Predicting Gasoline RON Loss by Machine Learning

Abstract

In order to improve gasoline quality and reduce the environmental pollution caused by gasoline combustion, we aim to study the nonlinear relationship between octane loss and various gasoline components in gasoline refining process, further optimize the gasoline refining process, and reduce the content of sulfur and olefin in gasoline products while minimizing the octane loss. Firstly, we used the nonlinear data reduction algorithm t-SNE to preprocess the data of each component in gasoline and clarify each characteristic parameter; then we constructed the Multiple Nonlinear Regression (MNR) model to analyze the correlation and influence of each characteristic data on octane value; finally, we selected NSGA-II multi-objective genetic algorithm. The NSGA-II multi-objective genetic algorithm was selected for in-depth optimization of the model. The multiple linear regression model was also constructed for comparison experiments, and the advantages and disadvantages of the predicted results of the two models were analyzed, as well as the feasibility of extension. In the experiments with industrial data samples published in Sinopec Gaoqiao Petrochemical real-time database (Honeywell PHD) and LIMS experimental database, it was found that the multivariate nonlinear regression model has more superiority in achieving the purpose of the study, the value of harmful content in gasoline was reduced by 32.7%, and the predicted data of each component of gasoline was more in line with environmental protection requirements. Therefore, the multivariate nonlinear model has the highest reasonableness in predicting the octane loss values.