开发了结构性质关联模型（QSPR），实现了基于烃类化合物的结构特征预测黏温特性的功能。搜集了254种烃类化合物不同温度下的黏度数据，选择改进的Andrade方程来描述烃类化合物的黏温特性曲线，并通过对实验数据进行回归，获得了化合物的Andrade方程参数B和T₀。在此基础上，采用分子质量和15个基团作为分子的结构特征参数，建立神经网络模型预测Andrade模型参数B和T₀，平均相对误差分别为3.59%和1.27%。通过所预测的Andrade模型参数计算化合物的黏温性能，预测值与实验数据相比绝对平均误差为0.42 mPa·s。

In this paper,a quantitative structure property relationship (QSPR) model is developed to predict the viscosity vs.temperature profile based on structural features of hydrocarbon compounds.The viscosity data sets of 254 hydrocarbon compounds at different temperatures are collected.An improved Andrade equation is chosen to describe viscosity vs.temperature curves of hydrocarbon compounds and two Andrade equation parameters (named B and T₀) of compounds are obtained through regressing the experimental data.On this basis,an associated model is constructed between structural parameters of compounds and Andrade equation parameters.Molecular weights and 15 groups are served as the structure characteristic parameters of molecular to establish an artificial neural network (ANN) model to estimate Andrade model parameters B and T₀ in terms of structural information of compounds.The average relative errors are 3.59% and 1.27%,respectively.The viscosity vs.temperature performances of compounds are calculated based on the predicted Andrade model parameters and the absolute mean deviation of predicted values is 0.42 mPa·s compared with experimental data.