Description
Elman networks and time series modelling
In Matlab an Elman network can be created using function newelm as follows (type “help newelm” in command window for help in detail):
newelm(input,target,[N1,...,N(K-1)])
where input and output represent the input and target of training set, [N1, …, N(K-1)] denote the number of units of each hidden layer (as the number of units of output layer is determined by target).
The demo code models the seriesx 0.75×2 0.2x (x x 1) using Elman n n1 n2 1 2
network, with one hidden layer of 15 units (K = 15) and 300 epochs in training. Understand the demo code and try different number of training/testing data size, different number of hidden layers and units of each layer, and different number of epochs.