网联车对抗神经网络跟驰模型

针对当前混行交通流场景下网联车对前车速度变化的实时性、安全性和车队稳定性较差的状况,提出一种由生成模型和辨别模型构成的网联车生成式对抗网络车辆跟驰模型(GANVFM).其中,生成模型提取跟驰参数中的前车速度、跟驰车速和相对车距计算生成加速度;辨别模型对生成模型生成的加速度参数进行相似度计算,并通过更新函数加以更新.采用...

A Connected and Autonomous Vehicle Following Model Based on Generative Adversarial Network

In view of the poor real-time performance and safety as the responses of connected and autonomous vehicles(CAVs) to the speed change of leading vehicle and the low stability of CAV platoon under the current mixed traffic flow situation,a generative adversarial nets vehicle following model(GANVFM) composed of generation model and discrimination model is proposed for CAVs.The generation model extracts the vehicle flowing parameters such as the leading vehicle speed,the following vehicle speed and the vehicle spacing to calculate the generated acceleration,while the discrimination model calculates the similarity of the acceleration parameters generated by generation model and updates both the generation and discrimination models by updating function.Then the realtime performance and safety of CAVs and the stability of vehicle platoon are analyzed by using man square deviation σ for speed and acceleration,rear-end collision predicting factor γn and vehicle following state factor φn as corresponding indicators.The results show that the GANVFM has the smallest γn and σ,and the real-time performance and safety of GANVFM to the speed change of leading vehicle are high.With the increase of the permeability rate δ of CAVs,the φn reduces,the fleet length shortens,and the fleet stability improves.