Utilizing Distillation Curve and Refractive Index to Predict Property of Hydrocarbon Mixtures

In this study, the property prediction models for density, surface tension, kinematic viscosity, dynamic viscosity, and flash point of diesel were developed by the feedforward neural network method utilizing the fuel distillation curve (T₁₀-T₉₀) as light and heavy composition distribution characteristics and the refractive index as the average composition characteristic. The applicability of the method across hydrocarbon mixtures of diesel and gasoline-kerosene-diesel fractions was explored. 92 surrogate diesel samples, ranging from binary to septenary mixtures, were prepared, and their refractive index, distillation curve, density, flash point, and surface tension were measured. The property prediction models were developed using 100 gasoline-kerosene and 92 diesel fraction samples. According to SHAP value analysis, the density of diesel is determined by the refractive index, while the flash point, surface tension, and viscosity depend on the light component (T₁₀), the refractive index and the middle boiling component (T₅₀), and both light and middle boiling components (T_30-T₅₀), respectively. The determinants of density and flash point of gasoline-kerosene-diesel fraction fuel are consistent with the model of diesel fraction. The secondary influencing factor of surface tension shifts to lighter composition (from T₅₀ to T₁₀), and the factors influencing viscosity are more complex, as it is affected by the middle boiling component.