论文总字数：67515字

摘 要

随着智能交通系统的发展，海量的公交运营大数据为多方位监测公交运行状态、提升公交运营服务水平提供了契机。但与此同时，如何从繁杂海量的公交运行数据中挖掘出有效的信息并做出科学的运营决策，成为当前智慧公交系统建设亟需解决的核心问题。而在高品质智慧公交服务的建设中，准确的公交行程时间信息的提供对减少乘客在站点等待时间、缓解乘客等待焦虑至关重要。

故本文为提高传统公交行程时间预测方法的准确度，对公交运行过程进行细致研究。首先，突破传统人工调查方法的弊端，融合多源公交大数据——公交自动定位AVL数据、乘客IC卡数据、线路站点静态数据，提取前方公交车辆的站间行程时间、站间距离、运营时段、上车乘客数量等影响因素，用于综合分析公交车辆运行时空动态变化特性。

而随着人工智能算法的发展，其广泛应用吸引了众多学者的注意。故本文为探究人工智能算法在公交行程时间预测问题上的适用性，使用多种机器学习模型——多元线性回归、支持向量机、随机森林、长短期记忆递归神经网络对公交车辆的行程时间进行预测，从多指标角度评估对比模型效果，探究不同模型的表现和实用性。

最后，以常州市客流走廊通江路为例进行实例分析，对快速公交线路和常规公交线路的站间行程时间均进行建模分析。结果表明，支持向量机模型在本文提出的四个模型中表现最优，其预测平均绝对误差在50秒以内，极大地改善了传统公交报站的误差，适用于实时公交行程时间预测问题。

关键词：公交行程时间预测，公交大数据，支持向量机模型，机器学习

Abstract

With the development of intelligent transportation system, massive data brought from transit operating provides opportunities for supervising the bus operating state and improving its level of service. But at the meantime, the core problem is how to find the useful information from the complex and massive data and make wise decisions correspondingly. To achieve high-quality bus services, accurate bus arrival time information is essential to reduce passengers’ waiting time at the station and ease their anxiety.

This paper researched transit operating process aiming to improve the accuracy of traditional methods for predicting transit travel time. Firstly, breaking through the drawbacks of traditional manual investigation methods, auto vehicle location data, auto passenger counting data and the static geographic data were matched and integrated. Then, four influencing characteristics, bus travel time, distance between stations, operating time and passenger volume, were extracted for analyzing the dynamic spatiotemporal states of transit operating.

Widespread applications of artificial intelligence algorithm also attract many researchers’ attentions. In order to explore the applicability of artificial intelligence algorithm for predicting transit travel time, this paper tried four machine learning models including multiple linear regression model, support vector machine model, random forests model and long-short term memory recurrent neural network model, which were evaluated by multiple metrics.

Lastly, taking the bus data of Tongjiang Road, Changzhou Passenger Flow Corridor, from May to June in 2017 as a case study, both BRT line and regular bus line were analyzed. The results show that the support vector machine model is the best among the four proposed models in this paper. The mean average absolute error is less than 50 seconds, which shows good performance for real-time bus arrival prediction.

KEY WORDS: transit arrival prediction, transit big data, support vector machine, machine learning

目 录

第一章 绪论 1

1.1 研究背景 1

1.2 国内外研究现状 2

1.2.1 智能公交系统的建设发展 2

1.2.2 公交行程时间预测研究综述 3

1.3 研究内容及技术路线 5

第二章 公交车辆运行大数据预处理与分析 7

2.1 公交多源数据介绍与预处理 7

2.1.1 线路站点数据 7

2.1.2 公交IC卡数据 8

2.1.3 公交AVL数据 9

2.2 公交行程时间提取与站点客流匹配 10

2.2.1 站间行程时间提取 10

2.2.2 站点车辆客流提取 11

2.3 本章小结 12

第三章 城市地面公交行程时间预测方法 13

3.1 公交行程时间预测方法概述 13

3.1.1 多元线性回归模型介绍 13

3.1.2 支持向量机模型介绍 13

3.1.3 随机森林模型介绍 14

3.1.4 长短期记忆递归神经网络模型介绍 15

3.2 模型评价指标 17

3.3 本章小结 17

第四章 地面公交行程时间预测实例分析 18

4.1 通江路交通现状与特征分析 18

4.2 影响公交行程时间的特征分析 19

4.3 公交站间行程时间预测 20

4.3.1 基础特性分析 20

4.4 实验设计及结果分析 22

4.4.1 多元线性回归模型结果 22

4.4.2 支持向量机模型和随机森林模型参数选择 24

4.4.3 长短期记忆递归神经网络参数选择 25

4.4.4 结果对比与分析 26

4.5 本章小结 30

第五章 总结与展望 32

5.1 论文总结 32

5.2 论文创新点 32

5.3 研究展望 33

致 谢 34

参考文献 35

附录A 37

剩余内容已隐藏，请支付后下载全文，论文总字数：67515字