Multi Step Ahead Prediction using TIMEDELAYNET

Hi ! I'm new in this field and i need help to perform a Multi Step Ahead Prediction using the Focused Time-Delay Neural Network (FTDNN) as seen here , http://www.mathworks.es/matlabcentral/answers/14970-neural-network-multi-step-ahead-prediction , but with a Timedelaynet .

Is it possible to have some script to understand and solve this problem ?

Thanks a lot !

Regards , Salvatore

Answer 1

[kira]

Hello,

I don't think is possible to do multiple steps ahead prediction using timedelaynets, recall that a timedelaynet it used for a time series like: y(t+1)=f(x(t),x(t-1),...,x(t-d)), therefore, to do predictions you will need the next values of x, that we usually don't have.

You can do multiple steps ahead predictions with narnets, since they model: y(t+1)=f(y(t),....,y(t-d)), therefore you use the same output of the net to do the predictions.

You can do multiple steps ahead predictions with narxnet, since they model: y(t+1)=f(x(t),....,x(t-dx),y(t),....y(t-dy)), as long as you have the next values of x.

However, recall that the time delay has the form: d1:d2, where d2 is the time delay, but d1 is usually 1, so you are doing 1 step ahead prediction. So it you use d1>1, you'll doing multistep ahead prediction, but nothing ensures it will work ok.