PREDICTIVE MODELING OF SYNGAS COMPOSITION AND GAS YIELD IN CATALYTIC BIOMASS REFORMING USING A MIMO-ANN FRAMEWORK

Abstract

The catalytic reforming of biomass represents an efficient thermochemical pathway for converting solid feedstock’s into combustible gaseous products such as CO, H₂, CO₂, and CH₄. Accurate prediction of these individual gas fractions is essential for optimizing process performance and designing advanced biomass-to-energy systems. In this study, a multiple-input multiple-output (MIMO) Artificial Neural Network (ANN) framework is developed to model and forecast gas yields generated under catalyst-assisted biomass conversion. A dataset of 300 plus experimentally characterized fuel samples, obtained from an operational gasification setup, forms the basis of the model training and evaluation. The ANN architecture incorporates 11 physicochemical and process-related input variables, including elemental composition (C, H, N, S, O), moisture content, ash, temperature, volatile matter, lower heating value, and equivalence ratio and predicts five key outputs: CO, CO₂, CH₄, H₂, and total gas yield. Network training is performed using the Levenberg-Marquardt (LM) and Bayesian Regularization (BR) algorithms to assess their comparative effectiveness. Model performance, evaluated through mean squared error metrics and regression analysis, demonstrates that the LM-trained ANN achieves superior predictive accuracy relative to BR. Overall, the developed ANN models exhibit strong agreement with experimental measurements, highlighting their potential as reliable predictive tools for catalytic biomass reforming processes.