Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!fas-news.harvard.edu!newspump.wustl.edu!news.ecn.bgu.edu!vixen.cso.uiuc.edu!howland.reston.ans.net!news-e1a.megaweb.com!newstf01.news.aol.com!uunet!in1.uu.net!ftpbox!mothost!schbbs!news
From: huang_r@sat.mot.com (RP Huang)
Subject: Re: Ga to optimize fuzzy controller
Reply-To: p26814@email.mot.com
Organization: Motorola SATCOM
Date: Thu, 3 Aug 1995 23:37:59 GMT
Message-ID: <1995Aug3.233759.7135@schbbs.mot.com>
References: <286002A5_000DF5E0.009943B2FC091E33$8_1@UK.AC.LUTON.VAX2>
Sender: news@schbbs.mot.com (SCHBBS News Account)
Nntp-Posting-Host: 170.1.246.108
Lines: 56

In article 1@UK.AC.LUTON.VAX2, JQIU@VAX2.LUTON.AC.UK writes:
> Hi, friends. Does anyone have experience in using GA to optimize fuzzy logic
> controllers for mobile robot, especially in composing multiple fuzzy
> behaviours? As it is a dynamic problem, I want to ask how to use GA to optimize
> it.
> Any help welcome.
> 
> Cheers
> 
> Jancheng Qui
> 
> 

This is my $0.02:

There are two aspects on adapting fuzzy rule based systems:
- adapt the fuzzy membership sets and fuzzy hedges
- adapt the fuzzy rules

Adapting membership set or hedges should affect every rule that
uses the set or hedge. Typically membership set and hedge can be
adapted with Genetic Algorithm. However, depends on the representation
of the data (real value or integer), you may want to use special
mutation operator. If the length of the membership set can be
varied, then borrow ideas from EP (Evolutionary Programming) and ES
(Evolutionary Strategies) that deal with dynamic length and free
representation.

If the fuzzy rules are represented in tree structure, it is very
easy to use Genetic Programming to adapt it. For example, a fuzzy
rule:    IF A IS Very Hot AND B Is quite Cold
         THEN C is High
can be represented by following tree:
              Rule
             -------------------
             IF             Then
             /                 \
            AND                Is
           /   \              /  \
        very   quite         C    High
       /   \    /   \
      A   Hot   B    Cold

Where very, quite, and Is are fuzzy hedge operators.
My preference is to add mutation to the Genetic Programming, so
it is possible to mutate "very" to "somewhat" without affecting
any other rules.

I wrote an Adaptive Fuzzy Control Simulation program based on the
above ideas, except that the fuzzy hedges are not included. You can
find the sharware in simtel - oak.oakland.edu:/msdos/ai/gaf200b.zip.

RP Huang
p26814@email.mot.com


