Machine learning prediction of bio-oil characteristics quantitatively relating to biomass compositions and pyrolysis conditions

• Published in: Fuel (Guildford) Vol. 312; p. 122812
• Main Authors: Zhang, Tonghuan, Cao, Danyang, Feng, Xin, Zhu, Jiahua, Lu, Xiaohua, Mu, Liwen, Qian, Hongliang
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 15.03.2022; Elsevier BV
• Subjects: Bio-oil; Biomass; Calorific value; Carbon; Chemical composition; Error analysis; Heating rate; Learning algorithms; Machine learning; Oil; Predictions; Pyrolysis; Regression analysis; Pyrolysis; Biomass; Machine learning; Bio-oil
• ISSN: 0016-2361; 1873-7153
• DOI: 10.1016/j.fuel.2021.122812

•Data of bio-oil characteristic related to pyrolysis were collected systematically.•Regression of viscosity, CV, H/C and O/C was realized via machine learning.•Bio-oil was analyzed from three aspects: quantity, quality and compositions. It is crucial to predict the characteristics of pyrolytic bio-oil accurately for its application, but the prediction results are greatly influenced by biomass compositions and pyrolysis conditions. In this work, different biomass compositions analysis (chemical compositions, ultimate and proximate analysis) and pyrolysis conditions (particle size, heating rate and pyrolysis temperature) were successfully used as input to analyze the characteristics of bio-oil by machine learning method. The model based on ultimate analysis is better for regression analysis of the yield, viscosity and oxygen-carbon ratio (O/C) of bio-oil. The model based on chemical compositions is better for regression analysis of calorific value and hydrogen-carbon ratio (H/C) of bio-oil. Moreover, relative error analysis and scatter diagrams were used to analyze the predicted results. In addition, the analysis of partial dependence diagram shows the influence of various factors and the interactions on the target variables. This study provides feasible thinking for the prediction of the characteristics of bio-oil obtained by biomass with different compositions under different pyrolysis conditions.