Intelligent Analysis of Direct Coal Liquefaction Diesel Components by Near-Infrared Spectroscopy
Direct coal liquefaction (DCL) diesel constitutes over 60% of DCL products, yet its cetane number (30–40) falls short of the automotive diesel standard (≥45). Rapid and accurate compositional analysis is essential for optimizing properties via component blending. Traditional gas chromatography offers high accuracy but is unsuitable for online industrial monitoring. Near-infrared (NIR) spectroscopy enables rapid, non-destructive analysis, but spectral interpretation is complex. This study integrates NIR spectroscopy with machine learning (ML) to construct a spectral-composition database for DCL diesel. Feature extraction using correlation coefficient and mutual information methods screened key wavelength variables, reducing dimensionality from ~1800 to ~200 wavelengths. Three ML models—Lasso, SVR, and XGBoost—were compared. Excluding spectral data with absorbance >1 significantly improved model accuracy, increasing test set R² from 0.85 to 0.96. After feature extraction, the optimal variable count was 177, enhancing computational efficiency. Among models, SVR-MI-0.9 (mutual information feature selection) achieved the best performance, with training and test set R² values exceeding 0.98, enabling precise prediction of paraffin, naphthene, and aromatic contents. This research provides a robust methodology for intelligent online quality monitoring. An intelligent NIR spectroscopy data analysis software was independently developed based on the established model. Compared with comprehensive two-dimensional gas chromatography, the software reduced analysis time by over 98%, with absolute prediction error below 0.2%. Thus, rapid analysis of DCL diesel components was successfully realized.