RFID Path Data Clustering Analysis And Frequent Pattern Mining

RFID(Radio Frequency Identification)is a kind of uncontact automatic recognition anddata access technology, now has been widely used in supermarket, supply chain management,library management, car management and so on. A large amount of data generated in RFIDapplication system, but RFID path data refer to the path data generated by RFID-taggedobjects in their movement process. From large amount of data extrat usful information hasbecome especially important. It can help user understand the movement regular pattern andtrend of the objects, and also can help business arrangement, adjustment and provide usefulinformation for abnormal moving. The main research content includes:RFID path similaritymeasurement, RFID paths clustering and frequent paths mining.This paper focus on the path data genearated from simulative RFID supermart.Acquiring RFID path data from supermarket, take EPC as a key values, and by the timerelevance to integrate the paths information. To save storage space, we can compressed theorganized path data from time and location demensions.In RFID path data clustering analysis, the path similartity calculation is the foundation ofthe path clustering. This paper uses bitmap to calculate the path similartity from location andtime respectively. According to RFID path data has data stream characteristics, this paperproposes Multi-granularity Time Sliding Window Clustering(MTSWClustering) algorithm.Sliding window model is used, the clustering algorithm divide into online layer and offlinelayer on the basis of time granularity. The online layer clustering algorithm based on RFIDevevt type, and off line layer clusters are generated accoding to all micro-clusters by itssimilarity.In frequent path mining, we propose a method of mining the frequent path data thatTCFPM (Top-k Closed Frequent Path Mining)in RFID path data stream sliding windows. Wedesign TCFP-Tree to store the current data in sliding window and candidates of the closedpath. The algorithm uses a depth-first search to mining closed path. In the path mining,patternmerge and skipped the prefix model have been used to make method efficient, anddynamically adjusting the mining threshold and pruing threshold, the checking of closed pathis performed by a hash map,which leads to effective mining of top-k frequent closed path.This paper uses RFIDTango, it can simulate supermarket and generate RFID path data foralgorithm to use. We verify the effectiveness of the MTSWCluster algorithm and TCFP-TreeMining algorithm by emperiments. The experiment results show that both algorithms can savetime and sace resourses effectively.