道路交通系统作为城市生命线系统之一，正承受来自内外部双重的运行干扰，交通安全和交通拥挤越来越突出。其中恶劣天气下城市路网交通拥挤是典型的交通灾害之一，给人们的生产生活和社会安全稳定造成了极大的隐患。近年来虽然国家在突发事件应急管理、气象灾害防灾减灾等方面做了许多努力，也有了较大的进步，但在应对恶劣天气城市交通灾害，特别是路网交通拥挤灾害时仍暴露出基础理论薄弱、处置方法单一落后等问题。总之，交通拥挤灾害管理需要有新思路和新方法。

       因此本研究以恶劣天气下路网交通拥挤预防与疏解决策信息需求为导向，运用风险概念来构建恶劣天气与路网交通拥挤灾害的关系，通过设计不同时段的路网交通拥挤风险评估与预警方法，完成从“交通拥挤后评价”到“交通拥挤前评价”的转变，最大程度地弥补现有的交通管理评价、实时交通拥挤态势分析在考虑恶劣天气影响方面不足的现实，也为进一步探索恶劣天气影响下的交通系统防灾减灾、响应天气的交通管理与控制策略等问题提供了素材和参考。

       本文重点研究了恶劣天气下城市交通拥挤形成一般机理，恶劣天气下交通拥挤风险评估与预警研究体系，恶劣天气下路网交通拥挤风险评估指标与方法、恶劣天气下交叉口-路段-路网三级交通拥挤风险预警指标和方法、恶劣天气下网络交通流预测关键技术共五个问题。

       1）在形成机理方面，结合机理分析方法论提出了恶劣天气下交通拥挤形成的一般机理，根据起主导作用的交通系统构成要素的不同，原理性机理可以细分为基于人要素的交通拥挤机理，基于车要素的交通拥挤机理，基于路要素的交通拥挤机理，基于管理要素的交通拥挤机理和基于耦合作用的多要素交通拥挤机理共5种。每一细分原理性机理都对应一个概念模型。概念模型主要由要素、事件（因素）、动作、阶段等记号组成。

       2）在评估与预警研究体系方面，把恶劣天气下城市路网交通拥挤风险定义为在恶劣天气干扰下城市道路交通运输系统运行出现（大范围长时间）路网交通拥挤或路网交通瘫痪等现象的可能性及后果。总结现有风险衡量相关文献，针对路网交通拥挤风险提出了分解模型和集成模型共2类风险量化模型。随后建立了由交通拥挤风险评估与预警定义、流程和指标体系等构成的研究体系。

       3）在评估指标与方法方面，针对路网，从天气危险性、交通管理能力和交通供需水平共3方面进行分解，构建了四层14个基础指标的路网拥挤风险评估指标体系。由于一些定性或定量指标具有明显模糊性的特点，选择了模糊层次分析法进行综合评价。

       4）在预警指标与方法方面，对于交叉口的拥挤风险预警指标，选择了饱和流率、损失时间、交叉口流量下降比例共3个指标；对于路段的拥挤风险预警指标，选择了自由流车速、通行能力和路段流量下降比例共3个指标；对于路网层面的拥挤风险预警指标体系，确定由各关键交叉口和路段的预警结果等微观类指标，以及路网平均车速、路段拥挤比例等宏观类指标共同组成。由于恶劣天气下路网交通拥挤发生是一个小概率事件，警兆指标历史数据严重匮乏，所以选择了灰色关联度法这种适合贫信息环境下简单高效的预警方法。

       5）在网络交通流预测关键技术方面，一是建立了恶劣天气下考虑出行安全需求的城市路网交通分配模型。安全需求是指用户路径选择时对遭遇交通事故的厌恶心理关注。首先整合路段和信控交叉口处的安全成本和出行时间到路径广义费用中，然后提出了一个等价于弹性需求下随机用户分配均衡条件的变分不等式模型并设计了启发式求解算法。通过案例数值分析发现：引入交通安全需求或交叉口延误均会造成分配结果的差异；均衡的弹性需求结果受到路径广义费用和信号配时的影响，但对于前者更加敏感。二是考虑到影响车速因素本身的模糊性以及影响作用非线性变化特性，抽取了交通量、占有率和降水强度作为输入变量，以车速作为输出变量，构建了一个三输入单输出的自适应神经网络-模糊推理系统。利用上海市快速路的交通流与气象数据进行训练与应用，结果表明该方法的预测均方根误差为3.82 km/h，预测平均百分比误差为5.28%。这表明该方法能应用在雨天下车速预测中并有较好效果。

   Urban road transport system, part of city lifeline systems, is bearing with increasing inside and outside disturbances so that transportation hazards continually occur. For example, network traffic congestion under adverse weather is such one typical phenomenon that threats both personal daily activities and social security greatly. Even though many efforts are made recently on emergency management of public incidents and meteorological hazards, traffic congestion management under adverse weather is still facing some difficulties, to name but a few, poor basic theory, outdated treatment practices etc. In one word, traffic congestion hazard management requires new thoughts and new methods.

   So this study aims to assist network traffic congestion management decisions by providing traffic information. It utilizes risk concept to connect adverse weather with network traffic congestion, and develops traffic congestion risk assessment and early-warning methods. It can be divided into five primary parts, which are weather-related traffic congestion formation mechanism, weather-related traffic congestion risk assessment and early-warning system, weather-related traffic congestion risk assessment indictor and method, weather-related traffic congestion risk early-warning indictor and method, and key technologies about network traffic flow prediction under adverse weather.

   1) Generalized traffic congestion formation mechanism under adverse weather is proposed based on mechanism analysis method. It is further divided into five kinds like human-dominated, vehicle-dominated, road-dominated, management-dominated and coupling multi-factor dominated mechanism. Each mechanism is explained by one conceptual model, which contains donations like element, incident (factor), action, and stage.

   2) Defination and measurement of weather-related network traffic congestion risk are given and discussed at first. In light of so-called decomposition model and integration model, risk assessment and risk early-warning are also defined respectively, and they are distinguished from five aspects like orientation stage, function, timeliness, alternative measurement models, and application scenario. At last research system of weather-related network traffic congestion risk is built up.

   3) Regarding road network, a traffic congestion risk assessment indictor system, which includes fourteen basic indicators in four levels, is built by discomposing three aspects: weather hazard, transport management ability and transport demand\supply level. And because some indicators are essentially fuzzy, fuzzy analytic hierarchy process method is used for risk evaluation.

   4) Three indictors like saturation rate, start-up loss time, and decline percent of intersection volume are selected for intersection congestion risk early-warning. And another three indicators like free-flow speed, capacity and decline percent of link volume for road segment congestion risk early-warning. Road network traffic congestion risk early-warning indictor system is consisted of microscopic indictor and macroscopic indictor. Microscopic indictors include early-warning index of critical intersection and road, whereas macroscopic indictors includes network average speed, and percent of mileage of congested segments. Considering low occurrence rate and limited historical data, the simple but effective grey correlation analysis that only demands little poor information is used as early-warning method.

   5) The first key technology is weather-oriented urban traffic assignment model used for flow predcition. By integrating safety costs and time cost on link and at intersection yields a new generalized path cost formula. Then one variational inequality model is constructed and one heuristic solution is developed. Finally case study suggests that, both safety consciousness and delay at intersection influence traffic flow pattern at equilibrium, and traffic demand at equilibrium is more sensitive to safety cost than to signal timing. The second key technology is real-time speed prediction method. Because factors influencing speed are fuzzy pe se and they interact non-linearly, one three-input-one-output adaptive neural fuzzy inference system is built based on traffic flow and meteorological data. Shanghai’s case study shows that this prediction method’s root mean square error is 3.82km/h, and its mean absolute percent error is 5.28%. It concludes that proposed fuzzy neural network enables to predict speed satisfyingly under rainy weather.