Rapid characterization of SAF based on ATR-FTIR and machine learning models

Air transportation plays an emerging role in carbon emission of transportation section, leading to a worldwide attention about the development of SAF. Due to the heterogeneous feedstock supply of SAF, it is necessary to characterize SAF properties for quality control. The study proposes a rapid characterization method for SAF by integrating Attenuated Total Reflectance Fourier Transform Infrared Spectroscopy (ATR-FTIR) with machine learning models. The input to the model consists of high-dimensional infrared spectral data. Key performance indicators for SAF include flash point, H/C ratio, LHV, C content, and H content. Model parameters were optimized based on prediction accuracy and correlation coefficients. A Support Vector Regression (SVR) model, an Artificial Neural Network (ANN) model, and a Random Forest (RF) model were established using Principal Component Analysis (PCA) and compared. The results showed that, under optimal parameters, the lowest Mean Relative Error (MRE) for flash point, H/C ratio, LHV, C content, and H content were 1.01%, 3.34%, 1.65%, 0.57%, and 2.97%, respectively. The average R 2 was 0.94. Additionally, the analysis of the optimal principal components (PC) in the infrared spectra revealed that PC10 and PC19 had the greatest impact on prediction performance.