Comparison of genetic algorithm and neural network approaches for the drying process of carrot

Abstract

Drying kinetic of carrot was investigated considering different drying conditions, in this study. The drying experiments were performed at four levels of drying air temperatures of 60–90°C, together with three levels of air flow velocities of 0.5–1.5m/s, and also three levels of thickness 0.5–1cm. Four different mathematical models available in the literature were fitted to the experimental data. Among the considered mathematical drying models, modified Page model, was found to be more suitable for predicting drying of carrot. In order to optimize mathematical models obtained by using regression analysis, genetic algorithm was used. In all stages of the mathematical modeling, genetic algorithms were applied. In addition, a feed-forward artificial neural network was employed to estimate moisture content of carrot. Back propagation algorithm, the most common learning method for the feed-forward neural networks, was used in training and testing the network. Comparing the r (correlation coefficient), r2 (coefficient of determination), χ2, and SSR (sum of squares of the difference between the experimental data and fit values) values of the four models, together with the optimized model by using genetic algorithms and the feed-forward neural network based estimator, it was concluded that neural network represented drying characteristics better than the others.