Arrival Time Prediction And Application In Intelligent Transportation System Based On Deep Learning

With the continuous increase of vehicle holdings,urban road congestion is becoming more and more common.Building an Intelligent Transportation System based on massive traffic data to improve urban traffic conditions is an urgent problem that needs to be resolved.Arrival time prediction,as an important basic task of Intelligent Transportation System,is very important for the development of Intelligent Transportation System.However,in actual use,on the one hand,the accuracy of the existing arrival time prediction methods are not high enough,and on the other hand,model safety issues(such as adversarial examples)also affect the accuracy of this task.In order to better meet the actual use needs,this paper studies the characteristics of traffic mode and trajectory spatial and temporal data types in the following three aspects:1)A Path-aware arrival time prediction is proposed based on the nearest neighbor adjacent road section.Firstly,divide the whole trajectory into multiple intersecting sections,and use historical trajectory data that similar to timegrainity to capture the characteristics of each section.Then,combine the topology of the road network to extract the trajectory data of the nearest neighbor adjacent sections,and use velocity characteristics as auxiliary information to alleviate the data sparseness problem.Experiments show that this method can effectively improve the accuracy of arrival time prediction results.2)A prediction of the arrival time of the path-blind adversarial transfer is proposed.The feature representation in the Path-blind method is insufficient to accurately predict the time.In this paper,the complete trajectory information contained in the Path-aware method,combined with the rich feature representation obtained by combining three different spatio-temporal features,is used to enhance the data features in the Path-blind method is realized through the adversarial training.Experiments show that this method can improve the arrival time prediction performance of blind paths.3)Use adversarial examples attack arrival time forecast.According to the existing adversarial examples generation method to test the robustness of the arrival time prediction model.Firstly,the white box methods are used to generate adversarial samples based on model gradient information,and then the black box methods are used to modify the sample data content to achieve attack.Experiments show that both methods can greatly reduce the generalization ability of the model.