城市公共自行车站点需求预测及调度优化方法研究

发布时间：2018-02-22 12:19

本文关键词： 公共自行车需求预测调度优化 BP神经网络关联规则 K-medoids 聚类蚁群算法　出处：《北京交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：城市公共自行车兼具承担公共交通"最后一公里",服务短距离出行和服务旅游景区及校园三大主要功能,是新时代人文、科技、绿色交通的产物。面对目前诸多城市出现的自行车站点无车可借,无位可还的现象,究其主要原因,是调度的不及时所致。因此,如何提前准确预知各站点出行需求及安排合理的调度路径方案对于解决调度的滞后性具有重要意义。由于国内城市可获取的PBS(public bicycle system)运营数据较少,因此本文选取美国纽约市的出行数据进行PBS出行需求及调度优化的研究。在对纽约市的公共自行车出行需求分析时,发现其出行规律和国内城市相似——工作日均存在早晚高峰的现象。另外,加上数据样本较大,可靠性强,因此基于纽约市出行数据的预测方法同样适用于国内城市。本文具体的研究工作包含以下几个方面:(1)对公共自行车的系统功能定位进行了阐述,并结合美国纽约市2016年6月份146万条公共自行车出行数据对出行需求进行了主客观因素分析,发现周末与工作日出行规律明显不同。(2)简单介绍了 BP神经网络理论,提出一种新的考虑日期属性的站点租还车需求预测方法。在利用BP神经网络预测时,输入变量的日期属性应与输出变量保持一致,即同为工作日或同为周末。使用SQL Server数据库和Matlab软件对实例进行了预测,效果较好,预测相对误差均较低,最低预测相对误差可达到1%。(3)概述了车辆调度问题的定义、分类及常用的优化算法,并将城市公共自行车的调度问题分为调度分区和区域内调度两步。首先对整个城市的自行车站点进行了调度区域的划分,在此提出了一种新的将关联规则与聚类分析相结合的划分方法,通过三次聚类,将城市所有站点划分成不同的静态调度区域。其次,对于每个静态区域内有调度需求的站点建立调度模型,目标是使调度车数量最少和总调度时间最短。并介绍了带2-opt局部搜索的蚁群算法求解模型的步骤与方法。(4)以美国纽约市21个站点数据为例,进行调度路径的优化仿真。通过MapGIS软件转换坐标和Matlab编程求解,发现带2-opt局部搜索的蚁群算法比常规蚁群算法总调度时间更短,效果更好。
[Abstract]:Urban public bicycles have the three main functions of undertaking the "last kilometer" of public transport, serving short distance travel, serving scenic spots and campus. They are the humanities and science and technology of the new era. The product of green traffic. Facing the phenomenon that there is no car to borrow or return to the bicycle station in many cities at present, the main reason is that the dispatch is not in time. Therefore, How to predict the travel demand of each station in advance and how to arrange a reasonable scheduling path plan is of great significance to solve the lag of scheduling. Because of the lack of PBS(public bicycle system operation data available in domestic cities, Therefore, this paper selects the travel data of New York City to study the PBS travel demand and scheduling optimization. It is found that the travel pattern is similar to that of domestic cities and that there are morning and evening peaks on weekdays. In addition, the data samples are large and reliable. Therefore, the prediction method based on New York City travel data is also applicable to domestic cities. Combined with the data of 1.46 million public bicycle trips in New York City on June 2016, this paper analyzes the subjective and objective factors of travel demand, and finds that the travel law of weekends is obviously different from that of weekdays.) the BP neural network theory is briefly introduced. In this paper, a new forecasting method of station rental and return demand considering date attribute is proposed. When using BP neural network, the date attribute of input variable should be consistent with that of output variable. Using SQL Server database and Matlab software to predict the example, the result is better, the relative error of prediction is lower, and the lowest relative error of prediction can reach 1 / 3) the definition of vehicle scheduling problem is summarized. The scheduling problem of urban public bicycle is divided into two steps: the scheduling partition and the intra-area scheduling. Firstly, the scheduling area of the bicycle station in the whole city is divided. In this paper, a new partitioning method combining association rules with clustering analysis is proposed, which divides all stations into different static scheduling areas through three clusters. Establish a scheduling model for each site with scheduling requirements within a static region, The aim is to minimize the number of dispatching vehicles and the shortest total scheduling time. The steps and methods of solving the model of ant colony algorithm with 2-opt local search are introduced. The optimization simulation of scheduling path is carried out. Through MapGIS software coordinate transformation and Matlab programming, it is found that the total scheduling time of the algorithm with 2-opt local search is shorter and the effect is better than that of the conventional ant colony algorithm.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491.225

【参考文献】