基于POMDP的电梯群控调度策略  

A POMDP Based on Elevator Group Control Scheduling Strategy

作　　者：彭诚[1] 姚进发 董正山 PENG Cheng;YAO Jinfa;DONG Zhengshan(Institute of Automotive Engineering,Hefei Technology College,Hefei,Anhui 238010,China;Research Institute of Ruijie Networks Co.,Ltd.,Fuzhou,Fujian 350002,China;College of Mathematics and Data Science(Software College),Minjiang University,Fuzhou,Fujian 350108,China)

机构地区：[1]合肥职业技术学院汽车工程学院,安徽合肥238010 [2]锐捷网络股份有限公司研究院,福建福州350002 [3]闽江学院数学与数据科学学院(软件学院),福建福州350108

出　　处：《闽江学院学报》2022年第5期104-111,共8页Journal of Minjiang University

基　　金：安徽省重点自然科学项目(KJ2021A1403)。

摘　　要：针对电梯群组系统的随机性和复杂性,以离散事件动态系统和分布式部分可观马尔可夫决策过程为理论基础,将电梯群组的调度问题建模为基于事件驱动的部分可观马尔可夫决策模型,并利用多智能体强化学习算法求解最优调度策略。仿真实验结果表明,所提出的事件驱动Dec-POMDP调度模型不仅能够较好地描述电梯群组系统环境中的不确定性,而且能够避免维度灾难带来的策略优化难题,所得到的最优策略具有较好的调度性能。To cope with the stochasticity and complexity of the elevator group system,this paper presents the scheduling problem of elevator group as an event-driven Markov decision model based on the theory of discrete event dynamic system(DEDS)and distributed partially observable Markov decision process(Dec-POMDP),and solves the optimal scheduling strategy by using multi-intelligent reinforcement learning algorithm.The simulation results show that the proposed event-based Dec-POMDP scheduling model can not only better describe the uncertainty in the system environment,but also avoid the policy optimization difficulties caused by dimensional disasters.The resulted optimal policy has better scheduling performance.

关 键 词：电梯群控系统 离散事件动态系统 马尔可夫决策过程 强化学习 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]