Research On Data Anomaly Detection And Data Quality In Wireless Sensor Networks

A wireless sensor network (WSN) is a self-organizing network which is composed of micro nodes with capabilities in sensing, data processing, storage and communication, etc.. WSNs have gained considerable attentions from research community, industrial sectors and government agencies in recent years. WSNs have a widespread use in areas like military, natural disaster monitoring, digital healthcare, structural monitoring, home monitoring, etc..The characteristic of data centric is more and more striking with the increased number of real deployments of WSNs. To be a successful application, it is critical for a WSN to acquire data from deployment environment and find valuable information out of the data. For this purpose two categories of problems should be focused and solved, that is:how to design effective anomaly detection mechanisms to find anomaly information or anomaly status out of the data; how to design the evaluation system and metrics for data quality so as to track the data quality from the sensing module, which is the origin of the data. Besides, all kinds of design constrains, especially energy limitation, should be considered when we solve these two categories of problems.With the above-mentioned background, this dissertation focuses on data anomaly detection and data quality issues and carries out some in-depth research. The main contents and achievements are as follows:●We propose a distributed fault detection scheme for wireless sensor networks. Faulty sensor nodes are identified based on the similarity measure of neighboring nodes and dissemination of the decision made at each node. This scheme makes full use of temporal and spatial correlations among neighboring sensor nodes, and it can find some kinds of typical sensor data faults efficiently. Based on the result of simulation, we discuss the hybrid detection scheme which combines two or more simple detection algorithms to improve the overall performance. ●We propose a linear pattern based unspecific event detection scheme for wireless sensor networks. This scheme combines the merits of threshold based and pattern based event detection mechanisms, and is optimized for low cost, low computing capability node which also has the limitation of energy consumption per unit time. We show how this scheme outperforms threshold based detection scheme in a continuously changing environment by simulation.●We research on the data quality issue in WSN. We combine the achievements from data quality and WSN research communities, analyze the quality requirement of WSN, formulate the relationship between data quality and QoS in WSN, and define a set of data quality dimensions that should be concerned. Data correction dimension and timeliness dimension and their influences on specific WSN applications are discussed as two case studies. In the case study for data correction dimension, an algorithm is proposed for efficient gross error removal based on Dixon test. In the case study for timeliness dimension, data jitter and its influence on fault node detection are discussed.