aDaimlerChrysler, Research and Technology, Alt-Moabit 96a, HPC U118, 10559 Berlin, Germany
bSAP Deutschland AG & Co. KG, Neurottstrasse 15, 69190 Walldorf, Germany cDaimlerChrysler, 800 Chrysler Drive, Auburn Hills, MI 48326, USA
ABSTRACT
The optimization literature has traditionally concentrated more on the power of optimization algorithms and not so much on their accessibility. In this paper we develop a methodology to automate optimization and thus to ease access to the multitude of optimization methods and their specific parameter settings. This will help engineers to solve more of their non-trivial optimization problems without (expensive) mathematically skilled help.
We show a prototypical system with this 'scheduling' functionality: it analyses a user-specified optimization problem, finds its optimizer-relevant properties, selects the 'best' of the available methods, sends the problem to the selected optimizer, and starts it with the appropriate parameters.
Preliminary, yet promising, results show that the basic idea works in principle and that future research in this area has high potential.
Keywords:
Automated optimization; Optimization problem classification; Optimization scheduling