Hi,

How do I start with neural modeling of dynamical system? I've tryed some things but nothing works.

thanks,
kgf

Can you supply more details about what you're trying to do?

-Will Dwinnell
http://will.dwinnell.com

Hi,
You could start with collecting data from your system. what kind of network are you planning? you can approximate the system with a linear model.

Hi,

I'm trying to model inverse pendulum. The model can be linearized, but I need to approximate the nonlinear model. Eventually I would like to approximate the real object itself and prove that neural model is better than analitical one.

I experimented with approximation of linear models (2nd, 3rd, 4th order) with good success. When I try ro approximate the nonlinear model of the pendulum I never get anywhere near good results. Even with much bigger NN.

my approach: given any initial state NN predicts new state after 0.05s. The new state is then used as new initial state.

Thanks alot

kgf