Interpreting the neural networkfor prediction of fermentation of thick juice from sugar beet processing

Abstract

Methods that can provide adequate accuracy in the estimation of variables from incomplete information are desirable for the prediction of fermentation processes. A feed-forward back-propagation artificial neural network was used for modelling of thick juice fermentation. Fermentation time and starting sugar content were usedas input variables, i.e. nodes. Neural network had one output node (ethanol content, yeast cell number or sugar content). The hidden layer had nine neurons. Garson's algorithm and connection weights were used for interpreting neural network. The inadequacy of Garson's algorithm can be seen by comparing with the results of regression analysis, which indicates that the influence of the fermentation time is higher. A better agreement of the results was obtained using network connection weights, a method that can be used to determine the relative importance of input variables.

Keywords

• bioethanol
• fermentation process
• neural networks
• Garson equation
• connection weights