苗旭
城市混合路网（Mixed Networks）是由城市中的快速路网与普通道路网两个路网构成的路网。随着我国各大城市快速道路建设步伐的加快，混合路网已成为我国发达地区城市道路网系统的主要形式。快速路网的建成，为我国城市居民出行及运输活动发挥了重要作用。然而，随着交通需求的快速增长，城市快速道路作为比普通道路更为优质的道路资源，吸引了更多的交通流量，造成了交通需求过度集中于快速路及其衔接节点和网络，进而产生匝道拥堵及其引发的快速路主线拥堵。因此混合路网不仅没有发挥应有的优势，反而在快速路与普通道路衔接区域即出入口匝道附近形成了高峰期间交通拥挤的瓶颈区域，且围绕着这些衔接区域，交通拥挤向快速路系统和普通道路系统迅速扩散，诱发大范围网络交通阻塞。因此，在混合路网的层面上对路网交通流进行疏导和调控十分必要。为保障缓堵策略的合理性与有效性，需要充分了解混合路网的交通拥挤产生及演化规律。本研究首先完善固定检测器采集的交通流数据质量;其次构建面向于交通动态管理的道路资源评价指标并分析出入口匝道对该指标的影响；然后构建快速路和普通道路资源利用时空相关性的评价指标及关联分析模型；最后解析混合路网衔接区交通拥挤影响因素及在两个路网不同道路利用程度下交通拥挤的演化规律。
基于真实数据进行规律解析，数据的完整性及可靠性则尤为关键。本研究在细致地梳理五层异常数据判断规则的基础上，针对数据缺失问题，提出了一种缺失数据综合修复方法。该方法将检测器采集的交通流数据分成两部分，一部分是周期性趋势，采用简单平均值法（Simple Average Method, SAM）描述，另一部分是残差值，用动态选择解释变量的支持向量回归算法（Dynamic Variable Support Vector Regression, DV-SVR）进行回归。该方法一方面充分考虑了检测器数据的周期变化规律，利用周期变化特性提升数据修复精度；另一方面基于检测器缺失数据与其相关数据的时空相关性，动态选择数据修复模型解释变量，提高修复精度的同时避免了数据连续缺失导致修复误差逐步传递累积的现象。
为满足动态交通管理的需求，本研究提出了动态道路利用率指标，该指标分别考虑交通流失效前一定可接受失效概率的车道运行通行能力，以及交通流失效后不同时间占有率下的车道运行通行能力。相较于传统的饱和度指标采用固定车道通行能力的做法，本研究提出的动态道路利用率指标充分考虑了交通状态的时变性，更加合理和全面。另外，为更清楚地阐述出入口匝道对快速路主线交通流的影响，本研究基于大量真实数据分别对出入口匝道上下游断面及无匝道断面的车道运行通行能力、车道交通流量分布特征及动态道路利用率进行对比分析，并针对失效后的车道运行通行能力建立占有率-通行能力的二次多项式模型；针对车道交通流量分布特征建立占有率-车道流量对数分布模型。
快速路与普通道路两个路网存在协作及博弈关系，交通流量在两个路网之间进行转移，因此研究两个路网之间的交通流时空关联特征十分必要。本研究首先基于快速路及普通道路两个路网动态道路利用率指标，介绍了时序关联度的四种量化方法，即统计分析的相关系数法、模糊分析法、灰色关联分析法及动态弯曲距离方法。在此基础上，本研究考虑两个路网数值关联及变化趋势的差异性，提出来用均值水平及趋势变化程度两个指标组成的模式向量对路网子序列进行描述，并进一步考虑序列延迟特征提出一种新的关联度计算模型。
分析交通拥挤影响因素是解析交通拥挤产生机理的关键。本研究在对快速路入口匝道衔接区和出口匝道衔接区交通拥挤影响因素进行总结的基础上，建立快速路入口匝道衔接区和出口匝道衔接区附近的快速路主线交通流生存概率面板回归模型，充分讨论了各影响因素对交通流失效的影响程度进行。最后，在快速路与普通道路不同道路利用率情况下，解析入口匝道和出口匝道衔接区交通拥挤传播特征的差异。
本研究通过对混合路网衔接区拥挤演化特征的认知和把握，基于数据解析现象的理念，构筑基于动态道路利用率评价指标的拥挤识别、基于模式向量的路网时空关联特征解析体系，并在此基础上讨论了混合路网衔接区拥挤产生及演化与两个路网道路利用率的关系。限于数据、现场实验等条件和交通拥挤特征的复杂性和变化性，研究尚有诸多不足和缺陷，特别是未能解析除衔接区外更大范围的混合路网演化规律，此部分内容可以作为后续研究的关注点。
关键词：混合路网，检测数据修复，交通状态评价，时空关联模型，衔接区拥挤演化
Mixed networks is a road network structure composed of freeway network and ordinary road network. With the acceleration of the pace of rapid road construction in major cities in China, mixed networks has become the main form of road network system in developed region cities and major provincial capitals in eastern China. The construction of the expressway network plays an important role in the travel and transportation activities of urban residents in China. However, with the rapid growth of traffic demand, urban rapid road, as a better road resource than ordinary road, has attracted more traffic, resulting in the excessive concentration of traffic demand distribution on freeway and its connecting nodes and networks, which is prone to ramp congestion and the gridlock on the main expressway. Therefore, instead of giving full play to its advantages, mixed networks has formed the main area of traffic congestion in peak period near the connection area of freeway and ordinary road, that is, the entrance and exit ramp, and around these connection areas, traffic congestion spreads rapidly to the freeway system and ordinary road system, causing large-scale network traffic congestion. Therefore, it is necessary to dredge and regulate the traffic flow in the mixed road network. In order to ensure the rationality and effectiveness of the congestion mitigation strategy, it is necessary to fully understand the generation and evolution of traffic congestion in mixed network. In this study, the quality of traffic flow data collected by the fixed detector was improved firstly. Secondly, an indicator of road resources oriented to dynamic traffic management was constructed and the influence of ramps was analyzed. And then the evaluation index and correlation analysis model of the temporal and spatial correlation of expressway and ordinary road resource utilization were constructed. Finally, the influencing factors of traffic congestion in the junction area of mixed road network and the evolution law of traffic congestion under different levels of road utilization are analyzed.
Based on the rule analysis of real data, the integrity and reliability of the data is very important. This paper proposes a comprehensive repair method for missing data on the basis of carefully combing the five level judgment rules suitable for the abnormal date of this study. In this method, the detector data is divided into two parts, one is periodic trend, which is described by simple average method, the other is residual value, which is predicted by dynamic variable support vector regression, this method takes full account of the periodic variation rule of the detector data, and improves the accuracy of data correction by using periodic variation characteristics. Based on the spatiotemporal correlation between the missing data of the detector and its related data, this method dynamically selects the explanatory variables of the data repair model to improve the repair accuracy and avoid the gradual transmission and accumulation of repair errors caused by continuous absence of data.
In order to meet the needs of dynamic traffic management, this study proposes a dynamic road utilization index, which respectively considers the road operation capacity with certain acceptable failure probability before traffic flow failure, and the road operation capacity with different time occupancy rate after traffic flow failure. Compared with the traditional saturation index, which adopts fixed lane capacity, the dynamic road utilization index proposed in this study fully considers the time-varying of traffic state, which is more reasonable and comprehensive.In order to more clearly explain the impact of ramps on the lane operating features at urban expressway bottleneck，the lane operational capacity and lane traffic flow distribution characteristics of the upstream and downstream sections of the ramps were compared and analyzed in detail based on a large number of real data. And occupancy-capacity quadratic polynomial model is established for operational capacity after traffic flow breakdown, power function model is established for lane traffic flow distribution.
There is a cooperative and game relationship between freeway and ordinary road networks, and the traffic flow transfers between them. It is very important and necessary to study the correlation characteristics between the two networks. Based on the dynamic road utilization rate of freeway and ordinary road networks, this study introduces four quantitative methods of time series correlation degree, namely correlation coefficient of statistical analysis, closeness degree of fuzzy analysis, grey correlation degree of grey analysis and dynamic bending distance method. On this basis, considering the difference of numerical correlation and slope correlation between the two road networks, this study proposed to use the pattern vector to describe the series of road networks, and further considering the delay characteristics of the series, proposed a new correlation degree calculation model.
The key to analyze the mechanism of traffic congestion is to analyze the influencing factors of traffic congestion. On the basis of summarizing the influencing factors of freeway on ramp connection area and off ramp connection area, a panel regression model for survival probability of expressway mainline traffic flow near the on-ramp and the off-ramp junction area was established.This study fully discusses the effects of various influencing factors on traffic flow breakdown, and compares the regression results with the traditional multiple linear regression model.
Based on the concept of data analysis phenomenon, through the cognition and grasp of the evolution characteristics of congestion in mixed networks connection area, this study constructs congestion identification based on dynamic road utilization rate evaluation index and analysis system of spatiotemporal correlation characteristics of road networks based on dynamic pattern matching, and on this basis, the relationship between the congestion generation and evolution characteristics of the mixed networks connection area and the road utilization rate of the two road networks is discussed. Limited to the complexity and variability of data, field experiments and traffic congestion characteristics, there are still many deficiencies and defects in the study, especially the failure to analyze the evolution rule of a wider range than junction area of mixed networks, which is an important content of the follow-up study.
Key Words: mixed networks, modify missing data, traffic condition evaluation, spatio-temporal correlation model, traffic congestion evolution in mixed road networks junction areas
