How to use a trained narx network with only input variable ?

%?How to use a trained narx network with only input variable

% Asked by Leon on 17 Aug 2012 at 7:14

> I am using the NARX function of Neural Network Toolbox. In the User's

> Guide, Neural Network Toolbox, R2012a. Page 3-23, there is an example as

> below. (u is the input and y is the output).

> load magdata

> [u,us] = mapminmax(u);

> [y,ys] = mapminmax(y);

> y = con2seq(y);

> u = con2seq(u);

> d1 = [1:2];

> d2 = [1:2];

> narx_net = narxnet(d1,d2,10);

> narx_net.divideFcn = '';

> narx_net.trainParam.min_grad = 1e-10;

> [p,Pi,Ai,t] = preparets(narx_net,u,{},y);

Write the 4 explicit algebraic equations that yield the 4 outputs [p,Pi,Ai,t] in terms of the 2 open loop inputs u and y

> narx_net = train(narx_net,p,t,Pi); % Ai is empty

Note that open loop narx_net is explicitly defined by t

> yp = sim(narx_net,p,Pi);

> e = cell2mat(yp)-cell2mat(t);

> plot(e);

> narx_net_closed = closeloop(narx_net);

Note that closed loop narx_net is NOT explicitly defined by t

> y1=y(1700:2600);

> u1=u(1700:2600);

> [p1,Pi1,Ai1,t1] = preparets(narx_net_closed,u1,{},y1);

Write the 4 explicit algebraic equations that yield the 4 outputs [p1,Pi1,Ai1, t1] in terms of the 2 open loop inputs u1`and y1, In particular, concentrate on Ai1 and t1.

> yp1 = narx_net_closed(p1,Pi1,Ai1);

Note that yp1 is explicitly defined by t1 ONLY through the 2 initial feedback delay conditions stored in Ai1.

> plot([cell2mat(yp1)' cell2mat(t1)']);

> Now I would like to use this function to first train the network (so the > first part of the example) and then to use the trained (closed) network > to calculate y with only u as an input variable. What is seen in the second > part of the example that they still use part of the output (y1) as input > (beside the normal input: u1).

> How can I do use the trained (closed) narx_network to calculate y1 with > only u1 as input?

It has already been done above. Although y1 is an input to preparets, only the 2 delay initial conditions consistent with d2 = [1 2 ] are used.

1. Type the command "narx_net" and store the listing in a file.

2. Type the command "narx_net_closed" and store the listing in another file.

3. Compare the differences in the two listings and you should get a better feel for what is accomplished by the closeloop command.

Hope this helps.

Greg