基于混合压缩感知的分簇式感知网络数据收集研究

发布时间：2018-09-04 18:26

【摘要】：现如今,无线感知技术已经融入到我们生活,学习,工作和娱乐当中。而在无线感知数据收集网络中低耗损,低延时数据传输一向是研究热门方向。节点的数据采集与压缩,高效的路由策略,数据处理,都包含对网络低功耗的分析。本文分析了传统分簇式感知网络数据收集方法与基于混合压缩感知的分簇式感知网络数据收集方法。通过分析现有分簇式感知网络数据收集算法,注意到网络中数据传输延迟与吞吐量是衡量算法好坏的关键指标。因此,本文提出了基于马尔科夫模型选择移动sink路径的数据收集方法。首先对监控的无线传感器网络进行区域划分,然后应用马尔科夫模型预测节点位置,其次根据预测得到的节点数目期望值对网络区域赋予优先级,再对各区域进行数据融合,最后得到最优的移动sink节点运动轨迹。本方法能适用于不同规模的网络,具有节省网络能量、延长网络生存周期、降低网络整体延时、并防止数据溢出等优点。其次,提出一种类BP神经网络的分簇传感网络数据收集方法。首先对监控的无线传感器网络进行初始化处理,依节点的GPS信息寻找网络的地理中心位置,其次根据节点的位置信息选举网络簇头节点并构建簇,再由已经完成的分簇网络建立BP神经网络模型,最后根据网络的总传输跳数动态调整网络的分簇数量,至网络达到最优簇头数目的稳态。本方法能适用于不同规模大小的网络,以最优分簇数量收集网络数据,具有减少网络能耗、延长网络生命周期、降低网络延时等优点。为了减少分簇式传感器网络中的数据传输量并均衡网络负载,提出了一种采用混合压缩感知进行数据收集的方法。首先选定各临时簇中离簇质心最近的一些节点为候选簇头节点,然后基于已确定的簇头节点到未确定的候选簇头节点的距离依次确定簇头,其次各普通节点选择加入距离自己最近的簇,最后贪婪构建一棵以sink节点为根节点并连接所有簇头节点的数据传输树,对数据传输量高于门限值的节点使用CS压缩数据传输。仿真结果表明当压缩比率为10时,数据传输量比Clustering without CS和SPT without CS分别减少了75%和65%,比SPT with Hybrid CS和Clustering with Hybrid CS分别减少了35%和20%;节点数据传输量标准差比Clustering without CS和SPT without CS分别减少了62%和81%,比SPT with Hybrid CS和Clustering with Hybrid CS分别减少了41%和19%。
[Abstract]:Nowadays, wireless sensing technology has been integrated into our life, study, work and entertainment. In wireless perceptual data collection networks, low-latency data transmission has always been a hot research direction. Node data acquisition and compression, efficient routing strategy, data processing, including the analysis of the low power consumption of the network. In this paper, we analyze the traditional clustering perceptual network data collection method and the clustering perceptual network data collection method based on hybrid compression perception. By analyzing the existing clustering perceptual network data collection algorithms, it is noted that the data transmission delay and throughput in the network are the key indicators to evaluate the algorithm. Therefore, this paper proposes a method of data collection based on Markov model to select mobile sink paths. Firstly, the monitored wireless sensor network is divided into regions, then the node position is predicted by Markov model, and then the network area is given priority according to the expected value of the number of nodes predicted, and then the data fusion of each region is carried out. Finally, the optimal trajectory of moving sink nodes is obtained. The method can be applied to networks of different scales, and has the advantages of saving network energy, prolonging network lifetime, reducing overall network delay and preventing data overflow. Secondly, a clustering sensor network data collection method based on BP neural network is proposed. Firstly, the monitored wireless sensor network is initialized to find the geographical center of the network according to the GPS information of the node, and then the cluster head node is selected according to the location information of the node and the cluster is constructed. Then the BP neural network model is established from the completed clustering network. Finally, according to the total transmission hops of the network, the clustering number of the network is dynamically adjusted to the steady state of the optimal cluster head number of the network. The method can be applied to networks of different size and size, collect network data with the optimal number of clusters, and has the advantages of reducing network energy consumption, prolonging network life cycle and reducing network delay. In order to reduce the amount of data transmission in cluster sensor networks and to balance the network load, a method of data collection using hybrid compression sensing is proposed. First, some nodes in each temporary cluster are selected as candidate cluster heads, and then the cluster heads are determined based on the distance from the determined cluster heads to the undetermined candidate cluster heads. Secondly, each common node chooses to join the nearest cluster. Finally, a data transmission tree with sink node as the root node and connecting all cluster head nodes is constructed, and the data transmission is compressed by CS for the node whose data transmission amount is higher than the threshold. The simulation results show that when the compression ratio is 10:00, The volume of data transmission is 75% and 65% less than Clustering without CS and SPT without CS, 35% and 20% less than SPT with Hybrid CS and Clustering with Hybrid CS, respectively. The standard deviation of node data transmission is 62% and 81% less than Clustering without CS and SPT without CS, respectively, compared with SPT with Hybrid CS. And Clustering with Hybrid CS decreased by 41% and 19%, respectively.
【学位授予单位】：湘潭大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212.9;TN929.5

【参考文献】