光华讲坛——社会名流与企业家论坛第6549期
主题：Optimizing CAV platoon movements in a signalized road network for travel and energy efficiency优化信号化道路网络中的自动驾驶车辆编队运动以提高出行和能源效率
主讲人：加州大学戴维斯分校 张红军（Michael Zhang）教授
主持人：管理科学与工程学院 肖峰教授
时间：8月4日 14:30-16:30
举办地点：腾讯会议，会议ID：225-505-930
主办单位：管理科学与工程学院 人工智能与管理科学研究中心 科研处
主讲人简介：
Dr. H. Michael Zhang is a Professor in the Department of Civil and Environmental Engineering at the University of California Davis. He is also a faculty member in the Graduate Programs of Applied Mathematics and Transportation Technology and Policy at UC Davis. Dr. Zhang's current research focuses on applications of systems theory to transportation systems analysis and operations. Specific topics include traffic flow theory, traffic control, transportation network models, and intelligent transportation systems such as the application of wireless communications and grid computing technology to distributed, on-demand traffic management.
Professor Zhang is an Area Editor of the Journal of Networks and Spatial Economics and Associate Editor of Transportation Research Part B：Methodological and Transportmetrica A: Transport Science.
张红军（Michael Zhang），加州大学戴维斯分校土木与环境工程系教授，加州大学戴维斯分校应用数学及交通技术与政策研究生项目导师。目前主要研究为系统理论在交通系统分析及运营中的应用。具体方向为交通流理论、交通控制、交通网络模型和智能交通系统（如无线通信和网格计算技术在分布式、按需交通管理中的应用）。
张教授为Journal of Networks and Spatial Economics领域主编，Transportation Research Part B：Methodological和Transportmetrica A: Transport Science副主编。
内容简介：
Vehicular emissions and traffic congestion around the world have been deteriorating due to rapid urbanization and increase in car ownership. The worsening traffic burdens drivers with higher operating costs and longer travel times, and exposes pedestrians to higher concentrations of pollutants such asPM,���������,������2. One promising technology to reduce traffic congestion is connected autonomous vehicles (CAVs), a technology that enables self-driving and information sharing between drivers and infrastructure. With advanced wireless technologies offering extremely low latency, platooning control of CAVs can be realized to improve traffic efficiency and safety. However, conventional platooning control algorithms require complex computations and therefore are not perfectly suited to real-time operations. To overcome this challenge, this project focuses on designing an innovative learning framework for platooning control capable of reducing fuel consumption through the four basic platoon manipulations: split, acceleration, deceleration, and no-op. We integrate reinforcement learning (RL) with neural networks (NNs) to be able to model non-linear relationships between inputs and outputs for a complex application. The experimental results show a decreasing trend of the fuel consumption and a growing trend of the reward, and demonstrate that the proposed DRL platooning control is effective to reduce fuel consumption and improve mobility.
随着城市化快速发展、汽车持有量增加，世界各地车辆排放问题和交通拥堵问题日益恶化。越来越糟糕的交通状况令司机们承担更高的运营成本，忍受更长的出行时间，也使得行人暴露在更高浓度的污染中，如PM,���������,������2。自动驾驶车辆（CAVs）即为一项极具前景的改善交通拥堵状况的技术，该技术可实现自动驾驶及司机与设施之间的信息共享。借助先进的无线技术提供的极低延迟，自动驾驶车辆编队控制可提升交通效率及安全性。然而，传统的编队控制算法需复杂的计算，因此并不完全适合实时操作。为克服这一难题，本项目侧重于设计一个创新的编队控制学习框架，通过四种基本的编队操作（分割、加速、减速和无操作）来减少燃料消耗。我们将强化学习(RL)与神经网络(NN)相结合，以便能够对复杂应用的输入和输出之间的非线性关系进行建模。实验结果显示，燃料消耗呈下降趋势，回报呈上升趋势，并证明所提出的DRL编队控制能有效降低燃料消耗和提升出行效率。
