Fast Characterization of Biomass Pyrolysis Oil Via Combination of Atr-Ftir and Machine Learning Models

Bio-oil is a promising carbon neutral and sustainable alternative for liquid fossil fuels, but the difficulties in property characterization is hindering its further development. Accordingly, this study proposed a fast characterization method of bio-oil via combination of attenuated total reflection flourier transformed infrared spectroscopy (ATR-FTIR) and machine learning model. The model framework contains a principal component analysis (PCA) section to dimensionally reduce the spectral data, and a support vector regression (SVR) section to obtain the property results of bio-oil. The performance and the working mechanism of the model framework were comprehensively discussed. The results showed that PCA was capable for enhancing the predicting accuracy, where principal components 6 and 7 contained little spectral information but play the most important roles. The SVR model with linear kernel was accurate with high correlation performance as the regression segment for predicting the properties of bio-oil. Under optimal parameters, the predicting accuracies for unsaturated concentration, effective hydrocarbon ratio, low heating value, C content, H content, and O content reached 91.98%, 97.44%, 99.50%, 98.65%, 98.56%, and 97.88%, respectively. This study is promisingly potential for the quality control of bio-oil production industry, and is hoped to significantly promote the development and application of liquid biofuels