Machine learning approach for the prediction of biomass pyrolysis kinetics from preliminary analysis

The pyrolytic behavior of lignocellulosic biomass is highly complex, and its kinetic behavior varies with operating conditions and the type of biomass. To reduce timescales, cost and rigorous calculations associated with new set of experimentation used for the estimation of kinetic parameters, model-based predictions are recommended. In the present work, Artificial Neural Network (ANN) based machine learning models are developed to predict the biomass pyrolysis kinetics. Data sets of thermogravimetric analysis and feedstock characterization from a diverse range of biomass were used to develop and test the networks. Four models were developed in this study based on proximate analysis (ANN-1), ultimate analysis (ANN-2), combined proximate and ultimate analysis (ANN-3) and the combined proximate, ultimate, and biochemical analysis (ANN-4). A total of 704 kinetic datasets were extracted and recalculated with the Coats-Redfern Method from which 662, 585, 465 and 133 datasets were used to develop models sequentially. The developed models, in particular ANN-3 and ANN-4 have shown a competitive prediction capability (R-2 similar to 0.99, RRMSE <10.0%, and MAE < 0.071). Relative importance of each input (biomass properties & heating rate) on outputs (kinetic parameters) was also studied. Biochemical analysis was found to have higher contribution (similar to 38%) in comparison to ultimate (similar to 29%) followed by proximate analysis (similar to 22%) and the pyrolysis kinetics were found to be affected by the heating rate to the extent of similar to 10%. The developed models were found to be accurate enough in predicting the pyrolysis kinetics for any new biomass feedstocks based on preliminary analysis.