基于流量模式的Q-学习路由及其连接调度  

Q-Learning Routing and Link Scheduling Based on Traffic Mode

作　　者：姚铭明 曹霑懋[1] 黄启嵩 单志龙[1] YAO Mingming;CAO Zhanmao;HUANG Qisong;SHAN Zhilong(School of Computer, South China Normal University, Guangzhou 510631, China)

机构地区：[1]华南师范大学计算机学院,广州510631

出　　处：《华南师范大学学报（自然科学版）》2021年第4期107-114,共8页Journal of South China Normal University(Natural Science Edition)

基　　金：国家自然科学基金项目(61671213);广州市科技计划项目(202007040006)。

摘　　要：为解决无线网状网中因多条路径同时传输数据而引起网络性能降低的问题,提出了一个基于流量的Q-学习路由与调度方案(QRST):针对每一个路由请求,首先采用强化学习中的Q-学习算法寻找路径;然后根据找到的路径结合信道分配完成组合调度,以启发式的方法尽可能为每个时隙使用网络资源分配路径的连接.并在不同网络资源配置和多种流量请求下进行虚拟计算实验,以验证该方案的正确性和有效性.实验结果表明:与COSS方案和AODV方案相比,采用QRST方案的无线网状网在吞吐量、激活链路数量和传输完成时间等网络性能上有较好的表现.The interference and resource congestion caused by multiple concurrent flows may cause sharp perfor-mance degradation of wireless mesh networks.In order to solve the problem,a Q-learning Routing and Scheduling concerning Traffic(QRST)scheme is proposed.Firstly,the Q-learning algorithm is used to find the path for each routing request.Then the combined scheduling is completed according to the path finding and channel allocation,and the connection of paths is allocated with cyber source for every slot in a heuristic way.In order to verify the co-rrectness and effectiveness of the scheme,virtual computing is performed under different network resource configurations and multiple traffic requests.The experimental results show that,compared with COSS and AODV,wireless mesh network using the QRST scheme has better performance in terms of throughput,activated link number and transmission completion time.

关 键 词：无线网状网 路由 强化学习 Q-学习 多并发流 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]