A hybrid approach combining mechanism-guided data augmentation and machine learning for biomass pyrolysis

Abstract

Process accuracy modelling requires comprehensive thermodynamics and kinetic knowledge, which is not feasible for complex biomass pyrolysis. Here, we propose a hybrid modelling strategy that combines mechanism-guided modelling (Aspen Plus) with data-driven machine learning (ML) models to accurately predict biomass pyrolysis products. Specifically, we establish the pyrolysis equilibrium model based on the first principles and Gibbs free energy minimization approach; then, the equilibrium model is leveraged to augment data by selecting points with smaller residuals; finally, the augmented data and experimental data serve as input for the ML models training. The results revealed that the hybrid model exhibited superior performance, with average R2 and MAPE values of 0.981 and 0.266, respectively, for predicting biochar yield and heating value. Conclusively, this work introduces new insight and strategy to accelerate the engineered application of bioenergy via modelling biomass pyrolysis and screening biomass.