烃类沸点的定量构效关系研究

应用CODESSA软件计算296种烃类物质的分子结构描述符,分别用启发式回归（HM）和最佳多元线性回归（B-MLR）筛选计算出的所有分子描述符,并建立沸点的线性回归模型。用B-MLR方法筛选出的4个描述符作为支持向量机（SVM）的输入建立了非线性模型。预测结果表明：所建立的模型稳健,泛化能力强,预测误差小。非线性模型（R2=0.9905,RMSE=10.2295）的性能优于线性回归模型（HM：R2=0.9819,RMSE=14.0606;B-MLR：R2=0.9842,RMSE=13.1058）,预测效果令人满意。     

基于QSPR方法的烃类物质苯胺点预测

烃类物质在石油工业中有着非常广泛的应用.在石油工业中常用苯胺点来衡量有机溶剂的溶解性能.基于定量结构-性质相关性(QSPR)原理,根据分子结构计算反映分子结构信息的结构参数,应用遗传函数算法从大量结构参数中优化筛选出与烃类物质苯胺点最为密切相关的结构参数作为表征相应化合物结构特征的分子描述符,采用多元线性回归方法对分子描述符与苯胺点之间的定量函数关系进行关联,建立了预测烃类物质苯胺点的理论模型.最后,对模型进行了内部及外部验证来检验模型的可靠性.在此基础上,对所建立的预测模型进行机理解释,分析了影响烃类物质苯胺点的主要结构因素及其影响规律.研究表明,所建模型具有较高的稳定性和预测能力.     

基于QSPR方法的脂肪醇化合物闪点预测

为提高脂肪醇化合物闪点预测精度,提出基于定量结构-性质关系(QSPR)原理的脂肪醇化合物闪点预测方法。应用Dragon软件计算出91种脂肪醇的分子描述符,利用遗传函数算法(GFA)从1 481个描述符中筛选出3个与脂肪醇闪点关系最密切的分子描述符。分别用多元线性回归(MLR)方法和支持向量机(SVM)方法进行建模,并采用内部验证和外部检验的方式对模型的拟合度、预测性等性能进行验证。结果表明:预测集的MLR方法和SVM方法的平均绝对误差(AAE)分别为2.870 K和2.706 K;均方根误差(RMSE)为3.451 K和3.371 K。SVM模型在精度上略优于MLR模型,而MLR模型更为简单和方便。     

芳香族硝基化合物爆速的定量构效关系预测

为提高芳香族硝基化合物爆速的预测精度,提出基于定量结构-性质相关性(QSPR)的芳香族硝基化合物爆速的理论预测方法。应用Dragon软件计算21种芳香族硝基化合物的分子描述符,利用遗传-多元线性回归方法(GA-MLR)从大量描述符中筛选出4个与爆速关系最为密切的分子描述符,建立相应的4参数线性理论预测模型,并采用内部和外部验证方式,验证模型的拟合能力、稳健性和预测能力。结果表明:训练集和测试集的平均绝对误差(AAE)分别为0.048和0.208km/s,均方根误差(RMSE)分别为0.058和0.302 km/s,所建模型具有较好的稳健性和预测性能。     

应用原子类型AI指数预测烃类燃烧热

以基于Xu指数的原子类型AI指数作为分子结构描述符,表征了80个液态烃的分子结构特征,并分别结合人工神经网络和多元线性回归方法,对这80种液态烃的燃烧热进行定量结构-性质相关性建模和预测研究。结果表明,基于Xu指数的原子类型AI指数能很好地表征烃类物质的分子结构特征。所建的最佳预测模型为基于Xu指数的原子类型AI指数多元线性回归模型,模型复相关系数为0.999,对测试集的平均预测相对误差为0.637%,模型预测值与实验值的一致性令人满意。     

咪唑类离子液体捕集CO_2性能的定量结构-性质相关性(QSPR)研究

CO_2是主要的温室气体,大量CO_2的存在严重影响着人类的生存环境和生态平衡,而咪唑型离子液体具有独特的气体溶解性,在CO_2的捕集分离中有很好的应用前景。基于定量结构-性质相关性(QSPR)原理,研究了咪唑类离子液体捕集CO_2的性能与其结构参数之间的内在定量关系。应用遗传算法获得与捕集量最为密切相关的一组描述符作为输入参数,随后,分别采用多元线性回归算法及支持向量机结合粒子群优化算法建立了咪唑类离子液体捕集CO_2的性能与其描述符之间的线性和非线性模型。多元线性回归算法得出训练集和测试集的复相关系数分别为0.765和0.814,支持向量机算法得出训练集和测试集的复相关系数分别为0.987和0.933。对预测模型进行了评价验证以及稳定性分析,结果表明,2种模型具有良好的稳定性能和预测能力。     

应用电性拓扑状态指数预测烷烃自燃点

建立了一个基于人工神经网络的定量结构-性质相关性模型，用于52种烷烃化合物自燃点的预测研究。应用原子类型电性拓扑状态指数作为表征分子结构特征的描述符。该指数既能表征分子的电子特性，又反映其拓扑特征，同时易于计算，并有较强的同分异构体区分能力。采用误差反向传播（BP）神经网络方法对烷烃自燃点与电性拓扑状态指数间可能存在的非线性关系进行拟合。将52种烷烃样本随机划分为训练集（30种）、验证集（8种）和测试集（14种），并通过“试差法”确定网络的最优参数。运用最佳网络结构[64—1]对实验样本进行模拟，结果表明，多数样本的自燃点预测值与实验值符合良好，对于测试集，平均预测绝对误差为8．4℃，均方根误差为11．8，优于多元线性回归方法和传统基团贡献法所得结果。该方法的提出为工程上提供了一种根据分子结构预测有机物白燃点的有效方法。     

基于新型拓扑指数的咪唑类离子液体捕集CO2性能预测

大量温室气体CO_2的存在严重影响环境,而咪唑型离子液体具有独特的气体溶解性,在CO_2捕集方面的应用较为广泛。基于定量结构-性质相关性(QSPR)原理提出了一种新的描述符——拓扑指数(Topological Index,TI)描述符,研究了咪唑类离子液体捕集CO_2的性能与其拓扑指数描述符之间的内在定量关系。应用遗传算法获得与捕集量最为密切相关的一组拓扑指数描述符,将其与温度和压力一起作为输入参数,分别采用多元非线性回归算法及支持向量机算法建立了咪唑类离子液体捕集CO_2性能与其拓扑指数描述符之间的非线性模型。通过多元非线性回归算法得出训练集和测试集的决定系数分别为0.771和0.754,由支持向量机算法得出训练集和测试集的决定系数分别为0.990和0.981。对预测模型进行了评价验证及稳定性分析,结果表明,两种模型均具有良好的稳定性能和预测能力。根据拓扑指数描述符所建立的预测模型为工程应用提供了一种预测咪唑类离子液体捕集CO_2性能的有效方法。     

基于遗传算法的支持向量机预测有机物自燃点的研究

根据定量构效关系(QSPR)原理,研究自燃点(AIT)与其分子结构间的内在定量关系。以265种有机化合物作为样本集,随机选择238种作为训练集,27种作为测试集,用遗传算法(GA)进行变量选择,分别建立多元线性回归(MLR)模型和支持向量机(SVM)模型研究有机物的自燃点与其分子结构间的关系。通过分析,发现造成模型预测效果不佳的原因是试验数据本身存在问题。通过对2个模型的比较,结果为GA-SVM模型明显优于GA-MLR模型,说明自燃点与其分子结构间具有很强的非线性关系。     

QSPR在混合物典型燃爆特性预测中的研究进展

为全面了解定量结构-性质关系（QSPR）方法在混合物燃爆特性预测中的研究现状，展望其发展趋势，综述其在混合物闪点、爆炸极限与自燃温度预测中的国内外研究进展，分析预测目标参数的选择、数据收集、描述符计算和筛选以及模型建立和验证等方面的不足与研究方向。结果表明:QSPR在混合物燃爆特性预测中尚处于起步阶段，当前研究的首要限制是混合物燃爆特性参数实验数据样本不足，关键点及难点是混合物结构的准确表征，未来研究应关注的重点是大量数据源统一的数据样本的获取方法、非加和性混合物分子描述符的计算方法以及机器学习等非线性建模方法。     

Flash points measurements and prediction for binary miscible mixtures

Flash point is one of the most important parameters used to characterize the potential fire and explosion hazards for flammable liquids. In this study, flash points of twenty eight binary miscible mixtures comprised eighteen flammable pure components with different compositions were measured by using the closed cup apparatus. The obtained experimental data are further employed to develop simple and accurate models for predicting the flash points of binary miscible mixtures. Based on the vapor–liquid equilibrium theory, the normal boiling point, the standard enthalpy of vaporization, the average number of carbon atoms, and the stoichiometric concentration of the gas phase were selected as the dominant physicochemical parameters that were relevant to the overall flash point property of liquids. With these parameters for pure components as well as the compositions of mixtures, the new form of characteristic physicochemical parameters for mixtures were developed and used as the input parameters for the flash point prediction of mixtures. Both the modeling methods of multiple linear regression (MLR) and multiple nonlinear regression (MNR) were employed to model the possible quantitative relationships between the parameters for mixtures and the flash points of binary miscible mixtures. The resulted models showed satisfactory prediction ability, with the average absolute error for the external test set being 2.506 K for the MLR model and 2.537 K for the MNR model, respectively, both of which were within the range of the experimental error of FP measurements. Model validation was also performed to check the stability and predictivity of the presented models, and the results showed that both models were valid and predictive. The models were further compared to other previously published models. The results indicated the superiority of the presented models and revealed which can be effectively used to predict the FP of binary miscible mixtures, requiring only some common physicochemical parameters for the pure components other than any experimental flash point or flammability limit data as well as the use of the Le Chatelier law. This study can provide a simple, yet accurate way for engineering to predict the flash points of binary miscible mixtures as applied in the assessment of fire and explosion hazards and the development of inherently safer designs for chemical processes.     

A new method for the prediction of flash points for ternary miscible mixtures

The flash point is one of the most important physicochemical parameters used to characterize the fire and explosion hazard for flammable liquids. The flash points of ternary miscible mixtures with different components and compositions were measured in this study. Four model input parameters, being normal boiling point, the standard enthalpy of vaporization, the average number of carbon atoms and the stoichiometric concentration of the gas phase for mixtures, were employed and calculated based on the theory of vapor–liquid equilibrium. Both multiple linear regression (MLR) and multiple nonlinear regression (MNR) methods were applied to develop prediction models for the flash points of ternary miscible mixtures. The developed predictive models were validated using data measured experimentally as well as taking data on flash points of ternairy mixtures from the literature. Results showed that the obtained average absolute error of both the MLR and the MNR model for all the datasets were within the range of experimental error of flash point measurements. It is shown that the presented models can be effectively used to predict the flash points of ternary mixtures with only some common physicochemical parameters.     

Development of machine learning based prediction models for hazardous properties of chemical mixtures

Lower flammability limit (LFL), upper flammability limit (UFL), auto-ignition temperature (AIT) and flash point (FP) are crucial hazardous properties for fire and explosion hazards assessment and consequence analysis. In this study, a comprehensive prediction model set was constructed by using expanded chemical mixture databases of chemical mixture hazardous properties. Machine learning based gradient boosting quantitative structure-property relationship (GB-QSPR) method is implemented for the first time to improve the model performance and prediction accuracy. The result shows that all developed models have significantly higher accuracy than other regular QSPR models, with the 5-fold cross-validation RMSE of LFL, UFL, AIT, and FP models being 1.06, 1.14, 1.08, and 1.17, respectively. All developed QSPR models can be used to estimate reliable chemical mixture hazardous properties and provide useful guidance in chemical mixture hazard assessment and consequence analysis.