Neural Network Toolbox

Network Initialization

This problem requires that the Elman network detect a single value (the signal), and output a single value (the amplitude), at each time step. Therefore the network must have one input element, and one output neuron.

• R = 1;% 1 input element
  S2 = 1;% 1 layer 2 output neuron
  

The recurrent layer can have any number of neurons. However, as the complexity of the problem grows, more neurons are needed in the recurrent layer for the network to do a good job.

This problem is fairly simple, so only 10 recurrent neurons are used in the first layer.

• S1 = 10;% 10 recurrent neurons in the first layer
  

Now the function newelm can be used to create initial weight matrices and bias vectors for a network with one input that can vary between -2 and +2. We use variable learning rate (traingdx) for this example.

• net = newelm([-2 2],[S1 S2],{'tansig','purelin'},'traingdx');
  

Problem Definition

Network Training

© 1994-2005 The MathWorks, Inc.