一种应用强化学习的自适应无线传感器网络路由算法  

An Adaptive Wireless Sensor Network Routing Algorithm Using Reinforcement Learning

作　　者：江勇[1] 赵倩[1,2] 

机构地区：[1]清华大学深圳研究生院,广东深圳518055 [2]清华大学计算机科学与技术系,北京100084

出　　处：《小型微型计算机系统》2013年第8期1723-1727,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60972011)资助;国家"九七三"重点基础研究发展计划项目(2009CB320502)资助

摘　　要：数据收集是无线传感器网络的重要应用之一,其主要的工作过程可以概括为传感器节点将感知的信息通过一定的路径传送到无线网关节点进行进一步分析处理的过程.在数据收集时,由于人们无法预知事件触发的地点,常常将传感器均匀布置在监测的场所中,但是信息收集的地点往往是不均匀分布的,这就导致了一部分节点会因处在事件频发地段而持续的工作,而另一些节点却始终不会工作.为了解决这个问题,提出一个应用加强学习算法的自适应无线路由策略.在该路由策略中,路由的过程被当作分布式智能节点加强学习的过程.每一个传感器节点都是一个独立的智能节点,可以通过参数化的选择概率和回报来决定自己的下一跳地址.该策略的目的是使长时间不工作的节点代替长时间工作的节点传输数据,以达到平均节点能耗,延长整体网络寿命的效果.最后的仿真结果说明我们的路由策略可以有效的分散数据传输,延长网络寿命.Data gathering has been an important research area in wireless sensor networks. The operation of data gathering is the sys- tematic transmission of sensed data to a sink for further processing. In data collection, because it＇s unlikely to predict the exact place that the events are triggered, sensors are always uniformly deployed among the monitored place. But events often have uneven distri- bution, which leads to the situation that some sensor nodes will work continuously, while some other nodes do not work at all. To solve this problem, we propose an adaptive wireless sensor network routing algorithm which uses reinforcement learning technology. In this strategy, routing process is treated as a process that distributed intelligent nodes do reinforcement learning. Each sensor node is a stand-alone intelligent node and determines the next-hop address by the parameterized probability and rewards. The routing scheme can make inert nodes take the place of busy-working nodes to transfer data, so that it can average nodes＇ energy consumption and pro- long the whole network life time. At last, the simulation results prove that our scheme can efficiently decentralize data transmission and prevent nodes from early dying.

关 键 词：无线传感器网络 路由 机器学习 加强学习 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]