Runoff Prediction Using Artificial Neural Network …
35
Fig. 6 Feed-forward back propagation network diagram
Fig. 7 Diagram of cascade-forward back propagation network
propagation Lavenberg–Marquardt training algorithm (“trainlm”) was applied for
network learning. In the training phase, simultaneous calculations were carried out
in the forward direction, and then to adjacent layers, error values were propagated.
This network has two hidden layers with log-sigmoid and tan-sigmoid nonlinear
transfer functions, and a purelin linear function in the output layer.
Cascade-Forward Back Propagation (CFBP) Networks
This model, presented in Fig. 7, is almost identical to feed-forward networks. Distribution of weights with each successive layer along with the individual input connections makes a difference from the previous one. A single hidden layer using a logsigmoid transfer function with a combination of an output layer of purelin function
has been used in this study.
35
Fig. 6 Feed-forward back propagation network diagram
Fig. 7 Diagram of cascade-forward back propagation network
propagation Lavenberg–Marquardt training algorithm (“trainlm”) was applied for
network learning. In the training phase, simultaneous calculations were carried out
in the forward direction, and then to adjacent layers, error values were propagated.
This network has two hidden layers with log-sigmoid and tan-sigmoid nonlinear
transfer functions, and a purelin linear function in the output layer.
Cascade-Forward Back Propagation (CFBP) Networks
This model, presented in Fig. 7, is almost identical to feed-forward networks. Distribution of weights with each successive layer along with the individual input connections makes a difference from the previous one. A single hidden layer using a logsigmoid transfer function with a combination of an output layer of purelin function
has been used in this study.