Research On Node Localization In Wireless Sensor Network Based On Semi-supervised Learning

With the development of big data and cloud computing technology,wireless sensor network(WSN)has entered the era of big data.WSN is a new type of wireless communication network.The main goal of WSN is to obtain the data of network,these data come from sensor nodes,and different locations of nodes represent different meanings.Therefore,the location of node is an important parameter of WSN,and the node localization is an important task of WSN.With the development of machine learning,semi-supervised learning can be used into WSN localization,in this way,the sensitivity of algorithm to the proportion of beacon nodes can be reduced and the high localization accuracy can be obtained.In this thesis,the localization algorithm model of supervised learning and semi-supervised learning are established based on the thinking of DV-Hop algorithm,and the localization performance of different algorithms are compared.Firstly,the basic concept of WSN node localization and the principle of range free localization algorithms are introduced,the simulation of three range free localization algorithms are analyzed,and the average localization error and localization coverage rate of three localization algorithms are compared under different proportion of beacon nodes.Secondly,the thinking of getting the hops of DV-Hop algorithm is introduced into support vector machine(SVM),and the model of multi-classification SVM localization algorithm based on hops is established.The SVM localization algorithm is based on the thinking of "one to many",the WSN area is divided into a plurality of grids.The grids number of beacon nodes and the hop vectors of all nodes are used as SVM training parameters,the mapping relations of them are trained for the algorithm model.The position coordinates of unknown nodes are predicted by the trained model.The results of simulation show that the localization accuracy of the SVM localization algorithm is high compared with the DV-Hop localization algorithm and the O-DV-Hop improved algorithm when the communication radius of node is large,the proportion of beacon nodes is high,and the length of grid is small.Finally,the SVM localization algorithm and the k nearest neighbor algorithm of machine learning algorithm are combined to establish the model of semi-supervised SVM(SSL)localization algorithm.The SSL localization algorithm is based on collaboration training,two localization models are trained at the same time,the same marking results of unknown nodes are used as new beacon nodes.The parameters of new beacon nodes are put into the trained model by the SSL localization algorithm,the localization model is updated until all nodes are located.The results of simulation show that compared with the SVM localization algorithm,not only the localization accuracy of SSL algorithm is improved,but also the sensitivity of algorithm to the proportion of beacon nodes is reduced.