摘要
电梯群控调度是一类开放、动态、复杂系统的多目标优化问题.目前应用于群控电梯调度的算法主要有分区算法、基于搜索的算法、基于规则的算法和其他一些自适应的学习算法.但已有方法在顾客平均等待时间等目标上并不能够达到较好的优化性能.本文采用强化学习技术应用到电梯群控调度系统中,使用CMAC神经网络函数估计模块逼近强化学习的值函数,通过Q-学习算法来优化值函数,从而获得优化的电梯群控调度策略.通过仿真实验表明在下行高峰模式下,本文所提出的基于CMAC网络强化学习的群控电梯调度算法,能够有效地减少平均等待时间,提高电梯运行效率.
Elevator group control is a multi-objective optimization problem in an open, complicated and dynamical system. Currently,many algorithms have been applied in elevator group control, such as zoning approaches, search-based approaches,rulebased approaches and other adaptive approaches. However these methods fail of achieving the optimal performance in the average wait time. In this paper, the reinforcement learning technology is applied in the elevator group control system. The CMAC neural network is used to approx the value function of reinforcement learning and Q-learning algorithm is used to optimize the value function,thereby the optimal control policy of the elevator group control is achieved. The simulation experiment shows that the elevator group control using reinforcement learning with CMAC can reduce the average wait time efficiently in the down peak Waffle.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2007年第2期362-365,共4页
Acta Electronica Sinica
基金
国家自然科学基金(No.60475026)
国家杰出青年科学基金(No.60325207)
国家973重点基础研究发展规划(No.2002CB312002)
关键词
电梯群控调度
强化学习
CMAC神经网络
函数估计
elevator group control
reinforcement leaming
CMAC netral network
function approximation
