Trajectory Pattern Mining And Its Application In Vehicle Trajectory Prediction

In recent years,the trajectory data mining and knowledge discovery has become a hot research,and are widely used in intelligent transportation,personalized recommendation services and other areas.More and more researchers recognize that research about the regulation of vehicles is significant.Mining the historical trajectory pattern of the vehicle can analyze the movement and behavioral patterns of the vehicle,so the vehicle travel path can be predicted and this will provide more valuable and accurate personalized service for drivers.Therefore,mining vehicle trajectory pattern has practical significance and application value.Frequent pattern mining algorithm is an important data mining algorithms.Frequent pattern mining is widely used in mining transaction database and many other types of databases.However,the traditional Frequent-pattern Growth algorithm(referred to FP-Growth algorithm)is useless when dealing with large-scale data.For this weakness,this paper take advantage of the merit of Hadoop,which is parallel computing,distributed storage,fault-tolerant processing mechanism and high scalability,to improve and deserialize FP-Growth algorithm.The new algorithm uses Hadoop framework to achieve this goal..This paper has mainly completed the following three jobs.First,for original FP-Growth algorithm's weakness that it repetitively scans the database and generate a lot of conditional pattern bases and conditional pattern tree,the paper presents support-count sheets and pruning strategies to reduce the time-consuming and also reduce the number of scans of non-frequent items,thereby improving the efficiency of the serial algorithm;secondly,deserialize the FP-Growth algorithm to deal with the limitations in processing large-scale data by using Map-Reduce programming model.This also extends computing power to solve the problem that the algorithm occupies too much memory space;and finally,use the parallel MR-FPG algorithm to mine frequent patterns from traffic trajectories.The result of experiments shows that the efficiency of parallel algorithm is better than that of original algorithm.