文章编号： 1672-5328（2025）05-0090-12

耿天霜1，孟丹青1，戴骏晨2，凌小静2
(1. 南京地铁建设有限责任公司，江苏南京210008；2.中咨城建设计有限公司江苏分公司，江苏南京210012)

摘要： 区域出行在频率和规律性方面显著区别于城市出行，需构建针对性的出行需求模型，而该类 模型相较于传统方法的具体提升效果及低成本实现路径尚缺乏充分实证研究。以南京都市圈为例， 基于手机信令和小样本调查数据，重点针对出行生成与出行分布两个阶段开展建模方法研究，并通 过划分训练集与测试集进行模型验证与对比分析。提出一种融合小样本调查数据与出行大数据的建 模流程：首先通过调查数据标定离散选择模型，再借助手机信令数据校正泊松回归模型，在保证精 度的同时保留对个人特征的解释力。研究表明：未引入个人特征及起终点交互变量时，目的地选择 模型性能甚至弱于重力模型；而加入交互变量后，其表现显著优于重力模型，嵌套Logit 模型中 Nest 结构的合理构建可进一步优化效果；所提出的方法在出行生成和分布阶段精度较传统模型分别 提升24%和18%。

关键词： 区域出行需求；离散选择模型；出行生成；出行分布；出行频率；目的地选择；交互效 应；南京都市圈

中图分类号： U491

文献标识码：A

Key Technologies for Constructing Regional Travel Demand Models: A Case Study of the Nanjing Metropolitan Area

GENG Tianshuang1, MENG Danqing1, DAI Junchen2, LING Xiaojing2
(1. Nanjing Metro Construction Co., Ltd., Nanjing Jiangsu 210008, China; 2. Jiangsu Branch, CIECC Urban Construction Design Co., Ltd., Nanjing Jiangsu 210012, China)

Abstract： Regional travel differs substantially from urban travel in terms of frequency and regularity, thus requiring a specialized travel demand model. However, current empirical research is insufficient regarding the specific improvements and cost-effective implementation pathways of such models compared to conventional modeling approaches. Using the Nanjing Metropolitan Area as a case study, this paper focuses on modeling methodology research on the trip generation and trip distribution stages based on mobile signaling data and small-sample survey data. Model validation and comparative analysis are conducted by dividing the data into training and testing datasets. The paper proposes a modeling framework that integrates small-sample survey data with large-scale travel data: a discrete choice model is first calibrated using survey data, followed by a Poisson regression model corrected through mobile signaling data, thereby ensuring the accuracy while preserving explanatory power with respect to individual attributes. Findings indicate that, when individual attributes and origin- destination interaction variables are excluded, the performance of the destination choice model is weaker than that of gravity models. After incorporating interaction variables, however, the model significantly outperforms gravity models, and a reasonable construction of nesting structures within nested Logit models can further enhance the results. The proposed approach improves the modeling accuracy in trip generation and distribution stages by 24% and 18%, respectively, compared with traditional models.

Keywords： regional travel demand; discrete choice model; trip generation; trip distribution; travel frequency; destination choice; interaction effects; Nanjing Metropolitan Area