Research On Several Data Mining Algorithms For Massive RFID Data

Recently, the Radio Frequency Identification technology is developing fast, the prices of readers and tags are decreasing greatly, and the accuracy of reading is increasing obviously. With the advent of RFID technology, manufactures, distributors, and retailers will be able to track the movement of individual object throughout the supply chain. Large retailers like Wal-mart, Target, and Albertsons have already begun implementing RFID systems in their warehouses and distribution centers, and they are requiring their suppliers to attach RFID tags to products at the pallet and case levels. With the price level going down, people can expect tags to be placed at the individual item level for many products. The main challenge then becomes how can companies handle and interpret the enormous volume of data that an RFID application will generate. The mass data have been stored in the database or data warehouse. In the face of this "data explosion" situation, how to extract valuable information from massive data has become particularly important. With the emergence and development of data mining techniques, this problem has been solved by people. The potential and useful information and knowledge are extracted from the massive, incomplete, noise, fuzzy and random practical data by data mining techniques and their analytical tools.The RFID technology is used for tacking the moving items in supply chain, so the most important data in RFID system is the data of moving, called path data or track data. The main task for RFID data mining is mining frequency paths. In this paper, we??earch the path data, and propose some methods for mining high frequency path in database. Base on these frequency paths, the users can understand the moving trends, optimize the supply chain, and find out abnormal moving. The main research content includes: frequency paths mining, frequency multi-dimensional paths mining, and distributed frequency paths mining. The main contributions and innovations of this dissertation are as follows:1) The methods for mining frequency patterns and sequential patterns can't mine frequency paths with high efficiency. In this paper, we divide the path data into several sequences and based on existing frequency patterns and sequential patterns mining methods, we propose high performance methods mining frequency paths.2) Base on frequency paths mining methods proposed in this paper and multi-dimensional sequential patterns mining methods, we propose two strategies, one is embed multi-dimensional information into path then mine them as a whole; the other is combine iceberg cubing and frequency paths mining. These strategies can be used in different situations.3) We proposed distributed frequency paths mining method. The supply chain is distributed in different places, so the RFID system is also distributed. If the system gathers data into one serve and mines, the volume of data transmit through network is massive and the capabilities of other serves are wasted. In this paper, we use all serves mine frequency paths and store the result in lexicographic path tree, and transmit these trees among different serves. Basing on the regular and simple sequences of merged trees, we will have global frequency paths. This method not only decreases the transmit volume of data, but also uses all serve capabilities.4) The RFID system keeps on generating data, and gives true data mining results to users, the system must update its data mining results. We propose an update method to solve this task. The volume of paths data are massive, so we use path coding to compress the data.