Predictive Analytics for Octane Number: A Novel Hybrid Approach of KPCA and GS-PSO-SVR Model

Abstract

Octane number is the most important indicator of reflecting the combustion performance, and a great deal of research has been devoted to improving it. In this paper, a new analytical framework is proposed to predict octane number, kernel principal component analysis (KPCA) is used to reduce the dimension of the variables in the process of Fluid Catalytic Cracking (FCC), support vector regression (SVR) is used to construct the gasoline octane number prediction model and the particle swarm optimization algorithm (PSO) is used to select the optimal combination of parameters for the model. The experiments show that the octane number can be improved under a given production environment with a guaranteed desulfurization effect of gasoline products. Furthermore, several key attributes that have a significantly positive or negative correlation with the improvement of gasoline product quality are identified through computing the feature score. The findings can help engineers adjust operational variables to obtain a series of high-quality products.