Optimizing Rescheduling Intervals Through Using Multi-Armed Bandit Algorithms

Abstract

Well scheduling in oil and gas production in a virtual enterprise is a distributed and online scheduling problem. For such a scheduling problem, planned schedules are subject to unexpected disruptions or under- or over- estimated completion times. To reduce the impact of these uncertain events, schedule revision is necessary to keep the current schedule feasible and optimal in productivity. However, even though frequent schedule revisions may maximize the number of well tasks, it can also increase machine setup and transportation costs. This indicates the necessity of designing a systematic strategy for determining when to carry out schedule revisions. There is no trivial solution to this problem. In this research, we propose an approach to rescheduling interval determination through using a reinforcement learning - multiarmed bandit model. A set of experiments is conducted in a multiagent simulation environment. The results of the experiment demonstrate the effectiveness of the proposed approach in detecting optimal rescheduling intervals.