Research On Data Cleaning For Non-deterministic RFID Data Stream

In recent years, with the gradually maturity of RFID technology, RFID (radio frequency identification) has gain much more attention for its inherent characteristics like simplest and fast speed, and has been widely used in many areas, such as intellectualized mansion, supply chain, asset tracing and health care. However, the inaccurate of raw RFID data restricts the further development of RFID technology, in order to provide high-quality RFID data, the cleaning of RFID raw data is necessary. RFID non-deterministic data including: negative reading, active reading and redundant reading.This paper studied the three kinds of RFID non-deterministic data, and then proposed two cleaning method for different application environments, the work are as follows:For the RFID application that does not existed complex relationship, we proposed an adaptive sliding window cleaning structure-VSBFC, the structure is divided into three layers. In the window adjustment layer, proposed an adaptive sliding window based on data velocity aiming at the non-uniform speed of RFID data stream, in the process of window size adjusted, taking into account the velocity of data stream and the average reading rate of data in window. In the Redundant reading processing layer, proposed a redundant reading cleaning method based on pseudo-events that could effectively remove the redundant data, In the negative reading and active reading processing layer, proposed a one-time cleaning method for negative reading and active reading which significantly reduced the middle data.For the RFID application that existed complex relationship, we proposed a complex relationship based date cleaning structure-CRRC, the structure is divided into four layers. In the dynamic graph construction layer, we proposed a method to encode complex relationship into dynamic graph, and introduce the update algorithm of graph. In the uncertainty reasoning layer, proposed a negative reading cleaning approach based on contain-relationship and adjacent-relationship, In the graph trim layer, proposed an active reading cleaning approach based on the connected component of graph. In the composite data generate layer, introduced six kinds of component data, and gives the generation method of composite data. Experimental results show that the proposed two cleaning methods applied in their respective areas both have good performance.