This article addresses the problem of dynamic job scheduling on a single machine with Poisson arrivals, stochastic processing times and due dates, in the presence of sequence-dependent setups. The objectives of minimizing mean earliness and mean tardiness are considered. Two approaches for dynamic scheduling are proposed, a Reinforcement Learning-based and one based on Fuzzy Logic and multi-objective evolutionary optimization. The performance of the two scheduling approaches is tested against the performance of 15 dispatching rules in four simulation scenarios with different workload and due date pressure conditions. The scheduling methods are compared in terms of Pareto optimal-oriented metrics, as well as in terms of minimizing mean earliness and mean tardiness independently. The experimental results demonstrate the merits of the proposed methods.
- Dispatching rules
- Fuzzy Logic
- Multi-objective evolutionary optimization
- Reinforcement Learning
ASJC Scopus subject areas