🚀 go-pugleaf

In article <1e9hpgINNc8j@iskut.ucs.ubc.ca>, lug@balsam.ppc.ubc.ca (Guoqiano Lu) writes:
|> 
|> I am using Simulink to simulate continuous-time self-tuning algorithms. Generally it works well except
|> sometimes there are some crazy peaks in the output which is supposed to be smooth. It seems to me
|> that these are due to the numerical intrgration algorithms that SimuLink used. Any one has the same
|> experiences or any suggestions to overcome this problem? 

I had some similar problems when simulating a system which had both
very fast and very slow dynamics, as well as high levels of noise
and disturbance inputs.  I had to `tune' the integration parameters
of max & min integration step size and integration error tolerance.
I did this heuristically, varying them until I `optimized' simulation
speed and `smoothness'.
   I unfortunately know little about this area and would appreciate
any formalism someone could bring to this topic.

  The fact that you are using a self-tuning algorithm has me a little
suspicious.  Are you sure your peaks in the output are not a result
of the self-tuning algorithm `losing track'  ?  
  Do you see similar peaks in the output if you use a non-adaptive
controller ?  Of course you may not, and it may still be the integration
algorithm having trouble handling the time variation of the tuning
algorithm.

--
Mike Stoughton
stote@den.mmc.com