Study On Traffic Prediction And Analysis Based On Deep Learning In VANETs

In recent year,with the rapidly increasing development of information technology and the transportation industry,the intelligent transportation systems(ITS)have gained extensive attention from people and research institution.As an important part of the intelligent transportation systems,the vehicular ad hoc network(VANET)have becoming a promising research area.VANET have its unique features,such as the high mobility of vehicles and the dramatically changing topology,which may lead to the occurrence of congestion in traffic and network communication.In VANET,the dynamically changing traffic flow result in the changed network traffic and then affect the performance of network in communication.In order to alleviate city traffic and avoid the network congestion,it should make a study on analysis and prediction about the traffic flow and network traffic in VANET,which has an importantly theoretical and practical significance.This paper a new improved deep learning model to predict the two traffics,which takes advantage of the new features constructed by analyzing the correlation between traffic flow and network traffic.Firstly,the simulation of VANET is conducted by traffic simulator and network simulator,then we make an analysis of correlation between traffic flow and network traffic.Meanwhile,the network performance is also evaluated.Then,we create the datasets about traffic flow and network traffic,which are added to new features by the correlation between them.Afterwards,the improved deep learning models,deep belief network(DBN),make some predictions about traffic flow and network traffic,and adjusting the parameters of model until to the best results.Finally,the best results are made compared with other machine learning models.Through the simulation and the result analysis,it is concluded that it has a positive correlation between traffic flow and network traffic,by which constructed features are helpful for the performance of prediction model.In the meantime,the analysis between traffic flow and network traffic provides valuable reference for designing a protocol including congestion control mechanism.In addition,in terms of the forecasting effect,the improved DBN is better than other typically machine learning models,which proves that deep learning is effective,reliable and advanced on traffic prediction.