论文总字数:59842字
摘 要
租赁自行车主要分为两种模式:有桩的公共自行车和无桩的共享单车。目前,两种模式各自为政,运营管理难以联动,高峰时期借还车需求存在明显的时空分布不均衡性,严重制约了租赁自行车的运营效率。鉴于此,本文基于公共自行车 IC 卡数据和共享单车 APP 数据,提出了租赁自行车协同调度子区划分方法,构建了租赁自行车短时需求预测模型,在此基础上研究了租赁自行车协同调度方法。
首先,运用 Python 软件对公共自行车 IC 卡数据和共享单车 APP 数据进行数据清洗和预处理,删除无效数据及逻辑错误数据;基于 DBSCAN 聚类算法,生成共享单车推荐停车点,类似于公共自行车站点;分别生成公共自行车和共享单车的借还车流量时间序列数据。
然后,考虑公共自行车和共享单车的借还流量差以及两种模式之间不能在站点互通,结合 K 均值算法和 DBSCAN 算法,提出了基于租赁自行车借还车流量的 DBSCAN 改进算法,动态划分调度子区。
其次,考虑到租赁自行车借还车流量信息具备时间序列数据的特性,分别构建长短时记忆神经网络模型(LSTM)和卷积-长短时记忆神经网络模型(ConvLSTM)预测租赁自行车站点借还车流量。结果表明,同时考虑了时空信息的 ConvLSTM 模型的预测精度和运行效率均优于 LSTM 模型。
最后,基于 ConvLSTM 模型预测结果,设置站点启动调度的阈值,结合调度成本、调入调出需求、自行车类别约束等因素,构建了动态协同调度路径优化模型,并提出相应的改良蚁群算法进行求解。结果表明,本文提出的调度模型和算法可以实现两类租赁自行车的协同调度,并能降低调度成本,提高运营效率。
关键词:租赁自行车,调度子区,短时需求预测,动态调度,蚁群算法
Abstract
Bikeshare is mainly divided into two modes: docked bikeshare and dockless bikeshare. At present, the two modes are independent of each other, and it is difficult to connect the operation management. During the peak period, there is an obvious imbalance in the spatio-temporal distribution in the demand, which seriously restricts the operational efficiency of bikeshare. In view of this, this paper proposes a method for sub-division of bikeshare based on the docked bikeshare IC card data and dockless bikeshare APP data and constructs a short-term demand forecasting model for bikeshare. Based on this, the cooperative bikeshare scheduling method is studied.
Firstly, the Python software is used to clean and preprocess the docked bikeshare IC card data and the dockelss bikeshare APP data, delete invalid data and logical error data. Generate a dockless bikeshare recommended parking spot, which is similar to a public docked bikeshare station, based on the DBSCAN clustering algorithm. The time-series data of the borrowing and returning traffic of the docked bikeshare and the dockless bikeshare are separately generated.
Then, considering the difference of borrowing and returning traffic between docked bikeshare and dockless bikeshare, and the inability to intercommunicate between the two modes at the station. Combined with K-means algorithm and DBSCAN algorithm, an improved DBSCAN algorithm based on bikeshare loaning traffic is proposed, which dynamically divides the dispatcher.
Secondly, considering the characteristics of time-series data of bikeshare traffic information, construct long-short-time memory neural network model (LSTM) and convolution-long-short-term memory neural network model (ConvLSTM) to predict the traffic of bikeshare stations. The results show that the ConvLSTM model considering both spatio-temporal information has better prediction accuracy and operational efficiency than the LSTM model.
Finally, based on the ConvLSTM model prediction results, the threshold of the site start-up scheduling is set, and the dynamic cooperative scheduling path optimization model is constructed based on the factors such as scheduling cost, call-in and call-out requirements, and bicycle category constraints. To solve this problem, an improved ant colony algorithm is proposed. The results show that the scheduling model and algorithm proposed in this paper can realize the coordinated scheduling of two types of bikeshare, and can reduce the scheduling cost and improve the operation efficiency.
KEY WORDS: bikeshare, scheduling sub-area, short-term demand forecasting, dynamic scheduling, ant colony algorithm
目 录
摘要 I
Abstract I
第一章 绪论 1
1.1 研究背景及其意义 1
1.2 国内外研究现状 1
1.3 研究主要内容和技术路线 4
第二章 租赁自行车数据的获取与处理 7
2.1 南京市租赁自行车概况 7
2.2 数据来源与结构 8
2.2.1 公共自行车数据结构 8
2.2.2 共享单车数据结构 9
2.3 数据预处理 9
2.3.1 公共自行车数据预处理 9
2.3.2 共享单车数据预处理 12
2.4 本章小结 14
第三章 基于 DBSCAN 改进算法的调度子区动态划分方法 16
3.1 调度子区划分方法 16
3.1.1 K-Means 算法介绍 16
3.1.2 基于流量的 DBSCAN 改进算法 17
3.2 调度子区划分结果 20
3.3 本章小结 23
第四章 基于ConvLSTM模型的租赁自行车短时需求预测方法 24
4.1 短时需求预测方法选择 24
4.2 长短时记忆神经网络模型(LSTM)简介 25
4.3 卷积-长短时记忆神经网络模型(ConvLSTM)简介 28
4.3.1 全连接-长短时记忆神经网络(FC-LSTM) 28
4.3.2 CNN 神经网络 29
4.3.3 ConvLSTM 神经网络 31
4.4 LSTM 和 ConvLSTM 神经网络的建模实现 32
4.4.1 LSTM 的建模实现 32
4.4.2 ConvLSTM 的建模实现 37
4.4.3 预测结果对比 41
4.5 本章小结 42
第五章 租赁自行车动态协同调度方法 43
5.1 调度子区内调度站点选择 43
5.2 调度车辆路径优化问题 44
5.3 协同调度车辆路径优化模型建立 45
5.3.1 模型假设 45
5.3.2 变量定义 46
5.3.3 模型建立 46
5.4 蚁群算法求解模型 47
5.4.1 基本原理及算法流程 48
5.4.2 协同调度路径优化模型下的改良算法 50
5.5 实例分析 51
5.5.1 调度子区选取 51
5.5.2 调度站点及需求确定 52
5.5.3 蚁群算法求解调度路径 53
5.6 本章小结 55
第六章 总结与展望 56
6.1 研究内容和主要结论 56
6.2 创新点 57
6.3 未来研究展望 57
参考文献 59
本科期间个人成果 63
致 谢 65
绪论
研究背景及其意义
如今,在绿色交通、共享交通等理念的普及下,租赁自行车这一具备有效缓解交通拥堵与节能减排等特点的绿色交通方式,在中国,乃至世界范围内得到了广泛关注。租赁自行车主要用于中短途旅行,能够解决“最后一公里”的交通问题,同时租赁自行车能够弥补轨道交通和地面常规公交难以到达的街区。该模式不仅满足出行方式的多元化的需求,促进绿色交通的发展,而且有助于改善居民身体健康、提升居民生活品质。
剩余内容已隐藏,请支付后下载全文,论文总字数:59842字
该课题毕业论文、开题报告、外文翻译、程序设计、图纸设计等资料可联系客服协助查找;