The Study Of Key Technologies For Uncertain RFID Stream Data Management

With the development of RFID technology and reduction of RFID system cost, RFID has been applied into more and more fields, such as the supply chain management and smart monitoring. It is even expected to have great impacts on humans’ production and life in the future. How to manage and analyze the real-time RFID data efficiently, is one of the most important issues for determining the prospects of RFID applications, and thus arouses extensive attention in academia as well as industry. RFID readers will collect large amounts of RFID readings with time, location and status information of monitored objects. The data are generated in the form of streams, with the features of temporality, rich semantic, heterogeneity and huge amounts. Especially, recent studies indicate RFID data have typical uncertainty. How to model and manage data considering the uncertainty is the hot issue in the RFID data management area.This dissertation concludes the uncertain phenomena and issues of RFID data management and analyzes the state of arts about RFID uncertain data management. We classify RFID uncertainty into subjective uncertainty and objective uncertainty. On account of the uncertainty, novel models and methods are proposed for data pre-processing, complex event processing and spatio-temporal query, and thus a distinctive process framework is built. These techniques can efficiently improve the capability and robustness of RFID data management, and furthermore support RFID applications in complex circumstances.Specifically, this dissertation conducts in-depth studies into several RFID stream data management techniques, including RFID physical data cleaning, logic trajectory imputation, complex event processing and scheduling, probabilistic spatial range query and probable k-NN query. The key problems about RFID data pre-processing, complex event processing and spatio-temporal query are covered. Specifically, the major work includes:(1) For RFID physical data cleaning, considering RFID physical data uncertainty, we propose to clean data based on the correlations of RFID object group members. Efficient correlation model is built and adaptive incremental maintenance strategies are proposed by compressing the original graph model. The proposed model is more suitable in the smart monitoring scenarios where the concept of group is involved. Compared to the purely window-smoothing techniques, our technique will gain better accuracy when groups exist.(2) For RFID logic data imputation, we propose an efficient trajectory imputation strategy. By constructing the probabilistic trajectory event model, the aware regions can be interpolated efficiently. Because the imputation granularity is enhanced from physical data to the logic layer of aware region trajectory points, better efficiency can be gained. Furthermore, utilizing the region missing ratios and the trajectory statistics for different monitored objects, various RFID monitoring object trajectory imputation strategies are proposed. Also, taking stay interval into account, we propose the improved methods to enhance the imputation accuracy. This framework is suitable for the scenarios where the trajectory regulation of monitored objects can be captured.(3) For RFID data complex event processing and scheduling, we model the RFID event arrival probabilistically and propose the unified consumption-mode-aware complex event processing automata model. Furthermore, by considering deadline satisfying model and processing-disordering-oriented resource optimization allocation model, deadline-constrained complex event processing strategy and scheduling method are proposed. Compared to the best-effort manner, for the applications with fixed real-time requirement, higher deadline satisfying can be gained in general.(4) For RFID spatial range query over monitored objects, considering the location information uncertainty, we look into the probabilistic moving spatial range query techniques. The monitored objects’ locations in the regions or on the paths are modeled probabilistically and the evaluation methods for the range query are proposed. By introducing the concept of triggered moving range query, several incremental maintenance strategies are suggested, and thus improve the query efficiency and accuracy in different scenarios.(5) For RFID k-NN query over monitored objects, we study the probable k-NN query techniques in RFID applications deeply. RFID-based semi-constrained space model is built and corresponding distance evaluation methods are proposed. Furthermore, we propose three k-NN query estimation models, and design efficient indexes to improve the query efficiency.In conclusion, this dissertation aims at the specific features and challenges of RFID data management and studies the key technologies of RFID data management, covering the techniques of data pre-processing, complex event processing and spatio-temporal query. And thus, efficient and robust RFID data management performance can be offered for real-time RFID data stream applications.