Research On The Methods Of Outlier Detection In Wireless Sensor Networks

Wireless sensor networks are composed of multiple sensor nodes with data processing,storage and communication functions,which are self-organized and multi-hop.It is a cutting edge research hotspot that involves multi-disciplinary and highly integrated information in the way of sensing and processing the wireless network of monitoring object information in the network coverage mode.At present,WSNs have been widely used in industrial and agricultural control,military,environmental monitoring and intelligent city and other fields.The main purpose of wireless sensor network deployment in the real environment is to obtain the required data information and transmit it to the network owner,so that it can make the corresponding decision through the data information.However,the sensor node equipment is prone to damage,lack of energy or exhaustion and other failures,as well as deployed in hostile circumstances are manipulated and many other factors,will cause WSNs sensing data outlier,so that the quality of sensing data degradation,which inevitably lead decision makers to make wrong judgments.Therefore,it is of great scientific research and application value to study the outlier detection method of sensing data in wireless sensor networks.Resource limitation is a special feature of wireless sensor networks,such as energy,storage,computing power,and communication bandwidth.These characteristics lead to data dynamic,data correlation,the acquisition of known tag data,the distinction of outlier data,the way of outlier detection,the computational complexity and the like are different from the general network,so the conventional outlier detection method Can not be directly applied to wireless sensor networks.In this paper,we focus on the key issues such as accuracy,computational complexity and communication complexity of WSNs.The main research contents and innovations are as follows:1.Aiming at the problem that the sensor data of a single sensor node in a wireless sensor network is fixed due to a fixed fault,random noise failure,gain or offset fault.The temporal and spatial correlation of node sensing data in wireless sensor networks and the relationship between the sensing data of neighbors and the distance between them are studied.An improved outlier detection method based on distance weighting is proposed.The distance weighted value algorithm between node and neighbor node sensing data is constructed by Euclidean distance,which improves the accuracy of outlier detection and reduces the false positive rate of normal data.2.For the case where the outlier data is generally less than the normal data sample and is seriously asymmetric,and the characteristics of the wireless sensor network itself are limited.An outlier detection method for wireless sensor networks based on support vector data description(SVDD)is studied.However,the SVDD method has a high computational complexity problem,so a method to reduce the computational complexity of SVDD training stage and decision stage is proposed.First,the computational complexity of the training phase is reduced by using sequential minimal optimization(SMO)algorithm of training set reduction and second order approximation.Then,by analyzing the expression of decision function,the approximate complexity of SVDD hypergonal sphere in the original feature space is obtained,which effectively reduces the computational complexity of decision stage.Finally,the sensing data of the real wireless sensor network is verified.The results show that this method can effectively reduce the computational complexity and ensure the accuracy of outlier detection in wireless sensor networks.3.The independent and identically distributed characteristics of the sensing data in the wireless sensor networks with limited resources lead to the fact that the outliers may occur independently on a single attribute.Combining the spatiotemporal correlation of the sensing data,an outlier detection method based on spatiotemporal and attribute correlation SVDD(STASVDD)for wireless sensor networks is studied.In order to reduce the computational complexity of the method,a lightweight outlier detection method is proposed to optimize the quadratic programming problem in STASVDD by using the core set of ideas.At the same time,the distributed approach for outlier detection to reduce the wireless sensor network communication complexity.Finally,the real wireless sensor network dataset validate the method of detection accuracy guaranteed under the premise of computational complexity and communication complexity is effectively reduced.4.According to the needs of practical application,the application of outlier detection method of wireless sensor network in industrial monitoring system is studied.Firstly,according to the needs of industrial enterprises and the analysis of key technologies,hardware devices such as wireless sensor network nodes and gateways are designed,and the embedded software of data acquisition is realized.The wireless sensor network monitoring system of industrial enterprises is constructed.Then,three kinds of outlier detection methods proposed in this paper are validated by using the sensing data obtained by the system.