Thread Subject: Mapminmax in SIMULINK

On Jun 26, 7:27 am, dedy <masded...@gmail.com> wrote:
> I have a system with PI controller. I want to replace this controller
> with neural network (NN). So, the input of PI controller i use as
> input of NN. The output of PI controller i use as target of NN.
>
> I am using the premnmx function to preprocess the training data before
> the neural network can be trained:
>
> [pn,minp,maxp,tn,mint,maxt]=premnmx(P,T)
> net=train (net,pn,tn);
>
> then to transform the neural network, i use
> gensim (net,Ts)
>
> I use the result of gensim to replace the old controller in simulink.
> But why the output of NN controller doesn't sama as target (when it
> trained). The result of NN controller worse than PI controller..
>
> is it need processing block (like mapminmax) before processed by NN
> controller in simulink ?
> if its need block, how to use it with NN in simulink ?
> please help me.

How did you train, validate and test the net before
incorporating into simulink?

Size of training, validation and test sets?
Number of hidden nodes?

Size of training, validation and test set errors?

Hope this helps.

Greg