基于分子表面电荷密度分布与机器学习的混合物设计方法研究

doi:10.11949/0438-1157.20190795

化工学报 ›› 2020, Vol. 71 ›› Issue (S1): 282-292.doi: 10.11949/0438-1157.20190795

基于分子表面电荷密度分布与机器学习的混合物设计方法研究

毛海涛¹(),王璐¹,许志颖²,解万翠²,都健¹,张磊¹()

^1.大连理工大学化工学院化工系统工程研究所，辽宁大连 116024
^2.青岛科技大学海洋科学与生物工程学院，山东青岛 266042

收稿日期:2019-07-10 修回日期:2019-09-13 出版日期:2020-04-25 发布日期:2020-05-22
通讯作者: 张磊 E-mail:haitaomao0730@foxmail.com;keleiz@dlut.edu.cn
作者简介:毛海涛（1994—），男，硕士研究生，haitaomao0730@foxmail.com
基金资助:
国家自然科学基金项目(21808025)

Mixture product design based on molecular surface charge density distribution and machine learning

Haitao MAO¹(),Lu WANG¹,Zhiying XU²,Wancui XIE²,Jian DU¹,Lei ZHANG¹()

^1.Institute of Chemical Process Systems Engineering, School of Chemical Engineering, Dalian University of Technology, Dalian 116024, Liaoning, China
^2.College of Marine Science and Biological Engineering, Qingdao University of Science & Technology, Qingdao 266042, Shandong, China

Received:2019-07-10 Revised:2019-09-13 Online:2020-04-25 Published:2020-05-22
Contact: Lei ZHANG E-mail:haitaomao0730@foxmail.com;keleiz@dlut.edu.cn

摘要/Abstract

摘要：

由于混合物性能的可调控性，当前市场对其关注与日俱增。对于这类产品，基于模型的设计方法由于具有高效性以及普适性，相较于其他产品设计方法得到了更快的发展。但是对于很多性质，如气味、颜色等，准确且普适的模型尚不可得。因此，本文提出了一种基于分子表面电荷密度分布描述符（S描述符）和机器学习模型的混合物设计方法，采用描述符表征产品、再通过机器学习模型将其与性质关联，直接用于混合物产品设计。具体地，根据给定的产品性质需求，机器学习模型直接预测/设计混合物产品的S描述符；然后以欧几里德距离为指标，在给定的数据库中筛选出S描述符满足要求的候选混合物组成。最后，对候选混合物及其组分性质进行实验验证，完成设计。本文以香精的混合替代物设计作为算例，设计得到丙酸叶醇酯的两种混合香精替代物，通过实验对混合物进行了验证。结果表明，混合替代物的气味及其组分的各理化性质均与丙酸叶醇酯相近，证实本文所提出方法的有效性。

关键词: 系统工程, 产品设计, 神经网络, 分子表面电荷密度分布, 混合物设计, 香精

Abstract:

Modern business pays increasing attentions to mixture products due to its adjustable characteristics. For the design methods towards such kinds of products, the development of model-based design methods is faster than others, because of its efficiency and wide application. However, the models for some properties, like odor and color, with acceptable accuracy or general application range are still not available. Therefore, an application methodology of machine learning (ML) with molecular surface charge density distribution (Sdescriptors) for mixture product design is proposed in this study, where descriptors are employed to represent the product and ML is responsible for correlating them to the target properties, for the purpose of designing product directly. Specifically, machine learning model is expected to predict Sdescriptors of candidate products according to the assigned property value, and ingredients are screened out using Euclidean-based method according to the predicted descriptors. Finally, the properties of the candidate mixtures and its ingredients are verified by experiments. This methodology is introduced using a case study of mixture substitution fragrance design for cis-3-hexenyl propionate and two mixture fragrances are obtained ultimately. The odor properties of mixtures and physicochemical properties of their components are similar to the target, which highlights the effective of the proposed method.

Key words: systems engineering, product design, neural network, molecular surface charge density distribution, mixture design, fragrance

中图分类号:

TQ 021.8

毛海涛, 王璐, 许志颖, 解万翠, 都健, 张磊. 基于分子表面电荷密度分布与机器学习的混合物设计方法研究[J]. 化工学报, 2020, 71(S1): 282-292.

Haitao MAO, Lu WANG, Zhiying XU, Wancui XIE, Jian DU, Lei ZHANG. Mixture product design based on molecular surface charge density distribution and machine learning[J]. CIESC Journal, 2020, 71(S1): 282-292.

图/表 13

图1

图2

图3

表1

基团贡献法以及阈值"

组分性质	基团贡献法公式	阈值
$δ$ /MPa^1/2	$δ = 1000 H 298 V a p - R T V 298 m 12$	$δ ≥ 158.9$
$T b$ /K	$T b = 244.7889 l n ∑ i ∈ G n i T b, i$	$T b ≥ 333$
$T f$ /K	$T f = 150.0218 + ∑ i ∈ G n i T f, i$	$T f ≥ 323$
$K o / w$	$l n K o / w = 0.4876 + ∑ i ∈ G n i K o / w, i$	$l n K o / w ≤ 2.91$
$L C 50$ /(mol·L^-1)	$- l n L C 50 = 2.18 + ∑ i ∈ G n i L C 50, i$	$- l n L C 50 ≤ 3.70$

表1

表2

图4

图5

表3

图6

表4

丙酸叶醇酯替代香精的设计结果"

参数	丙酸叶醇酯	替代混合物1			替代混合物2
参数	丙酸叶醇酯	4-异丙基苯甲醇	左旋香芹酮	混合物偏差	2-甲基戊酸	2-乙基丁酸烯丙酯	混合物偏差
CAS No.	33467-74-2	536-60-7	6485-40-1	—	97-61-0	7493-69-8	—
体积分数	1	0.2	0.8	—	0.4	0.6	—
溶解度 $δ$ ^①(298 K,水)/(mg/L)	158.9	1687	367.1	472.18	15000	157.3	5935.48
沸点T_b^①(101.325 kPa) /K	453-455	512.66	499.47	47.11-49.11	468.36	449.96	2.32-4.32
闪点T_f^①/K	333	498.15	465.15	138.75	364.26	327.59	9.258
K_o/w^①	2.909	2.37	2.71	-0.267	1.8	2.972	-0.4058
LC₅₀^②/(mol·L^-1)	3.36	3.25	3.39	0.002	2.45	4.03	0.038

表4

图7

表5

图8

参考文献 34

1	Rodríguez O, Gomes P, Mata V, et al. Chapter 1 - A Product Engineering Approach in the Perfume Industry [M]//Teixeira M A. Perfume Engineering. Oxford: Butterworth-Heinemann, 2019: 1-13.
2	Wibowo C, Ng K M. Product-oriented process synthesis and development: creams and pastes[J]. AIChE Journal, 2001, 47(12): 2746-2767.
3	Fung K Y, Ng K M. Product-centered processing: pharmaceutical tablets and capsules[J]. AIChE Journal, 2003, 49(5): 1193-1215.
4	Gani R, Brignole E A. Molecular design of solvents for liquid extraction based on UNIFAC[J]. Fluid Phase Equilibria, 1983, 13(83): 331-340.
5	Joback K G. Designing molecules possessing desired physical property values[D]. Massachusetts: Massachusetts Institute of Technology, 1989
6	Conte E, Gani R, Ng K M. Design of formulated products: a systematic methodology[J]. AIChE Journal, 2011, 57: 2431-2449.
7	Kontogeorgis G M, Michele M, Ng K M, et al. An integrated approach for the design of emulsified products[J]. AIChE Journal, 2019, 65: 75-86.
8	张磊, 刘琳琳, 都健. 替代燃油的计算机辅助设计方法[J]. 化工进展, 2018, 37(6): 2438-2444.
	Zhang L, Liu L L, Du J. A computer-aided design methodology for tailor-made surrogate fuels[J]. Chemical Industry and Engineering Progress, 2018, 37(6): 2438-2444.
9	Hornic K. Multilayer feedforward networks are universal approximators[J]. Neural Networks, 1989, 2: 359-366.
10	Raccuglia P, Elbert K C, Adler P D, et al. Machine-learning-assisted materials discovery using failed experiments[J]. Nature, 2016, 533: 73-76.
11	苏荣欣, 邹龙花, 齐崴, 等. 酪蛋白-胰酶水解历程分子量变化模拟与三维表征[J]. 化工学报, 2013, 64(1): 346-351.
	Su R X, Zou L H, Qi W, et al. Simulation and 3D plot of molecular weight distribution of released peptides from pancreatic hydrolysis of casein[J]. CIESC Journal, 2013, 64(1): 346-351.
12	黄凯, 陈勇, 母志为, 等. 基于人工神经网络和遗传算法的甲烷制氢催化剂设计[J]. 化工学报, 2016, 67(8): 3481-3490.
	Huang K, Chen Y, Mu Z W, et al. Catalyst design for production of hydrogen from methane based on artificial neural network and genetic algorithm[J]. CIESC Journal, 2016, 67(8): 3481-3490.
13	安爱民, 刘云利, 张浩琛, 等. 微生物燃料电池的动态性能分析及其神经网络预测控制[J]. 化工学报, 2017, 68(3): 1090-1098.
	An A M, Liu Y L, Zhang H C, et al. Dynamic performance analysis and neural network predictive control of microbial fuel cell[J]. CIESC Journal, 2017, 68(3): 1090-1098.
14	林生岭, 徐绍芬, 王俊德, 等. 钙钛矿型La_xSr_1-_xNi_1-_yCo_yO₃光电催化活性研究[J]. 化学学报, 2005, 63(5): 385-390.
	Lin S L, Xu S F, Wang J D, et al. Study on photo-electro catalytic activity of perovskite type oxides La_xSr_1－_xNi_1－_yCo_yO₃[J]. Acta Chimica Sinica, 2005, 63(5): 385-390.
15	Zhang L, Mao H, Liu L, et al. A machine learning based computer-aided molecular design/screening methodology for fragrance molecules[J]. Computers & Chemical Engineering, 2018, 115: 295-308.
16	Sanchez-Lengeling B, Aspuru-Guzik A. Inverse molecular design using machine learning: generative models for matter engineering[J]. Science, 2018, 361: 360-365.
17	Klamt A, Schueuermann G J. COSMO: a new approach to dielectric screening in solvents with explicit expressions for the screening energy and its gradient[J]. Journal of the Chemical Society, Perkin Transactions II, 1993, 5: 799-805.
18	Rossiter K J. Structure-odor relationships[J]. Chemical Review, 1996, 96: 3201-3240.
19	Klamt A, Reinisch J, Eckert F, et al. Polarization charge densities provide a predictive quantification of hydrogen bond energies[J]. Physical Chemistry Chemical Physics, 2011, 14(2): 955-963.
20	Lin S T, Sandler S I. A priori phase equilibrium prediction from a segment contribution solvation model[J]. Industrial & Engineering Chemistry Research, 2002, 41(5): 899-913.
21	Klamt A, Eckert F, Arlt W. COSMO-RS: an alternative to simulation for calculating thermodynamic properties of liquid mixtures[J]. Annual Review of Chemical and Biomolecular Engineering, 2010, 1: 101-122.
22	Kang X, Liu X, Li J, et al. Heat capacity prediction of ionic liquids based on quantum chemistry descriptors[J]. Industrial & Engineering Chemistry Research, 2018, 57(49): 16989-16994.
23	Kang X, Zhao Z, Qian J, et al. Predicting the viscosity of ionic liquids by the ELM intelligence algorithm[J]. Industrial & Engineering Chemistry Research, 2017, 56(39): 11344-11351.
24	Palomart J, Torrecilla J S, Ferro V R, et al. Development of an a priori ionic liquid design tool (Ⅱ): Ionic liquid selection through the prediction of COSMO-RS molecular descriptor by inverse neural network[J]. Industrial & Engineering Chemistry Research, 2009, 48(4): 2257-2265.
25	Keller A, Vosshall L B. Olfactory perception of chemically diverse molecules[J]. BMC Neuroscience, 2016, 17: 55.
26	周志华. 机器学习[M]. 北京: 清华大学出版社, 2016: 23-53.
	Zhou Z H. Machine Learning[M]. Beijing: Tsinghua University Press, 2016: 23-53.
27	刘方, 徐龙, 马晓迅. BP神经网络的发展及其在化学化工中的应用[J]. 化工进展, 2019, 38(6): 2559-2573.
	Liu F, Xu L, Ma X X. Development of BP neural network and its application in chemistry and chemical engineering[J]. Chemical Industry and Engineering Progress, 2019, 38(6): 2559-2573.
28	de Bruyne M, Foster K, Carlson J R. Odor coding in the Drosophila antenna[J]. Neuron, 2001, 30(2): 537-552.
29	Tamir A. In Applications of Markov Chains in Chemical Engineering[M]. Amsterdam, Netherlands: Elsevier, 1998.
30	Marrero J, Gani R. Group-contribution based estimation of pure component properties[J]. Fluid Phase Equilibria, 2001, 183: 183-208.
31	Hukkerikar A S. Development of pure component property models for chemical product-process design and analysis[D]. Denmark: Technical University of Denmark, 2013.
32	马琦, 伯继芳, 冯莉, 等. GC-MS结合电子鼻分析干燥方式对杏鲍菇挥发性风味成分的影响[J]. 食品科学, 2019, 40(14): 276-282.
	Ma Q, Bo J F, Feng L, et al. Effect of drying method on volatile components of pleurotus eryngii analyzed by combined use of GC-MS and electronic nose[J]. Food Science, 2019, 40(14): 276-282.
33	Reid R C, Prausnitz J M, Poling B E. The Properties of Gases & Liquids[M]. New York: McGrawHill, 1988.
34	Lee B I, Kesler M. A generalized thermodynamic correlation based on three-parameter corresponding states[J]. AIChE Journal, 1975, 21: 510-527.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

传感器名称	检测气味类型	检出限/ (mg/kg)
W1C	有机化合物	10
W5S	氮氧化物	1
W3C	氨类	10
W6S	氢气	0.1
W5C	烷烃与非极性有机化合物	1
W1S	甲烷	100
W1W	含硫有机化合物	1
W2S	酒精	100
W2W	无机硫	1
W3S	有机化合物以及脂肪族有机化合物	10

参数	数值	参数	数值
可食用	32.42	馊味	25.93
烘焙味	43.00	愉悦度	60.19
甜味	34.37	蒸气压/Pa	53.86
水果味	31.54	扩散系数/（m²/h）	0.16
花香味	29.95

试剂	说明
丙酸叶醇酯	北京迈瑞达科技有限公司，纯度>98%
2-甲基戊酸	北京迈瑞达科技有限公司，纯度>98%
2-乙基丁酸烯丙酯	北京迈瑞达科技有限公司，纯度>97%
4-异丙基苯甲醇	河南郑州阿尔法化工有限公司，纯度>99%
左旋香芹酮	河南郑州阿尔法化工有限公司，纯度>97%
95%乙醇	天津市富宇化工有限公司

[1]	贺鹏程, 庄莉, 胡亮, 刘刚, 王瑞琪, 包亚强. 板翅式换热器压力特性工程计算方法[J]. 化工学报, 2020, 71(S1): 172-178.
[2]	方黄峰, 刘瑶瑶, 张文彪. 基于LSTM神经网络的流化床干燥器内生物质颗粒湿度预测[J]. 化工学报, 2020, 71(S1): 307-314.
[3]	贺彦林, 田业, 顾祥柏, 徐圆, 朱群雄. 基于正则化的函数连接神经网络研究及其复杂化工过程建模应用[J]. 化工学报, 2020, 71(3): 1072-1079.
[4]	张璐, 张嘉成, 韩红桂, 乔俊飞. 基于模糊神经网络的污水处理生化除磷过程控制[J]. 化工学报, 2020, 71(3): 1217-1225.
[5]	李晨莹, 刘琳琳, 张磊, 顾偲雯, 都健. 不确定性下基于多工况优化的可控性换热器网络综合[J]. 化工学报, 2020, 71(3): 1154-1162.
[6]	黄正梁, 王超, 李少硕, 杨遥, 孙婧元, 王靖岱, 阳永荣. 基于深度学习的气液固三相反应器图像分析方法及应用[J]. 化工学报, 2020, 71(1): 274-282.
[7]	张楠, 陈龙祥, 胡芃. 混合工质临界性质的推算研究[J]. 化工学报, 2019, 70(S2): 1-7.
[8]	曹晨鑫, 杜玉鹏, 王昕, 王振雷. 基于Ms-LWPLS的化工过程网络化性能分级评估方法[J]. 化工学报, 2019, 70(S1): 141-149.
[9]	陈虎, 陈倩, 刘长军, 黄卡玛, 龙卓. 基于SIW的介电系数宽带测量装置[J]. 化工学报, 2019, 70(S1): 182-185.
[10]	王羽鹏, 梁俊伟, 罗向龙, 李逸帆, 陈健勇, 陈颖. 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.
[11]	廉小亲, 王俐伟, 安飒, 魏伟, 刘载文. 基于SOM-RBF神经网络的COD软测量方法[J]. 化工学报, 2019, 70(9): 3465-3472.
[12]	柴伟, 郭龙航, 池彬彬. 污水处理厂出水水质变量区间预测建模[J]. 化工学报, 2019, 70(9): 3449-3457.
[13]	章聪, 江锦波, 彭旭东, 赵文静, 李纪云. 近临界区CO₂物性预测模型对比与修正[J]. 化工学报, 2019, 70(8): 3058-3070.
[14]	乔俊飞, 贺增增, 杜胜利. 基于混合评价指标的自组织模糊神经网络设计研究[J]. 化工学报, 2019, 70(7): 2606-2615.
[15]	王志甄, 邹志云. 基于神经网络的pH中和过程非线性预测控制[J]. 化工学报, 2019, 70(2): 678-686.

基于分子表面电荷密度分布与机器学习的混合物设计方法研究

Mixture product design based on molecular surface charge density distribution and machine learning

RichHTML

PDF (PC)

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 34

相关文章 15

Metrics

本文评价

推荐阅读 10