Sequential Pattern Mining Of Moving Objects Trajectories

As the GPS(Global Positioning System) has become more and more widely used, in consideration of using moving objects to locate tracking, it is of significant value to analysis the behavior trajectory of moving objects gained by GPS equipment. As an important method of data mining, sequence pattern mining has been paid much attention to.The data set used in this experiment comes from the taxi mobile data from San Francisco America in May 2008, which was distributed mainly in the San Francisco bay area, containing about 536 taxi GPS coordinate data.Firstly, due to the movement about the two aspects of time and space trajectory information, GPS data has lots characteristics, such as, large volume of data, periodic characteristics, serious noise interference, missing data and so on. The method how to effectively preprocess the trajectory data and gain the useful information, will directly affect the result of sequence pattern mining and the correctness of the matching predictions. This paper will introduce methods in particular, such as, data interpolation, data clustering, data normalization, feature representation and grey association analysis. Besides, by optimizing the track points, the detection and combination methods used for GPS trajectory make it more efficiently for feature extraction and generate reliable trajectory sequence.Secondly, sequential pattern mining algorithm is mainly divided into two categories. One is typical representative, such as AprioriAll algorithm, the GSP algorithm and SPADE algorithm, which are based on the basic ideal Apriori algorithm. And the other one is based on the optimization of pattern growth, including Freespan algorithm, Prefixspan algorithm, etc. This paper learns the advantages of the growth mode of the latter model which does not produce the candidate sequences, and at the same time optimizes algorithm through effectively pruning the candidate sequences to effectively improve the efficiency and correctness of pattern mining algorithm in the end.Finally, in the face of the problem about the inaccurate matching of the sequence pattern, FreSeqMatching algorithm is proposed in this paper. By using the definitions and concepts about sequence class and sequence focus, combining the pattern mining algorithms with the timely feedback of the matching results and the timely adjustments of support conditions, the method effectively improves the accuracy of the sequence pattern matching and the prediction accuracy of moving objects track.