This study first aims to determine the physical characteristics of bioethanol from its dielectric properties during the production process. For this purpose, bioethanol samples were produced from sugarcane molasses under different fermentation conditions such as temperature and reaction time. The main permittivity properties, including dielectric constant and loss factor at various frequencies, were measured as the inputs, whereas flash point and octane number, as significant physical parameters, were considered as output variables. Linear regression (LR) models were developed to predict the relationships between predictors and targets. Next, four learning algorithms – multivariate regression splines (MARS), M5 tree, multilayer perceptron (MLP), and support vector regression (SVR) – were deployed to improve the models’ performance. The results from LR models revealed that the flash point had a direct relationship with the dielectric constant and loss factor. However, the octane number was indirectly proportional to the dielectric properties. Also, analysis of the testing data set of learning algorithms indicated that the best algorithm for predicting the responses was MARS, whereas the SVR model performed with the lowest accuracy. The findings from this paper suggest that dielectric spectroscopy is a valuable approach for estimating the physical features of bioethanol.

Keywords