Stochastic Algorithms for Optimization and Application to Job-Shop-Scheduling

Combinatorial optimization problems are often used to test heuris-tics. Among heuristics, stochastic ones deserve particular consideration being generally meta-heuristics that aim at performing reasonnably well on a wide spectrum of problems. Among them, evolutionary algorithms have recently appeared. Emphasis have been put on them by researches that have shown that they are able to solve eeciently a wide range of problems, among them combinatorial optimization problems. It is however not easy to compare meta-heuristics between each others, and with other more classical heuris-tics. However, this kind of cross-paradigm comparisons are much worthy. In this paper, we perform a comparison of three kinds of algorithms (multi-start hill-climber, Tabu search, and evolutionary algorithm) on the job-shop-scheduling problem using a series of well-known standard benchmarks. Not surprisingly, hybrid algorithms perform better than pure algorithms. However , it is noteworthy that a multi-start hill-climber often outperforms the other algorithms.