Research On Traffic Prediction Driven By Trajectory Data

In recent years,with the rapid development of social economy and urbanization,the population and vehicles have also increased sharply,resulting in more and more serious traf-fic congestion.In order to improve transportation efficiency and alleviate various impacts caused by traffic congestion,accurate urban road-level traffic prediction is particularly im-portant.At present,large-scale trajectory data are collected and stored by enterprises.These data not only describe the real-time state of traffic network,but also contain the pattern law of traffic evolution,which provides an unprecedented opportunity for traffic condition anal-ysis.In this regard,this thesis mainly studies the traffic prediction technologies driven by trajectory data,hoping to make accurate prediction of the whole city with the help of the trajectory data.Urban traffic prediction needs to not only consider the temporal pattern,but also pay attention to the traffic evolution on the road network.At the same time,there are periodic regularity in the traffic data and traffic are vulnerable to external environmental factors such as weather,holidays and others.These complex laws and patterns cannot be described by simple rules and traditional methods.Deep learning has a high ability to characterize com-plex things,but the existing methods ignore the potential impact of road network structure and other factors.In view of these,this thesis proposes a novel deep learning model.The model considers both spatio-temporal variations and the topological structure of urban road network.It learns the characteristics of spatial and temporal traffic evolution restricted by the underlying road network in the near future.And it corrects the prediction results by periodic and environmental characteristics.In addition,for learning the potential association be-tween various variables of traffic conditions,multi-task learning mechanism is introduced.This method not only utilizes the information of relevant variables,but also improves the generalization ability of the model,so that the accuracy of the traffic prediction problem on the urban road network is further improved.In this thesis,the proposed model is experimentally verified on two real datasets.The results show that the proposed method has a good effect on urban road-level traffic predic-tion.The model proposed in this article makes certain contributions in both research aspect and practical aspect.At the same time,an urban traffic prediction prototype system driven by trajectory data is developed.