Abstract: Aiming at the detection difficulty of gasoline Research Octane Number (RON) in fuel sales enterprises, a RON prediction method based on random forest regression algorithm was proposed. Based on the fuel quality database, with gasoline olefin content, aromatic content, oxygen content, distillation range (T10, T50, T90 and FBP) and density as independent variables and RON value as the dependent variable, the random forest regression prediction models of NO.92 gasoline, NO.95 gasoline and ( NO.92+ NO.95) gasoline were established. The results showed that the prediction accuracy of models for the NO.92 and the NO.95 gasoline was better, and the coefficient of determination (R2) of the two models both reaches 0.95. After the fuel quality upgraded, the prediction models maintained high accuracy, reliability, and adaptability. Compared with the mid-infrared spectral detection method, the absolute error of more than 84% prediction results of the random forest regression model was less than 0.7, and its accuracy was significantly better than that of the mid-infrared spectral detection method. This prediction model can be helpful for the quality monitoring of gasoline RON of fuel sales enterprises.

Key words: gasoline, research octane number, random forest, regression, prediction