MATLAB: Narx delays problem & multistep ahead predictions

delay problemmulti-step predictionnarxnarxnetnncorr

Attached includes my datasets and NARX network architecture. The data is comprised of two datasets (training and test subsets )and I am trying to make multi-step predictions for the external test data subset by using the internal training data subset. As all the data curves have exponential growth,Narx model is confused with this type of growth and gave very inaccurate results. Then, I used the difference equation for exponential growts (a=x(i)/x(i-1)) to transform the data in a form that NARX can make meaningful predictions and it worked well in the model. However, now, the results are not always coherent and can come up with low performance of training. I suppose the problem is about Crross correlation between neural network time series (nncorr). I tried to find a solution from Greg's answers and tutorials and found the following code

X = zscore(cell2mat(x));

T = zscore(cell2mat(t));

[ I N ] = size(X)

[ O N ] = size(T)

crosscorrXT = nncorr(X,T,N-1);

autocorrT = nncorr(T,T,N-1);

crosscorrXT(1:N-1) = []; % Delete negative delays

autocorrT(1:N-1) = [];

sigthresh95 = 0.21 % Significance threshold

sigcrossind = crosscorrXT( crosscorrXT >= sigthresh95 )

sigautoind = autocorrT( autocorrT >= sigthresh95 )

inputdelays = sigcrossind(sigcrossind <= 35)

feedbackdelays = sigautoind(sigautoind <= 35)

feedbackdelays(1)=[] % Delete zero delay

In the original code with the simple data set, feedback delays only results with 1,2 and 3 and entered as 1:3 but in my case the results are much more different than that.

feedbackdelays=[0.227215651811241,0.233970150901862,0.284917894683197,0.264096206558765,0.393205223678322,0.574922519886786,0.294921921143733,0.270700384072885];

What is the point that I am missing? Can anyone tell me what can I do with the code above?

Best Answer

The version of the code you are using is both dated and error prone. Check both the NEWSGROUP and ANSWERS for the latest version.

Also: You are mistaking correlation values for correlation lags.

Hope this helps.

Thank you for formally accepting my answer

Greg

Related Solutions

MATLAB: Auto correlation and cross correlation with NARXnet to find ID and FD

Look for my recent NEWSGROUP posts (2015)

 greg nncorr 2015

Especially

 8 Jul 2015 OPENLOOP NARNET TUTORIAL EXAMPLE
 2 Jun 2015 SIGNIFICANT AUTOCORRELATION DELAYS OF THE SIMPLENAR_DATASET
 13 Apr 2015 NARNET TUTORIAL ON MULTISTEP AHEAD PREDICTIONS: MESSAGES 1, 7, 11

Hope this helps.

Thank you for formally accepting my answer

Greg

MATLAB: NARXnet with Multi input

> My first question is, am I do it right in finding the ID and FD for multi-input?

In general yes. However there are corrections to be made.

>IF I have 10 input then I will have to do the similar things for 10 times?

You don't HAVE to do anything. However, since the MATLAB documentation is useless w.r.t. advice, I've come up with this idea which makes sense to me (and I hope, others).

>> Are there a better way of doing this? Would greatly appreciate any commments!

I have heard of partial correlations where correlations are chosen taking into account the correlations that have been chosen before them.

I have not pursued this approach due to time constraints AND the success I have had with OL solutions. (According to my recent post in the NEWSGROUP, I am still confused and frustrated without a straightforward CL solution).

>> Secondly, there are so many statistically significant for the data, how do I select what to include? There are a few hundreds of them (for e.g X1 and X2 has more than 500 significant)! I don't quite understand how to do, any advice would be much appreciated.

I try to use the smallest subset that will yield a sufficiently small NMSE. So far, it is just a trial and error process.

> After finding the statistically significant, I then select the minimal subset of delay and use it to calculate the size of the hidden nodes. I remember Greg said that if more than 1 input, some equations need to be modified and I am pretty sure this is the reason why my results are poor but I am not sure which equation need to be modified,

The ones that quickly come to mind are VAR and ZSCORE. Maybe NNCORR ...

> even if so, I may not understand why I need to change it since the numbers of data point for both inputs are the same and that the NID is any references for both input, so are NFD, O, I etc.

I agree. But don't forget, this is just something that I came up with because I could not find a relevant reference.

Later I became aware of partial correlations. However, I haven't had the time. If you have nothing to do the next few weeks you might want to pursue that (;>}.

More later,

Greg

Related Question