Prediction of diesel cetane number based on physicochemical properties and hydrocarbon compositions using XGBoost model

Diesel fuel is a complex mixture of petroleum compounds, and its most important physical parameter is the cetane number (CN). The traditional diesel CN measurement and analysis methods are time-consuming, costly, and easy environmental pollutants. Meanwhile, because of the complex diesel hydrocarbon composition, complex nonlinear coupling relationships exist between diesel hydrocarbon composition and diesel properties. These relationships considerably affect the stability and generalization ability of diesel CN prediction models. In this study, an improved XGBoost model was developed to predict the CN of diesel. The model is based on the physicochemical properties and different hydrocarbon compositions of diesel. The mean absolute error and mean squared error of independent test sets were 0.293 and 0.143, respectively, and the R2 was 0.964. The experimental results indicated that the physicochemical properties and different hydrocarbon compositions of diesel can be used to predict its CN. Moreover, the complex relationships among the constituents can be ascertained using this XGBoost algorithm. Consequently, this method effectively predicts the CN of the diesel fuel, which could greatly decrease the cost of the experiments and accelerate the research and development of high performance diesel.