TDNN multistep prediction with unknown future data for the target

Hi, everyone!

I am a little bit confused how to use a time delay neural network for multistep ahead prediction.

By my system characteristics, I must use a time delay neural network and not others. So, from previous measurement, I know the input and target time series, where:

Xdata(1:500) (input)
Tdata(1:500) (target)

Let's create the neural network:

net = timedelaynet(1:10,10);
[Xs,Xi,Ai,Ts] = preparets(net,Xdata,Tdata);
net = train(net,Xs,Ts,Xi,Ai);

I can clearly understand all the procedure until this point. However, I do not know how to prepare the new input data to be predicted and to use the net. I mean, in the future, I will just know the input data. So, how I could predict it? For example:

Xnew1(1:50);
Y1 = net(Xnew1,Xi,Ai);

Ans further, for a new data:

    Xnew2(51:100);
    Y2 = net(Xnew2,Xi,Ai);

Would it be the same? With the same Xi and Ai?

Thanks for helping!

Answer 1

[Greg Heath]

In general, PREPARETS will yield the correct inputs. However, for TIMEDELAYNET, just use common sense:

 Ai = {}    % There is no feedback

 Xi = Xnew2(:,1:10);

 Xnew2s = Xnew2(:, 10:end);

Hope this helps.

Thank you for formally accepting my answer

Greg

PS: See my tutorials

[Hugo Mendonça]

Thank you, Greg!

As I have seen, you are always helpful!

BTW, very god tutorial about NARNET. Maybe, in a very close future, I might use NARNET.

Thank you again!