Estimación de la humedad crítica para el secado de manzanas (Pyrus malus) utilizando redes neuronales artificiales

Abstract

The use of artificial neural networks as a prediction tool in food science is becoming increasingly important. This study seeks to determine the error of the prediction of the Critical Humidity in the drying of apple (Pyrus malus) using neural networks. First, a base matrix was developed consisting of 29 experimental physical parameters and one output variable. These variables were selected through experimental data of drying curves for apple (Pyrus malus) from 30 similar works already developed. Next, the MATLAB® software was used to design the artificial neural network RNA using the four-layer conformed RP back propagation algorithm, where in the first input layer it has 15 static receptor neurons, in the second layer that is hidden it has of 10 hidden neurons, the third layer is sum and has a dynamic neuron, the final layer has a static output neuron. The RNA RP estimated the Critical Humidity obtaining 90 percent of the results with calculation errors less than 1 percent and 10 percent with errors less than 2.38 percent with a network training that ended in 7 stages. Finally, the results showed that the critical humidity was numerically estimated for apple (Pyrus malus) drying through neural models. This type of model of neurons represents. The statistical analysis calculated a reliability of 99.99 percent of the responses obtained from the artificial neural network and the real value.