Construction And Application Of Traffic Knowledge Graph Based On Multi-source Data Fusion

With the development of intelligent transportation,the data in the field of transportation presents explosive growth.The intelligence of transportation systems is often data-driven,so the fusion of multi-source data is the first problem to be solved.In addition,pure data-driven analysis ignores empirical knowledge and cannot achieve the desired results.Therefore,the fusion of multi-source data and the introduction of empirical knowledge are essential for building intelligent transportation systems.As an important branch of artificial intelligence technology,knowledge graph is often used to fuse multi-source data to build a large-scale knowledge base.In the area of intelligent traffic,constructing traffic knowledge graph can not only facilitate the query and statistics of traffic data,but also provide rich knowledge and diverse information for traffic analysis and prediction.Against this background,the author firstly builds a traffic knowledge graph based on the Neo4 j graph database to fuse multi-source data.Different from the traditional knowledge graph,the traffic knowledge graph contains a lot of time-related dynamic data,which makes the knowledge graph too large and affect the query speed.Therefore,the author stores these dynamic parts related to time as attributes,and the static parts are space-related elements such as roads and POIs,which can be stored as entities.In this way,it can greatly reduce the edges in the knowledge graph and improve query efficiency.On the other hand,the time and space parts can be separated,which is convenient for mining the correlation between the two during traffic prediction.After the building of traffic knowledge graph,the cross-database knowledge query will become more convenient.In order to enrich the knowledge graph,the author also carries on the rule-based reasoning on it.This thesis also studies the application of traffic knowledge graph.The author builds a spatio-temporal graph on traffic knowledge graph and uses the mining method based on the spatio-temporal graph convolution network to make traffic prediction.The present optimization of spatio-temporal graph convolution network focus on improving its network structure,so that the model can mine the information better.However,these methods ignore the improvement of spatio-temporal graph itself,which makes it difficult to continue to improve performance by optimizing the network structure.The existing spatio-temporal graphs are usually established by the distance between checkpoints and can only represent the close relationship of checkpoints in the road network.In fact,the traffic conditions between checkpoints can also be related to other information such as POI,direct adjacency,and multi-hop adjacency.Such diversified information is widely found in traffic knowledge graph.Therefore,the author uses the information in knowledge graph to establish the spatio-temporal graph,and automatically mines the correlation of checkpoints,so that the spatio-temporal graph can be more comprehensive.Through experiments on real traffic data sets,the prediction model in this thesis is superior to other existing models.This fact verifies that the application of traffic knowledge graph can improve the effect of traffic prediction.