This paper describes a method to control Autonomous decentralized Flexible Manufacturing System (AD-FMS) by using a memory to determine a priority ranking for competing hypotheses. The aim is to increase the reasoning efficiency of a system the author calls reasoning to anticipate the future (RAF) which controls automatic guided vehicles (AGVs) in AD-FMSs. The system includes memory data of past production conditions and AGV actions. The system was applied to an AD-FMS that was constructed on a computer. The results showed that, compared with conventional reasoning, this reasoning system reduced the number of hypothesis replacements until a true hypothesis was reached.