Accompany Vehicle Discovery Algorithm And Research Based On Big Data Of Traffic Trajectory

In the field of intelligent transportation,there are two types of real-time traffic big data: one is Automatic Number Plate Recognition(ANPR)data and the other is Global Positioning Satellite(GPS)data,and mining the accompanying relationship of vehicles based on these two types of data has been a hot topic.Because intelligent transportation applications require extremely high real-time,and the development of social media makes all kinds of text information flooded in the traffic network,how to efficiently use the traffic big data and related media information to discover the accompanying vehicles from it becomes a challenge.In order to solve the related problems,for ANPR data,this thesis proposes an accompanying vehicle mining model based on semantic similarity of trajectories;An accompanying vehicle mining model based on a distributed stream processing framework is constructed for GPS data.The specific work is as follows:1.Aiming at the problem that spatio-temporal trajectories can’t effectively characterize vehicle intentions,Our study takes the semantics of trajectories as the entry point,integrates the spatio-temporal characteristics of vehicle trajectories,semantic features,and proposes an accompanying vehicle mining algorithm based on semantic similarity.Firstly,we propose a trajectory semantic vectorization representation method based on dynamic time slicing of trajectories based on context,fusing spatio-temporal features of trajectories and textual information.Then,a Bi-GRU model based on "trajectory pairs" is proposed to build a forward and backward sub-network by the set of trajectory pairs composed of actual and sampled trajectories,parameter transfer and sharing are realized during the training process to better track compensation.Finally,considering the sensitivity of the Attention mechanism to local features,a more accurate representation of the trajectory is obtained by using Attention to weight the key nodes affecting the shape of the trajectory,and thus effectively discovering the accompanying vehicle groups.2.A model for real-time mining of accompanying vehicles from GPS data is proposed using the Spark steaming distributed stream processing framework to address the real-time requirements of mining accompanying vehicles from GPS traffic big data.Firstly,this thesis improves the Apriori algorithm based on broad-first search,which combines the algorithm advantages of multiple angles and different angles,enhances the ability of computational performance,and enhances the set of energy items of breadth in the data.Then,the proposed algorithm is applied to the Spark distributed cluster,using Spark RDD to store data,using the Spark Streaming framework to simulate streaming data through a sliding window,and executing related advanced functions provided by Spark Streaming for distributed parallel mining processing.and accompanying vehicles are instantly discovered from GPS streaming data.The experimental results show that the method proposed in this thesis can more effectively discover local and global accompanying patterns,effectively overcome the interference of the multiple selectivity of traffic paths on the accompanying pattern mining,and improve the accuracy and immediacy of accompanying vehicle mining.The proposed model is effective in the field of companion vehicle mining.