A reinforcement learning approach to support setup decisions indistributed manufacturing systems

摘要

A reinforcement learning approach to specifying payoffs for setupgames is presented. Setup games are normal form, noncooperative gamesused by heterarchical machine controllers to evaluate reconfigurationdecisions. While past work utilizing heuristic measures to approximatethe effect of setup decisions has demonstrated promising performance,the lack of an accurate long-term model of system dynamics in theseheuristic approaches limits their usefulness. The reinforcement learningapproach iteratively learns the long term costs of setup decisions,accounting for both immediate decision effects and the effects of likelydownstream decisions