Robust modelling development for optimisation of hydrogen production from biomass gasification process using bootstrap aggregated neural network
Hannah O. Kargbo, Jie Zhang, Anh N. Phan
Abstract
In this study, a robust model using bootstrapped aggregated neural network (BANN) was developed for optimising operating conditions of a two-stage gasification for high carbon conversion, high hydrogen yield and low CO2. The developed BAAN model predicted accurately (R2 of 0.999) the gas composition and the 95% confidence bounds for model predictions on unseen validation data indicated good prediction reliability for various feedstock. The BANN was also used to predict the optimum operating condition for hydrogen production from waste wood (1st stage temperature of 900 °C, 2nd stage temperature of 1000 °C, steam/carbon molar ratio of 5.7) to achieve high hydrogen (71–72 mol%), gas yield (98–99 wt%) and low CO2 (17–18 mol%). The optimal conditions were tested in the laboratory and the experimental results agreed well with the predicted data with an error of 0.01–0.05. Sensitivity analysis revealed that an increase in temperatures for both stages and high steam/carbon ratio favoured the H2 production and carbon conversion.