How can i use sim function in trained neural network

i have input 7*2601 and target 1*2601 data. I trained Narx network and got the some plots. I want to do forecasting with new data so i created new matrix with 7*1 matrix and put some data. But when i call "pre = net(testdata')';" its always gives the same errors which are

"Error using network/sim (line 271) Number of inputs does not match net.numInputs.

Error in network/subsref (line 16) otherwise, v = sim(vin,subs{:});"

I read all of Narx , multistep predictions, open-close loop predictions but cant solve it. Any ideas ? Thank you.

code:

>>

Data = xlsread('data.xlsx'); Target = xlsread('Target.xlsx'); testdata= xlsread('testdata.xlsx'); X = tonndata(Data,false,false); T = tonndata(Target,false,false); trainFcn = 'trainlm';

inputDelays = 1:2; feedbackDelays = 1:2; hiddenLayerSize = 100; net = narxnet(inputDelays,feedbackDelays,hiddenLayerSize,'open',trainFcn);

%net.inputs{1}.processFcns = {'removeconstantrows','mapminmax'}; %net.inputs{2}.processFcns = {'removeconstantrows','mapminmax'};

[x,xi,ai,t] = preparets(net,X,{},T);

net.divideFcn = 'dividerand'; % Divide data randomly net.divideMode = 'time'; % Divide up every sample net.divideParam.trainRatio = 70/100; net.divideParam.valRatio = 15/100; net.divideParam.testRatio = 15/100;

net.performFcn = 'mse';

[net,tr] = train(net,x,t,xi,ai);

y = net(x,xi,ai); e = gsubtract(t,y); performance = perform(net,t,y);

trainTargets = gmultiply(t,tr.trainMask); valTargets = gmultiply(t,tr.valMask); testTargets = gmultiply(t,tr.testMask); trainPerformance = perform(net,trainTargets,y); valPerformance = perform(net,valTargets,y); testPerformance = perform(net,testTargets,y);

view(net)

pre = net(testdata')';

[Greg Heath]

Run your code on one of the MATLAB Time-Series examples and we can compare results.

 help nndatasets
 doc nndatasets

 load('simpleseries_dataset')%Simple time-series prediction dataset.
   load('simplenarx_dataset')%Simple time-series prediction dataset.
   load('exchanger_dataset')%Heat exchanger dataset.
   load('maglev_dataset')%Magnetic levitation dataset.
   load('ph_dataset')%Solution PH dataset.
   load('pollution_dataset')%Pollution mortality dataset.
   load('refmodel_dataset')%Reference model dataset
   load('robotarm_dataset')%Robot arm dataset
   load('valve_dataset')%Valve fluid flow dataset.

whos

So, which one do you want to pick?

Greg

Answer 1

[Greg Heath]

 Data     = xlsread('data.xlsx'); 
 Target   = xlsread('Target.xlsx'); 
 testdata = xlsread('testdata.xlsx');
 X        = tonndata(Data,false,false); 
 T        = tonndata(Target,false,false); 

 whos  % 1. Check class and size

 % 2. How many training equations?
 % What is the average target variance to be modeled?

 trainFcn       = 'trainlm';
 inputDelays    = 1:2; 
 feedbackDelays = 1:2;

%3. What makes you think these delay values are any good? What values correspond to significant auto and cross correlations?

 hiddenLayerSize = 100;

%4. 100 hidden units probably generates too many unknown weights. How many equations vs unknowns?

 net = narxnet(inputDelays,feedbackDelays,hiddenLayerSize,'open',trainFcn);

 %net.inputs{1}.processFcns = {'removeconstantrows','mapminmax'};
 %net.inputs{2}.processFcns = {'removeconstantrows','mapminmax'};

 [x,xi,ai,t] = preparets(net,X,{},T);

% 5. Add additional subscript 'o' to denote 'openloop'. % 6. Preferable to use uppercase for cells, lower case for doubles

 net.divideFcn = 'dividerand'; % Divide data randomly

% 7. UH...OH! If you do this the spacing within the trn/val/tst subsets will be nonuniform.

 net.divideMode             ='time'; % Divide up every sample 
 net.divideParam.trainRatio = 70/100;
 net.divideParam.valRatio   = 15/100; 
 net.divideParam.testRatio  = 15/100;
 net.performFcn = 'mse';

%8. Why waste space explicitly specifying default values???

 [net,tr] = train(net,x,t,xi,ai);

% 9. Forgot the final states on the LHS

 y           = net(x,xi,ai);
 e           = gsubtract(t,y); 
 performance = perform(net,t,y);

%. 10. Performance is not normalized w.r.t. average target variance. So how can you tell whethter it is good or not?

 trainTargets     = gmultiply(t,tr.trainMask); 
 valTargets       = gmultiply(t,tr.valMask); 
 testTargets      = gmultiply(t,tr.testMask);
 trainPerformance = perform(net,trainTargets,y); 
 valPerformance   = perform(net,valTargets,y);
 testPerformance  = perform(net,testTargets,y);

% 11. Same comment as 10

 view(net)

%12. view right after train

 pre = net(testdata')';

% 13. No input states obtained from output states in train output

% 14. Why haven't you reviewed my posts in the NEWSGROUP and ANSWERS? All of the above points are covered.

Hope this helps.

Thank you for formally accepting my answers

Greg