Decomposition-based job-shop scheduling with constrained clustering
Scheduling is a crucial problem appearing in various domains, such as manufacturing, transportation, or healthcare. In most problem definitions, the goal is to schedule a given set of operations on available resources to complete the operations as early as possible. Unfortunately, most scheduling problems cannot be solved efficiently. Therefore, the research of suitable approximation methods is of primary importance.
This work suggests a novel approximation approach based on problem decomposition with data mining methodologies. This study proposes a constrained clustering algorithm to group the operations into clusters corresponding to time windows in which these operations must be scheduled. The decomposition process depends on two main phases. The first phase is to extract features to predict the sequence of the operations on each resource. These features are extracted either from the problem itself or from solutions obtained by other heuristics. The second phase is to develop a constrained clustering algorithm to assign each operation into a time window. We solved the problem using the Answer Set Programming. Evaluation results show that our proposed outperformed other heuristic schedulers in most cases, where features, like Remaining Processing Time, Machine Load, and Earliest Starting Time, contributed significantly to the solution quality.