How to set up weights/bias of the NARX neural network

I want to set up the range of weights of NARX network in [-0.8,0.8], how can i set up it? The number of the output neurons is 2.

    my code is that 
       ID=1:2; FD=1:2; H=[7,15];
        net=narxnet(ID,FD,H);
        [inputs,inputstates,layerstates,targets]=preparets(net,T,{},P);
        net.inputs{1}.processFcns = {'mapstd','removeconstantrows','mapminmax'};
        net.inputs{2}.processFcns = {'mapstd','removeconstantrows','mapminmax'};
        net.initFcn='initlay';
        net.layers{2}.initFcn='initwb';
        net.layers{3}.initFcn='initwb';
        net.layerWeights{2,1}.initFcn='randsmall';
        net.layerWeights{3,2}.initFcn='randsmall';
This method can initialize weights to be [-0.1,0.1].And it's right.
But when i want to use the following method to get the initial range [-0.8,0.8],it's wrong. The code is shown in below. 

        net=narxnet(ID,FD,H);
        [inputs,inputstates,layerstates,targets]=preparets(net,T,{},P);
        net.inputs{1}.processFcns = {'mapstd','removeconstantrows','mapminmax'};
        net.inputs{2}.processFcns = {'mapstd','removeconstantrows','mapminmax'};
        net.initFcn='initlay';
        net.layers{2}.initFcn='initwb';
        net.layers{3}.initFcn='initwb';
        net.layerWeights{2,1}.initFcn=tonndata(0.8*rands(25,7));
        net.layerWeights{3,2}.initFcn=tonndata(0.8*rands(2,25));
*How can i do it?*

Answer 1

[Greg Heath]

You are concentrating on the wrong stuff. The most important things are (assuming a SINGLE hidden layer)

 1. Obtain a good set of crosscorrelation delays for ID (e.g., using NNCORR)
 2. Obtain a good set of autocorrelation delays for FD (e.g., using NNCORR)
 3. Given ID and FD, find the smallest successful value for H considering 
    ~ 10 different random weight initialization trials for each H candidate.

MATLAB defaults will take care of the rest. Search for my openloop examples

 greg nncorr narxnet
 greg narxnet

Then consider closing the feedback loop

 greg closeloop

Hope this helps.

Thank you for formally accepting my answer

Greg