A Multiscale Framework for Forecasting Highway Traffic Flow

交通堵塞是日常生活中一件令人讨厌的事。它们也阻碍了我们的经济：英国工业联合会（CBI）估计，道路交通拥堵造成的延误每年给英国企业造成高达200亿英镑的损失。据预测，在2000年至2015年期间，英国的道路交通将增长30%，因此，拥堵问题似乎只会变得更糟。因此，国际上在使用信息和通信技术来管理交通以缓解拥堵方面做出了巨大的努力——这个广泛的领域被称为智能交通系统（ITS）。普通的高速公路司机应该已经熟悉ITS了。示例包括1；M25伦敦轨道上的受控高速公路项目（在交通拥挤时设置临时减速限制）；2. 伯明翰M42公路的主动交通管理（在繁忙时期，硬肩变成了一条普通的行驶车道）；和3。“排队”的警告标志现在在英国的高速公路网上几乎无处不在。在这种远程信息处理基础设施上的投资非常可观——仅主动交通管理就投入了大约1亿英镑。以上描述的每一个ITS应用程序的核心都有详细的数学和计算机模型来预测交通如何流动以及队列如何形成和消散。然而，这些模型远非完美，本研究的目的是通过研究支撑这些模型的基础科学来改进这些模型。这是一个所谓的多尺度挑战，因为有一个完整的层次结构的不同层次的细节模型，从模拟个体驾驶员行为的模拟模型，到在交通流量和可压缩气体之间进行类比的宏观模型。本研究将建立找出哪些模型是好的，哪些是坏的方法。此外，它将使用现代“机器学习”技术来结合良好的模型，从而使基于计算机的交通预测具有类似人类的人工智能。

Traffic jams are an annoying feature of everyday life. They also hamper our economy: the CBI has estimated that delays due to road traffic congestion cost UK businesses up to 20 billion annually. UK road traffic is forecast to grow by 30% in the period 2000-2015, so it seems that the congestion problem can only get worse. There is consequently an intense international effort in using Information and Communication Technologies to manage traffic in order to alleviate congestion --- this broad area is known as Intelligent Transport Systems (ITS). Regular motorway drivers will already be familiar with ITS. Examples include 1. the Controlled Motorways project on the M25 London Orbital (which sets temporary reduced speed limits when the traffic gets heavy); 2. Active Traffic Management on Birmingham's M42 (where the hard-shoulder becomes an ordinary running lane in busy periods); and 3. The `Queue Ahead'warning signs which are now almost ubiquitous on the English motorway network. The investment in this telematics infrastructure has been very significant --- about 100 million pounds for Active Traffic Management alone.Each of the ITS applications described above has at its heart detailed mathematical and computer models that forecast how traffic flows and how queues build up and dissipate. However, these models are far from perfect, and the purpose of this research is to improve the models by working on the fundamental science that underpins them. This a so-called multiscale challenge, since there is a whole hierarchy of models of different levels of detail, ranging from simulation models that model the behaviour of individual drivers, up to macroscopic models that draw an analogy between the flow of traffic and compressible gas. This research will establish methods for finding out which models are good and which ones are bad. Moreover, it will use modern `machine learning' techniques to combine good models so that computer-based traffic forecasting has human-like artificial intelligence.