Using Artificial Neural Networks to Estimate Xylose Conversion and Furfural Yield for Autocatalytic Dehydration Reactions

This study developed a feed-forward artificial neural network (ANN) model with a single hidden layer to predict xylose conversion and furfural yield from autocatalytic reactions in various organic solvent systems. It is known that the reaction severity, a function of temperature and time, and the polarity of the solvent, as determined by the Hansen Solubility Parameters, affect the reaction, and therefore, these two parameters were chosen as the independent variables for the investigation. Reactions were performed between a severity of 3.53 and a severity of 5.20 with solvent polarities ranging from 7 to 16. The ANN model performance was determined by prediction error indices and the Akaike information criterion, which resulted in the best model having six hidden nodes. The ANN confirmed that significantly higher xylose conversions and furfural yields were seen in mixtures with polarities above 15 at severities greater than 4.33 compared to lower polarities and severities. The estimated and predicted values from the model were all within the 95% prediction confidence band region, indicating that the model has accurate prediction capabilities within the data range used to develop the model.