how to approximate two subnetworks at the same time, f and g functions from NARMA L2 Controller

Hello, I'm trying to use NARMA L2 Controller to solve my Problem. I have two Inputs signals, one is control Signal, the other one can't be controlled but can be predicted. My question is how to approximate these two functions (subnetworks) at the same time with the same data? f = F(y(k),y(k-1),u(k)=0,u(k-1)), I understand this one, but g is the partial derivative F/u and u(k)= 0. My solution is to simulate the System with the same signals twice, first u(k) = 0 and get y(k+1) second time with u(k) = u get y_new(k+1), and use (y_new(k+1) - y(k+1))/u as the Training object of function g. I get a good result, but this should not be the idea, and this can't be realized in real experiments. So can anybody explain how to approximate These two functions at the same time with the same data sets

Thanks Hao

[Hao Wu]

or simply say, how to Train two correlated functions at the same time using matlab, I use "feedforwardnet" to set the Input and Output, but how to Combine These two Networks without knowing the ideal Outputs for each but only the final Output.

Answer 1

[Hao Wu]

Hello, can anyone tell me, how to train this Network with two subnetworks?

Thanks