Research Of Low Speed Car-following Technology Under Urban Traffic Congestion Environment

At present,urban traffic congestion is serious,and it is already one of the most difficult problems in urban governance.Traffic congestion means that vehicles need to start and brake frequently.This will not only make the driver tired and irritated,but also prone to traffic accidents;It will greatly increase the vehicle’s exhaust emissions,thereby reducing the city’s air quality and increasing the city’s "heat island effect." The automatic car-following technology not only can reduce the driver’s driving intensity and relieve the driver’s emotions,but also can make the whole traffic flow orderly,which help reduce the traffic congestion time and improve the urban traffic conditions.Therefore,research and development of such driving technology has important theoretical significance and application value.The paper mainly conducts the following research:(1)The establishment of the theory-driven low-speed car-following model: This paper analyzes and compares the Gipps model,the Full Velocity Difference(FVD)model and the Intelligent Driver Model(IDM).The difference in safety between the three models is not great,and the IDM model is much better than the other two models in terms of comfort.Therefore,based on the IDM model,this paper establishes a theoretically driven low-speed car-following model and the problem of the slow start of the original IDM model vehicle has been optimized.(2)The establishment of data-driven low-speed car-following model: This paper uses the low-speed car-follow data set obtained by certain processing of the US NGSIM(Next Generation Simulation)traffic data set as the data source.Based on Radial Basis Function Nerual Network(RBFNN)this paper establishes a data-driven low-speed car-following model,and the model prediction results can be accurately close to the true value.At the same time,the model is compared with the IDM low-speed car-following model.In order to combine the advantages of the two,the IDM-RBFNN low-speed car-following model is established by the optimal weighting method.The model achieves better prediction than the single model.(3)Research on low-speed car-following decision algorithm for vehicles: This paper uses Deep Deterministic Policy Gradient(DDPG)algorithm and Trust Region Policy Optimization(TRPO)in Deep Reinforcement Learning(DRL)algorithm.The two algorithms are improved by the Control Barrier Functions(CBF)method.The decision performance of each algorithm is analyzed in the low-speed car-following experiment.Finally,the best performance of TRPOCBF algorithm is obtained.Therefore,the algorithm is chosen as the low-speed follow-up decision algorithm for subsequent simulation experiments.(4)Design and analysis of vehicle low-speed heel simulation experiment: This paper builds a simulation experiment platform for vehicle low-speed heel decision based on Carsim and Unity3 D,and simulates the movement of the preceding vehicle in the experimental scene according to the real data in the low-speed car-following data set.The TRPO-CBF algorithm is used to control the chiseling movement of the smart car to the preceding car.The IDM-RBFNN low-speed car-following model is used as the decision-making goal of the algorithm.Finally,the visualization of the algorithm decision-making training process is realized,and the TRPOCBF algorithm can effectively converge to the target strategy.