Crutial Techniques For Group Behavior Pattern Mining And Analysing On Big Data Of Temporal-Spatial Trajectory

Mining and analyzing temporal-spatial trajectory big data are popular in many applications at present. The key problems include the dynamic semantic fusion for heterogeneous temporal-spatial trajectories in big data, the fast mining computation on both total and incremental dis-tributed temporal-spatial trajectory big data in WAN. Considering the requirements in the public security field, this dissertation focuses on the mining and analyzing of group behavior patterns based on temporal-spatial trajectory big data. The main contributions of this dissertation are:(1) The semantic fusion on heterogeneous centralized big data in LAN. It is difficult to set the public semantic reference model in advance for heterogeneous centralized temporal-spatial trajectories from big data in LAN, the PACO (Parallel Ant Colony Optimization) method is proposed by adapting the ACO (Ant Colony Optimization) algorithm based on the MapReduce framework. With the execution of critical clustering phases and adaptively clustering in parallel, the public semantic reference model is generated automatically and the clustering is implement-ed fast.(2) The semantic fusion on heterogeneous distributed big data in WAN. For a dynamic vari-ety of data sources and the applications of group behavior pattern mining, a distributed parallel computation framework DPF (Distributed & Parallel Frame) is proposed based on MapReduce. In terms of DPF, the DPKM (Distributed & Parallel kmeans) method is developed by adapt-ing from the kmeans algorithm for the dynamic semantic fusion on heterogeneous distributed temporal-spatial trajectory mata-data in WAN. With distributed computation, the migrating of temporal-spatial trajectory big data is avoided, which decreases the total computation time sig-nificantly and improves the efficiency.(3) Mining group behavior patterns on temporal-spatial trajectory big data. Considering the typical features like large amount, distributed storage and high migrating costs of the temporal-spatial trajectory big data in WAN, DPACO (Distributed & Parallel Ant Colony Optimization) method is adapted from the ACO Algorithm based on DPF. DPACO is applied to the group behavior pattern mining problem on totally distributed temporal-spatial trajectory big data in WAN. Avoiding the migrating of big data decreases the total computation time significantly. By applying clustering on total temporal-spatial trajectory big data, the consequences caused by data sampling and dimension reducing are avoided, which keeps the accuracy of clustering.(4) Mining group behavior pattern on incremental temporal-spatial trajectory big data. S-ince temporal-spatial trajectory big data is not only big in total amount but also in incremental amount of a certain period, DPIACO (Distributed & Parallel & Incremental Ant Colony Op-timization) is developed from ACO based on DPF. The clustering process is divided into the historical total phase and the several period incremental phase, which are executed continuous-ly. DPIACO is applied to the group behavior pattern mining problem on incremental distributed temporal-spatial trajectory big data in WAN. By fixing the existing results of the periodic incre- mental clustering and realizing the parallel clustering phases in WAN, the repetition of clustering and copy migrating of big data in WAN are avoided, which improve the computation efficiency and keep the accuracy of clustering.