Abstract
This study aimed to simplify the complex mechanisms of biomass pyrolysis by developing a quasi-single reaction model suitable for simulation applications. To reduce the time and cost associated with experimental determination of pyrolysis products, a multiple-input multiple-output Artificial Neural Network (MIMO-ANN) was employed to predict 11 key variables (the atomic subscripts and stoichiometric coefficients of biomass, char, and gaseous products) in the reaction equation. The input variables were based on proximate and/or ultimate analyses of biomass and the pyrolysis temperature. Several combinations of output and input selection methods were investigated. The best performance was achieved using an ANN model that grouped outputs into three categories (biomass, char, and gas components) and used ultimate analysis with temperature as inputs. This approach produced strong agreement with experimental data from five local biomasses tested at three temperatures. The root mean squared percentage errors (RMSPEs) for all predicted variables ranged from 0.26 % to 27.94 %. Notably, the gas component predictions showed relatively higher RMSPEs (9.18 %–27.94 %), likely due to experimental and modeling limitations. Overall, the proposed ANN framework with the optimal output–input selection method demonstrated reliable predictive performance and offers a practical tool for supporting pyrolysis simulations across a wide range of biomass types.
Publication details
- Authors: Phuakpunk, K., Chalermsinsuwan, B., Assabumrungrat, S.
- Published in: (2025) Alexandria Engineering Journal, 130, pp. 175-197.
- Year: 2025
- DOI: 10.1016/j.aej.2025.09.003