Behavioral Models for Microscopic Traffic Simulation

本研究旨在开发新一代微观交通模拟，利用高性能计算方法，利用驾驶行为模型捕捉驾驶员的策略行为。创新的智能交通系统（ITS）需要灵活的仿真工具来进行实验，包括动态交通管理技术、交通控制和路由算法。还需要更可靠的模拟工具来研究交通影响，如拥堵、安全、能源消耗和空气污染。这些应用程序需要详细而准确的驾驶员行为模型。大型网络的在线和离线应用程序以及日益复杂的模型和流量管理系统需要使用更有效的计算方法。最著名的驾驶行为模型是加速（或汽车跟随）和变道模型。最先进的模型将得到增强，以包括更现实的行为，如：主动预期行为：驾驶员通过预测未来的交通状况并采取行动，创造机会来进行他们想要的机动。这与现有的驾驶员被动地对当前交通状况做出反应的模型形成了对比。扩展视野：驾驶员的决策是基于他们扩展的社区的交通状况，而不是简单地跟随领导者或对相邻车辆做出反应。相互依赖的决策：相互依赖存在于不同的决策之间（例如变道对加速的影响）和一个决策内部（例如过去变道对未来的影响）。这些依赖关系在现有的模型中被忽略了：变道模型假设一旦驾驶员决定变道，他/她会被动地考虑交通流中出现的可用间隙——驾驶员不会为了变道而调整他/她的加速度。因此，合并、编织和类似设施的能力没有被正确捕获。高性能计算实现的发展将扩展微观交通模拟的适用性。使用粗粒度并行和分布式计算平台将允许使用公共领域处理器间通信软件库开发可移植代码。我们将回顾最新的网络分解方法，从而为后续研究奠定基础。该项目的重点将是开发和校准捕捉上述智能行为原则的创新驾驶行为模型。未来，经过校准的模型将在麻省理工学院开发的微观交通模拟实验室MITSIMLab中实施。将进行模拟来测试这些模型并提出进一步的改进建议。该合同是在运输工业工程探索性研究（ETI）项目招标下获得的。

This research aims to develop a new generation of microscopic traffic simulation with driving behavior models that capture strategic behaviors of drivers and utilize high performance computational methods. Flexible simulation tools are needed to conduct experiments of innovative Intelligent Transportation Systems (ITS) including dynamic traffic management technologies and traffic control and routing algorithms. More reliable simulation tools are also needed to study traffic impacts such as congestion, safety, energy consumption and air pollution. These applications require detailed and accurate models of drivers' behavior. Online and offline applications to large networks and increasingly complex models and traffic management systems require the use of more efficient computational methods. The most notable driving behavior models are acceleration (or car following) and lane-changing models. State-of-the-art models will be enhanced to include more realistic behaviors such as: Proactive anticipatory behavior: Drivers create opportunities to undertake their desired maneuvers by anticipating future traffic conditions and acting upon them. This is contrasted with existing models in which drivers passively react to present traffic conditions. Extended field of view: Drivers' decisions are based on traffic conditions in their extended neighborhood as opposed to simply following a leader or reacting to the adjacent vehicles.Interdependent decisions: Interdependencies exist across different decisions (e.g. the effect of lane changing on acceleration) and within one decision over time (e.g. the effect of past lane changes on future ones). These dependencies are ignored in existing models: lane-changing models assume that once a driver decides to change lanes, s/he will passively consider available gaps as they appear in the traffic stream - the driver does not adjust his/her acceleration in order to change lanes. As a result, capacities of merging, weaving and similar facilities are not captured correctly.The development of high performance computing implementations will extend the applicability of microscopic traffic simulation. Using coarse-grained parallel and distributed computing platforms will allow for the development of portable codes using public domain inter-processor communication software libraries. We will review the state-of-the-art in network decomposition methods, thus creating a basis for follow-up research. The focus of the project will be the development and calibration of innovative driving behavior models capturing the intelligent behavior principles mentioned above. Calibrated models will, in the future, be implemented in MITSIMLab, a microscopic traffic simulation laboratory developed at MIT. Simulations will be conducted to test the models and to suggest further refinements.This award is made under the Exploratory Research on Engineering the Transport Industries (ETI) program solicitation.