Algorithm comparison by automatically configurable stochastic local search frameworks: A case study using flow-shop scheduling problems

Abstract

The benefits of hybrid stochastic local search (SLS) methods, in comparison with more classical (non-hybrid) ones are often difficult to quantify, since one has to take into account not only the final results obtained but also the effort spent on finding the best configuration of the hybrid and of the classical SLS method. In this paper, we study this trade-off by means of tools for automatic algorithm design, and, in particular, we study the generation of hybrid SLS algorithms versus selecting one classical SLS method among several. In addition, we tune the parameters of the classical SLS method separately and compare the results with the ones obtained when selection and tuning are done at the same time. We carry out experiments on two variants of the permutation flowshop scheduling problem that consider the minimization of weighted sum of completion times (PFSP-WCT) and the minimization of weighted tardiness (PFSP-WCT). Our results indicate that the hybrid algorithms we instantiate are able to match and improve over the best classical SLS method. © Springer International Publishing Switzerland 2014.