Comparing two models to generate hyper-heuristics for the 2d-regular bin-packing problem

Abstract

The idea behind hyper-heuristics is to discover some combination of straightforward heuristics to solve a wide range of problems. To be worthwhile, such combination should outperform the single heuristics. This paper presents two Evolutionary-Computation-based Models to produce hyperheuristics that solve two-dimensional bin-packing problems. The first model uses an XCS-type Learning Classifier System which learns a solution procedure when solving individual problems. The second model is based on a GA that uses a variable-length representation, which evolves combinations of condition-action rules producing hyper-heuristics after going through a learning process which includes training and testing phases. Both approaches, when tested and compared using a large set of benchmark problems, perform better than the combinations of single heuristics. The testbed is composed of problems used in other similar studies in the literature. Some additional instances of the testbed were randomly generated.

Optimal job packing, a backfill scheduling optimization for a cluster of workstations

“…We use the Alea parallel and distributed job scheduling simulator built on top of the GridSim simulator [19,20]. The Alea simulator is an event-based modular simulator, composed of independent entities like a centralized scheduler, job generator, job submission system, and the resources.…”

DVRP: a hard dynamic combinatorial optimisation problem tackled by an evolutionary hyper-heuristic

“…5.1). Unlike most of the evolutionary-based hyper-heuristic approaches proposed in the literature (Han and Kendall 2003;Ross et al 2004;Marín-Blázquez and Schulenburg 2006;Ersoy et al 2007;Burke et al 2007;Terashima-Marin et al 2007;Bader-El-Den and Poli 2008;Kumar et al 2008;Pillay 2008;Terashima-Marín et al 2008;Tay and Ho 2008), and in accordance with the new and unified classification of hyper-heuristics described in Burke et al (2010), we have proposed a hybrid heuristic selection approach whose general representation corresponds to a sophisticated, yet easy to understand, sequence of constructiveperturbative pairs of low-level heuristics which gradually construct and improve partial solutions.…”

“…In the above categories, several evolutionary approaches have successfully been applied to a great amount of NP-hard problems (Han and Kendall 2003;Krasnogor and Gustafson 2004;Ross et al 2004;Özcan 2006;Bader-El-Den and Poli 2008;Ersoy et al 2007;Garrido and Riff 2007;Burke et al 2007;Poli et al 2007;Terashima-Marin et al 2007;Kumar et al 2008;Pillay 2008;Tay and Ho 2008). Considering the nature of the heuristic search space, the major contributions in heuristic selection methodologies are genetic/memetic algorithms and evolutionary classifier systems.…”

“…Genetic algorithms have been proposed either to generate sequences of constructive heuristics for the problem at hand or to evolve and find (generalised) patterns based on previous knowledge, in a similar way as classifier systems induce rules. Both genetic algorithms and classifiers systems have been proposed in domains such as trainer scheduling (Han and Kendall 2003), offline packing , cutting stock (Terashima-Marin et al 2007), and examination timetabling (Pillay 2008).…”

Providing a memory mechanism to enhance the evolutionary design of heuristics

“…Previous work has also shown that human competitive heuristics can be evolved for two dimensional strip packing problems [19], and heuristics designed by evolution for the three dimensional packing problem are compared to a human designed heuristic in [20]. A genetic algorithm is used to evolve hyper-heuristics for the two dimensional packing problem in [21], [22], [23], [24]. The evolved individual contains criteria to decide which packing heuristic to apply next, based on the properties of the pieces left to pack.…”