Chemical Engineering Communications, Vol.206, No.2, 247-253, 2019
Prediction of upper flammability limits for fuel mixtures using quantitative structure-property relationship models
Three efficient and accurate QSPR models for predicting upper flammability limits (UFLs) of hydrocarbon mixtures were built solely based on chemical structures and mole fractions of pure compounds. Firstly, experimental UFLs of 78 blended fuels were gathered from a single reference and molecular descriptors of the nine pure chemicals were calculated by Gaussian 09. Then support vector machine (SVM) analysis was carried out for the development of QSPR models and three totally different external validation strategies were applied for testing the true external predictive capabilities of the models. It was found that all the models perform well, and hence, they are qualified for predicting UFLs of blended hydrocarbon fuels without experimental data. The applicability domains (ADs) of the three novel models were also discussed and all the points are within the AD area. This is the first time that QSPR methodology and quantum chemistry were combined to study the UFL property for blended hydrocarbon fuels. The developed models can serve as an excellent alternative for determining the UFLs of blended hydrocarbons, thereby eliminating any form of danger existing in experimental measurements.
Keywords:Fuel mixtures;Quantitative structure-property relationship;Support vector machine;Upper flammability limits