Incident-aware Resilient Traffic Management for Urban Road Networks (InTURN)

想象一下，你正在尽可能快速可靠地通过城市交通网络，同时避免所有拥堵和干扰。再想象一下，您在这样做的时候，没有持续地、无处不在地向大型数据驱动型企业发送位置和上下文信息，从而使您的所有个人隐私信息蒸发。在本提案中，我们将研究如何将这样的系统实现为自适应和自组织（SASO）系统。城市交通是SASO系统的一个具有挑战性的试验台：巨大的交通量与潜在的动态和时变行为以及事件和负面环境影响等干扰相结合，需要新的综合控制和管理策略。在过去的十年中，已经提出了几种交通信号灯控制、渐进信号系统和路线引导的方法，包括我们自己的初步工作：有机交通控制（OTC）。然而，现有的系统通常仅限于对观察到的交通状况做出反应，而不考虑事件等干扰（例如车祸、建筑工作或卡车卸载）。在本提案中，我们希望通过智能机制来克服这些限制，以增加交通控制和管理解决方案的弹性。我们使用机器学习方法检测异常交通状况并识别事件。这些事故将根据其预计持续时间、预计严重程度以及对其他交叉路口和道路要素的预期影响自动分类。我们进一步利用交通控制的互联特性，通过开发对检测到的事件进行合作验证的技术——这也允许检测受干扰的传感器。事件分类受制于一种自主学习机制，该机制可以在运行时自我改进和改进决策。为了最终利用确定的事件信息，我们研究了一个基于OTC的集成和强大的交通管理系统，该系统1)适应并自我改进交通灯信号策略，2)建立和维护交通响应渐进信号系统，以及3)动态引导驾驶员通过底层道路网络。结果将优于现有的解决方案，在旅行时间，红灯前停车次数和减排方面。

Imagine that you are passing an urban traffic network as fast and reliable as possible while simultaneously avoiding all congestions and disturbances. Imagine also that you are doing this without continuously and ubiquitously sending location and context information to large data-driven enterprises evaporating all your personal privacy information. In this proposal, we will investigate how such a system can be realised as a self-adaptive and self-organising (SASO) system. Urban traffic is a challenging testbed for SASO systems: Massive traffic volumes in combination with the underlying dynamics and time-variant behaviour as well as disturbances such as incidents and negative environmental effects demand for novel integrated control and management strategies. Within the last decade, several approaches for traffic light control, progressive signal systems, and route guidance have been presented, including our own preliminary work: Organic Traffic Control (OTC). Existing systems, however, are typically limited to only reacting to observed traffic conditions and they do not consider disturbances such as incidents (e.g. car accidents, construction work, or un/loading of lorries). In this proposal, we want to overcome these limitations by means of intelligent mechanisms to increase the resilience of traffic control and management solutions. We detect abnormal traffic conditions and identify incidents using machine learning approaches. These incidents are automatically classified by means of their estimated duration, their anticipated severity, and the expected influence on other intersections and road elements. We further take advantage of the interconnected character of traffic control by developing techniques for a cooperative validation of detected incidents – which also allows for detecting disturbed sensors. The incident classification is subject to an autonomous learning mechanism that self-improves and refines the decisions at runtime. In order to finally take advantage of the determined incident information, we investigate an integrated and robust traffic management system based on OTC that 1) adapts and self-improves the traffic light signalisation strategy, 2) establishes and maintains traffic-response progressive signal systems, and 3) dynamically guides drivers through the underlying road network. The result will outperform existing solutions in terms of travel times, number of stops in front of red traffic lights, and emission reductions.