A question about Neural Plant Model in Neural Predictive Controller demo.

Hello.

There is one thing about the matlab neural predictive controller example that I am not sure of. So basically the very first thing to do would be 'building' a neural plant model, right? Let's say my plant has one input and one output, I have already gathered a training data set. But why is it that the inputs on the neural network consist of delayed inputs and outputs only and there is no normal input value?

I mean, shouldn't I feed my network with the very first input value from my training data set and compare the network output with my training data set output value? Please explain this matter to me if you will.

Yours truly, Mat

Answer 1

[Greg Heath]

Good point. That is one of my gripes.

 In particular, for TIMEDELAYNET and NARXNET, the default input delays should include 0:

 net1 = timedelaynet( 0:2 );

 net2 = narxnet( 0:2, 1:2 );

 net3 = narnet( 1:2 );

Obviously, you cannot have a 0 feedback delay. If you try it, it will work in the open-loop configuration but will BOMB when you close the loop.

Hope this helps.

Thank you for formally accepting my answer

Greg