Designing and Optimizing a BP Neural Network to Model a Thin-Layer Drying Process

In the present study, the application of a BP neural network for prediction of the moisture content of barberry fruit (berberis vulgaris) during drying was investigated. The important parameters namely, pretreatment (heat shocking, olive oil + K2CO3 and NAT), air drying temperature (60, 70 and 80oC), air drying velocity (0.3, 0.5 and 1 m/s) and time (s) were considered as the input parameters and moisture content as the output of the ANN. Experimental data obtained from thin-layer drying process, were used for training and testing the network. Several criteria such as training algorithm, learning rate, momentum coefficient, number of hidden layer, number of neurons in each hidden layer and activation function were given to improve the performance of the ANN. The best training algorithm was LM with the least MSE value. Optimum values of learning rate and momentum for the ANN with GDM training algorithm were set at 0.5 and 0.7 respectively. The optimal topologies were 4-20-1 and 4-25-5-1 with MSE values of 0.00318 and 0.001 with logsig activation function. Also with tansig activation function, the optimal topologies were 4-20-1 and 4-15-15-1 with the MSE values of 0.00293 and 0.00130. There was no significant difference between two activation function in optimal topologies. There was good correlation between the predicted and experimental values in optimal models.