home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!pipex!warwick!kinguni2!ceres!pd_s001
- From: pd_s001@ceres (J Karwatzki)
- Newsgroups: comp.ai.neural-nets
- Subject: Re: help
- Date: 20 Nov 1992 09:42:56 GMT
- Organization: Kingston University
- Lines: 30
- Message-ID: <1eibv0INN7ds@mercury.kingston.ac.uk>
- References: <sc.8.721955240@basil.eng.monash.edu.au>
- NNTP-Posting-Host: ceres.kingston.ac.uk
- X-Newsreader: Tin 1.1 PL4
-
- sc@basil.eng.monash.edu.au (Cong Shiguo) writes:
- : Hello,
- :
- : I am tring to train a neural network controller. I met a problem
- : which I don't know how to solve it. In controller cases, we must train the
- : neural nets to follow the desired outputof the plant to controlled. The
- : reference input to the system can be any time-varying signals, such as step
- : function or sinusoidal waveform with different frequencies. My question is
- : What kind of signal can be used as training reference? I tried to use step
- : input and white noise. The nn controller works well to its training reference
- : but doesn't converge to those it didn't learn.
- :
- : Could anybody give me some help please? Many thanks.
- :
- :
- :
- --
- Such an approach will only be useful if all inputs to the system can be
- categorised and, hence, taught to a neural network. Since this is
- extremely unlikely in most cases the nn approach is probably not suited
- for such an application. What may be much more useful would be the ability
- to categorise disturbances to the plant. A nn could be taught to
- recognise typical disturbances (ignoring others) and take appropriate
- corrective action as they occur.
- John Karwatzki
- Kingston University
- Kingston
- Surrey
- UK
- karwatzki@kingston.ac.uk
-