北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2007, Vol. 30 ›› Issue (4): 5-9.doi: 10.13190/jbupt.200704.5.zhangyj

• 论文 • 上一篇    下一篇

采用强化学习的自治联合会话接纳控制

张永靖 唐恬 陈杰 张平   

  1. 北京邮电大学
  • 收稿日期:2006-08-09 修回日期:2006-09-13 出版日期:2007-08-30 发布日期:2007-08-30
  • 通讯作者: 张永靖

Autonomic Joint Session Admission Control Using Reinforcement Learning

ZHANG Yongjing,TANG Tian,CHEN Jie   

  1. School of Telecommunication Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
  • Received:2006-08-09 Revised:2006-09-13 Online:2007-08-30 Published:2007-08-30
  • Contact: ZHANG Yongjing

摘要:

提出了一种基于强化学习的联合会话接纳控制算法,用于可重配置系统中异构无线接入技术之间分布式自治的联合资源优化。通过将Q学习引入接纳控制算法,并根据各无线技术自身的特性,调整不同会话类型的反馈强化信号,能够驱使各无线接入技术吸纳更适合自己的业务,形成合理的业务分布,从而提高系统的资源利用效率。仿真结果表明,具有重叠覆盖的各无线接入技术通过这种“试错”的在线学习方式,能够收敛到较优化的接纳控制策略,在降低系统的总体呼叫阻塞率的同时获得更低的切换掉话率和更高的收益。

关键词: 可重配置, 联合会话接纳控制, 强化学习, Q学习, 分布式, 自治

Abstract:

In this paper, a reinforcement learning based joint session admission control algorithm is proposed to realize the autonomic and distributed joint resource optimization between the heterogeneous radio access technologies (RAT) in a reconfigurable system. By introducing Q-learning into the admission control algorithm and adjusting the strength of the reinforcement signals for different types of sessions considering the inherent characteristics of different RATs, RATs are driven to absorb the suitable traffic for a proper service distribution, which improves the efficiency of resource utilization. The simulation results show that, through the “trial-and-error” on-line learning process, overlapping RATs can converge to the optimized admission control policies that reduce the overall blocking probability while achieve lower handover dropping probability as well as higher revenue.

Key words: reconfigurable, JOSAC, reinforcement learning, Q-learning, distributed, autonomic

中图分类号: