Modelling of biomass gasification for fluidized bed in Aspen Plus: Using machine learning for fast pyrolysis prediction

Abstract

The potential offered by biomass to upgrade into more valuable products via gasification is now being widely recognized globally. Due to difference of pyrolysis conditions, conventional Aspen modelling is challenged for bubbling fluidized bed(BFB) biomass gasification. In this work, a novel approach is developed for Aspen biomass gasification in BFB, combined with machine learning. Machine learning is utilized for biomass fast pyrolysis char and gas prediction. A sub-model for pyrolysis products evolution lumping equilibrium is then established via element balance calculation for predicted fast pyrolysis products compositions. Subsequent gasification in gasifier is controlled kinetically. Evaluation and discussion have been carried out on the method feasibility and precision of gasification products prediction in current model. Comparative analysis with six sets of experimental data reveals that most relative errors of syngas composition are controlled within ± 20 %, with half of them falling within ± 10 %. New model demonstrates satisfying accuracy and adaptability for different feedstock, attributed to application of machine learning in fast pyrolysis products prediction. Sensitivity analysis confirms current model’s capability to simulate trends of syngas compositions under varying gasification conditions correctly. Modules contribution analysis indicates that further promotion of accuracy can be achieved by refining tar cracking prediction and element equilibrium. Through present method, modelling for feedstock whose pyrolysis kinetics are unknown is not limited to thermodynamic equilibrium and can obtain higher accuracy and feedstock scalability. It provides original insight for more reasonable Aspen modelling and comprehensive usage of Aspen-machine learning combination.