Research On Worm Nodes Detection And Outliers Detection In Wireless Sensor Network

With the continuous development of Internet of Things and wireless communication technology,Wireless Sensor Networks(WSNs)are widely used in data collection and analysis in the real world.WSN is a typical ad hoc network which posesses a large number of sensor nodes and base stations.Sensor nodes are responsible for collecting data measurments and communicating with base stations,they are small and energy-limited.WSN needs realtime data measurments to make decisions,However,due to the deployment of harsh enviroment,the limited energy,there exist a series of network security problems in WSN.The security problems of wireless sensor network are mainly divided into two parts.On the one hand,how to identify and remove malicious nodes in the network and ensure the right control of the network is particularly important.Due to the deployment of harsh environment,sensor nodes are easily compromised by attackers.Up util now,schemes for mobile compromised node and replica node attacks in WSN are very mature,but few researches on worm compromised nodes have been taken.Worm attacks have the greatest damage to wireless sensor network,they could infect sensor nodes and paralyze the network quickly.On the other hand,in order for the base station to make an accurate judgment,it is important for the sensor node to transmit reliable data.Therefore,it is necessary to detect and eliminate outliers caused by noise or internal errors of sensor nodes in the network.However,due to the limited computing resources and memory resources of sensor nodes,the proposed detection schemes cannot achieve a compromise between resources and detection efficiency.In view of the above two points,this paper has carried on the related research.Based on the characteristics of worm propagation and sequential probability ratio test,a SPRT-Biased-Random worm detection method is proposed combining with biased sampling and random value sampling to accelerate worm node detection in WSN.Experimental results show that SPRT-Biased-Random method can efficiently detect worm propagation in WSN and find all worm nodes in 5 to 18 time slots,so that the number of nodes eventually infected in the sensor network remains between 2 % and 5 %.Based on the PCA algorithm and Mahalanobis distance,an Improved Distributed PCA-Based Outlier Detection Method(IDPCA)is proposed for the outlier detection in WSN.By exploiting the double detection mechanism,IDPCA greatly improves the detection performance of outliers and reduces the communication overhead in the network.Experimental results show that IDPCA can achieve an outlier detection rate of 96 %-97 % and maintain a false alarm rate about 2 %.Moreover,the method can trace the source of outliers and enable network managers to locate network problems quickly.In order to solve the problem that IDPCA scheme can not be well applied to non-linear data measurements detection,this paper proposes an improved distributed kernel principal components analysis based on Mahalanobis kernel function,which performs well in detecting non-linear ouliers by mapping the data measuments to a high-dimensional space.The Mahalanobis distance is proposed to measure the similarity of data vectors,by leveraging the correlation among the measured data vectors and training a proper kernel parameter,about 98% outlier detection rate and 2% false alarm rate can be achieved by IDKPCA model.