Learning From Scenarios for Stochastic Repairable Scheduling

Decision-making can be challenging due to the stochastic nature of real-world processes. This complexity is evident in various contexts, such as manufacturing, where uncertain processing times make it challenging to meet strict customer deadlines. Formulating Constrained Optimization (CO) models for these problems is common, but unknown parameter values during decision-making add challenges, because wrong estimates of the parameters can lead to infeasibilities. In practice, such infeasibilities are repaired when reality unfolds. For instance, in a manufacturing system, tasks may be postponed due to delays in earlier stages to maintain the factory's flow. Various repair policies and schedule definitions are used across different contexts. Historical data, represented as scenarios of unknown parameters like task duration, is often available. Simple averaging of these scenarios is a common yet naive approach that ignores uncertainty.