Abstract: Reasonable design of the length of the acceleration lane can effectively alleviate the frequent traffic bottleneck problem in the expressway merging area. Therefore, a data-driven method was used to study the length of the acceleration lane in the expressway merging area. Firstly, the unmanned aircraft equipment was used to measure the traffic data of the expressway merging zone. The measured data was analyzed from the perspectives of traffic flow characteristics and vehicle convergence behavior, and the driving behavior of vehicles in the merging zone was obtained. Then, according to the characteristics of the traffic flow in the merging zone, the clustering of the dataset was performed. The generative adversarial network (GAN) was used to train car following and lane changing models for vehicles with different merging behaviors in the merging area, which was compared and analyzed with the measured data and the built-in model of SUMO simulation software. Finally, the generative adversarial network model was applied to simulate the traffic environment, and the indicators such as the speed, traffic density and traffic conflict rate were selected to establish the reward evaluation function, so as to obtain the recommended design value of acceleration lane length. The research results show that: through two measures such as early deceleration of mainline vehicles and lane changing towards the inner lane, collaborative lane changing can be achieved to avoid vehicles merging into the ramp. Compared with the built-in model of SUMO software, the generative adversarial network model is closer to the actual situation. The length of the single-lane parallel acceleration lanes obtained by the simulation is recommended to be 280 m, 240 m, and 200 m respectively under the conditions of 100 km/h, 80 km/h, and 60 km/h.

Key words: traffic engineering; merging area; acceleration lane; follow-up lane change model; generating adversarial networks; traffic simulation

摘要： 设计长度合理的加速车道能有效地缓解快速路合流区频繁出现的交通瓶颈问题，因此采用数据驱动方法对快速路合流区的加速车道长度进行研究。利用无人机设备测取了快速路合流区的交通数据，从交通流特性及车辆汇入行为这两个角度对实测数据进行分析，得到了合流区车辆的驾驶行为；根据合流区交通流特点，对数据集进行聚类分析，使用生成对抗式网络训练不同合流区汇入行为车辆的跟驰换道模型，并与实测数据和SUMO仿真软件中内置模型进行对比分析；应用生成对抗式网络模型进行交通环境仿真，选取速度、交通密度、交通冲突率指标建立奖励评价函数，得出了加速车道长度设计的推荐值。研究结果表明：采用主线车辆提前减速和向内侧车道换道这两种手段，可实现协同换道避让匝道汇入的车辆；相比SUMO软件内置模型，生成对抗式网络模型更加贴近实际情况；仿真得出的单车道平行式加速车道长度分别在100、 80、 60 km/h情况下的推荐值为280、 240、 200 m。

关键词: 交通工程；合流区；加速车道；跟驰换道模型；生成对抗式网络；交通仿真