基于多模态数据流的无线传感器网络异常检测方法

An Anomaly Detection Method of Wireless Sensor Network Based on Multi-Modals Data Stream

伴随着无线通信技术的不断发展和广泛应用,信息物理融合系统(Cyber-Physical System,CPS)作为物联网领域的最新研究方向成为近年来研究者广泛关注的热点.无线传感器网络(Wireless Sensor Networks,WSN)作为CPS系统物理空间的主要感知网络,若有效提高对应感知数据的准确性和可靠性,可及时准确地发现突发事件、监测网络工作状况,因此对传感器网络节点数据流进行异常检测,发现其中的异常数据并确认其来源具有重要意义.该文在无线传感器网络多模态数据流研究的基础之上,提出了一种对传感器异常数据进行检测以及监测节点自身工作状态的方法,该方法不仅应用了无线传感器网络中的时空相关性原理,还更进一步,研究了同一节点中多模态数据流之间的相干性,并以此作为理论基础,利用多维数据和滑动窗口模型对异常数据及其来源进行检测和验证.该文的方法可以分为3个步骤:首先,利用滑动窗口中的历史数据对传感器数据流进行异常数据的检测;其次,利用节点的空间相关性对异常的来源进行确认和识别;最后,对由于测量误差导致的异常值进行筛选,使输入CPS的数据进一步的精确化.通过实验对比,该文的方法对传感器异常数据的检测率保持在95%;在不同数据维度的条件下,对四维数据集的检测率比单维数据集提高了3%.

With rapid development of the wireless communication technology,cyber physical system（CPS）that is a significant research orientation in internet of things has become a hot topic recently.As the main sensing network of physical space for CPS,improving the data accuracy and reliability of wireless sensor network（WSN）can effectively recognize emergent events and monitoring the situation of networks.Therefore,it is significant to detect abnormal data and identify their source.In order to accomplish this object,a novel anomaly detection and node-self monitoring method of wireless sensor networks based on multi-modals data stream that takes spatial-temporal correlation of different sensor nodes along with association of multi-modals data properties into account is proposed.Thus,abnormal data in streams can be detected while the source of anomalycould be verified effectively based on multi-dimension data model and sliding window.The method proposed in this paper is divided into three steps：Firstly,abnormal data in sensor stream can be detected by historical relativity of data sets based on sliding window.Then,it is essential to identify the source of abnormal data and verify what makes the anomaly by spatial correlation of sensor nodes.Finally,the anomaly resulted from measurement error should be filtered and data set could be further sanitized into CPS.Experiment results demonstrated that the proposed method can obtain 95% accuracy of the anomaly detection while the accuracy of anomaly detection rate in four-dimensional data stream is 3% higher than that of the single dimension data stream for different dimension data.