基于半监督学习的无线传感器网络节点定位问题研究

发布时间：2018-01-16 11:15

本文关键词：基于半监督学习的无线传感器网络节点定位问题研究　出处：《西南交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着大数据和云计算技术的发展,无线传感器网络(Wireless Sensor Network,WSN)已经步入大数据时代。作为新型的无线通信网络,WSN的主要目标是获取网络环境中的数据,这些数据来自于传感器节点,节点所处的位置不同,数据代表的意义也就不同。因此,节点的位置信息是WSN的重要参数,节点定位是WSN的一项重要任务。随着机器学习技术的发展,将半监督学习思想引入WSN节点定位中,可以减小算法对信标节点比例的敏感度,并且可以获得较高的定位精度。本文基于DV-Hop算法思想,建立了监督学习和半监督学习的定位算法模型,并对算法的定位性能进行了比较。本文首先介绍WSN节点定位的基本概念以及无需测距定位算法的定位原理,并对三个无需测距定位算法进行仿真分析,对比三个算法在不同信标节点比例下的平均定位误差和定位覆盖率。其次,本文将经典DV-Hop定位算法获取跳数的思想引入到支持向量机(Support Vector Machine,SVM)中,建立基于跳数的多分类SVM定位算法模型。该SVM定位算法根据"一对多"的构造思想,将WSN网络区域等分为多个网格,将信标节点的网格编号以及所有节点的跳数向量作为支持向量机训练参数,训练网格编号与跳数向量的映射模型,并通过训练好的模型预测未知节点的位置坐标。仿真结果表明,在节点通信半径较大,信标节点比例较高,网格划分长度较小的情况下,相比经典DV-Hop算法以及O-DV-Hop改进算法,基于跳数的多分类SVM算法的定位精度较高。最后,本文将基于跳数的多分类SVM算法与机器学习算法中的k近邻算法相结合,建立基于协同训练的半监督SVM(SSL)定位算法模型。该SSL定位算法同时训练两个定位模型,取标记结果一致的节点作为新的信标节点,并将该信标节点的参数输入定位模型中进行训练,不断更新定位模型,直到全部节点定位完成。仿真结果表明,相比基于跳数的多分类SVM算法,SSL算法不仅定位精度有所提高,并且降低了算法对信标节点比例的敏感度。
[Abstract]:With the development of big data and cloud computing technology, wireless sensor network (WSN) wireless Sensor Network. WSNs have entered the era of big data. As a new wireless communication network, the main goal of WSN is to obtain the data in the network environment. These data come from sensor nodes and the nodes are located in different positions. Therefore, the location information of nodes is an important parameter of WSN, and node location is an important task of WSN. With the development of machine learning technology. Introduction of semi-supervised learning into WSN node localization can reduce the sensitivity of the algorithm to the beacon node ratio and obtain high positioning accuracy. This paper is based on the DV-Hop algorithm. The location algorithm model of supervised learning and semi-supervised learning is established, and the localization performance of the algorithm is compared. Firstly, this paper introduces the basic concept of WSN node localization and the positioning principle of the location algorithm without ranging. And the simulation analysis of three location algorithms without ranging, compared with the three algorithms in different beacon node proportion of the average positioning error and location coverage. Secondly. In this paper, the classical DV-Hop localization algorithm is introduced to support Vector Machine (SVM). A multi-class SVM localization algorithm model based on hops is established. According to the idea of "one-to-many" construction, the WSN network region is divided into several meshes. The mesh number of the beacon node and the hop vector of all nodes are taken as the training parameters of the support vector machine, and the mapping model between the training grid number and the hopping vector is proposed. The simulation results show that when the communication radius of nodes is large, the proportion of beacon nodes is higher, and the length of mesh division is small. Compared with the classical DV-Hop algorithm and O-DV-Hop improved algorithm, the multi-class SVM algorithm based on hops has higher positioning accuracy. Finally. In this paper, the multi-class SVM algorithm based on hops is combined with the k-nearest neighbor algorithm in the machine learning algorithm. A semi-supervised SSL location algorithm model based on cooperative training is established, which trains two localization models at the same time and takes the nodes with the same tagging results as the new beacon nodes. The parameters of the beacon node are input into the localization model to train and update the location model until all nodes are located. The simulation results show that compared with the multi-classification SVM algorithm based on hops. The SSL algorithm not only improves the location accuracy, but also reduces the sensitivity of the algorithm to the beacon node ratio.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN929.5;TP212.9

【参考文献】