Premium
Machine learning prediction of hydrocarbon mixture lower flammability limits using quantitative structure‐property relationship models
Author(s) -
Jiao Zeren,
Yuan Shuai,
Zhang Zhuoran,
Wang Qingsheng
Publication year - 2020
Publication title -
process safety progress
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.378
H-Index - 40
eISSN - 1547-5913
pISSN - 1066-8527
DOI - 10.1002/prs.12103
Subject(s) - quantitative structure–activity relationship , boosting (machine learning) , support vector machine , random forest , hydrocarbon , mean squared error , machine learning , property (philosophy) , artificial intelligence , linear regression , flammability , cross validation , decision tree , flammability limit , computer science , mathematics , statistics , chemistry , combustion , organic chemistry , philosophy , epistemology
Lower flammability limit (LFL) of hydrocarbon mixture is a critical property for fire and explosion hazards. In this study, by using experimental LFL data of hydrocarbon mixture from a single reference, quantitative structure‐property relationship (QSPR) models have been established using four machine learning methods, namely, k ‐nearest neighbors, support vector machine, random forest, and boosting tree. The K ‐fold cross‐validation method, which has significant advantages over the traditional validation set approach, is implemented for QSPR model evaluation. Prediction errors and accuracy are assessed and compared with traditional multiple linear regression. The results show that models generated by machine learning methods have a significantly lower root mean square error than traditional methods in both training and test data sets. This is the first time that machine learning‐based QSPR models are developed for prediction of hydrocarbon mixture LFL, and the models are proven to be highly predictable and reliable.