Forecast with GRNN

Hello,

I would like to make a multi step ahead prediction with GRNN network. My idea is: I make a one step ahead prediction and with a for loop I train the network again with the earlier predicted data. So I get a multi step forecast. I made the code, but the prediction is always a horizontal line! I do not know what is the problem!

Here is my code:

%--------------Paraméterek------------------
adatSzam = 500;
N = 20;

%--------------Adatok betöltése, előkészítése--------------
load('C:\Neuroforex\Adatok\HMA35nnagy.mat');
imputSeries1 = HMA35nnagy(1:adatSzam)';

for i=1:N

      X = [1:length(imputSeries1)]
      T = imputSeries1;

      %--------------GRNN előállítása------------------
      spread = 0.1;
      net = newgrnn(X,T,spread);
      view(net);
      trainOutput = net(X);

      %-------------Becslés beállítása----------------
      predictSeries = [length(imputSeries1)+1];
      yPred = net(predictSeries);

      %-----------------Plot 1 lefutásra-------------------------
      hold on;
      %Bemeneti adatok plottolása
      plot(X,T,'k','markersize',2,'color','black')
      outputline = plot(X,trainOutput,'o','markersize',2,'color','blue');
      %Becslés plottolása
      %plot(predictSeries,[yPred(1:end-1),nan(1,1)],'linewidth',1,'color','red')
      predict = plot(predictSeries,[nan(1,length(predictSeries-1)),yPred],'o','markersize',2,'color','green');
      %plot([nan(1,length(imputSeries1)),yPred(end:end)],'linewidth',1,'color','green');
      title('Becslés eredménye')
      xlabel('Gyertyaszám')
      ylabel('Keresztárfolyam')

      %-----------Jelenlegi becslés hozzáadása a bemeneti adatokhoz--------
      imputSeries1 = [imputSeries1, yPred(end:end)];

end

Picture of the predict:

https://picasaweb.google.com/lh/view?q=matlab+prediction&uname=113002498701823817763&psc=G&filter=1#5680080904755224882

Thanks Adam

Answer 1

[Greg Heath]

Newgrnn is not appropriate for timeseries prediction.

Use newfftd.

help newfftd

doc newfftd

Search the archives

newfftd

heath newfftd

Hope this helps.

Greg

[Adam Farkas]

Hi Greg,

I have already made a NARX network, but my forecasted results weren't good enough. Thats's why I started to search for other solutions. I found an article with GRNN prediction:

http://www.neural-forecasting-competition.com/downloads/NN3/methods/44-NN3_WeizhongYan.pdf

I already tried to define delays in grnn.m file. So my question is: Is there any way to predict with GRNN?

Thanks Adam