Study Of Urban Traffic Congestion Areas Prediction By Combining Knowledge Graph And Deep Learning

In large and medium-sized cities around the world,traffic congestion has been becoming more and more serious,which causes obviously increases of both travel time delays and vehicle emissions.The intelligent transportation system provides an effective way for relieving traffic congestion and improving traffic private safety.Urban traffic congestion predict is one of key techniques for intelligent transportation system.Based on traffic big data,this work studies urban traffic congestion area prediction methods.The main work of this dissertation is as follows:1.Map matching algorithm researchFor the study of urban traffic congestion area prediction method,it is necessary to use the combination of bus trajectory data and link speed data to characterize the label of the model,that is whether the congestion occurs.We need to use the map matching algorithm to process the trajectory data to match the trajectory to the corresponding road segment.In order to improve the accuracy of matching,a random forest multi-classification strategy is introduced and the matching problem is treated as a classification problem,which can make full use of massive historical trajectory data.In order to further accelerate the matching speed to meet the requirement of real-time matching,this paper uses the distributed processing mechanism and then proposes a method based on distributed random forest multi-classification DRFMM to deal with the map matching problem of trajectory data.The real Hefei road network and taxi t:rajectory data were utilized to valid our method.Compared with the classical point-line matching algorithm and neural network classification method,the proposed map matching algorithm based on distributed random forest multi-classification DRFMM effectively improves the matching accuracy and speed.2.Urban traffic congestion area prediction modelThis paper proposes a model that combines knowledge graph and spatiotemporal convolutional neural network(KG-ST-CNhN)to collaboratively predict urban congestion areas.Specifically,by discretizing and semanticizing multi-source heterogeneous urban traffic big data,we construct urban knowledge graph and graph convolution network is introduced to further extract features from urban knowledge graph,which are treated as the input of spatiotemporal convolution neural network.The method regards the whole traffic network as an image to effectively capture the spatial correlation between the regions,and by analyzing the temporal correlation of the traffic data,we construct multiple convolution networks to capture the temporal features from different time periods.Therefore,urban traffic congestion areas were accurately predicted.After that,in the real traffic data of Beijing,compared with the other four prediction models,the validity of our proposed KG-ST-CNN prediction model is verified.This method effectively improves the accuracy and recall of congestion prediction.3.Design and implementation of urban traffic congestion areas prediction systemBased on the distributed random forest multi-classification matching algorithm and the urban congestion areas prediction model combined knowledge graph and spatiotemporal convolutional neural network proposed in this paper,the congestion prediction system is designed and implemented as a subsystem of urban traffic intelligent control system.The algorithm and model optimize geolocation service in the service layer and decision analysis module in the application layer.