[Help-glpk] solving similar MIP problems

[Erik Rantapaa]

This is more of a general question about solving MIPs rather than about using
GLPK, but I thought I would ask in case anyone had something to say about it.

I have a situation where I am basically solving the same MIP repeatedly.
The problems are not always exactly the same, but they are largely the
same modulo a small number of modified/added constraints or a
modified objective function. In many cases, a solution from one of the problems
might be a solution (although not necessarily optimal) to another problem.

The question I have is whether the work done in solving one of the problems
can be applied to more quickly find a solution to another very similar problem.
Some things I have in mind are:

  * deciding what variable to branch on based on "past performance" in
    previous runs.

  * using a previous solution as a starting point for solving a new problem
    to more quickly find an initial feasible solution.

Is anyone familiar with any techniques or research in this area?

Thanks,

ER



                
__________________________________ 
Do you Yahoo!? 
Yahoo! Mail - now with 250MB free storage. Learn more.
http://info.mail.yahoo.com/mail_250