
18	Walters W P, Barzilay R. Applications of deep learning in molecule generation and molecular property prediction[J]. Accounts of Chemical Research, 2021, 54(2): 263-270.
19	Zhang J, Wang Q, Su Y, et al. An accurate and interpretable deep learning model for environmental properties prediction using hybrid molecular representations[J]. AIChE Journal, 2022, 68(6): e17634.
20	Altintas C, Altundal O F, Keskin S, et al. Machine learning meets with metal organic frameworks for gas storage and separation[J]. Journal of Chemical Information and Modeling, 2021, 61(5): 2131-2146.
21	Moghadam P Z, Rogge S M J, Li A, et al. Structure-mechanical stability relations of metal-organic frameworks via machine learning[J]. Matter, 2019, 1(1): 219-234.
22	Wang S H, Cheng M, Luo L, et al. High-throughput screening of metal-organic frameworks for hydrogen purification[J]. Chemical Engineering Journal, 2023, 451: 138436.
23	Schweidtmann A M, Rittig J G, König A, et al. Graph neural networks for prediction of fuel ignition quality[J]. Energy & Fuels, 2020, 34(9): 11395-11407.
24	Liu R C, Liu R Z, Liu Y F, et al. Design of fuel molecules based on variational autoencoder[J]. Fuel, 2022, 316: 123426.
25	Alshehri A S, Tula A K, You F Q, et al. Next generation pure component property estimation models: with and without machine learning techniques[J]. AIChE Journal, 2022, 68(6): e17469.
26	Yan X, Lan T, Jia Q Z, et al. A norm indexes-based QSPR model for predicting the standard vaporization enthalpy and formation enthalpy of organic compounds[J]. Fluid Phase Equilibria, 2020, 507: 112437.
27	Aouichaoui A R N, Fan F, Abildskov J, et al. Application of interpretable group-embedded graph neural networks for pure compound properties[J]. Computers & Chemical Engineering, 2023, 176: 108291.
28	Wang Y Y, Wang J R, Cao Z L, et al. Molecular contrastive learning of representations via graph neural networks[J]. Nature Machine Intelligence, 2022, 4(3): 279-287.
29	Morris C, Ritzert M, Fey M, et al. Weisfeiler and leman go neural: higher-order graph neural networks[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2019, 33(1): 4602-4609.
30	Su Y, Wang Z H, Jin S M, et al. An architecture of deep learning in QSPR modeling for the prediction of critical properties using molecular signatures[J]. AIChE Journal, 2019, 65(9): e16678.

数据集	模型输入：模型输出	数量
#1	SMILES：T_c (℃)	2508
#2	SMILES + T：H_eva (J/mol)	44717
#3	SMILES + T：C_p_g (J/(mol·℃))	44717
#4	SMILES + T：C_pl (J/(mol·℃))	44717

数据集	模型输入：模型输出	数量
#1	SMILES：T_c (℃)	2508
#2	SMILES + T：H_eva (J/mol)	44717
#3	SMILES + T：C_p_g (J/(mol·℃))	44717
#4	SMILES + T：C_pl (J/(mol·℃))	44717

特征	描述	维度
原子类型	原子类型（C, Cl, F, O）	4
是否成环	原子是否在成环体系中	1
是否芳香性	原子是否是芳香体系的一部分	1
杂化	sp, sp², sp³, sp³d, sp³d²	5
键数目	原子相连的键数目（0~6）	7
氢原子数目	原子相连的氢原子数目（0~3）	4

特征	描述	维度
原子类型	原子类型（C, Cl, F, O）	4
是否成环	原子是否在成环体系中	1
是否芳香性	原子是否是芳香体系的一部分	1
杂化	sp, sp², sp³, sp³d, sp³d²	5
键数目	原子相连的键数目（0~6）	7
氢原子数目	原子相连的氢原子数目（0~3）	4

特征	描述	维度
键类别	单键、双键、叁键或芳香键	4
共轭	键是否共轭	1
成环	键是否属于环结构	1
立体	无、任意、E、Z、顺式、反式	6

图神经网络预测烃类工质的热力学性质

Prediction of thermodynamic properties of hydrocarbon working fluids by graph neural network models

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 10

参考文献 30

相关文章 15

编辑推荐

Metrics

本文评价

指标	训练/验证/测试	训练/测试^[30]
均方根误差，RMSE	13.2/14.4/13.6	—
平均绝对误差，MAE	9.1/10.1/9.7	22.5/22.8
决定系数，R²	0.98/0.98/0.97	—

指标	训练/验证/测试	训练/测试^[26]	训练/测试^[14]
均方根误差，RMSE	896.3/958.0/971.8	—	2470.0
平均绝对误差，MAE	342.9/393.3/403.0	2499.0	—
决定系数，R²	0.99/0.99/0.99	0.97	0.97

指标	训练/验证/测试
指标	C_p_g	C_pl
均方根误差，RMSE	2.5/3.3/2.5	3.1/3.6/3.4
平均绝对误差，MAE	1.4/1.7/1.6	1.6/1.9/1.9
决定系数，R²	1.00/1.00/1.00	1.00/1.00/1.00

序号	SMILES	T_c/℃	H_eva@0.9T_c/(J/mol)	C_p_g@0.9T_c/(J/(mol·℃))	C_pl @0.9T_c/(J/(mol·℃))
1	FC(F)CC(C)C(F)(F)F	172.4	15988.5	194.5	227.7
2	FC(F)C1CCCC1(F)F	218.4	18425.0	239.4	268.0
3	CC1CCC(F)C1C(F)(F)F	273.2	22472.0	269.5	305.5
4	CCCCC(F)CC(C)(C)F	334.6	22170.4	362.0	329.5
5	CCCC(C)(F)CCC(C)C	347.9	23476.0	392.1	362.1
6	CCC(CF)(CC(C)C)CC	357.8	24531.6	393.1	368.3
7	CC(C)CCCCC1CC1F	366.4	27816.5	378.8	346.4
8	FC1CCCC2CC2(F)CC1	374.8	25305.4	327.0	312.6
9	FC1CC(CC1)CCCC#C	383.7	25063.7	350.4	329.9
10	CC1CCC2CCCC2=C1F	396.5	26709.4	330.9	355.6

[1]	温凯杰, 郭力, 夏诏杰, 陈建华. 一种耦合CFD与深度学习的气固快速模拟方法[J]. 化工学报, 2023, 74(9): 3775-3785.
[2]	诸程瑛, 王振雷. 基于改进深度强化学习的乙烯裂解炉操作优化[J]. 化工学报, 2023, 74(8): 3429-3437.
[3]	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.
[4]	尹刚, 李伊惠, 何飞, 曹文琦, 王民, 颜非亚, 向禹, 卢剑, 罗斌, 卢润廷. 基于KPCA和SVM的铝电解槽漏槽事故预警方法[J]. 化工学报, 2023, 74(8): 3419-3428.
[5]	徐野, 黄文君, 米俊芃, 申川川, 金建祥. 多源信息融合的离心式压缩机喘振诊断方法[J]. 化工学报, 2023, 74(7): 2979-2987.
[6]	郭雨莹, 敬加强, 黄婉妮, 张平, 孙杰, 朱宇, 冯君炫, 陆洪江. 稠油管道水润滑减阻及压降预测模型修正[J]. 化工学报, 2023, 74(7): 2898-2907.
[7]	于源, 陈薇薇, 付俊杰, 刘家祥, 焦志伟. 几何相似涡流空气分级机环形区流场变化规律研究及预测[J]. 化工学报, 2023, 74(6): 2363-2373.
[8]	高学金, 姚玉卓, 韩华云, 齐咏生. 基于注意力动态卷积自编码器的发酵过程故障监测[J]. 化工学报, 2023, 74(6): 2503-2521.
[9]	李艳辉, 丁邵明, 白周央, 张一楠, 于智红, 邢利梅, 高鹏飞, 王永贞. 非常规服役超临界锅炉的微纳尺度腐蚀动力学模型建立及应用[J]. 化工学报, 2023, 74(6): 2436-2446.
[10]	姚晓宇, 沈俊, 李健, 李振兴, 康慧芳, 唐博, 董学强, 公茂琼. 流体气液临界参数测量方法研究进展[J]. 化工学报, 2023, 74(5): 1847-1861.
[11]	黄磊, 孔令学, 白进, 李怀柱, 郭振兴, 白宗庆, 李平, 李文. 油页岩添加对准东高钠煤灰熔融行为影响的研究[J]. 化工学报, 2023, 74(5): 2123-2135.
[12]	陈科, 杜理, 曾英, 任思颖, 于旭东. 四元体系LiCl+MgCl₂+CaCl₂+H₂O 323.2 K相平衡研究及计算[J]. 化工学报, 2023, 74(5): 1896-1903.
[13]	贠程, 王倩琳, 陈锋, 张鑫, 窦站, 颜廷俊. 基于社团结构的化工过程风险演化路径深度挖掘[J]. 化工学报, 2023, 74(4): 1639-1650.
[14]	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.
[15]	张江淮, 赵众. 碳三加氢装置鲁棒最小协方差约束控制及应用[J]. 化工学报, 2023, 74(3): 1216-1227.