线性回归的分布式压缩采样算法  

Distributed compressive sampling algorithm based on linear regression

作　　者：张波[1] 刘郁林[1] 常博文[1] 张建新[1] 

机构地区：[1]重庆通信学院DSP研究室,重庆400035

出　　处：《重庆邮电大学学报（自然科学版）》2014年第2期207-213,共7页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：教育部新世纪优秀人才支持计划(NCET-11-0873);重庆市自然科学基金重点项目(CSTC2011BA2016);重庆市基础与前沿研究计划项目(CSTC2013jcyjA40045)~~

摘　　要：为减少无线传感器网络数据传输量,进而延长网络的生命周期,研究了一种联合线性回归和压缩感知的分布式采样方法。依据节点数据的相关性对网络进行分簇,将感知数据显著线性相关的传感器节点划分到同一簇中。以此为基础,提出了一种基于线性回归的分布式压缩采样算法,该算法联合运用线性回归和压缩感知理论重构节点数据,实现了低速率采样条件下节点数据的高精度重构。对实测温度数据进行仿真实验,结果表明,与等间隔采样相比,该算法减少了71%的采样值个数。In order to reduce the data traffic in wireless sensor networks （WSNs） and maximize the network lifetime, we proposed a distributed sampling method through joint linear regression and compressed sensing （CS）. Firstly, a dynamic clustering algorithm based on inter-signal correlation was proposed. The proposed algorithm can divide nodes that sample data are extremely significant linear correlation into the same cluster. Then, a distributed compressive sampling algorithm was proposed. Using linear regression and CS, the proposed algorithm can reconstruct the data of nodes with high accuracy by significantly reducing the sampling rate. Our results based on real temperature data sets indicate that a reduction in sen-sor sampling by up to 71% can be achieved compared to periodically sampling.

关 键 词：无线传感器网络 压缩感知 线性回归 分簇 预测 

分 类 号：TN911[电子电信—通信与信息系统]