Researches On Traffic Speed Prediction Method Based On Multi-source Data Fusion In VANET

With the acceleration of urbanization in China and the continuous growth of residents ' motor vehicle ownership,the traffic pressure borne by the urban road network is increasing day by day,and a series of traffic problems,such as traffic congestion and traffic accidents,have arisen,and seriously restrict the road capacity and the efficiency of residents ' travel.By monitoring and inducing the running state of the road,the Urban intelligent transportation system can effectively alleviate the traffic congestion problem and improve the road traffic efficiency.Accurate and real-time road traffic condition prediction is the key to realize traffic control and induction in intelligent transportation system,and it can also provide decision support for driving route planning.The Internet of vehicles provides wireless communication between vehicles and between vehicles and nearby roadside equipment,by obtaining information about their own vehicles,other vehicles and the external environment,it is easy to form massive data that reflects traffic conditions in the Internet of Vehicles.Compared with the lack of news data in traditional transportation system,how to integrate the rich and varied information in the Internet of Vehicles to obtain more accurate traffic state estimation,has practical significance for improving urban traffic congestion and improving road utilization.This paper proposes a multi-source data fusion middleware framework(MDFOSGi)for the Internet of Vehicles,which takes advantage of the service-oriented advantages of the OSGi(Open Service Gateway initiative)technology framework and enable vehicles to dynamically obtain information about data services that published in the roadside unit by traffic participants.Different data fusion algorithms are encapsulated for different applications,and the vehicles,roads,environment and other related service data are processed and fused by using the vehicle's own abundant storage and computing resources,which produce local decisions and provide data service support for the upper application.At the same time,two data fusion algorithms and a model combination algorithm for different characteristics are proposed for traffic prediction related applications,and encapsulated in this middleware to provide more accurate data services.Most of the current traffic state prediction based on data fusion only combines the traffic data collected by multiple detectors,and ignores the influence of social and environmental factors on traffic speed.This paper proposes a single-segment traffic speed prediction algorithm based on multi-source data fusion.Based on the analysis of the influence of holiday and climate factors on traffic speed,the relevant data are divided into traffic data and external impact data,and the feature set is divided into subsets by cluster analysis,and then the multi-branch LSTM section traffic speed prediction model is constructed by using different subsets.In the prediction stage,the traffic speed prediction is performed based on the real-time data extraction corresponding LSTM model.Secondly,aiming at the large spatial spread of traffic flow in the peak period,a combination prediction algorithm for peak period and a DBN-LSSVR prediction model based on feature fusion of neighbor links are proposed.Firstly,the gray relation analysis is used to select the traffic data of the neighboring road segments with high relevance,and the feature extraction is carried out.and then the fused feature input least squares support vector regression model is predicted.And combine the predicted results with the single-segment prediction results to improve the prediction accuracy of the traffic speed during the peak period.The experimental analysis on the real data sets shows that the proposed road traffic speed prediction method based on data fusion is less affected by external disturbance factors,improves the accuracy of traffic speed prediction,and has satisfactory prediction effect.