Trajectory Knowledge Graph Construction And Destination Prediction

Nowadays,the accumulation of trajectory data provides an opportunity to explore the movement patterns of vehicles in cities.The trajectory data contains various different spatial objects,such as origin,destination,the intersection it passes through,and so on;And as time series data,trajectories contain both temporal and spatial information of trajectory points.In addition,the road network structure and weather conditions as auxiliary factors can also have an impact on the analysis of trajectory data.How to organize and integrate these multi-source heterogeneous data to analysis and mining of trajectory data,and apply it to real-world scenarios is significant.Such as traffic guidance,location prediction and recommendation,social movement perception and behavior analysis.Knowledge graph contains ontologies,attributes,and relationships between ontology.As a modeling method,knowledge graph use graph to describe the relationship between knowledge and entities in the world.It can effectively organize and integrate multi-source heterogeneous data,making it more convenient to serve various data analysis and mining tasks.Therefore,this thesis constructs a trajectory knowledge graph based on trajectory data,road network and weather auxiliary information,then predict destination based on trajectory knowledge graph.Accurate and efficient destination prediction plays an important role in personalized advertising services,public security management and urban transportation intelligent planning.The main work of this article is as follows:(1)Construct trajectory knowledge graphs using bottom-up and top-down methods respectively.First,three parts(trajectory,lane,intersection)in the trajectory data are stored in three tables of the relational database;Then,data extraction is carried out through mapping rules,transforming the two-dimensional table data into triplet data of the knowledge graph,fused with the triplet formed by road network and weather data to form an ontology structure.It was observed that the knowledge graph constructed from bottom to top showed a star shaped structure centered on the trajectory,while the correlation between lanes and intersections within the trajectory was not reflected.Therefore,through the top-down method to construct trajectory knowledge.According to the domain knowledge,the ontology structure of the trajectory knowledge graph is determined.Proposed four spatial ontologies(i.e.,origin,lane,intersection and destination),ontology attributes,and relationship between ontologies.Then,organizing and integrating trajectory data,road network data,and weather data through the ontology structure,trajectory knowledge is constructed.Finally,the trajectory knowledge graphs constructed by the two methods are stored in the graph database Neo4 j for the convenience of data extraction and application.(2)Trajectory knowledge graph-based destination prediction(TOP)method.Since the four spatial entity in the top-down trajectory knowledge graph can express the trajectory semantic characteristics of the trajectory from origin to the destination;In addition,the relationship between spatial entity implies the topological structure of the road network.Therefore,based on the knowledge graph constructed from top to bottom,a new destination prediction method,trajectory knowledge graph-based destination prediction(TOP)is proposed.Specifically,the relational graph convolutional network(R-GCN)is used on the trajectory knowledge graph to obtain the embedding representation of the spatial entities;trajectories are transformed into embedding trajectories utilizing the embedding representation of the spatial entities;GRU is employed to learn the temporal features on the embedded trajectories;finally,external metadata,such as workdays,holidays,and departure time are fused together to predict the destination.A large number of experiments were conducted on the real dataset to verify the effectiveness of the proposed method.