Font Size: a A A

Research On Node Localization Method In Wireless Sensor Networks Based On Support Vector Machines

Posted on:2015-08-30Degree:MasterType:Thesis
Country:ChinaCandidate:G H LinFull Text:PDF
GTID:2298330431490229Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
Node localization in wireless sensor networks is one of the most important functions.Validity of other data closely related with the accuracy of node self-localization. With thedevelopment of wireless sensor network applications recently, new applications pose newchallenges for node localization. However, due to the node’s own conditions, there is not auniversal positioning algorithm until now. Through the research of the basic knowledge ofnode localization in this paper, combined with support vector machine (SVM) and leastsquares support vector machine (LS-SVM) relevant theory, and improvements on it. Theimproved method is applied to the node positioning technology. This article make somecontribution to how to improve the rapidity and robustness of node localization and howto achieve mobile node localization. Includes the following aspects:1) Diferent combinations of regularization parameters and kernel parameters playa pivotal role on the exactness and rapidity of the LS-SVM. Existing methods of tuningthe parameters for LS-SVM have some disadvantages, such as difculty of getting globaloptimum and poor convergent speed. In this paper, we manufacture the parameter designof LS-SVM using quantum particle swarm optimization (QPSO) and apply the designedLS-SVM into node localization. According to the initial values obtained from QPSO, theoptimal parameter values are obtained using the leave-10-out cross validation method.Simulation results show that the proposed method is compared to the coupling simulatedannealing (CSA) LS-SVM and the improved particle swarm optimization (IPSO) LS-SVM.2) This paper studies and improves the weighted least squares support vector machine(WLS-SVM). We propose a modifed robust least squares support vector machine (MRLS-SVM) and employ it to the node localization of wireless sensor networks, so as to improvethe accuracy and robustness. During the process of node localization, we frst use theWLS-SVM and a iterative weighting scheme to predict the position of virtual node, thencombine with location information and the values of sample to gain a set of data asa new sample. After that, the model parameters can be obtained and the location ofunknown node is predicted. Finally, simulation example shows that the proposed methodcan achieve the node localization more accurate than the LS-SVM and WLS-SVM.3) The mobile node localization is divided into three models depending on the typeof the moving node, and modeling node localization for each model. Through the estab-lishment of Nonlinear AutoRegressive with eXogenous Inputs model to predict the futurelocation node for some mobile node localization systems which must have high timeliness.Simulation output evidence that the MRLS-SVM algorithm can be applied to three kindsof positioning model. It also shows the forecasting model built in this paper can efectivelypredict the future trajectory of mobile node in the next few cycles.
Keywords/Search Tags:least squares support vector machine (LS-SVM), node localization, Quantum Particle Swarm Optimization (QPSO), robustness, Mobile node
PDF Full Text Request
Related items