Real-time network control strategies such as congestion pricing have been used in a number of metropolitan areas around the world for traffic congestion mitigation. Recent advances in Global Navigation Satellite System (GNSS) technology have led to increasing interest in distance- or usage-based road pricing as an effective alternative to traditional facility-, cordon- and area-based pricing that typically rely on fixed infrastructure. In this paper, we propose the use of feature-variant clustering methods, OPTICS and HDBSCAN*, as a systematic approach for tolling zone definition to operationalize distance-based tolling schemes. Subsequently, we develop a framework for predictive distance-based toll optimization to evaluate network performance for the various tolling zone definitions derived from the aforementioned feature-variant clustering methods. In this framework, for a specific tolling zone definition, tolling function parameters are optimized using a simulation-based Dynamic Traffic Assignment (DTA) model operating within a rolling horizon scheme. Predictive optimization is integrated with the guidance information generation. Behavioral models capture drivers’ responses to the tolls in terms of trip cancellation and choices of mode, route and departure time. Experiments on the real-world Expressway and Major Arterials network of Singapore demonstrate improved effectiveness of distance-based toll optimization given tolling zone definitions derived from feature-variant clustering, compared to fixed cordon-based pricing, adaptive cordon-based pricing, as well as distance-based pricing with ad hoc tolling zone definitions. Further, the results indicate that the use of the marginal link cost tolls as a clustering feature produces the most robust tolling zone definitions and yields significant improvements in social welfare over ad hoc zone definitions and cordon-pricing. Finally, experiments on the Boston CBD network also demonstrate the effectiveness of distance-based toll optimization schemes on urban traffic networks.

诸如拥堵定价之类的实时网络控制策略已在全球许多大都市地区用于缓解拥堵。全球导航卫星系统（GNSS）技术的最新进展已引起人们对基于距离或使用量的道路定价的兴趣日益浓厚，以取代传统的基于设施，警戒线和区域的定价，后者通常依赖于固定基础设施。在本文中，我们建议使用特征变量聚类方法OPTICS和HDBSCAN *作为收费区定义的系统化方法，以实施基于距离的收费方案。后来，我们开发了一种基于预测距离的收费优化的框架，以评估从上述特征变量聚类方法得出的各种收费区定义的网络性能。在此框架中，对于特定的收费区定义，使用在滚动范围内运行的基于仿真的动态交通分配（DTA）模型来优化收费功能参数。预测性优化与制导信息生成集成在一起。行为模型记录了驾驶员在出行取消以及对模式，路线和出发时间的选择方面对通行费的反应。在新加坡的真实世界的高速公路和主要干线网络上进行的实验表明，鉴于基于特征量聚类的收费区定义，基于距离的收费优化的有效性有所提高，与基于固定警戒线的定价，基于自适应警戒线的定价以及具有特殊收费区定义的基于距离的定价相比。此外，结果表明，使用边际链接费用通行费作为聚类功能，可以产生最可靠的通行费区定义，并且在社会福利方面比自定义区定义和警戒线定价显着提高。最后，在波士顿中央商务区网络上进行的实验也证明了基于距离的收费优化方案在城市交通网络上的有效性。结果表明，将边际链接费用通行费用作聚类功能可产生最可靠的通行费区定义，并比特设区的定义和警戒线定价带来显着的社会福利改善。最后，在波士顿中央商务区网络上进行的实验也证明了基于距离的收费优化方案在城市交通网络上的有效性。结果表明，将边际链接费用通行费用作聚类功能可产生最可靠的通行费区定义，并比特设区的定义和警戒线定价带来显着的社会福利改善。最后，在波士顿中央商务区网络上进行的实验也证明了基于距离的收费优化方案在城市交通网络上的有效性。