基于海量IC卡数据的乘客出行网络及动力学研究

发布时间：2018-04-29 08:06

本文选题：复合网络 + 复杂网络　；参考：《西南大学》2017年硕士论文

【摘要】：在现代社会中公共交通系统一直充当着重要的角色,无论在实践中还是学术中,交通问题极大地引起了各界的注意。伴随着交通工具的不断普及,乘客出行量逐年增加,公交与地铁换乘不协调、交通拥挤及城市公共交通系统整体运营效率低等问题日益凸显。如何提升城市公共交通系统的运输效率,已成为交通领域的热点话题。在如今大数据的时代背景下,通过对海量、多样化的交通数据进行挖掘,不仅能够分析出公共交通网络的拓扑特征,同时也能挖掘出乘客的出行行为规律,对提升公交与地铁之间的高效配合以及公共交通的综合运输能力具有重要的现实意义。首先,本文通过构建公交-地铁复合网络,并分析了复合交通网络的拓扑性质及鲁棒性;其次,利用基于多头绒泡菌仿生模型改进的粒子群算法,对地铁乘客出行网络进行社团划分;最后,采用非负矩阵分解-自回归模型,对地铁乘客动态起讫(Origin Destination,OD)矩阵进行预测。本文的主要贡献如下:(1)实现构建公交-地铁复合网络,同时对比分析复合网络与子网络的拓扑特性及鲁棒性:通过采用两种建模方式(Space L方式、Space P方式)构建了公交-地铁复合站点网络及公交-地铁复合换乘网络,并对比分析两种复合网络与其相应模式下公交子网络与地铁子网络的拓扑特征值。此外,对比分析复合网络与子网络在不同攻击模式下的鲁棒性指标,即最大连通子图相对大小、平均路径长度、网络直径、网络性能参数等鲁棒性指标的变化情况。并以中国西部某市的公交网络及地铁网络数据进行实证分析,结果表明:该市公交-地铁网络复合网络、公交子网络、地铁子网络都是小世界网络,且具有无标度特性。复合网络在随机攻击模式下的鲁棒性较强,然而在目标攻击模式下较弱;对于这两种攻击模式,公交-地铁复合站点网络的鲁棒性均优于公交子网络和地铁子网络。(2)实现对地铁乘客出行网络的社团划分:通过引入多头绒泡菌模型得到目标函数即网络模块度的粗略解,并以此作为初始解,结合粒子群算法对模块度函数进行优化求解,从而完成对乘客出行网络的社团划分。以中国西部某市交通IC卡信息为基础,构建地铁乘客出行网络,并同粒子群算法进行对比,实验结果表明:在对加权网络进行社团挖掘时,基于多头绒泡菌网络模型改进的粒子群算法在解的可行性方面有了明显提升。(3)实现基于非负矩阵分解-自回归的算法来对地铁乘客动态OD矩阵进行预测:首先,通过非负矩阵分解得到地铁乘客的出行特征量,然后基于非负矩阵分解得到系数矩阵建立自回归模型,从而完成对地铁乘客出行流量的预测。并以中国西部某市的地铁乘客流量数据为基础,通过与K近邻、C4.5、朴素贝叶斯、随机森林等回归算法进行对比,实验结果表明,该算法的预测准确率有显著提升。
[Abstract]:In modern society, the public transportation system has been playing an important role, whether in practice or academic, traffic problems have attracted great attention from all walks of life. With the continuous popularization of transportation, passenger travel volume increases year by year, bus and subway transfer is not coordinated, traffic congestion and the urban public transport system as a whole low efficiency and other problems become increasingly prominent. How to improve the transport efficiency of urban public transport system has become a hot topic in the field of transportation. Under the background of big data's times, through mining massive and diversified traffic data, not only can the topological characteristics of public transport network be analyzed, but also the travel behavior rules of passengers can be mined. It is of great practical significance to improve the efficient cooperation between public transportation and subway and the comprehensive transportation capacity of public transportation. First of all, this paper constructs the bus-subway complex network, and analyzes the topological properties and robustness of the complex traffic network. Secondly, using the improved particle swarm optimization algorithm based on the bionic model of polycephalus, Finally, the non-negative matrix factorization-autoregressive model is used to predict the dynamic origin derivation of subway passengers. The main contributions of this paper are as follows: (1) realizing the construction of a bus-subway composite network. At the same time, the topological characteristics and robustness of the composite network and the sub-network are compared and analyzed. By using two modeling methods, the bus-subway compound station network and the bus-subway complex transfer network are constructed by adopting two modeling methods: the space L mode and the space P mode. The topological eigenvalues of two kinds of complex networks and their corresponding modes are compared and analyzed. In addition, the robustness indexes of composite network and subnetwork under different attack modes are compared and analyzed, such as the relative size of maximum connected subgraph, average path length, network diameter, network performance parameters, and so on. Based on the data of public transport network and subway network in a certain city in western China, the results show that the bus subway network, bus subnetwork and subway subnetwork are small world networks, and have scale-free characteristics. The robustness of compound network in random attack mode is stronger than that in target attack mode. The robustness of the bus-subway complex station network is better than that of the bus sub-network and the subway sub-network. It realizes the community division of the subway passenger travel network. The objective function, that is, the rough solution of the network modularity, is obtained through the introduction of the multi-headed bacterial model. As the initial solution, the modular degree function is solved optimally with particle swarm optimization (PSO), and the community partition of passenger travel network is completed. Based on the IC card information of a certain city in western China, the subway passenger travel network is constructed and compared with the particle swarm optimization algorithm. The experimental results show that: when mining the weighted network, The improved particle swarm optimization algorithm based on the multi-headed Particle Swarm Optimization (PSO) model has significantly improved the feasibility of the solution. (3) based on the non-negative matrix factorization and autoregressive algorithm to predict the dynamic OD matrix of subway passengers: first of all, The travel characteristic quantity of subway passengers is obtained by non-negative matrix decomposition, and then the autoregressive model is established based on non-negative matrix decomposition to predict the travel flow of subway passengers. Based on the data of subway passenger flow in a certain city in western China, the prediction accuracy of this algorithm is obviously improved by comparing it with K-nearest neighbor C4.5, naive Bayes, random forest and other regression algorithms.
【学位授予单位】：西南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U12;O157.5

【参考文献】