Comparison between artificial neural networks and mathematical models for estimating equilibrium moisture content in raisin

Abstract

Empirical models and Artificial Neural Networks (ANNs) were utilized for the prediction of Equilibrium Moisture Content (EMC) in raisin.  Six empirical models including GAB, Smith, Henderson, Oswin, Halsey and D’Arsy-watt were applied for this estimation.  Two types of Multi Layer Perceptron (MLP) neural networks entitled Feed Forward Back Propagation (FFBP) and Cascade Forward Back Propagation (CFBP) were used.  In order to train the input patterns, two training algorithms consist of Levenberg-Marquardt (LM) and Bayesian regularization (BR) were used.  Thermal and relative humidity limits were 30-80℃ and 10.51%-83.62%, respectively.  The best result for mathematical models belonged to D’Arsy-Watt with R² and the mean relative error of 0.9943% and 10.84%, respectively.  The best outcome for the use of ANN also appertained to FFBP network with LM training algorithm, topology of 2-3-3-1 and threshold function order of TANSIG-TANSIG-PURELIN.  With this optimized network, R² and the mean relative error was 0.9969% and 8.32%, respectively.  These results show the supremacy of ANN, in comparison with empirical models.  In order to predict the EMC in raisins, empirical models can therefore be replaced with the ANN.

Keywords: ANN, back propagation, sorption isotherm, EMC, Iran

Citation: Chayjan R. Amiri, and M. Esna-Ashari.  Comparison between artificial neural networks and mathematical models for estimating equilibrium moisture content in raisin.  Agric Eng Int: CIGR Journal, 2010, 12(1): 158-166.