future value prediction using narx

I have a trained series-parallel configuration NARX neural network with %inputs one output value. I have data of 4months 'Aug-nov'.

Now i want to predict the future values of the next week i.e Dec first week.

How to do that???

will using sim help??? How??? Any reference???

[SK]

i have similar problem. did you get it resolved? if yes could you share the code please?

Answer 1

[Greg Heath]

Scanty info.

 [ Xc Xci Aci Tc ] = preparets(netc,X,{},T);

 [ Yc Xcf Acf ] = net(Xc, Xci, Aci);

where (Xci,Aci) and (Xcf,Acf) are initial and final input and layer states, respectively.

With NARX, you cannot get a valid future output unless you have a valid future input.

If you do not have a valid future input you can try to predict one using NARNET.

Also you could use NARNET to predict the future output!

Obviously, if you do both you can compare and even consider combining.

Hope this helps,

Greg

[Jeong_evolution]

Hi, Greg.

I have same problem.

In your answer, Where can I insert "future input"(=future value) ?

In code, must insert target? But, future value does not have target.

Please answer, Greg.

[Greg Heath]

See my revised ANSWER.

[NICKTHEGREEK]

Is it possible to create our future inputs with a Narnet and then through a narxnet find our future prediction?

[Greg Heath]

Yes, PROVIDED the AUTOcorrelation of the input is sufficient.

In serious work I look at the AUTOcorrelation functions of BOTH input and target in addition to their crosscorrelation.

It tends to give a better feel for the data.

Hope this helps.

Greg

[Fatma HM]

@Greg Heath hi, I want to ask you about NARX did you have any idea about it ? I’m using NARX to forecasting my datas and calculate the RMSE but the results of RMSE still bad. Plz if you have any idea how to get RMSE lower tell me I really tried everything but I did not find any good results

[Seda]

Where can I insert "future input"(=future value) ?