Modelling of yields in torrefaction of olive stones using artificial intelligence coupled with kriging interpolation

Abstract

A predictive model is developed using an artificial neural network (ANN) to calculate the solid-liquid, and gas yields (wt %) from the torrefaction of olives stones, based on the material and process parameters. These parameters are average olive stone particle size, reaction temperature and reaction time. Ordinary Kriging interpolation is coupled with ANN to improve the experimental data resolution by increasing the data points used in building the ANN models. This coupling improved the ANN prediction accuracy (R2) by 11.1%, 13.5%, and 1.0% in training and 27.3%, 8.5%, and 14.8% in validation of the solid, liquid and gas yields, respectively. Also, the mean absolute deviations of the models significantly improved after the coupling. The prediction profiles show a linear relationship between the solid and liquid yields and a nonlinear relation for the gas yields in terms of the material and process parameters. Average olive stone particle size showed the highest effect on the yields due to the improvement in heat transfer with the exposed surface area of the olive stones leading to a faster reaction rate.