Research On Distributed Localization Algorithm Of Wireless Sensor Network

Recently,wireless sensor network(WSN)technology have been widely investigated.WSN have been applied in many fields,such as military applications,healthcare applications,and environment monitoring.Among them,node positioning technology is very important.Sensor nodes have limited cost,resources,deployment environment,and so on.It is impossible to install positioning module for all sensor nodes.Therefore,a small number of nodes install positioning modules to determine the locations of unknown nodes in the WSN.A lot of localization algorithms have been proposed.The centralized localization algorithm needs all nodes to send data to a central receiver,the computed positions are sent back to respective nodes.Due to the high computational complexity and high communication cost,this technique cannot be scalable to large network sizes.The distributed localization algorithm does not require a central receiver.The nodes themselves complete computations and obtain their positions by communicating with each other.This technique reduces the cost,have low computational complexity,and are easy to scale.However,the technique uses less information and has low localization accuracy.In this thesis,distributed localization algorithms for large scale sensor networks are thoroughly studied.(1)The sensor network localization problem can be formulated into a highly nonlinear nonconvex optimization problem.The problem is hard to solve in large scale sensor networks.For this reason,a distributed algorithm based on modified Newton's method is proposed to solve the problem.The algorithm includes network partitioning and distributed algorithm.Firstly,the network is divided into several overlapping subregions according to the nodes positions and the distance information between the sensors.The localization problem of subregions is formulated into an unconstrained optimization problem.Then distributed algorithm is used to determine nodes positions in subregions and merge the subregions.Among them,the positioning result of the triangulation is used as the initial iterate,and the Hesse matrix is modified to be positive definite matrix,and then the modified Newton's method is used to solve the problem.Simulation results indicate that compared with the existing methods,the proposed algorithm has higher scalability,higher localization accuracy and can meet the needs of nodes localization in large scale network.(2)In order to ensure the positioning accuracy and further reduce the computational complexity of distributed localization algorithm,this thesis proposes a distributed algorithm based on Barzilai-Borwein gradient method to solve the problem.Firstly,the undirected graph is decomposed into partially overlapped subgraphs,and the optimization problem is reconstructed.Then the optimization problem is decomposed into a series of small scale subproblems for iterative solutions.Each iteration is performed in two steps.Firstly,the position estimation from maximum likelihood estimation as the initial iterate for Barzilai-Borwein gradient method to solve the subproblem.The gradient method has less computational cost,and greatly speeds up the convergence.Secondly,the fusion is carried out by averaging the estimated locations within overlapped subgraphs.The simulation results show that compared with the existing methods,the proposed distributed algorithm has higher scalability,higher localization accuracy,lower computational complexity,higher robustness,and can be used for localization in large scale sensor networks.