Chemical reaction neural networks for semi-detailed kinetic schemes including the influence of inorganics and secondary reactions during cellulose pyrolysis

The demand for renewable energy and chemicals is increasing and biomass, as one of the most important sources of renewable energy and green carbon, is attracting significant attention in recent years. To understand and model biomass pyrolysis, detailed kinetic schemes are necessary. Conventional kinetic analysis requires a lot of prior knowledge and chemical insights. Due to the complex processes involved in biomass pyrolysis, conventional kinetic analysis methods face noticeable challenges, especially when considering the heterogeneity of the material, the influence of inorganic species, and the coupling of transport limitations with chemical pathways. Benefiting from available experimental data and rapidly progressing machine-learning techniques, a data-driven method based on the chemical reaction neural network (CRNN) is employed in the present study to determine a new semi-detailed kinetic scheme for the pyrolysis of cellulose, accounting for the impact of inorganic species (potassium, K) and mild transport limitations. Experiments have been conducted with different initial masses (5 and 20 mg), heating rates (2.5, 5, and 10 K min −1 ), with/without K (catalyst), with/without lid (external transport limitations). The obtained kinetics are physically interpreted and finally validated against conventional data-fitting methods. The CRNN leads to a semi-detailed kinetic scheme for modeling purposes. • A novel AI methodology to identify kinetics including catalytic and secondary reactions. • Novel kinetic schemes for cellulose pyrolysis process under various working conditions. • Our methodology was validated against various state-of-art data-fitting methods. • The data-driven model was further optimized with prior knowledge.