Study On Data Preprocessing Techniques In Rfid Complex Applications

As a new technology integrated with signal processing, wireless communication, embedded calculation and data management, RFID technology is being widely used in more and more areas, such as supply chain management, object tracking, quick disbursement and so on. However, RFID technology adopts wireless radio frequency signal to communicate, which is easily interfered with environment, so there are many missed readings, erroneous readings, duplicates and data out of order in time when collecting data in RFID applications, which influences the accuracy of query results for event detection badly and limits the development of RFID applications. Therefore, the preprocessing over RFID data is the prerequisite of assuring high quality of query results.For the goal of solving the issues proposed above, this paper focuses the preprocessing strategy over "dirty" data generated in RFID applications.Firstly, on the basis of triple tuple over RFID data, the paper proposes a data abstraction algorithm which transforms RFID data from data level to logic area level. This algorithm is used to compress data where lots of redundant data are deleted and some missed readings are considered. After that, a tuple may be considered as a simple event. Experimental results show that through abstraction the amount of data is extremely cut down. In this way, system resource is greatly saved for further data cleaning.Secondly, in order to solve the missed reading problem-the main type of "dirty" data in RFID applications, this thesis proposes three interpolating algorithms based on data abstraction, namely rapacity algorithm, mink-similar algorithm and allk-similar algorithm. Above all, a dynamic probabilistic event model is established by statistically studying arriving events and computing the missing rate of each logic area. Then, on the basis of this model, missed events are interpolated by searching their most similar events using different searching strategies. These three algorithms increase data accuracy largely, and eliminate the influence of erroneous data to query quality. Theoretical analysis and abundant experiments prove the effectiveness and efficiency of proposed data interpolating algorithms.Lastly, this thesis improves the above interpolating algorithms by adding the factor of time. It mainly develops probabilistic event model by introducing temporal model, and thus two improved strategies of original interpolating algorithms are proposed, namelyβ* improved algorithm andβ+ improved algorithm.β* improved algorithm adopts histogram graph distribution to estimate time, andβ+ improved algorithm adopts Euclidean distance to estimate time. In different cases, these two algorithms behave well separately. Experiments show that improved data interpolating algorithms have the superiority on accuracy of processing results.