The Prediction and Optimization of Biochar Yield Based on the Feedstock Characteristics and Pyrolysis Parameters Using Machine Learning

Lignocellulosic biomass, as one of the most abundant and promising renewable raw materials producing biofuels, have attracted extensive attention for the dual pressures from increasing energy demand and environmental pollution caused by fossil fuel. The feedstock was derived from the living or waste plants such as woody and agricultural residue. Biochar was the solid by-product of biomass conversion, which has been applied in multidisciplinary due to the higher surface area, microporous structures, and containing functional groups. Generally, the functional groups on the biochar surface area decreased and the surface area increased along with the pyrolysis temperature. Biochar with different characteristics could be applied for different fields. For example, the higher surface area and microporous structure was the most important factor for CO2 capture, while the adsorption of biochar for heavy metals was significantly influenced by the functional groups and containing mineral. Accordingly, improving the biochar yield was one of the important research fields based on biomass properties. Machine learning method was used in the study to predict the biochar yield based on the structural and elementary components of biomass, the particle size of biomass and the pyrolysis parameters. The relative contribution of the influencing factors to biochar yield was also estimated. The results demonstrated the influence of pyrolysis temperature was overwhelming for biochar yield. The partial dependence plots were also performed to gain insights into the relationship between biochar yield and each feature or the synergy effect of any two features on the biochar yield, which could guide the selection of pyrolysis parameters based on different biomass properties.