Data Outlier Detection And Recovery In WSN Based On Graph Signal Processing

Along with Wireless Sensor Network(WSN)widely applied in many fields,WSN data processing has attracted more and more attention in signal processing.However,the WSN data outlier and missing often appear due to the limitations of sensor nodes' power,storage space,and computing power,also the unstable network transmission.WSN outlier detection and recovery problem have become one of the fundamental problems in WSN data processing.In recent years,a novel signal processing theory--Graph Signal Processing(GSP),provides a new framework for network data processing with topological structure.It has become a current research topic to detect and recovery outlier more efficiently based on graph signal processing in WSN.This thesis first studies how to detect and label the outlier node data of WSN.In terms of the relationship between WSN topology and network data,a WSN outlier detection algorithm is proposed based on graph-spectral analysis of sub-graphs.This algorithm provides a novel way to detect outlier in WSN.The first step of the algorithm is extracting the high-frequency component of the signal using a high-pass graph filter.Secondly,the network is decomposed into a set of sub-graphs,and then the specific frequency components of the signal in sub-graphs are filtered.The third step is to label the suspected outlier center nodes of sub-graphs based on the threshold of the filtered sub-graphs signal.Finally,the outlier nodes in the network are determined by comparing the set of nodes of each sub-graph with the set of suspected outlier nodes.Experimental results show that compared with the existing outlier detection algorithms in networks,the proposed algorithm not only has a higher detection probability of outlier but also has a higher labeling rate of outlier nodes.This thesis also studies the recovery of WSN data.In terms of the characters of time-vary and network topological to recovery data,a WSN data recovery algorithm based on joint graph model analysis is proposed.Firstly,the joint graph model of network data is established based on the smoothness characteristics of the data in the spatial domain and the time domain,and then the concept of joint graph total variation is defined based on the joint graph model of network data.The data recovery is formulated as an optimization problem based on the minimization of joint graph total variation.An iterative method is proposed to solve the problem.The experimental simulations show that the data recoveryalgorithm based on joint graph model analysis has a higher recovery accuracy and a higher iteration efficiency.