The modeling method for ANN was based on the principles
of the back propagation algorithm. In order to find out
an optimal configuration of neural network model, it is
necessary to combine many different ANN prototypes.
Consequently, six different training algorithms including
gradient descent with momentum, scaled conjugated gradient,
Levenberg–Marquardt, quasi-Newton, conjugated
gradient and variable learning rate were used for network
training. The number of hidden layer(s), the neurons in the
hidden layer(s) and the value of the training parameters for
every training algorithm were obtained by trial and error
method with considering of ANN performance. Also, three
types of transfer functions of tangent hyperbolic conversion,
sigmoid (tan sig) and linear motion (pure line) function
among layers were used for ANN models. For
selecting the number of hidden layers along with the right
number of neurons in the middle layers, comparison of
networks which had different number of neurons and also
different number of hidden layers were carried out