Neural Model of Osmotic Dehydration Kinetics of Fruits Cubes

The paper proposed to use a Recurrent Neural Network model (RNN) for process prediction of the osmotic dehydration kinetics of nature product cubes (apple, sweet potatoes and potatoes) at different operational conditions of temperature and concentration of the osmotic solution. The proposed RNN model has five inputs, three outputs and eight neurons in the hidden layer, with global and local feedbacks. The learning algorithm is a modified version of the dynamic backpropagation one. The learning and generalization mean squared errors are below 2%. The learning was performed in 701 epochs, 40 iterations each one. The statistical analysis confirms the good quality of the proposed RNN model.