Research On Prediction Of Urban Road Speed And Estimation Of Travel Time Based On Deep Learning

The development of national economy and social productivity runs through all aspects of people’s life.The development and use of transportation tools is the inevitable trend of human civilization and development,and the fundamental driving force lies in the pursuit of a better and convenient life.When every household is immersed in the pleasure of buying and using cars,the social impact can not be ignored.The rapid increase of traffic flow leads to frequent road congestion.Road traffic is facing severe challenges on how to ensure safety and improve traffic efficiency at the same time.The development of intelligent transportation system is urgent.Traffic road speed prediction refers to the prediction of road speed in the future based on road history or real-time information.As an important part of intelligent transportation system,it is crucial for route planning,travel time estimation,navigation and other applications.Although there are many methods to try to solve this problem,most of the road speed prediction methods regard the road network as a static graph,ignoring that the relationship between roads will change with the traffic conditions;At the same time,combined with the road network structure information and the massive data generated by vehicle operation,it can bring great help for the taxi order travel time estimation.However,most travel time estimation algorithms only regard the road as the only element of the road network,ignoring the intersection with complex traffic conditions.Based on deep learning method,this paper studies road speed prediction task and travel time estimation task in intelligent transportation system.The main research contents include:1.A traffic road speed prediction algorithm based on graph attention network.The algorithm first uses the recurrent neural network to learn the road condition information of the current road,then models the short-term and long-term road condition portraits of the neighbor roads,and then uses the graph attention network to weigh the influence of the neighbor roads to predict the speed of the target road section.The experimental results on a real Beijing dataset show that the proposed model performs better than the selected comparison algorithms in the road speed prediction task.2.A travel time estimation algorithm based on road network structure relationship.According to the real traffic network graph,this algorithm proposes a representation method of road network structure relationship,learns road representation and intersection representation of hidden direction,and integrates traffic flow information that can represent short-term and long-term road conditions in the calculation of road section travel time.Finally,the model uses attention mechanism instead of simple sum mode to calculate the time of vehicles passing through each road section and intersection.The experimental results on the taxi order datasets of Hefei,Anhui and Chengdu,Sichuan show that the proposed model is superior to the selected comparison algorithm.