Newsgroups: comp.constraints
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!yeshua.marcam.com!charnel.ecst.csuchico.edu!nic-nac.CSU.net!usc!howland.reston.ans.net!EU.net!julienas!news.fnet.fr!ilog!puget
From: puget@corvisart.ilog.fr (Jean-Francois Puget)
Subject: Re: Optimization in CLP languages
Message-ID: <1994Sep30.074814.26839@ilog.fr>
Sender: puget@corvisart (Jean-Francois Puget)
Nntp-Posting-Host: corvisart
Organization: ILOG S.A., France
References:  <36eo1mINN3hk@diable.upc.es>
Date: Fri, 30 Sep 94 07:48:14 GMT
Lines: 47

In article <36eo1mINN3hk@diable.upc.es>, creemers@ic.upc.es (Tom Creemers) writes:
|> The method of optimization most often used in CLP
|> languages (like CHIP) is Branch-and-Bound.  
|> I'm looking for references to other optimization
|> schemes inside the CLP framework.  Has anyone
|> heard of combinations CLP / Simulated Annealing or
|> CLP / Genetic algorithms?
|> Or more general: operators which transform one
|> solution into a better one according to some
|> optimization criterion, while respecting the
|> existing constraints?
|> 
|> ---
|> Tom Creemers
|> Instituto de Cibernetica (UPC/CSIC)
|> Diagonal 647, 2a planta
|> 08028 Barcelona
|> Tel: +34 3 401.66.59
|> Fax: +34 3 401.66.05
|> 

We have used simmulated annealing together with Ilog Solver to solve
a complex scheduling problem. This problem (the allocation of locomotive
engines for the French railways) can be seen as a contrained TSP problem.

Branch and bound is known to be inefficient for TSP because no good estimate of
the cost of a solution can be computed from a partial solution. On the contrary,
permutation-based search is very efficient, especially if it is controlled by
simulated annealing, or genetic algorithms.

Our application is divided in two parts:

A first solution is found using "classical" CLP techniques, nameley chronological
backtracking plus lookahead.

Then the solution is optimized by trying permutations. These permutations involve
a limited amount of nonmonotonicity that is enabled by the "resetDomain"
primitive of Ilog Solver.

A (not very good) paper describing this application has been presented to the
ASTAIR'92 conference and is available upon request.

-- 
  Jean-Francois Puget		 net : puget@ilog.fr
  ILOG S.A.
  2 Avenue Gallieni - BP 85	 tel : (33 1) 46 63 66 66
  94253 Gentilly Cedex FRANCE	 fax : (33 1) 46 63 15 82
