基于遗传函数近似法的液态烃类化合物热导率预测

GFA-based prediction of thermal conductivity of liquid hydrocarbon compounds

针对化合物热导率数据缺失,现有方法估算的热导率误差较大的问题,提出基于遗传函数近似法(GFA)建立估算液态烃类化合物热导率的多元线性定量构型关系(QSPR)模型。收集122种液态烃类化合物在不同温度下的972个热导率数据,用GaussView 6.1.1软件构建这些化合物的分子结构并用Gaussian 16 C01优化分子构型,再用Dragon 6.0软件计算筛选得到382个分子描述符。采用分层随机方法将数据集划分为训练集和测试集,基于训练集采用GFA算法建立了含有5个分子描述符烃类化合物热导率的预测模型。结果表明,测试集的相关系数的平方R(2)_(test)和均方根误差RMSE_(P)分别为0.9069和0.0061,说明该模型具有良好的拟合度和预测能力,同时适用性域(AD)表明预测模型具有较好的泛化能力和鲁棒性。

In order to address the lack of thermal conductivity data of compounds and large errors in thermal conductivity estimation by existing methods,a multivariate linear quantitative conformational relationship(QSPR)model based on genetic function approximation(GFA)was established for estimating thermal conductivity of liquid hydrocarbon compounds.972 thermal conductivity data of 122 liquid hydrocarbon compounds at different temperatures were collected,and the molecular structures of these compounds were constructed with GaussView 6.1.1 software and optimized with Gaussian 16 C01 for molecular conformation.The data was then screened with Dragon 6.0 software to obtain 382 molecular descriptors.A stratified random method was used to divide the data set into a training set and a test set,and a prediction model for the thermal conductivity of hydrocarbon compounds containing five molecular descriptors was developed based on the training set using the GFA algorithm.The results show that the squared correlation coefficient(R^(2)_(test))and root mean square error(RMSE)of the test set are 0.9069 and 0.0061,respectively,which indicates that the model has good fitting and external prediction ability.Moreover,the applicability domain shows that the prediction model has good generalization ability and robustness.