| Abstrakt: |
Fermentable sugar release is a crucial step in the fermentation of lignocellulosic biomass to biofuels. The efficiency of fermentable sugar release from pretreated biomass can be improved through acquiring pre-experimental information from predictive models that function as initial screening tools during lignocellulose bioprocessing. In this study, artificial neural networks, random forest, and decision tree were used to model fermentable sugar concentration (g/L) and combined severity factor (CSF) index from a pretreated starch-based lignocellulosic biomass-cassava peels. The input parameters for all the models include soaking temperature, soaking time, autoclave duration, hydrochloric acid concentration, and percentage solid loading. A backpropagation training algorithm was adopted for the ANN models and trained on experimental data while the random forest models were trained using a binary tree–based machine learning algorithm and the decision tree models were trained using an information entropy–based learning algorithm. For the fermentable sugar concentration, the highest coefficient of determination (R2> 0.99) was achieved with the decision tree model while the highest prediction accuracy of 82% was achieved with the ANN model (R2> 0.82). For the CSF index, the ANN model gave both the highest coefficient of determination (R2> 0.93) and prediction accuracy (78%). Thus, the ANN models gave better predictions of the fermentable sugar concentration and the combined severity factor index compared to the other intelligent models. The study, therefore, provides an insight into intelligent predictive modelling towards the screening of pretreatment strategies for starch-based lignocellulosic biomass valorization. |
Full text from SpringerLink