A Research on the Prediction of Superheat Limit Temperature of Hydrocarbons and Their Derivatives Based on QSPR

This research employed the quantitative structure-property relationship (QSPR) approach to predict the superheat limit temperature (SLT) of 64 hydrocarbons and their derivatives. Three models were constructed by using machine learning methods, namely a multiple linear regression (MLR) model, an extreme learning machine (ELM) model, and a support vector machine model based on particle swarm optimization (PSO-SVM). The multiple correlation coefficient (R²), the root mean square error (RMSE), and the mean absolute error (MAE) were adopted to evaluate the model fitting ability. The leave-one-out cross-validation coefficient (Q²_lOO) was employed to assess the model stability, the external validation coefficient (Q²_ext) was used to evaluate the model’s external prediction ability, and the Williams plot was drawn to assess the model’s generalization ability. The results demonstrated that the R², Q²_lOO, and Q²_ext of the training and test sets of the three models were all above 0.9, and the arm ratio of the majority of compounds (96.88%) in the Williams plot was within the warning value, indicating that all the three models were suitable for predicting the SLT of hydrocarbons and their derivatives. By comparing the performance parameters of the three models, the PSO-SVM model achieved the best performance across all parameters, suggesting that there exists a strong nonlinear relationship between the molecular structure of hydrocarbons and their derivatives and the SLT. The utilization of the QSPR method to predict the SLT of hydrocarbons and their derivatives can provide powerful theoretical support for the safe design and control of industrial operations.