Hybrid residual modelling of biomass pyrolysis

Predicting biochar yield and heating value poses a great challenge due to the complexity of the biomass pyrolysis process. Here, we propose a hybrid residual (HR) modelling approach by combining the residual of the pyrolysis equilibrium model and the Random Forest (RF) model. Then, the separate pyrolysis equilibrium model and the RF model were also established for comparison. These three models were evaluated in terms of R² and RMSE, and results revealed that the HR model exhibited the optimal performance with R² values of 0.890 and 0.956 for yield and higher heating value (HHV) of biochar, respectively. Permutation importance analysis and SHAP value analysis were conducted to rank the feature importance of the HR model. The results indicated that the ash, pyrolysis temperature, and residual are important for biochar yield; the ash, carbon, and residual are crucial for the HHV of biochar in the biomass pyrolysis process. Furthermore, simplified interpretable equations were developed based on the identified features. In conclusion, this work demonstrated through simple case studies that combining a mechanism model with a machine learning model can accurately establish complex process modelling and offer simplified equations.