Problem with network NarX need help for final step
2 views (last 30 days)
Show older comments
Good, I write again to see if anyone can help me as I left the last steps to fully utilize the network NarX. I expose the example I put in previous post: I've trained a network with exogenous variables NarX, I checked the accuracy, and I've done compared closeloop and now I want accuracy in predicting real-time, ie if I introduce today inputs (14/03/2012) and I want to get predictions in the next 10 days as you would? Should use the predicted outputs to make that prediction but how?. If there is some code to understand me would be helpful, or if there is any book, web .... where I could learn to perform this step. The accuracy, code, etc ... are published in previous post, so I'd like to know this step.
Thank you very much
3 Comments
Accepted Answer
Greg Heath
on 18 Mar 2013
Edited: Greg Heath
on 18 Mar 2013
If you look at the significant lags of the output autocorrelation function and the output/input crosscorrelation function, you can deduce how far ahead you might be able to predict AND the number of delays that you might have to use to accomplish that goal. In general,
y( : , t : t + d ) = f( x( : , t - id : t) , y( : , t - ld : t - 1))
For a special type of NARXNET scalar output prediction
y( t + d ) = f( x( : , t - id : t) , y( t - ld : t - 1))
for d = 10, id >= 0, ld >= 1
Whereas for a special type of TIMEDELAYNET prediction
y( t + d ) = f( x( : , t - id : t) )
PS: timedelaynet(ID) may be internally different than narxnet(ID,[])!
Hope this helps.
Thank you for formally accepting my answer
Greg
More Answers (1)
Greg Heath
on 16 Mar 2013
If you want to predict exactly 10 days ahead, if possible, design a net that will do that.
Iterating a net that will predict one or two days ahead doesn't seem to be the way to go (if you can avoid it).
Better yet, design a net with a 10-dimensional output that predicts ahead for a 10-day window.
Hope this helps.
Thank you for formally accepting my answer
Greg
See Also
Categories
Find more on Sequence and Numeric Feature Data Workflows in Help Center and File Exchange
Products
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!