Improving Schedule Robustness via Stochastic Analysis and Dynamic Adaptation

We study methods that improve scheduling robustness under uncertain disturbances and dynamic shop conditions. A new method is proposed based on the notion of preprocessrst-schedule-later by Wu et al. [19]. Preprocessing starts at the beginning of the planning period where we use a Lagrangean Relaxation of the scheduling model to form networkstructured job subproblems. For each job subproblem we introduce stochastic constraints which capture a priori information in the form of processing time uncertainty. This stochastic information comes at minimal computational expenses since the subproblems retain their special network structure. Using a subgradient search algorithm, we iteratively improve the lower and upper bounds of the scheduling instance. The preprocessing produces a partially resolved sequence or a Lagrangean Ranking. Actual schedule generation is performed dynamically over time using theLagrangean ranking. Intensive computational experiments show that the proposed preprocessing signi cantly outperforms its deterministic counterparts without extra computing burden, achieves robust performance under highly uncertain conditions, and could be used to improve the quality of dynamic dispatching.