Forecasting for Short-term Passenger Flow of Subway Based on Dynamic Graph Neural Ordinary Differential Equations
-
摘要: 随着城市轨道交通的快速发展，客流量的准确预测对于改善运营服务至关重要。为了解决当前地铁客流预测存在的时空特性挖掘不充分等问题，进一步提高预测的精度与效率，研究了基于动态图神经常微分方程模型（multivariate time series with dynamic graph neural ordinary differential equations，MTGODE）的地铁短时客流预测方法。该方法通彭颢1贺玉过学习地铁站点间的动态关联强度构建动态拓扑图结构，基于学习得到的动态图进行连续图传播以传递时空信息、挖掘客流的依赖关系，并采用残差卷积提取多时间尺度下的周期性模式，实现了对站点间时空动态的连续表征，克服了传统图卷积网络模型难以刻画动态空间依赖的局限性。此外，为了充分挖掘不同站点间客流分布的时空规律，综合利用北京地铁自动售检票系统（auto fare collection，AFC）刷卡数据、天气数据、空气质量数据以及车站周边用地属性数据构建多源融合的客流预测模型。通过选取地铁北京站和积水潭站-东直门站的历史数据开展进站客流和OD客流预测实验，结果表明：与多个基准模型相比，该模型在平均绝对误差、均方根误差和平均百分比误差这3个指标中均取得了更优的预测效果，相较最优基准模型扩散卷积循环神经网络（diffusion convolutional recurrent neural network，DCRNN）分别降低了9.93%，12.30%，9.23%，对地铁客流时空分布的拟合程度更好，模型具有更好的预测精度、稳定性和拟合能力。Abstract: With the rapid expansion of urban rail transit networks, accurate forecasting for passenger flows has become paramount for optimizing operational services. To solve the issue of the inadequate mining for the spatiotemporal characteristics in the forecasting of current subway passenger flow forecasting and to further enhance accuracy and efficiency of forecasting methods, a forecasting method for short-term subway passenger flow based on multivariate time series with dynamic graph neural ordinary differential equations (MTGODE) is proposed. The method constructs a dynamic topological graph structure by learning the dynamic correlation strength between subway stations. Continuous graph propagation is performed on the learned dynamic graph to transmit spatiotemporal information and capture the dependencies of passenger flows. Moreover, residual convolution is employed to extract periodic patterns at multiple time scales, enabling continuous representation of spatiotemporal dynamics between stations and overcoming the limitations of traditional graph convolutional network models in capturing dynamic spatial dependencies. Furthermore, to fully uncover the spatiotemporal patterns of passenger flow distribution among different stations, a multi-source fusion model for passenger flow forecasting is developed by comprehensively utilizing data from the Beijing subway's automatic fare collection system, weather data, air quality data, and surrounding land use attributes of stations. The proposed model was tested by forecasting inbound passenger flow and origin-destination flow using historical data from Beijing Station and Jishuitan Station-Dongzhimen Station. The experimental results demonstrate that the proposed model achieves superior performance compared to multiple benchmark models across three metrics: mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). Compared to the best-performing benchmark model, the diffusion convolutional recurrent neural network (DCRNN), the proposed model reduces MAE, RMSE, and MAPE by 9.93%, 12.30%, and 9.23%, respectively. It exhibits a better fit to the spatiotemporal distribution of subway passenger flows and possesses improved prediction accuracy, stability, and fitting capability.
-
Key words:
- rail transit /
- subway passenger flow /
- dynamic graph neural network /
- MTGODE model /
- deep learning
-
图 7 北京站不同模型进站客流预测结果
Figure 7. Beijing Station incoming passenger flow forecast results for different mode
图 8 北京站不同模型进站客流预测均方误差分布
Figure 8. Beijing Station incoming passenger flow forecast MSE distributions for different models
图 9 积水潭站-东直门站不同模型OD客流预测结果
Figure 9. Jishuitan Station-Dongzhimen Station OD passenger flow forecast results for different models
表 1 城市建设用地类型分类
Table 1. Classification of urban construction land use types
一级分类 二级分类 居住用地 居住用地 商业用地 商务办公用地 商业服务用地 工业用地 工业用地 交通用地 交通场站用地 机场设施用地 公共管理和服务用地 行政办公用地 教育科研用地 医疗卫生用地 体育与文化用地 公园与绿地用地 
下载: 导出CSV
表 2 基本参数设置
Table 2. Basic parameter settings
参数 数值 时间窗（time window）/个 10 批大小（batch size）/个 32 训练周期数（epoch）/次 500 卷积层数（layers）/层 3 卷积核大小（kernels） 1×1 神经元个数（neurons）/个 32，64 优化器选择（optimizer） Adam 初始学习率（learning rate） 0.001 学习率衰减因子（lr decay） 0.7 学习率衰减周期数（step size）/次 10 激活函数（activation function） tanh，ReLU 失活概率（drop prob） 0.1
下载: 导出CSV
表 3 北京站地铁站周边主要用地类型占比
Table 3. Proportions of major land use types around Beijing Station
主要用地类型 占比/% 交通场站用地 52.66 商务办公用地 25.56 居住用地 18.11
下载: 导出CSV
表 4 积水潭站周边主要用地类型占比
Table 4. Proportions of major land use types around Jishuitan Station
主要用地类型 占比/% 居住用地 56.74 行政办公用地 26.06 医疗卫生用地 7.41
下载: 导出CSV
表 5 直门站周边主要用地类型占比
Table 5. Proportions of major land use types around Dongzhimen Station
主要用地类型 占比/% 商务办公用地 33.30 商业服务用地 21.55 居住用地 14.40
下载: 导出CSV
表 6 积水潭站—东直门站不同模型OD客流预测精度对比
Table 6. Comparison of OD passenger flow forecast accuracy of the different models from Jishuitan Station to Dongzhimen Station
模型 MAE RMSE SVR 3.875 6.068 GRU 3.556 5.802 STGCN 2.306 3.412 DCRNN 1.958 2.705 STGODE 2.236 3.169 MTGODE 1.653 2.208
下载: 导出CSV
表 7 6种模型的各指标比较
Table 7. Comparison of metrics for six different models
模型 MAE RMSE MAPE/% R2 SVR 22.244 36.392 21.164 0.930 GRU 19.575 32.211 17.018 0.941 STGCN 16.008 23.955 13.140 0.963 DCRNN 14.838 23.264 11.818 0.968 STGODE 15.354 23.479 12.827 0.966 MTGODE 13.365 20.403 10.727 0.974
下载: 导出CSV
-
[1] 徐新颖. 基于卷积神经网络的城市轨道交通短时客流预测方法研究[D]. 福州: 福州大学, 2023.XU X Y. Research on urban rail transit short-term passenger flow forecasting method based on convolutional neural network[D]. Fuzhou: Fuzhou University, 2023. (in Chinese) [2] 光志瑞. 城市轨道交通节假日客流预测研究[J]. 交通工程, 2017, 17(3): 27-35. https://www.cnki.com.cn/Article/CJFDTOTAL-DLJA201703005.htmGUANG Z R. Research on urban rail transit passenger flow forecasting during holidays[J]. Journal of Transportation Engineering, 2017, 17(3): 27-35. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-DLJA201703005.htm [3] 张国赟, 金辉. 基于改进ARIMA模型的城市轨道交通短时客流预测研究[J]. 计算机应用与软件, 2022, 39(1): 339-344. doi: 10.3969/j.issn.1000-386x.2022.01.052ZHANG G Y, JIN H. Short-term passenger flow forecasting of urban rail transit based on improved ARIMA model[J]. Computer Applications and Software, 2022, 39(1): 339-344. (in Chinese) doi: 10.3969/j.issn.1000-386x.2022.01.052 [4] 张子翰. 城市轨道交通新线运营初期短时客流预测[D]. 南京: 南京理工大学, 2021.ZHANG Z H. Short-term passenger flow forecast at the initial stage of new urban rail transit line[D]. Nanjing: Nanjing University of Science and Technology, 2021. (in Chinese) [5] 陈小健, 唐秋生. 基于多模式灰色模型的地铁全网客流预测研究[J]. 交通科技与经济, 2019, 21(4): 16-20. https://www.cnki.com.cn/Article/CJFDTOTAL-KJJJ201904004.htmCHEN X J, TANG Q S. Research on metro network passenger flow forecasting based on multi-mode gray model[J]. Technology & Economy in Areas of Comm-unications, 2019, 21(4): 16-20. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-KJJJ201904004.htm [6] 郭文. 基于支持向量机的轨道交通短期客流预测方法研究[D]. 苏州: 苏州大学, 2020.GUO W. Research on short-term passenger flow forecasting method of rail transit based on support vector machine[D]. Suzhou: Soochou University, 2020. (in Chinese) [7] 刘祝娟. 基于PSO-SVR短时客流预测的北京地铁4号线&大兴线运营组织研究[D]. 石家庄: 石家庄铁道大学, 2023.LIU Z J. Research on Beijing subway line 4&daxing line operation organization based on PSO-SVR short-term passenger flow forecasting[D]. Shijiazhuang: Shijiazhuang Railway University, 2023. (in Chinese) [8] 谢鑫鑫. 基于EMD-KNN的城市轨道站点客流预测方法研究[D]. 苏州: 苏州科技大学, 2022.XIE X X. Research on urban rail transit station passenger flow forecasting method based on EMD-KNN[D]. Suzhou: Suzhou University of Science and Technology, 2022. (in Chinese) [9] 张恒, 秦振华, 肖为周, 等. 基于决策树模型的地铁线网短时OD客流预测[J]. 河北工业科技, 2023, 40(2): 146-154. https://www.cnki.com.cn/Article/CJFDTOTAL-HBGY202302010.htmZHANG H, QIN Z H, XIAO W Z, et al. Short-term OD passenger flow forecasting of subway network based on decision tree model[J]. Hebei Journal of Industrial Science and Technology, 2023, 40(2): 146-154. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-HBGY202302010.htm [10] WANG L, ZENG Y, CHEN T. Back propagation neural network with adaptive differential evolution algorithm for time series forecasting[J]. Expert Systems with Applications, 2015, 42: 855-863. doi: 10.1016/j.eswa.2014.08.018 [11] HAN Y, WANG S K, REN Y B, et al. Predicting station-level short-term passenger flow in a citywide metro network using spatiotemporal graph convolutional neural networks[J]. ISPRS International Journal of Geo-Information, 2019, 8(6): 243. doi: 10.3390/ijgi8060243 [12] WANG J L, ZHANG J, WANG X X. Bilateral LSTM: a two-dimensional long short-term memory model with multiply memory units for short-term cycle time forecasting in re-entrant manufacturing systems[J]. IEEE Transactions on Industrial Informatics, 2018, 14: 748-758. doi: 10.1109/TII.2017.2754641 [13] 肖明亮. 基于改进BP神经网络的地铁客流预测研究[D]. 南昌: 南昌大学, 2022.XIAO M L. Research on subway passenger flow forecast based on improved BP neural network[D]. Nanchang: Nanchang University, 2022. (in Chinese) [14] 王磊, 陆川, 蒲丹丹, 等. 基于改进卷积神经网络的地铁客流量预测算法设计[J]. 现代电子技术, 2021, 44(24): 87-91. https://www.cnki.com.cn/Article/CJFDTOTAL-XDDJ202124019.htmWANG L, LU C, PU D D, et al. Design of subway passenger flow forecasting algorithm based on improved convolutional neural network[J]. Modern Electronics Technique, 2021, 44(24): 87-91. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-XDDJ202124019.htm [15] 冯碧玉. 基于CNN-LSTM组合模型的城市轨道交通短时客流预测研究[D]. 南昌: 华东交通大学, 2021.FENG B Y. Research on short-term passenger flow forecasting of urban rail transit based on CNN-LSTM combined model[D]. Nanchang: East China Jiaotong University, 2021. (in Chinese) [16] YE J, ZHAO J, YE K, et al. How to build a graph-based deep learning architecture in traffic domain: a survey[J]. IEEE Transactions on Intelligent Transportation Systems, 2020, 23(5): 3904-3924. [17] LI Y, YU R, SHAHABI C, et al. Diffusion convolutional recurrent neural network: data-driven traffic forecasting[C]. International Conference on Learning Representations, Vancouver: ICLR, 2018. [18] YU B, YIN H T, ZHU Z X. Spatio-temporal graph convolutional networks: a deep learning framework for traffic forecasting[C]. 27th International Joint Conference on Artificial Intelligence, Freiburg: IJCAI, 2018. [19] 崔文岳, 谷远利, 赵胜利, 等. 基于有向图卷积与门控循环单元的短时交通流预测方法[J]. 交通信息与安全, 2023, 41(2): 121-128. doi: 10.3963/j.jssn.1674-4861.2023.02.013CUI W Y, GU Y L, ZHAO S L, et al. Short-term traffic flow forecasting method based on directed graph convolution and gated recurrent units[J]. Journal of Transport Information and Safety, 2023, 41(2): 121-128. (in Chinese) doi: 10.3963/j.jssn.1674-4861.2023.02.013 [20] 李亮. 基于神经常微分方程的时序数据分析与预测[D]. 成都: 电子科技大学, 2022.LI L. Analyzing and forecasting time-series data via neural ordinary differential equations[D]. Chengdu: University of Electronic Science and Technology of China, 2022. (in Chinese) [21] FANG Z, LONG Q, SONG G, et al. Spatialtemporal graph ode networks for traffic flow forecasting[C]. 27th ACM SIG-KDD Conference on Knowledge Discovery and Data Mining, Singapore: SIGKDD, 2021 [22] CHEN R T, RUBANOVA Y, BETTENCOURT J, et al. Neural ordinary differential equations[C]. 32nd Conference and Workshop on Neural Information Processing Systems, Montreal: NIPS, 2018. [23] 罗文慧, 董宝田, 王泽胜. 基于CNN-SVR混合深度学习模型的短时交通流预测[J]. 交通运输系统工程与信息, 2017, 17(5): 68-74. https://www.cnki.com.cn/Article/CJFDTOTAL-YSXT201705010.htmLUO W H, DONG B T, WANG Z S. Short-term traffic flow forecasting based on CNN-SVR hybrid deep learning model[J]. Journal of Transportation Systems Engineering and Information Technology, 2017, 17(5): 68-74. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-YSXT201705010.htm -
点击查看大图
图(9) / 表(7)
计量
- 文章访问数: 247
- HTML全文浏览量: 119
- PDF下载量: 18
- 被引次数: 0