Is it possible to solve a multi-input-multi-output control problem by using Matlab neural network toolbox

Hello friends,

My problem is about a Multi-input Multi-output control system. There are 14 inputs and 12 outputs and different inputs lead to different outputs. So it seems that I have many input-output pairs but I don't know the relationship between them, just like I don't know what's inside the black box. So I want to use MATLAB neural network toolbox to help me find their relationship. My concern is whether it is feasible to solve this problem with so many inputs and outputs.

Thank you.

[Himanshu Rathore]

I m using this "model refrence controller" block. As for now it demanding single i/p single o/p but I want 3-3 of them. how to make it working with multiple i/p & o/p, please help.

thank you

Answer 1

[Greg Heath]

Should be straightforward if you have enough data so that the number of training equations, Neq, is significantly larger than the number of unknown weights Nw.

[ I N ] = size(p) % I = 14

[ O N ] = size(t) % O = 12

Neq = N*O

For an I-H-O node topology,

Nw = (I+1)*H+(H+1)*O

and if training to convergence, require Neq >= Nw but desire Neq >> Nw. Or equivalently,

H <= Hub = (Neq-O)/(I+O+1) is required but

H << Hub is desired.

Otherwise, validation set early stopping or using a regularized objective function like MSEREG should be used.

Hope this helps.

Greg