Solving a scheduling problem consists of organizing a set of tasks, that is assigning their starting and ending times and allocating resources such that all constraints are satisfied.In this thesis, we propose new constraint programming approaches for two categories of NP-hard scheduling problems which are validated experimentally by the implementation of a set of new features within the constraint solver choco.
In shop scheduling, a set of n jobs, consisting each of m tasks, must be processed on m distinct machines. A machine can process only one task at a time. The processing orders of tasks which belong to a job can vary (global order, order per job, no order). We consider the construction of non-preemptive schedules of minimal makespan. We first propose a study and a classification of different constraint models and search algorithms. Then, we introduce a new flexible approach for these classical problems.
A batch processing machine can process several jobs simultaneously as a batch. The starting and ending times of a task are the ones of the batch to which they belong. The studied problem consists of minimizing the maximal lateness for a batch processing machine on which a finite number of tasks of non-identical sizes must be scheduled. The sum of the sizes of the jobs that are in a batch should not exceed the capacity b of the machine. We propose, within this context, a constraint model based on a decomposition of the problem. Then, we define a new optimization constraint based on the resolution of a relaxed problem enhanced by cost-based domain filtering techniques which improves the resolution.