Design Of Car Following Model Based On LTE-V Vehicle To Vehicle Communication Environment

With the development of connected vehicle,the future will be a long-term phase of traffic development in which connected vehicles and human-driven vehicles coexist.Vehicle to vehicle communication is an important supporting technology for future traffic development,which is used for information sharing of vehicle workshop to realize coordinated operation of traffic flow.The vehicle to vehicle communication environment has its own characteristics such as signal attenuation,which makes the car following behaviors have different operation rules from the traditional environment.Based on the analysis of traffic flow characteristics in the actual vehicle to vehicle communication environment,it is of great significance to study the car following model in the vehicle to vehicle communication environment.This paper proposes a car following model based on the LSTM network by using the data of car following driving in the actual vehicle to vehicle communication environment.Considering the influence factors of different urban road types,weather environment and time periods,the proposed model is compared with the classic GM model,FVD model and an improved FVD model in terms of prediction accuracy of distance and velocity.Firstly,the characteristics of car following behavior and the influence of vehicle to vehicle environment on car following behavior are described.Several car following models are briefly introduced,and the advantages and disadvantages of each model are analyzed.Then,the car following test and test data in vehicle to vehicle environment are introduced,and the data are processed by matlab.Next,the advantages and disadvantages and applicability of RNN algorithm and LSTM algorithm in deep learning and the basic principle of LSTM algorithm are briefly analyzed.After considering velocity,distance,acceleration,driver response time,leading vehicle status and communication delay,the car following model based on LSTM algorithm was modeled by introducing Tensorflow and other frameworks in python language environment,and the prediction results of velocity and diatance of the test set were obtained.Next,the GM,FVD and the improved FVD models are respectively calibrated in different urban roads,different weather conditions and different time periods by using matlab toolbox.Then,the prediction effects of the proposed model and the three traditional models on the distance and velicity are compared and analyzed.The results show that under different urban road types,different weather and different time periods,the prediction accuracy and goodness of fit of the proposed model are significantly better than the other three classical models.The prediction accuracy of the four models' velocity is higher than that of the distance between cars.FVD model and the improved FVD model have a significantly better description of the velocity than the distance.The prediction effect of these two models is not good when the fluctuation range of distance is large,and the fluctuation range of distance is smaller than the real situation.These two models have better prediction ability for data fragments with small fluctuation range of distance value.From the perspective of urban road types,different urban road types have no obvious influence on the FVD model and the improved FVD model.The model proposed and GM model in this paper have a poor prediction effect on high velocity road sections such as arterial road and expressway,compared with those sections with low velocity such as branch road and secondary arterial road.From the perspective of weather conditions,the prediction effect of the proposed model in rainy days and glare weather is less than that in normal weather.The prediction accuracy of FVD and the improved FVD model in rainy weather is lower than that in normal weather.The prediction effect of GM model varies little in different weather conditions,and its prediction accuracy for the two physical quantities of distance between cars and velocity also varies little.From the perspective of different time periods,there is no significant difference between the prediction accuracy of the proposed model.and GM model at different time periods.The effect of FVD model and the improved FVD model in predicting morning peak and evening peak is worse than that in normal peak period.