【作者】 马广英；

【导师】 李平；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2006， 博士

【摘要】 随着社会经济的发展，城市化、汽车化速度的加快，交通拥挤、交通事故、环境污染、能源短缺等问题已经成为世界各国面临的共同问题。无论是发达国家，还是发展中国家，都毫无例外地承受着不断加剧的交通问题的困扰。解决交通问题的传统办法是修建道路，但无论是哪个国家，可供修建道路的空间都越来越小。另外，交通系统是一个复杂的巨系统，单独从道路方面考虑，很难从根本上解决问题。在此背景下，出现了把交通基础设施、交通运载工具和交通参与者综合起来系统考虑，充分利用高新技术解决交通问题的智能交通系统(Intelligent Transportation System,ITS)。广义的智能交通系统包括道路交通管理系统、整个交通运输系统的规划、设计和运营管理的智能化。而交通规划、交通管理和交通控制都离不开OD矩阵(也称交通出行矩阵)这一基础数据，即需要知道路网上的交通需求。另外，OD矩阵也是交通仿真系统最直接、可靠的仿真输入数据。传统获得OD矩阵的方法是进行大规模抽样调查，其昂贵的费用和组织上的难度是可以想象的。而由于当前城市交通监控系统的普及，交通流量已经成为较为容易获取的信息。因此由路段交通量反推OD矩阵已成为获取OD矩阵较为可行且经济的方法之一。对于一般的城市路网，由路段交通量反推OD矩阵主要包括以下几个步骤：路段流量检测以及先验信息的获取；路网特征及交通分配矩阵的获得；按照特定的反推模型进行OD反推。上述几个步骤中影响OD反推精度的主要有：反推模型的准确性、先验信息的可靠性、路段检测流量的准确性和交通分配方法的合理性。另外模型的求解方法也是一个值得探讨的问题，求解的可行性与简便性关系着模型的适用程度。本文以结构相对简单、理论依据明确的极大熵反推模型为基础，结合OD反推中各个关键步骤，以OD反推精度最好为主要目标，在模型求解、路段检测点设置、交通分配方法等方面作了详细探讨，并以浙江大学自主研发的城域混合交通仿真与分析系统(SASUMT)为应用平台，利用OD反推技术为SASUMT提供方便可靠的仿真输入数据。主要研究内容总结如下：1．基于OD反推的极大熵模型，提出了遗传算法求解的方法。首先分析了极大熵模型的推导过程及理论依据；针对极大熵反推模型的特点，利用解约束优化问题的基本方法——拉格朗日乘子法，将其转变为非线性方程组的求解；根据传统求解方法——牛顿法的不足，提出了遗传算法求解的方法。该算法以非线性方程组的待求量为决策变量，方程组两端向量的均方差最小值为目标函数，初值在决策变量可行域内随机产生。通过实例分析，遗传算法具有较强的鲁棒性，较之牛顿法不存在对初始值要求近似、易产生局部收敛并含有矩阵求逆的情况，且当初始值偏离真实值较大时，遗传算法求解成功率远远高于牛顿法，验证了遗传算法在求解基于单目标优化的OD反推模型方面的可行性与可靠性。2．针对OD反推中路段检测点设置问题，结合最大可能相对误差概念，提出了改进的路段检测点设置原则：路径覆盖原则和最少检测点原则。在此基础上，建立了求解路段检测点分布的整数规划模型，给出了大规模网络的遗传算法求解步骤。实例分析表明，依据该模型求得的路段检测点设置组合，推算出的OD矩阵误差满足要求，相比已有的检测点设置模型，能够在节省检测费用的基础上最大限度的提高OD反推精度。同时，该模型原理简单明确、求解方便，相对以往的模型，不需要先验OD矩阵和交通分配矩阵，从而减少各项误差的影响，具有更好的适用性。3．针对拥挤网络中交通分配矩阵随OD矩阵变化的特点，提出了OD反推与交通分配交替进行的方法。对于有无先验矩阵的情况分别进行分析，提出了基于不同配流模型和算法的反推过程。在有先验矩阵时根据拥挤网络中平衡配流方法进行分配，并结合OD反推的过程反复修正分配矩阵和OD矩阵，直到能再现路段观测流量；对于无先验OD矩阵的情况，提出了采用概率分配模型获得初始交通分配矩阵的方法，用此分配矩阵和观测流量进行OD反推，得到用于平衡分配的先验OD矩阵，然后再按照有先验矩阵的情况进行OD反推，直到分配的流量与观测流量一致。最后通过实例分析了两种情况下的反推过程与结果，并对无先验OD的情况，与全有全无分配法获取先验OD的反推结果相比较，验证了本文中提出的概率加载法获取先验矩阵的可行性与可靠性。4．论述了作者参与研发的城域混合交通仿真与分析系统(SASUMT)的基本功能、框架及其主要模型，详细介绍了作者针对交通需求模型设计的OD矩阵估计软件包。该软件包不仅可以为SASUMT提供可靠的路口转向流量比数据，也可以结合交通配流软件包，为SASUMT提供路网OD矩阵数据。另外，OD矩阵估计软件包作为相对独立的模块，还可以为交通规划、交通管理等部门提供基本的OD分布预测。最后，结合SASUMT中的一个应用实例，利用OD反推技术进行推算，通过与实测数据的比较，验证了OD反推结果的可靠性，从而为交通仿真系统提供更方便、可靠的输入数据。5．最后，对全文的研究工作进行了总结，对OD反推问题进一步的研究工作提出了一些设想。更多还原

【Abstract】 With the development of the social economy and the speed-up of the urbanization and motorization, traffic jam, traffic accident, environment pollution and energy sources scarcity have been the common problems faced by all the countries in the world. Not only the developed countries but also the developing countries are all enduring the traffic problems. The traditional method resolving the traffic problems is to build roads, but the space of constructing roads has become less and less for anyone country. In addition, the traffic system is a complex and huge system; it is difficult to resolve the problem in essence by only considering the roads. Consequently, the Intelligent Transportation Systems (ITS), which systematically consider traffic basic establishment, traffic means of delivery and traffic participants, emerges as the times require. Generally, ITS include road traffic management system, the planning and design of the entire transportation system and the intelligentization of the transportation management. Traffic planning, traffic management and traffic control all cannot be separated from Origin-Destination (OD) matrices, namely, people needs to know the traffic demands on the road network. In addition, OD matrices are also the most direct and reliable input data for traffic simulation system. Traditional method obtaining OD matrices is to have a spot check in large-scale, but this way needs a vast expense and its organization is very difficult. With the popularization of the traffic monitor and control systems, the information of traffic flows could be obtained easily. As a result, estimating OD matrices by road traffic flows are one of the most economic and feasible methods to obtain OD matrices.For the ordinary urban networks, estimating OD matrix by road traffic flows mainly includes several steps as follows: checking and measuring road flow and obtaining apriori information; procuring road network characters and traffic assignment matrix; estimating OD matrix according to special estimate models. In the above steps, the main factors influencing the precision of OD estimation are: the accuracy of the estimate model, reliability of apriori information, accuracy of road flow checked and measured, and the rationality of the traffic assignment method. In addition, the solution to the model is also a problem worthy of discussion and the feasibility, simplicity and convenience of the solution influence the applicability of the model. Based on the maximum-entropy model and aimed at the OD estimation having the best precision, the solution to the model, traffic counting locations and obtaining traffic assignment matrix are discussed in detail in this dissertation with considering the several key steps ofOD matrix estimation. And then with the application example of the Simulation and Analysis System for Urban Mixed Traffic (SASUMT) designed by Zhejiang University, the input data are supplied for the simulation system by the OD estimation method. The maim research work is summarized as follows:1. Based on the maximum-entropy model estimating OD matrix, the solution by Genetic Algorithm (GA) is proposed. Firstly, the reasoning process and the theory basis of the maximum-entropy model are analyzed. According to the characteristic of the maximum-entropy model, the model is transformed into non-linear equations to resolve by the introduction of the elementary method resolving optimization problem with restrictions——Lagrange multiplier method. Because of the shortcomings of the traditional method——Newton’s method, the calculation method of OD matrix by Genetic Algorithm (GA) is put forward. In this method, the decision-making variables of GA are the unknown quantities of the non-linear equations, the target function for optimization is the minimum of the root-mean-square error between the computational value on the left and the real value on the right in equations, and the initial value is generated randomly in the feasible fields of the decision-making variables. The example shows that GA has the better robustness, and overcomes the disadvantages of the Newton’s method that strictly depends on initial values, doesn’t easily converge and needs to calculate inverse matrices. When the initial values are far from the real values, there are more probabilities solving OD matrix successfully by GA than by Newton’s method. Finally, the feasibility and reliability of GA method in solving OD matrix estimation models based on single-objective optimization are proved.2. In allusion to the problem how to determine the optimal number and locations of traffic counting points in estimating OD matrix by road flows, the improved location rules for traffic counting points are proposed according to the concept of maximum possible relative error: path covering rule and minimal traffic counting point rule. A corresponding integer programming model for determining the locations of traffic counting points satisfying these rules is established; and the approaches by GA for large-scale networks are proposed. The instance analysis shows that the OD matrix estimated by the number and locations of traffic counting points calculated from the above model can satisfy the precision demand. Compared with the other models for traffic counting point locations, this model can improve the precision of OD matrix estimated as soon as possible and at the same time, it can economize the costs of the traffic counting point locations. In addition, the model proposed in this dissertation has simple and explicit principle and can be readily calculated. It doesn’t need apriori OD matrix and traffic assignment matrix, and consequently the influences ofdifferent errors are decreased. Compared with the former models, it has the better applicability.3. According as that traffic assignment matrix changes with the OD matrix in congested networks, the method in which OD matrix estimation and traffic assignment are performed alternately is brought forward. The estimation process based on different traffic assignment model and algorithm is proposed through the analyses to the circumstances with apriori OD matrix or not. When there is apriori OD matrix, equilibrium assignment is adopted in congested networks, and assignment matrix and OD matrix are correct repeated combining the process of OD estimation until the observed road traffic flow is reproduced. And without apriori OD matrix, probability assignment model is adopted to obtain the initial traffic assignment matrix. The apriori OD matrix for equilibrium assignment is procured through OD matrix estimation using the initial assignment matrix and observed traffic flow. Then OD matrix estimation is carried out according to the case where there is apriori OD matrix, and till the assignment flows are consistent with the observed flows. Finally, the estimating process and result in two circumstances are analyzed in an actual example. For the circumstance without apriori OD matrix, the estimation results of the method obtaining initial OD matrix in probability load means are compared with that of the method obtaining initial OD matrix in all-or-nothing means, and the feasibility and reliability of the former proposed in this dissertation are proved.4. The elementary function, the frame and the main models of the Simulation and Analysis System for Urban Mixed Traffic (SASUMT) partly designed by the author are discussed. The OD matrix estimation software designed for the traffic demand model, which can provide not only the data of crossing turning flows, but also the data of road network OD matrix for SASUMT in combination with traffic assignment software, is introduced. In addition, as a relatively independent module, the OD matrix estimation software can supply basic OD distribution forecast for traffic planning and traffic management. Finally, the reliability of the OD estimate result is proved through the simulating analyses of real network applied in SASUMT, and thereby OD matrix estimation can supply convenient, reliable input data for the application of traffic simulation systems.5. Finally, the work of this dissertation is summarized and the prospect of further research in OD matrix estimation is also discussed.更多还原