High performance optimization of integrated transportation management methods for emergency evacuation

影响主要城市地区的人为灾害和自然灾害日益增加，需要对紧急疏散情况进行全面分析和规划，同时利用智能交通系统（ITS）的潜力加快疏散过程。 尽管在过去的几年里，人们高度重视研究这一问题，但仍然需要整合各种交通需求管理和供应控制策略，以协同和优化其对紧急疏散的潜在效益。最优性是指最大限度地实现一个目标，如最大限度地增加到达安全地点的撤离人员人数，最大限度地减少他们到达安全地点的旅行时间，或最大限度地减少撤离人员暴露于某些风险（如化学烟雾等）。 需求管理是指通过控制出发时间和目的地选择来控制在交通网络中从某些起点移动到某些目的地的疏散人数。供应管理是指通过交通信号控制、高速公路控制和动态路径选择来控制交通网络的效率。 综合疏散管理策略的理想输出是向疏散人员提供关于何时疏散、去哪里以及如何到达那里的信息和指导，以及一组基础设施控制参数，例如真实的时间信号定时计划、逆流车道分配（将指定车道的通常流向逆转）等。本研究的目的是：（1）将人员疏散问题转化为多目标优化问题，（2）提出了一套基于并行分布式遗传算法的求解算法，以及（3）在高性能计算（HPC）网格上实现该系统。这种方法的新奇源于疏散管理策略的前所未有的整合，基于人工智能的最先进的可扩展解决方案算法的发展以及新兴高性能多处理器硬件的计算能力的利用。

The increasing rate of man-made disasters and natural catastrophes affecting major urban areas requires comprehensive analysis and planning for emergency evacuation scenarios while harnessing the potential of Intelligent Transportation Systems (ITS) to expedite the evacuation process. Despite the heightened attention to researching the subject over the past few years, the need still exists for integrating various transportation demand management and supply control strategies to synergize and optimize their potential benefits to emergency evacuation. Optimality refers to maximizing an objective such as maximizing the number of evacuees reaching safety, minimizing their travel times to safety or minimizing the exposure of evacuees to certain risks (e.g. chemical fumes etc.). Demand management refers to controlling the number of evacuees moving in the transportation network from certain origins to certain destinations by controlling departure time and destination choice. Supply management refers to controlling the efficiency of the transportation network using traffic signal control, freeway control and dynamic routing. The ideal output of an integrated evacuation management strategy is information and guidance to evacuees regarding when to evacuate, where to go and how to get there, in addition to a set of infrastructure control parameters such as, for instance, real time signal timing plans, contra-flow lane assignment (reversing the usual flow direction of a given lane) etc. The objectives of this research are to: (1) formulate the integrated evacuation problem as a multi-objective optimization problem, (2) develop a set of solution algorithms based on parallel distributed Genetic Algorithms, and (3) implement the system on a High Performance Computing (HPC) grid. The novelty in this approach is rooted in the unprecedented integration of evacuation management strategies, the development of state of the art extensible solution algorithms based on artificial intelligence and the harnessing of computing power of newly emerging high performance multi-processor hardware.