基于多元线性回归和反向传播人工神经网络预测离子液体的声速  

Prediction of Sound Speed of Ionic Liquids Based on Multiple Linear Regression and Back-Propagation Artificial Neural Networks

作　　者：季常征 万仁 时兆翀 彭昌军[1] 刘洪来[1] JI Changzheng;WAN Ren;SHI Zhaochong;PENG Changjun;LIU Honglai(School of Chemistry and Molecular Engineering,East China University of Science and Technology,Shanghai 200237,China)

机构地区：[1]华东理工大学化学与分子工程学院,上海200237

出　　处：《华东理工大学学报(自然科学版)》2025年第2期158-165,共8页Journal of East China University of Science and Technology

基　　金：国家自然科学基金(22078086,22378111)。

摘　　要：离子液体的声速可采用实验测定、半经验模型和理论研究方法获得,其中,定量结构-性质关系(QSPR)模型已受到广泛关注,但构造一个有效的QSPR模型取决于选择合适的分子描述符。本文采用片段活度系数类导体屏蔽模型(COSMO-SAC)获得离子液体电荷密度分布片段面积(Sσ)和空穴体积(VCOSMO)两个描述符,并分别采用多元线性回归(MLR)和反向传播人工神经网络(BP-ANN)构建了用于描述离子液体声速的线性QSPR模型u-MLR和非线性QSPR模型u-ANN,模型中包含了温度和离子液体相对分子量,所涉及的数据集包括171种离子液体的5 114个数据点。在总的离子液体声速数据集中,u-MLR和u-ANN的决定系数(R2)分别为0.970 6和0.999 5,平均绝对相对偏差(AARD)分别为1.59%和0.10%,均方根误差(RMSE)分别为30.68 m/s和4.12 m/s。结果表明,基于人工神经网络建立的u-ANN模型的预测效果明显优于基于线性回归方法建立的u-MLR模型的预测效果。Sound speed of ionic liquids(ILs)can be obtained using various methods,including experimental measurements,semi-empirical models,and theoretical research.Among these methods,the quantitative structureproperty relationship(QSPR)model has gained wide attention.However,constructing an effective QSPR model depends on the selection of appropriate molecular descriptors.In this study,there are two descriptors for model construction,namely,the charge density distribution area of ILs at a specific interval obtained using the conductor-like screening model for the segment activity coefficient(COSMO-SAC)and the cavity volume of ILs.Using the multiple linear regression(MLR)approach and the back-propagation artificial neural network(BP-ANN)method,two quantitative structure-property relationship(QSPR)models,namely u-MLR and u-ANN,were proposed respectively to describe the sound speed(u)of ILs.These models incorporated temperature and molecular weight of ILs.The evaluated data set comprised 171 ILs with 5114 data points.For the entire data set,u-MLR and u-ANN exhibited determination coefficients(R2)of 0.9706 and 0.9995,respectively.Furthermore,the calculated average absolute relative deviations(AARD)were 1.59%and 0.10%with root mean square errors(RMSE)of 30.68 m/s and 4.12 m/s,respectively.Therefore,the prediction performance of the u-ANN model based on back-propagation artificial neural networks is significantly better than that of the u-MLR based on multiple linear regression.

关 键 词：声速 离子液体 人工神经网络 多元线性回归 定量结构-性质关系 

分 类 号：TQ021.4[化学工程]