Integer programming-based decomposition approaches for solving machine scheduling problems

• Main Author: Sadykov, Ruslan
• Format: Others
• Language: en
• Published: Universite catholique de Louvain 2006
• Subjects: Interval-indexed formulation; Dantzing-Wolfe decomposition; Generic Benders decomposition; Time-indexed formulation; Branch-and-check; No-good cuts; Edge-finding
• Online Access: http://edoc.bib.ucl.ac.be:81/ETD-db/collection/available/BelnUcetd-06232006-155952/

The aim in this thesis is to develop efficient enumeration algorithms to solve certain strongly NP-hard scheduling problems. These algorithms were developed using a combination of ideas from Integer Programming, Constraint Programming and Scheduling Theory. In order to combine different techniques in one algorithm, decomposition methods are applied. The main idea on which the first part of our results is based is to separate the optimality and feasibility components of the problem and let different methods tackle these components. Then IP is ``responsible' for optimization, whereas specific combinatorial algorithms tackle the feasibility aspect. Branch-and-cut and branch-and-price algorithms based on this idea are proposed to solve the single-machine and multi-machine variants of the scheduling problem to minimize the sum of the weights of late jobs. Experimental research shows that the algorithms proposed outperform other algorithms available in the literature. Also, it is shown that these algorithms can be used, after some modification, to solve the problem of minimizing the maximum tardiness on unrelated machines. The second part of the thesis deals with the one-machine scheduling problem to minimize the weighted total tardiness. To tackle this problem, the idea of a partition of the time horizon into intervals is used. A particularity of this approach is that we exploit the structure of the problem to partition the time horizon. This particularity allowed us to propose two new Mixed Integer Programming formulations for the problem. The first one is a compact formulation and can be used to solve the problem using a standard MIP solver. The second formulation can be used to derive lower bounds on the value of the optimal solution of the problem. These lower bounds are of a good quality, and they can be obtained relatively fast.