Analysis of influencing factors on the properties of coal-to-direct liquefied diesel

To improve the quality of coal-to-direct liquefied diesel, with the help of machine learning method, the properties prediction model of coal-to-direct liquefied diesel was established, in which the chemical structure and diesel properties of each component of a coal-to-direct liquefied diesel were studied. The oil sample used was the finished diesel from a coal-to-direct liquefaction facility at Erdos in 2023 with an annual oil production of one million tons. Descriptive statistics and correlation analysis were conducted on the hydrocarbon composition of the oil sample and the properties of the diesel. It was found that the hydrocarbon composition was predominantly composed of paraffin and cycloparaffin, accounting for 96.57% of the total hydrocarbon composition, with the monocycloparaffin being the most abundant. The analysis of the diesel quality test results showed that the diesel met the commercial diesel quality specifications, with good combustion performance, low-temperature fluidity, and environmental performance. From Pearson correlation coefficient, it was found that some variables had a high degree of correlation. To avoid the impact of multicollinearity on the model interpretation, a tree model algorithm was chosen to establish the model. Random forest (RF) algorithm, light gradient boosting machine algorithm and extreme gradient boosting algorithm were individually used to establish the prediction model that can evaluate the physical characteristic properties of coal-to-direct liquefied diesel, such as density, kinematic viscosity and cetane number, respectively, and the fitting of each algorithm to the diesel combustion performance was compared and analyzed. The results show that the RF model has good fitting performance and high accuracy. On the training set, the determination coefficients (R²) of density, kinematic viscosity and cetane number were 0.946, 0.916 and 0.814, respectively, and the mean absolute percentage error were 0.073, 0.646 and 0.4, respectively. On the test set, the determination coefficients (R²) for density, kinematic viscosity, and cetane number were 0.976, 0.865, and 0.765, respectively, while the corresponding mean absolute percentage error were 0.48, 2.86, and 0.957, respectively. The analysis showed that the contents of paraffin, tricyclic alkane and alkyl benzene had significant effects on the density, kinematic viscosity and cetane number of coal-to-direct liquefied diesel, while the contents of naphthalene and tricyclic aromatic hydrocarbons had little effect on the above properties. Increasing paraffin content will reduce the density and kinematic viscosity of coal-direct liquefied diesel, but will help increase the cetane number of diesels. The increase in the tricyclic alkane content and the alkyl benzene content will increase the density and kinematic viscosity of coal-direct liquefied diesel, but will reduce the cetane number of diesels.