Abstract

In the scientific literature, the ratio between the total travel cost under a user equilibrium assignment and the total travel cost under a system optimum assignment is typically referred to as the Price of Anarchy (PoA), i.e., the level of inefficiency that can be eliminated by centralized control in routing. Recently, this concept has been attracting renovated attention due to the new opportunities offered by vehicles’ connectivity and automation. The new technologies could allow individual prescriptions centrally managed to achieve benefits in network performance. However, considering the infrastructure that such a system would need and the ethical implications it could have (related to privacy, equity, etc.), it is necessary to carefully quantify its actual benefits. Existing PoA related studies do not fully capture the essential realism of traffic dynamics and show contradictory results. Moreover, there are no studies that explore the relation of the PoA and driving behavior (e.g., reaction time, acceleration, deceleration, aggressiveness, etc.), which is very important for designing the algorithms of automated driving. In this light, the present paper investigates the PoA over a Braess-like network using the most detailed way to model traffic—the microscopic simulation with parameters defining the driving behavior. The magnitude of the PoA has been studied over a wide range of combinations of the model input parameters. Results show that the PoA can be much higher than that obtained in theoretical studies. In addition, results are used to reveal some PoA features of real traffic networks and propose further research directions.

摘要

在科学文献中，用户均衡分配下的总旅行成本与系统优化分配下的总旅行成本之间的比率通常称为无政府状态价格（PoA），即可以消除的低效率水平通过路由集中控制。最近，由于车辆的连接性和自动化提供了新的机会，这一概念已经引起了新的关注。新技术可以让个人处方集中管理，以提高网络性能。然而，考虑到此类系统所需的基础设施及其可能产生的伦理影响（与隐私、公平等相关），有必要仔细量化其实际收益。现有的 PoA 相关研究并未完全捕捉到交通动态的基本现实性，并显示出相互矛盾的结果。此外，目前还没有研究探讨 PoA 与驾驶行为（例如反应时间、加速、减速、攻击性等）的关系，这对于设计自动驾驶算法非常重要。有鉴于此，本文使用最详细的交通建模方法——使用定义驾驶行为的参数的微观模拟，研究了 Braess 类网络上的 PoA。已经在模型输入参数的广泛组合中研究了 PoA 的大小。结果表明，PoA 可以远高于理论研究中获得的 PoA。此外，