MPC for neural net predicted indirect control

Can anybody give a hint how to do this,

I have few inputs of time series data. The problem is if i train neural net i can't use target data because of lot of noise and uncertainty. I need to find maybe just few but exact moments when to trigger the control to hit the target for at least interval of few inputs. It's like rare event prediction problem. Is it possible do this using mpc where controller predicts next step and keeps trigger opened. Any reference or hints would be great.

Answer 1

[Greg Heath]

Modify your target data so that it is exactly what you want your output to be.

Hope this helps.

Greg

Answer 2

[Laokine]

I believe you misunderstood the question because of my fault.

So in mpc examples plant control goes for reference signal.

I need the task like in this scheme. http://imgur.com/9xo00 Such kind of NN train while in control test would give the results i cannot achieve, because my same targets can be on opposite conditions and SVM for example even doesn't converge. I need to find just scenario with 70+percent hit rate or etc

Currently i'm achieving good results while training SOM with genetic algos,but just for starting event guesture also it's very slow.