I 摘要 近年来共享单车快速兴起，但发展过程中存在许多问题，特别是单车供需不 匹配，这类问题造成资源错配和浪费，本文基于天气指标对区域单车的骑行指标 进行预测研究，研究中选取骑行量作为骑行指标，选取气温和下雨量作为天气指 标。 基于机器学习的神经网络具备诸多优点：学习能力、泛化能力、高适应能力 以及非线性映射能力，通过与线性回归、移动平滑法等几种传统方法做比较，本 文选取BP神经网络用于预测研究，并建立模型，以成都市成华区建设路区域的骑 行数据进行实证研究，通过网络参数的对比研究，逐步优化网络结构，同时对输 入指标进行优化，提升网络模型的效果，用这个优化模型对以“日”为粒度的骑 行量做了预测研究，并用五折交叉验证法进行了预测效果验证，与多元线性回归 模型和ARIMA模型进行比较，结果表明BP神经网络的定量预测效果明显好于后 两者。 最后对“小时”为粒度的骑行数据做了基于时空变量下的骑行特征研究，定 量分析了区域内不同地点不同时段下的天气因素对骑行量的影响特征，对该区域 内8个典型地点或06:00~23:59区间每小时的骑行量做了预测和对比研究。研究表 明，不同地点或不同时段受天气因素影响的特征差距很大，个别地点的骑行特征 与区域内总样本的骑行特征是截然不同的，这些研究结果为单车公司的车辆投放 和调度管理提供了更详实的参考依据。 关键词：天气，共享单车，骑行量，预测，BP神经网络 ABSTRACT II ABSTRACT In recent years, shared bicycles have been rising rapidly, but there are many problems in the development process, especially the mismatch between supply and demand of bicycles, which results in resource mismatch and waste. In this paper, based on the weather indicators, the riding indicators of regional bicycles are predicted. In the research, the ride volume is selected as the riding indicators, and the temperature and rain are selected as the weather indicators. Neural networks based on machine learning have many advantages: learning ability, generalization ability, high adaptability and nonlinear mapping ability. By comparing with several traditional methods such as linear regression and moving smoothing method, this paper selects BP neural network for prediction Research and build a model to conduct empirical research on the cycling data of the Jianshe Road area of Chenghua District, Chengdu. Through the comparative study of network parameters, the network structure is gradually optimized, and the input indicators are optimized to improve the effectiveness of the network model. with this optimization model，This paper made a prediction study on the riding volume with the "day" as the granularity, and the prediction performance was verified by the five-fold cross-validation method. Compared with the multiple linear regression model and ARIMA model, the results shows that the quantitative prediction performance of the BP neural network is significantly better than the latter models. Finally, the cycling characteristics of the cycling data with "hour" as the granularity were studied based on the spatio-temporal variables, and the influence characteristics of the weather factors on the ride volume at different locations or different periods in the region were quantitatively analyzed. The ride volume per hour at 8 typical locations or the interval between 06:00~23:59 was predicted and compared. The research shows that the characteristics of different locations affected by weather factors vary greatly, so as the characteristics of different periods of time, These research results provide a more detailed reference for the vehicle delivery and scheduling management of shared bicycle companies. Keywords: weather, shared bicycles, ride volume, prediction, BP neural network 目录 III 目录 摘要 ................................................................................................................................... I ABSTRACT ..................................................................................................................... II 目录 ................................................................................................................................ III 第一章 绪论 .................................................................................................................... 1 1.1 研究背景 ............................................................................................................ 1 1.2 研究目的和意义 ................................................................................................ 1 1.2.1 研究目的 ................................................................................................ 2 1.2.2 研究意义 ................................................................................................ 2 1.3 国内外研究现状 ................................................................................................ 3 1.3.1 国外研究现状 ........................................................................................ 3 1.3.2 国内研究现状 ........................................................................................ 5 1.3.3 存在的问题 ............................................................................................ 7 1.4 研究内容和方法 ................................................................................................ 8 1.4.1 研究内容 ................................................................................................ 8 1.4.2 研究路线 ................................................................................................ 8 1.5 预期的成果和创新成果 .................................................................................... 9 1.6 本章小结 ............................................................................................................ 9 第二章 共享单车和预测理论介绍 ............................................................................... 11 2.1 共享单车知识 ................................................................................................... 11 2.1.1 共享单车简介 ....................................................................................... 11 2.1.2 共享单车用户特征 ............................................................................... 11 2.1.3 共享单车骑行特征 .............................................................................. 14 2.2 预测方法模型与方法概要 .............................................................................. 16 2.2.1 多元线性回归模型 .............................................................................. 16 2.2.2 时间序列平滑法模型 .......................................................................... 17 2.2.3 ARIMA模型 ......................................................................................... 17 2.2.4 BP神经网络模型 ................................................................................. 18 2.3 BP神经网络原理 ............................................................................................ 20 2.3.1 神经元模型介绍 .................................................................................. 20 2.3.2 BP神经网络结构 ................................................................................. 21 目录 IV 2.3.3 学习原理 .............................................................................................. 22 2.4 本章小结 .......................................................................................................... 26 第三章 问题描述与数据准备 ...................................................................................... 27 3.1 问题描述 .......................................................................................................... 27 3.2 数据获取和预处理 .......................................................................................... 27 3.2.1 数据获取 .............................................................................................. 27 3.2.2 数据清洗和预处理 .............................................................................. 28 3.3 选择预测模型 .................................................................................................. 31 3.4 本章小结 ................................................................................