A Template for Designing Single-Solution Hybrid Metaheuristics
Résumé
Single-solution metaheuristics are among the earliest and
most successful metaheuristics, with many variants appearing
in the literature. Even among the most popular variants,
there is a large degree of overlap in terms of actual behavior.
Moreover, in the case of hybrids of different metaheuristics,
traditional names do not actually re
ect how the hybrids are
composed. In this paper, we discuss a template for singlesolution
hybrid metaheuristics. Our template builds upon
the Paradiseo-MO framework, but restricts itself to a predefined structure based on iterated local search (ILS). The
exibility is given by generalizing the components of ILS
(perturbation, local search and acceptance criterion) in order
to incorporate components from other metaheuristics.
We give precise definitions of these components within the
context of our proposed template. The template proposed is
exible enough to reproduce many classical single-solution
metaheuristics and hybrids thereof, while at the same time
being sufficiently concrete to generate code from a grammar
description in order to support automatic design of algorithms.
We give examples of three IG-VNS hybrids that
can be instantiated from the proposed template.