Biomass microwave pyrolysis characterization by machine learning for sustainable rural biorefineries

Microwave heating is a promising solution to overcome the shortcomings of conventional heating in biomass pyrolysis. Nevertheless, biomass microwave pyrolysis is a complex thermochemical process governed by several endogenous and exogenous parameters. Modeling such a complicated process is challenging due to the need for many experimental measurements. Machine learning can effectively cope with the time and cost constraints of experiments. Hence, this study uses machine learning to model the quantity and quality of products (biochar, bio-oil, and syngas) that evolve in biomass microwave pyrolysis. An inclusive dataset encompassing different biomass types, microwave absorbers, and reaction conditions is selected from the literature and subjected to data mining. Three machine learning models (support vector regressor, random forest regressor, and gradient boost regressor) are used to model the process based on 14 descriptors. The gradient boost regressor model provides better prediction performance (R2 > 0.822, RMSE <12.38, and RRMSE <0.765) than the other models. SHAP analysis generally reveals the significance of operating temperature, microwave power, and reaction time in predicting the output responses. Overall, the developed machine learning model can effectively save cost and time during biomass microwave pyrolysis while serving as a valuable tool for guiding experiments and facili-tating optimization.

Automated summary

This study uses machine learning to model the quantity and quality of products in biomass microwave pyrolysis. Through data mining and three machine learning models, it is found that the gradient boost regressor model provides better prediction performance. The significance of operating temperature, microwave power, and reaction time in predicting the output responses is revealed. This research is valuable for cost and time-saving, as well as guiding experiments and optimization.