Thèse soutenue

FR  |  
EN
Auteur / Autrice : Roman Potarusov
Direction : Gilles GoncalvesВистор КурейчикVictor Kureychik
Type : Thèse de doctorat
Discipline(s) : Science pour l’ingénieur. Génie informatique
Date : Soutenance en 2008
Etablissement(s) : Artois en cotutelle avec Taganrog, Université fédérale du sud

Mots clés

FR

Mots clés contrôlés

Résumé

FR  |  
EN

One-dimensional Bin Packing Problem (BPP) is well known combinatorial optimization problem. BPP in its general form is P-hard in the strong sense, so there is a little hope of finding even pseudo-polynomial time optimization algorithm for it. BPP is an interesting topic of research, because BPP is encountered in many industries, such as steel, glass and paper manufacturing. There are many other industrial problems that seem to be different, but have a very similar structure, such as capital budgeting, processor scheduling and VLSI design. BPP models several practical problems in computer science. Some examples are: table formatting, prepaging, file allocation. In this thesis the Hybrid Parallel Genetic Algorithm to solve 1-D BPP has been presented. Two evolution models (de Vries’ evolution model and Lamarck’s evolution model) have been adapted to solve the BPP. New problem-oriented genetic operators have also been developed. They never decrease the quality of solution and allow obtaining valid BPP solutions. Two effective local search algorithms are proposed. They allow improving of BPP solutions to get quasi-optimal and optimal packings. Computational experiments show that the presented algorithm gives quasi-optimal and optimal solutions for all benchmark instances in an acceptable amount of computing time, clearly showing the robustness of the proposed approach. In the case of quasi-optimal solutions the absolute deviation from reference solution is at most one bin. Future work could explore the possibility of designing more sophisticated architectures of genetic search with migration and applying the proposed approach to solve the Vehicle Routing Problem with Multiple Routes. BPP approach seems to be effective to distribute routes to vehicles.