| 考虑建成环境交互影响的共享单车需求预测 |
| |
| 引用本文： | 魏晋,安实,张炎棠.考虑建成环境交互影响的共享单车需求预测[J].科学技术与工程,2023,23(26):11424-11430. |
| |
| 作者姓名： | 魏晋 安实 张炎棠 |
| |
| 作者单位： | 上海市政工程设计研究总院集团第十市政设计院有限公司;哈尔滨工业大学 |
| |
| 基金项目： | 国家自然科学基金(52272332)；国家自然科学基金青年科学(72201080) |
| |
| 摘 要： | 共享单车的发展有利于交通的节能减排绿色发展。建成环境是影响共享单车出行需求的重要因素，然而很少有学者探究考虑其交互作用。为了准确分析建成环境中各影响因素的交互作用以达到精确预测共享单车出行需求的目的，本文使用了深圳市共享单车出行数据、兴趣点数据（point of interest，POI）、路网数据和公交线路数据等多源数据，采用梯度提升决策树(Gradient Boosting Decision Tree，GBDT)模型预测共享单车出行需求，并与BP(Back Propagation)神经网络模型预测结果进行比较；最后借助SHAP(SHapley Additive explanation)方法解释GBDT模型中各种影响因子对共享单车出行需求产生的影响，并分析各影响因素及其交互作用。实验结果表明：GBDT模型预测结果平均绝对误差为0.683，均方根误差为0.728，较BP神经网络模型预测准确性更高；通过SHAP方法发现自行车道密度、公交站点数等交通属性因素对于共享单车出行需求作用明显，土地利用中土地利用混合度不是简单线性作用且不同POI间存在复杂交互关系。可见通过借助GBDT模型和SHAP方法可以用来共享单车出行需求预测以及影响因素分析，从而为共享单车发展提出改善建议。
|
| 关 键 词： | 共享单车 需求预测 POI数据 梯度提升决策树 SHAP |
| 收稿时间： | 2023/1/29 0:00:00 |
| 修稿时间： | 2023/7/5 0:00:00 |
Prediction of Shared Bicycle Demand Based on Environmental Factor Interactions |
| |
| Wei Jin,An Shi,Zhang Yantang.Prediction of Shared Bicycle Demand Based on Environmental Factor Interactions[J].Science Technology and Engineering,2023,23(26):11424-11430. |
| |
| Authors: | Wei Jin An Shi Zhang Yantang |
| |
| Affiliation: | Shanghai Municipal Engineering Design and Research Institute Group Tenth Municipal Design Institute Co.;Herbin Institute of Technology |
| |
| Abstract: | The advancement of bike-sharing infrastructure is a pivotal contributor to the promotion of energy conservation and environmentally-friendly transportation. Undoubtedly, the constructed environment plays a vital role in determining the frequency and popularity of bike-sharing usage. Limited research has been conducted on this important interaction. In order to accurately analyze the interaction of various factors in the built environment to accurately predict the demand for shared bicycle trips, this study crawls through multiple sources of data such as shared bicycle trips, points of interest data, road network data and bus route data in Shenzhen, and uses the Gradient Boosting Decision Tree (GBDT) model to predict the demand for shared bicycle trips. Finally, with the help of the SHAP (SHapley Additive explanation) method, the influence of the various influencing factors in the GBDT model on the demand for bicycle sharing trips is explained, and the influencing factors and their interactions are analysed. Results show that the average absolute error of the GBDT model prediction results is 0.683 and the root mean square error is 0.728, which is higher than the prediction accuracy of the BP neural network model; The SHAP method reveals that traffic attributes such as the density of bicycle lanes and the number of bus stops play a significant role in the demand for shared bicycle trips, and that the land use mix is not simply linear and there are complex interactions between different POIs. The GBDT model and the SHAP method can be employed to predict the demand for shared bicycle trips and analyse the influencing factors, so as to suggest improvements for the development of shared bicycle trips. |
| |
| Keywords: | shared bicycle demand forecast POI data gradient boosting decision tree(GBDT) SHAP |
|
| 点击此处可从《科学技术与工程》浏览原始摘要信息 |
|
点击此处可从《科学技术与工程》下载全文 |
