problem with closeloop narx

I am spanish and i can't writte well english but i will try explain my problem. I want to create a RED NARX for predict values of future. I imported two records (inputs) and (targets). I prepared data with command preparets and i create the narxnet with one delay and 16 hiddenlayers. After, i train narxnet and simulated the data i got the outputs. But this outputs are delayed and i want values in future for example one day. I must use closeloop for obtain data in future. My problem is how iterate twice with closeloop because i does not got targets in future. how as would be the code of this part (closeloop) for iterate two times.

Many thanks

[Greg Heath]

I hope you mean one hidden layer with 16 hidden nodes.

Greg

[FRANCISCO]

Yes, my code is: xlsread p xlsread t

p1=p' t1=t'

inputSeries = tonndata(pn,true,false); targetSeries = tonndata(tn,true,false);

% Create a Nonlinear Autoregressive Network with External Input

net = narxnet(inputDelays,feedbackDelays,hiddenLayerSize);

% Prepare the Data for Training and Simulation [inputs,inputStates,layerStates,targets] = preparets(net,inputSeries,{},targetSeries);

% Train the Network [net,tr] = train(net,inputs,targets,inputStates,layerStates);

% Test the Network outputs = net(inputs,inputStates,layerStates); errors = gsubtract(targets,outputs);

okei ,I get up here

% Closed Loop Network netc = closeloop(net); [xc,xic,aic,tc] = preparets(netc,inputSeries,{},targetSeries); yc = netc(xc,xic,aic);

well, now, for re-iterate i must introduce (yc) in function preparets for simulate again close loop?? As would this comand?? [xc,xic,aic,tc] = preparets(netc,yc,{},{});?? because targets now i have not. This is my problem, i do not know the target in (t+1) and ,how can i predict the next value with closeloop?. Many thanks and sorry for my english

Answer 1

[Shashank Prasanna]

This is what an open loop looks like during training:

http://www.mathworks.com/help/nnet/ug/dynamic_narx_open.gif

Since it is NARX you will have one y(t) as the target and x(t) as the eXogenous input.

When you close the loop as follows:

http://www.mathworks.com/help/nnet/ug/dynamic_narx_closed.gif

you will ONLY need to provide the eXogenous inputs, y(t) will be predicted by the output.

Scroll down to see the code of how this is done:

http://www.mathworks.com/help/nnet/gs/time-series-prediction.html

If you don't have an eXogenous input x(t) then you need to be using NAR network.

[FRANCISCO]

Sorry for taking so long is that I lie a little mathworks page. Many thanks for your reply Benji Bowbow. i have one last question. I train the narxnet and simulated the network. After, i close loop and simulate the network again. For iterated the network two times i introduce this previous outputs in close loop and simulate again. I am going to explain with code.

%import data%

xlsread p xlsread t

p1=p' t1=t'

inputSeries = tonndata(pn,true,false); targetSeries = tonndata(tn,true,false);

% Create a Nonlinear Autoregressive Network with External Input

net = narxnet(inputDelays,feedbackDelays,hiddenLayerSize);

% Prepare the Data for Training and Simulation [inputs,inputStates,layerStates,targets] = preparets(net,inputSeries,{},targetSeries);

% Train the Network [net,tr] = train(net,inputs,targets,inputStates,layerStates);

% Test the Network outputs = net(inputs,inputStates,layerStates); errors = gsubtract(targets,outputs);

okei ,I get up here

% Closed Loop Network netc = closeloop(net); [xc,xic,aic,tc] = preparets(netc,inputSeries,{},targetSeries); yc = netc(xc,xic,aic);

well, now, for re-iterate i must introduce (yc) in function preparets for simulate again close loop?? As would this comand?? [xc,xic,aic,tc] = preparets(netc,yc,{},{});?? because targets now i have not

Answer 2

[Greg Heath]

You can iterate as long as you have exogeneous inputs. If your last exogeneous input value is at t = tf, your last predicted output is at t = tf + LDB where

LDB = max([ID,FD])

is the length of your delay buffer.

However, there is a limited amount of good predictions beyond the end of the training series caused by the accumulation of prediction errors. Therefore, it is imperative that you do not overtrain an overfit (Nw > Ntrneq) openloop net.

Hope this helps.

Thank you for formally accepting my answer

Greg

[FRANCISCO]

Good, I could give an example using code to view more clearly, for example get the ten values. taking as input the exogenous variables today (one day)?. I'm a little busy and I do not know how to make the prediction.

Thank you very much for your reply