Objective The aim is to establish a property prediction model for diesel fuel using compositional features, to realize efficient property prediction, and to analyze the effect of compositional features on properties.

Method Compositional features were calculated from 92 sets of simulated diesel fuel samples, and suitable descriptors were selected based on the correlation analysis between the compositional features (average relative molecular mass, hydrogen-carbon ratio, carbon number, unsaturation, average molecular aromatic hydrocarbon rings, average molecular cycloalkane rings, and average molecular total rings) and the properties (density, refractive index, kinematic viscosity, dynamic viscosity, surface tension, and integral of the distillation curve). The predictive model of compositional feature-property was established and optimized by multiple linear regression, the composition feature scales were unified by standardizing, and the relative importance of compositional features to properties was analyzed by re-modeling.

Result The initial model (hydrogen-carbon ratio and carbon number) showed good performance in predicting density, refractive index, and integral of distillation curve (determination coefficient R²＞0.85), but the predictions for viscosity and surface tension were not accurate enough. After optimization, the introduction of aromatic and cycloalkane ring numbers significantly improved the model performance, with R²＞0.90 for all properties, mean absolute percentage error δ_MAPE＜5%, and R²＞0.95 for density and refractive index. The analysis showed that the carbon number dominated the viscosity and distillation curves, and the aromatic and cycloalkane ring numbers had a significant effect on the density.

Conclusion Aromatic ring number, naphthenic ring number, and carbon number are the key descriptors affecting the basic physicochemical properties of diesel fuel, and models based on these descriptors can accurately predict such properties. This study provides a reliable quantitative analysis tool for diesel fuel composition design and performance optimization, which is of great significance in guiding the research and development of fuels.