Rule driven multi objective dynamic scheduling by data envelopment analysis and reinforcement learning

This paper presents a rule driven method of developing composite dispatching rule for multi objective dynamic scheduling. Data envelopment analysis is adopted to select elementary dispatching rules, where each rule is justified as efficient for optimizing specific operational objectives of interest. The selected rules are subsequently combined into a single composite rule using the weighted aggregation manner. An intelligent agent is trained using reinforcement learning to acquire the scheduling knowledge of assigning the appropriate weighting values for building the composite rule to cope with the WIP fluctuation of a machine. Implementation of the proposed method in a two objective dynamic job shop scheduling problem is demonstrated and the results are satisfactory.