Message-ID: <327837DC.59C1@livjm.ac.uk>
Date: Thu, 31 Oct 1996 05:23:40 +0000
From: Constantinos Vlachos <eeecvlac@livjm.ac.uk>
Organization: Liverpool John Moores University
X-Mailer: Mozilla 3.0Gold (Win95; I)
MIME-Version: 1.0
Newsgroups: comp.ai.genetic
Subject: Re: PID Tuning using GA
References: <54itj8$2ei@hermes.is.co.za>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
NNTP-Posting-Host: dialin-port6.livjm.ac.uk
Lines: 25
Path: cantaloupe.srv.cs.cmu.edu!rochester!cornellcs!newsstand.cit.cornell.edu!portc01.blue.aol.com!news-peer.gsl.net!news.gsl.net!news.sprintlink.net!news-peer.sprintlink.net!nntp-hub2.barrnet.net!news.Stanford.EDU!agate!sunsite.doc.ic.ac.uk!pfs1.livjm.ac.uk!

Rudiger von Varendorff wrote:
> 
> But then, if the process is modelled, why not use normal control
> theory / methods to design the PID parameters optimally, or to meet
> the required performance specifications. Using a GA may only produce a
> suboptimal solution.

When you say that the process 'is modelled', what kind of model do
you assume to have? If the process is linear (very rare) or you
linearize it (valid for only a small operating region) then I agree
with you, all the necessary analytic tools for the PID designs are
available. However, using GAs (or any other suitable optimization
method) enables the use of realistic, possibly highly non-linear
models which may include multiple noise sources, constraints such
as saturation / rate limit non-linearities, and many other properties
which prevent an analytic solution. If such models are used, then how
will you make the designs? Unfortunately, the Laplace and z-Transforms
are _linear_ transformations! Also, the optimisation approach enables
the use of complex objective functions that better capture the
performance objectives of the control problem (unlike the conventional
IAE, ITAE, ISE, ITSE, etc. which are often misleading indicators). The
main disadvantage of the optimisation approach as you pointed out is
that a process model IS required. I don't like 'approximate' methods but
very often this is the only design tool available.

Regards,

Constantinos Vlachos
