Neural Network Toolbox

Applin2: Adaptive Prediction

In application script applin2, a linear network is trained incrementally with adapt to predict a time series. Because the linear network is trained incrementally, it can respond to changes in the relationship between past and future values of the signal.

Problem Definition

The signal T to be predicted lasts 6 seconds with a sampling rate of 20 samples per second. However, after 4 seconds the signal's frequency suddenly doubles.

• time1 = 0:0.05:4;
  time2 = 4.05:0.024:6;
  time = [time1 time2];
  T = [sin(time1*4*pi) sin(time2*8*pi)];
  

Since we are training the network incrementally, we change t to a sequence.

• T = con2seq(T);
  

Here is a plot of this signal.

The input to the network is the same signal that makes up the target.

• P = T;
  

Thoughts and Conclusions

Network Initialization

© 1994-2005 The MathWorks, Inc.