Re: Adapting fuzzy sets automatically for weather prediction...?

Nick Outram (nicko@rome.cis.plym.ac.uk)
Sun, 19 Oct 1997 14:55:36 +0200 (MET DST)

Hello

There are many methods for adapting fuzzy sets, or even "evolving" fuzzy
sets/rules. Look in the IEEE Journal of Fuzzy Systems. There are some
good papers on "Rule Extraction" and Optimisation for fuzzy experts
systems. Genetic algorithms and simmulated anealing are popular
optimisation techniques used to optimise the fuzzy set centres, widths
etc... They are slow but good at finding good solutions.
Another altenative is to look at radial basis functions. Upon training a
RBF (you could use MATLAB with the Neural Network Toolbox for example if
you have it) you can map it to a set of fuzzy rules, thereby
"extracting" rules from data.
The whole business or deriving expert system fuzzy / crisp rules from
numerical examples is a very dynamic one and might be an avenue worth
persuing. Drop me a line if you want to discuss it further or ideally
post a reply here.

Regards

Nick Outram

Bjarne Hansen wrote:
>
> How can fuzzy expert systems be made to optimize their fuzzy sets
> automatically? Does someone have a suggestion or know of an instructive
> article?
>
> The reason I ask is that I am developing an expert system to help in
> weather prediction. The system uses fuzzy logic and case-based
> reasoning. I intend to incorporate some form of neural network to
> enable learning.
>
> In developing a prototype system, I designed reasonable fuzzy sets based
> on my knowledge of a particular airport’s climatology. The
> "climatology" consists of a 36-year long hourly record of weather data
> from the airport -- in compressed form, it is about 6 Megabytes of
> data. The system works for airport for which it was designed, but the
> system does not work so well when it is applied to other airports.
> Supposedly, the system would become portable if the fuzzy sets could
> adapt automatically to suit the climatology of other airports.
>
> A description of the prototype system is posted at:
>
> http://borg.cs.dal.ca/~bjarne/p1_5.html
>
> Thank you for your help,
>
> Bjarne Hansen,
> Computer Science Dept.
> Dalhousie University, Nova Scotia
>
> E-mail: bjarne@cs.dal.ca

-- 
Nick Outram
(Lecturer)
SECEE, University of Plymouth, UK

Tel +44 (0)1752 233513 / 232573; Fax +44 (0)1752 232583 mailto:nicko@cis.plym.ac.uk; http://www.cis.plym.ac.uk/cis/nicko/FormalPage/formal.htm