Time series prediction example using NARX network. a synthetic data is used to train the NARX net.
In this example, open-loop architecture is used during the training step because of two advantages. First, the resulting network is purely feedforward, and static backpropagation is used for training. Secondly, open-loop uses the actual values as input of the network and the result is more precise . open-loop can only be able to predict one step ahead and to overcome this limitation, after the training of the NARX network with open-loop architecture the network will be converted to the closed loop architecture to perform multi-step-ahead prediction ahead.