Machine learning prediction of biochar physicochemical properties based on biomass characteristics and pyrolysis conditions

Pyrolyzing waste biomass into functionalized biochar is aligned with the concept of the circular economy. The physicochemical properties of biochar are influenced by the type of biomass feedstock and pyrolysis parameters, necessitating significant time, energy, and resources for quantification. This study employed machine learning algorithms to predict the yield, elemental distribution, and degree of aromatization of biochar based on the physical and chemical properties, as well as the pyrolysis conditions of biomass. Support vector machines (SVM), multiple linear regression (MLR), nearest neighbor algorithm (KNN), random forest (RF), gradient boosting regression (GBR), and eXtreme Gradient Boosting (XGB) were comparatively analyzed. Among these algorithms, the XGB algorithm performed well in predicting biochar production and element distribution (R2>0.99). Furthermore, PCC and SHAP analyses revealed a strong positive correlation between pyrolysis temperature and the degree of aromatization in biochar. Therefore, selecting the appropriate ML model can aid in predicting the physicochemical properties of biochar from diverse biomass sources without the necessity for complex and energy-intensive pyrolysis experiments.