Is it possible to predict N future values with rbfnn?

I need help with forecasting values using newrb. My input looks like that:

P1=zeros(6,500);
Start=101;
tau=9;
P1(1,:)=P(Start:Start+500-1);
Start=Start+tau;
P1(2,:)=P(Start:Start+500-1);
Start=Start+tau;
P1(3,:)=P(Start:Start+500-1);
Start=Start+tau;
P1(4,:)=P(Start:Start+500-1);
Start=Start+tau;
P1(5,:)=P(Start:Start+500-1);
Start=Start+tau;
P1(6,:)=P(Start:Start+500-1);

so

size(P) = [6 500]

and target

size(T) = [1 500].

Then I choose sum-squared error goal, spread constant and use newrb

net = newrb(P1,T,eg,sc,mn,df);

and

ytrain = net(P1);).

So my question is how to write the script that predicts the N values from my Data ??

Answer 1

[Greg Heath]

1. Use the autocorrelation function to determine the significant nonzero delays of P; Use as many as needed.

2. For the purpose of an example, assume 1,2,and 4 are enough significant delays.

3. Create the 3-D input and 1-D output

   input = [ P( 1: 500-4) ; P(3: 500-2) ; P(4: 500- 1)];

   output = P(5: 500)

Hope this helps.

Greg

[Greg Heath]

This is easily extended to a multidimensional output.

However, why not just use NARNET?

[anurag kulshrestha]

Can you please elaborte how?