| 146 | 0 | 68 |
| 下载次数 | 被引频次 | 阅读次数 |
【背景】交通需求预测对于交通基础设施规划和资源分配至关重要。在时空需求预测中，模型性能在不同站点和时间段之间差异较大，因此在时间和空间维度上实现稳定精准预测构成了挑战。【目的】实现模型跨场景直接部署，提升预测模型在不同时空场景的实用性，以减少预测不确定对未来决策的影响。【方法】首先，提出了考虑图特征的时空交通需求集成预测模型（ST-SGAT），采用堆叠的集成学习策略融合不同的基模型和图注意力网络（GAT），ST-SGAT将长短时记忆网络、梯度提升决策树、多层感知机作为基模型捕获时间特征，并利用GAT对空间相关性进行编码，自适应分配关联权重，从而实现时空交通需求的稳健建模。随后，通过删除ST-SGAT的特定组件实现消融实验，系统分析各组件的贡献度与协同效应，并引入降水量作为外部特征，进一步探究模型在常规与突发气象场景下的鲁棒性与特征利用能力。【数据】研究数据来源于纽约市424个地铁站点688天的每小时客流量数据，预处理后生成2 h、3 h、6 h、12 h、24 h等多时间粒度数据集。【结论】ST-SGAT模型在交通时空需求预测中综合性能最优，时空适配性突出。时间维度上，于小时级至日级多粒度数据及工作日、周末数据中均展现最优综合预测精度；空间维度上，可自适应适配不同站点类型，泛化能力强。集成框架能够实现多组件协同互补，使模型在跨粒度、跨站点、跨时间类别的多场景中持续保持优异性能。同时，模型在常规与突发降水场景下均表现出更强的鲁棒性与特征利用稳定性，满足实际交通的多样化预测需求。
Abstract：[Background] Traffic demand forecasting is crucial for transportation infrastructure planning and resource allocation. In spatiotemporal demand forecasting， model performance varies significantly across different stations and time periods， posing a challenge to achieving stable and accurate predictions in both temporal and spatial dimensions. [Objective] To realize direct cross-scenario deployment of the model and enhance the practicality of forecasting models across diverse spatiotemporal scenarios， thereby reducing the impact of prediction uncertainty on future decision-making. [Method] A spatiotemporal traffic demand ensemble forecasting model (ST-SGAT) considering graph features is proposed， which adopts a stacking ensemble learning strategy to integrate multiple base models with Graph Attention Network (GAT). ST-SGAT employs Long Short-Term Memory， eXtreme Gradient Boosting， and Multilayer Perceptron as base models for capturing temporal features， while leveraging GAT to encode spatial correlations and adaptively assign association weights， thereby achieving robust modeling of spatiotemporal traffic demand. Subsequently， ablation experiments are conducted by removing specific components of ST-SGAT to systematically analyze the contribution degree and synergistic effects of each component， and precipitation is introduced as an external feature to further explore the model’s robustness and feature utilization capability under normal and sudden meteorological scenarios. [Data] The research data are derived from hourly passenger flow records of 424 subway stations in New York City over 688 days， which are preprocessed to generate multi-time granularity datasets (e.g.， 2 h， 3 h， 6 h， 12 h， 24 h). [Conclusion] Experimental results demonstrate that the ST-SGAT model achieves the optimal comprehensive performance in spatiotemporal traffic demand forecasting with outstanding spatiotemporal adaptability. In the temporal dimension， it exhibits the best overall prediction accuracy across multi-time granularity data (from hourly to daily) as well as data for weekdays and weekends. In the spatial dimension， it can adaptively fit different station types， showing strong generalization ability. The ensemble framework enables the synergy and complementarity of multiple components， allowing the model to consistently maintain excellent performance in multiple scenarios spanning different granularities， stations， and time categories. Meanwhile， the model exhibits stronger robustness and stability in feature utilization under both normal and sudden precipitation scenarios， thus meeting the diverse forecasting needs of practical transportation systems.
[1] BOUKERCHE A， WANG J. Machine learning-based traffic prediction models for intelligent transportation systems[J]. Computer Networks， 2020， 181: 107530.
[2] ROCCO DI TORREPADULA F， NAPOLITANO E V， DI MARTINO S， et al. Machine learning for public transportation demand prediction: a systematic literature review[J]. Engineering Applications of Artificial Intelligence， 2024， 137: 109166.
[3] WANG H， YI W， TIAN X， et al. Prescriptive analytics for intelligent transportation systems with uncertain demand[J]. Journal of Transportation Engineering， Part A: Systems， 2023， 149(12): 04023118.
[4] 赵鹏军， 陈霄依， 王祎勍， 等. 城市系统交通需求模拟预测技术框架构建及应用[J]. 地球信息科学学报， 2025， 27(3): 539-552.
[5] ZHU Y， GOVERDE R M P. Dynamic and robust timetable rescheduling for uncertain railway disruptions[J]. Journal of Rail Transport Planning & Management， 2020， 15: 100196.
[6] FILOM S， RAZAVI S. A learning-based robust optimization framework for synchromodal freight transportation under uncertainty[J]. Transportation Research Part E: Logistics and Transportation Review， 2025， 195: 103967.
[7] ISLAM S， AMIN S H， WARDLEY L J. A supplier selection & order allocation planning framework by integrating deep learning， principal component analysis， and optimization techniques[J]. Expert Systems with Applications， 2024， 235: 121121.
[8] GUO W， ATASOY B， VAN BLOKLAND W B， et al. Anticipatory approach for dynamic and stochastic shipment matching in hinterland synchromodal transportation[J]. Flexible Services and Manufacturing Journal， 2022， 34(2): 483-517.
[9] 王莹， 韩宝明， 张琦， 等. 基于SARIMA模型的北京地铁进站客流量预测[J]. 交通运输系统工程与信息， 2015， 15(6): 205-211.
[10] WANG D， WANG C， XIAO J， et al. Bayesian optimization of support vector machine for regression prediction of short-term traffic flow[J]. Intelligent Data Analysis， 2019， 23(2): 481-497.
[11] NEJADETTEHAD A， MAHINI H， BAHRAK B. Short-term demand forecasting for online car-hailing services using recurrent neural networks[J]. Applied Artificial Intelligence， 2020， 34(9): 674-689.
[12] LIU X， GHERBI A， LI W， et al. Multi features and multi-time steps LSTM based methodology for bike sharing availability prediction[J]. Procedia Computer Science， 2019， 155: 394-401.
[13] 谭雪， 张小强. 基于GRU深度网络的铁路短期货运量预测[J]. 铁道学报， 2020， 42(12): 28-35.
[14] LIU Y， WU C， WEN J， et al. A grey convolutional neural network model for traffic flow prediction under traffic accidents[J]. Neurocomputing， 2022， 500: 761-775.
[15] 唐进君， 段一鑫， 商淑杰， 等. 多特征融合时空图卷积的高速公路交通流量预测[J]. 公路交通科技， 2025， 42(6): 22-31.
[16] 胡晓伟， 吴则洋， 卢泓博， 等. 基于遗传算法优化深度神经网络的站点客流预测[J]. 交通运输工程与信息学报， 2025， 23(1): 72-84.
[17] XIAN J， YE Y， ZHANG W， et al. MDHGFN: Multiscale dual hypergraph fusion spatiotemporal network for traffic flow prediction[J]. Chaos， Solitons & Fractals， 2025， 201: 117228.
[18] ZHANG H， LIN Z， XIE H， et al. Two-way heterogeneity model for dynamic spatiotemporal traffic flow prediction[J]. Knowledge-Based Systems， 2025， 320: 113635.
[19] LIU J， WANG Y， ZHU J， et al. A multilayer spatiotemporal correlation-aware graph attention network for traffic flow prediction[J]. IEEE Transactions on Neural Networks and Learning Systems， 2025: 1-15.
[20] FENG C， CUI M， HODGE B M， et al. A data-driven multi-model methodology with deep feature selection for short-term wind forecasting[J]. Applied Energy， 2017， 190: 1245-1257.
[21] 杜培军， 慕号伟， 郭山川， 等. 集成学习遥感应用：进展与展望[J]. 遥感学报， 2025， 29(6): 1614-1635.
[22] MORETTI F， PIZZUTI S， PANZIERI S， et al. Urban traffic flow forecasting through statistical and neural network bagging ensemble hybrid modeling[J]. Neurocomputing， 2015， 167: 3-7.
[23] LEE D， KIM K. AdaBoost.RDT: AdaBoost integrated with residual-based decision tree for demand prediction of bike sharing systems under extreme demands[J]. IEEE Access， 2024， 12: 144316-144336.
[24] LEI J， CHU Z， WU Y， et al. Predicting vessel arrival times on inland waterways: a tree-based stacking approach[J]. Ocean Engineering， 2024， 294: 116838.
[25] 李浩， 卢朝阳， 谈翌平， 等. 基于MIC-iAFF-Stacking集成学习的航空器滑出时间预测[J]. 交通运输工程与信息学报， 2024， 22(4): 142-153. LI Hao， LU Chaoyang， TAN Yiping， et al. Aircraft departure taxi time prediction based on MIC-iAFF-Stacking ensemble learning[J]. Journal of Transportation Engineering and Information， 2024， 22(4): 142-153.
[26] 陈李越， 柴迪， 王乐业. UCTB：时空人群流动预测工具箱[J]. 计算机科学与探索， 2022， 16(4): 835-843. CHEN Liyue， CHAI Di， WANG Leye. UCTB: spatiotemporal crowd flow prediction toolbox[J]. Journal of Frontiers of Computer Science and Technology， 2022， 16(4): 835-843.
[27] HOCHREITER S， SCHMIDHUBER J. Long short-term memory[J]. Neural Computation， 1997， 9(8): 1735-1780.
[28] RUMELHART D E， HINTON G E， WILLIAMS R J. Learning representations by back-propagating errors[J]. Nature， 1986， 323(6088): 533-536.
[29] CHEN T， GUESTRIN C. XGBoost: A scalable tree boosting system[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco California USA: ACM， 2016: 785-794.
[30] TIAN K， GUO J， YE K， et al. ST-MGAT: Spatial-temporal multi-head graph attention networks for traffic forecasting[C]//2020 IEEE 32nd International Conference on Tools with Artificial Intelligence (ICTAI). Baltimore: IEEE， 2020: 714-721.
[31] WANG L， CHAI D， LIU X， et al. Exploring the generalizability of spatio-temporal traffic prediction: meta-modeling and an analytic framework[J]. IEEE Transactions on Knowledge and Data Engineering， 2023， 35(4): 3870-3884.
基本信息:
DOI：10.19961/j.cnki.1672-4747.2025.12.010
中图分类号:U293.6
引用信息:
[1]游欣,甘蜜,张义萌,等.考虑图特征的时空交通需求集成预测模型[J].交通运输工程与信息学报().DOI:10.19961/j.cnki.1672-4747.2025.12.010.
基金信息:
国家自然科学基金（52402521，U2568219，52372306，52232011）; 四川省自然科学基金（2025NSFSC1969）; 河北省自然科学基金（G2024105007）; 中央高校基本科研业务费（2682025CX056）
2026-03-09
2026-03-09
2026-03-09