Prediction of gasoline research octane number using multiple feature machine learning models

Abstract

Liquid hydrocarbons are typically used as fuels in the internal combustion engines due to their high volumetric energy density and ease of handling, storage, and transportation. Gasoline is the most extensively used fuel for the light-duty automobiles and passenger cars. Research octane number (RON), among other auto ignition related properties, is a primary indicator of the grade of the light-duty automobile fuels. Because measuring the RON is expensive and time-consuming, an alternative cost-effective grading method for engines is desired, for example, methods that in corporate machine learning. However, the physicochemical properties of the constituents of gasoline have a notable nonlinear relationship with the RON of the fuel. The accurate blending and production of commercial spark ignition engine fuels is limited by the lack of precise predictive models for RON. In this study, we developed a high-throughput method that accounts for the properties of gasoline and uses the Random Forest (RF) algorithm to predict the RON of gasoline. The maximum error of the RF model in RON prediction was 0.46, which was close to the range of experimental uncertainty. The experimental results indicated that the properties of gasoline can be used to predict its RON, and the complex relationships among the constituents can be analyzed using tools such as the RF algorithm.