Research Of Frequent Pattern Mining Technology And Its Application In Real-time Signal Processing

With the rapid development of information technology in various fields, frequent pattern mining technology has been widely applied in real life.For example in the biology, it is used for the disease prevention and treatment; in the financial industry, it is applied to prevent and avoid financial risks and in the military field, it is used to abnormal detection. Mining and analysis of time series data have become a hot research issue. Among them, frequent pattern mining is one of the basic problems. Therefore, the research of frequent pattern mining in the data stream is of higher significance.In this thesis, the application of frequent pattern mining technology in real time signal data is studied, it hopes to improve the quality of radar signal data by mining the frequent patterns of real time signal data, to discover intrusion behavior and to provide reliable information for equipment fault diagnosis and auxiliary military reconnaissance.Based on frequent pattern mining in time series data, in this thesis, the method of time sequence data preprocessing is studied and the existing maximal frequent pattern mining algorithms and closed frequent pattern mining algorithms are improved. Primary work includes:1. Research frequent pattern mining algorithms for static data sets. Based on pattern growth of the Prefixspan algorithm, an improved S-Prefixspan algorithm for mining maximal frequent patterns is proposed.Based on the Prefixspan algorithm, the proposed algorithm is improved by two points.The first point,given the existing Prefixspan algorithm in the process of mining needs to scan twice the database and will produce a large number of projection database to consume memory, a chain storage structure which is based on the idea of bitmap mapping is proposed. The position of the frequent items in the sequence is stored in the data structure. Through the data structure to reduce the construction of the projection database and needs scan only one time. It will save memory and improve the running efficiency. The second point, to enhance the effectiveness of search, in the process of generating frequent patterns, by the efficient pruning operation, it can further improve the search efficiency.2.Research frequent Pattern Mining in stream data. Based on the Moment algorithm, an improved OS-Moment algorithm for mining closed frequent patterns is proposed. The improved algorithm is mainly aimed at a large search space for theMoment algorithm in the process of mining, many useless results in the middle of the process, can only dig out the disorder sequence and other issues related to improvement.The improved algorithm uses the binary bits to represent the degree of support for each item, which is convenient to calculate the item set, and improves the operation efficiency. The sequence information of the maintenance item of the chain storage structure is designed to solve the shortcomings of the original algorithm; a new storage structure of index pattern tree is proposed to store closed frequent item sets, which can speed up the query of the result and the update of the node information. At the same time, in the mining process by reasonable pruning strategies it will avoid the generation of a lot of useless results,and to further improve the efficiency of the algorithm.3. In the radar real time signal data, the related experiments are done to analyze the improvement of the effectiveness of improved algorithm in time and space.