• • Excluding NIR spectral data with absorbance >1 increased test set R² from 0.85 to 0.96, a 12.9% improvement in predictive accuracy, critical for reliable online quality control in DCL diesel blending.
• • Feature extraction reduced wavelength variables from ~1800 to 177 (a 90.2% reduction), substantially improving computational efficiency without sacrificing accuracy, enabling real-time analysis in industrial settings.
• • The SVR-MI-0.9 model achieved training and test set R² values exceeding 0.98, demonstrating high precision in predicting paraffin, naphthene, and aromatic contents, essential for meeting cetane number specifications.
• • The developed intelligent software reduced analysis time by over 98% compared to comprehensive two-dimensional gas chromatography, with an absolute prediction error below 0.2%, facilitating rapid, non-destructive online monitoring.