马昊

      随着社会经济的发展和人民生活水平的普遍提高，众多城市小汽车保有量急速增长，所面临的交通主要问题都是“供不应求”。在这种局面下，主动交通管理应运而生，其目的是实现道路交通效率的极大化，最大限度的发挥道路基础设施的作用。它一方面需要适应当前交通流的变化，完成适应式控制系统实时控制策略；另一方面要及早采取相应措施，对交通流予以适当控制、引导，使其尽可能向良性循环的方向变化，这就需要前瞻性的交通状况预测。行程时间是表征道路交通状况的最重要的指标之一，其直观、简单明了的优点容易为广大用户所理解和接受。因此行程时间预测尤为重要，采用多源数据预测行程时间扩展了时间和空间的观测范围，进一步增强数据的可信程度，有助于获得更加准确的行程时间预测值。除此之外，对于出行者来说，一次出行的行程时间的长短及波动情况是影响他们选择出行路径及安排出行时间的关键因素，可靠度就是表征行程时间长短波动的指标，改善行程时间可靠度可以提高出行者的接受程度，也是诱导系统的一项重要指标。

      本文主要分为三个部分，分别为基于多源数据的行程时间估计、预测以及可靠度分析，主要工作如下：

      1、行程时间估计是下一步行程时间预测的基础，为了获得更准确的行程时间，本文按照各检测器的位置特点以及交通流特性，首先将行程时间划分为三个区段，充分发挥了各检测源的优势，分别用回归模型、改进BPR模型、改进波动论模型估计出各区段行程时间，最终获得整个路段总行程时间，采用实测数据验证，并与单一检测源估计方法作了对比，表明本文中所提出的估计方法有效地提高了估计精度。

      2、行程时间预测重点在于研究模型的设定，选取了RBF神经网络和支持向量机（SVM）这两种人工智能模型对行程时间进行预测，分别设置了三组不同输入变量作对比，结果表明将上游流量与历史行程时间作为输入变量，两种模型都能达到很好的预测效果。

      3、行程时间可靠度表征了行程时间的波动情况，本文将行程时间预测值引入到可靠度的概念中，定义了基于预测的行程时间可靠度概念，并用实测数据加以验证，表明此方法可以很好地反映下一时段行程时间在全天中所处的水平，进而判别下一时段交通拥挤状况。

关键词：城市道路，行程时间，估计，预测，可靠度，多源数据融合

    With the development of social economy and the improvement of living standard, the amount of vehicles keeps increasing rapidly, which makes the main problem of traffic become the unbalance between demand and supply. As a result, active traffic management (ATM) emerges and it aims at the improvement of traffic efficiency and performance of transportation facilities. It should be well adapted to traffic fluctuation to develop a real-time control strategy. Moreover, it should be able to take measures as early as possible to guide traffic flow in order to avoid future traffic jam or ease traffic jam, which requires real-time and exact traffic prediction. Travel time is one of the most important indicators to evaluate the traffic condition and it is well accepted because of its intuitiveness and simplicity. Using the multi-source data to predict travel time expands the measuring area and increases the credibility of data. And it makes travel time prediction more accurate, which is a fundamental part of ATM. In addition, the fluctuation of travel time is one of the key factors to influence the behavior of travelers when deciding the departure time and traffic mode. Reliability is used as an indicator to fluctuation of travel time. One of the key points of traffic guidance system is to improve the reliability of travel time to promote the acceptability to traffic guidance system, which is also an important part of ATM.

    This thesis is divided into three parts. Part 1 proposes a model of the estimation of travel time based on multi-source data. Part 2 focus on the prediction of travel time, followed by the analysis of reliability of travel time in Part 3. The main work is as follows.

    1.It’s obvious that the estimation of travel time is the fundamental part of the prediction. In order to obtain travel time, this thesis divided the travel time of a link into three parts considering the characteristics and locations of different detectors to make the most of each of the detectors. Regression model, improved BPR model and improved shockwave model were used respectively to estimate travel time of each part. According to the comparison of the results and the estimation based on single detection with test data, the proposed model improved the effectiveness of travel time estimation.

    2.The key point of prediction of travel time is the model specification. This thesis used both radical basis function (RBF) artificial neural networks and support vector machine (SVM) to predict travel time. Three groups of input variables were used to compare both methods and it turned out that both models achieved satisfactory results when using the upstream flow and historical data as input variables.

    3.Reliability is used as an indicator to fluctuation of travel time. This thesis introduced the prediction of travel time to the conception of reliability and gave the definition of the reliability of travel time based on the prediction of travel time. Based on the verification with test data, the proposed model was able to judge the congestion states in the next period.

Key Words:  Urban road, travel time, estimation, prediction, reliability, data fusion of multi-source