基于多任务学习的轨道交通短时客流预测研究

doi:10.3976/j.issn.1002-4026.20230038

摘要/Abstract

摘要：

为了精准预测轨道交通的短时客流量,有效缓解城市交通拥堵,提出了一种基于多任务学习的轨道交通短时客流预测模型,该模型采用残差卷积神经网络和嵌套式长短期记忆神经网络提取客流的时空相关性,引入注意力机制加强模块对特征的提取效果。考虑轨道交通运营的特点,模型进一步选取列车运行特征、轨道交通站点周边公交站点以及兴趣点数据作为外部特征,以提高轨道交通短时客流预测精度。基于北京地铁历史客流数据,在10、30、60 min等多时间粒度场景下进行实验。结果显示,该方法通过多任务学习的方式建模分析站点进出站客流之间的相互影响,提高了模型的预测性能和泛化能力,为城市轨道交通短时客流预测问题提供了新的思路。

关键词: 轨道交通, 客流预测, 多任务学习, 注意力机制, 深度神经网络

Abstract:

An accurate prediction of short-term subway passenger flowscan effectively alleviate traffic congestion and improve the quality of travel services for urban residents. Herein, we propose a multitask learning-based model for the prediction of short-term subway passenger flows, which uses a residual convolutional neural network (NN) and a nested long short-term memory NN to extract the spatio-temporal correlation of traffic patterns, and introduces an attention mechanism to enhance the feature extraction performance of the NNs. Considering the characteristics of subway operations, the model selects train operation features, bus stops around subway stations, and point of interest data as external features to improve the accuracy of the prediction. Based on the historical data of the Beijing Subway, experiments were conducted in multiple time granularity scenarios, such as 10, 30, and 60 min. The results showed that the methodsuccessfully modeled and analyzed the inflow-outflow interaction through multitask learning, improved the prediction performance and generalization ability of the model, and providednovel approaches for the prediction of short-term subway passenger flows.

Key words: subway, passenger flow prediction, multitask learning, attention mechanism, deep neural network

中图分类号:

U239.5

张含笑, 刘宇然, 刘媛, 牛子辰. 基于多任务学习的轨道交通短时客流预测研究[J]. 山东科学, 2024, 37(1): 95-106.

ZHANG Hanxiao, LIU Yuran, LIU Yuan, NIU Zichen. A multitask learning model for the prediction of short-term subway passenger flow[J]. Shandong Science, 2024, 37(1): 95-106.

图/表 11

图1

图2

图3

图4

图5

图6

表1

10 min时间步基线模型与STNN的预测性能比较"

预测模型	进站客流预测			出站客流预测
预测模型	δ_RMSE	$R 2$	δ_MAE	δ_RMSE	$R 2$	δ_MAE
Ridge	47.66	0.84	26.81	40.17	0.85	21.20
SGD	49.52	0.82	27.80	41.11	0.85	20.31
SVR	42.46	0.84	25.30	34.73	0.87	18.10
GBDT	39.76	0.85	22.13	32.76	0.88	17.69
LSTM	38.49	0.88	20.00	27.78	0.91	15.69
LSTMAtt	33.79	0.92	16.99	24.60	0.94	14.57
CNN-LSTM	29.81	0.95	15.44	23.54	0.95	13.82
STNN	24.02	0.96	13.56	21.11	0.97	12.57

表1

表2

30 min时间步基线模型与STNN的预测性能比较"

预测模型	进站客流预测			出站客流预测
预测模型	δ_RMSE	$R 2$	δ_MAE	δ_RMSE	$R 2$	δ_MAE
Ridge	266.84	0.80	120.45	240.41	0.81	110.80
SGD	253.98	0.78	126.04	244.16	0.80	117.90
SVR	193.46	0.85	93.43	152.45	0.82	93.98
GBDT	155.94	0.89	72.10	113.89	0.84	76.69
LSTM	91.27	0.90	59.13	101.62	0.87	58.01
LSTMAtt	85.08	0.94	43.46	73.61	0.94	43.30
CNN-LSTM	89.29	0.91	51.08	91.62	0.92	51.32
STNN	70.87	0.96	36.92	67.37	0.96	35.32

表2

表3

60 min时间步基线模型与STNN的预测性能比较"

预测模型	进站客流预测			出站客流预测
预测模型	δ_RMSE	$R 2$	δ_MAE	δ_RMSE	$R 2$	δ_MAE
Ridge	547.72	0.72	278.18	534.95	0.73	275.53
SGD	699.85	0.55	374.46	670.31	0.58	355.35
SVR	582.31	0.69	243.29	566.83	0.70	233.96
GBDT	354.29	0.88	162.89	353.80	0.88	174.01
LSTM	204.50	0.89	112.07	173.89	0.91	86.32
LSTMAtt	153.10	0.95	75.13	166.49	0.94	77.97
CNN-LSTM	185.75	0.92	102.11	167.55	0.94	82.40
STNN	136.17	0.96	64.33	142.72	0.96	65.63

表3

表4

多任务学习算法进站客流预测结果"

预测模型	30 min			60 min
预测模型	δ_RMSE	$R 2$	δ_MAE	δ_RMSE	$R 2$	δ_MAE
Ridge	266.84	0.80	120.45	547.72	0.72	278.18
SGD	253.98	0.78	126.04	699.85	0.55	374.46
SVR	193.46	0.85	93.43	582.31	0.69	243.29
GBDT	155.94	0.89	72.10	354.29	0.88	162.89
LSTM	91.27	0.90	59.13	204.50	0.88	112.07
LSTMAtt	85.08	0.94	43.46	153.10	0.95	75.13
CNN-LSTM	89.29	0.91	51.08	185.75	0.92	102.11
STNN	70.87	0.96	36.92	136.17	0.96	64.33
MTL- LSTM	88.44	0.91	49.86	150.50	0.93	83.95
MTL-CNN-LSTM	75.23	0.95	39.33	164.52	0.95	93.57
MTL-STNN	56.79	0.97	28.14	109.19	0.97	57.87

表4

图7

参考文献 18

[1]	中国城市轨道交通协会. 城市轨道交通2021年度统计和分析报告. [EB/OL]. [2022-04-22]. https://www.camet.org.cn/tjxx/9944
[2]	李梅, 李静, 魏子健, 等. 基于深度学习长短期记忆网络结构的地铁站短时客流量预测[J]. 城市轨道交通研究, 2018, 21(11): 42-46. DOI: 10.16037/j.1007-869x.2018.11.009.
[3]	LIU Y, LIU Z Y, JIA R. DeepPF: A deep learning based architecture for metro passenger flow prediction[J]. Transportation Research Part C: Emerging Technologies, 2019, 101: 18-34. DOI: 10.1016/j.trc.2019.01.027.
[4]	施俊庆, 李睿, 程明慧, 等. 基于动态时空神经网络模型的地铁客流预测[J]. 交通运输系统工程与信息, 2023, 23(2): 139-147. DOI: 10.16097/j.cnki.1009-6744.2023.02.015.
[5]	赵阳阳, 夏亮, 江欣国. 基于经验模态分解与长短时记忆神经网络的短时地铁客流预测模型[J]. 交通运输工程学报, 2020, 20(4): 194-204. DOI: 10.19818/j.cnki.1671-1637.2020.04.016.
[6]	LI Y, WANG X D, SUN S, et al. Forecasting short-term subway passenger flow under special events scenarios using multiscale radial basis function networks[J]. Transportation Research Part C: Emerging Technologies, 2017, 77: 306-328. DOI: 10.1016/j.trc.2017.02.005.
[7]	XUE G, LIU S F, REN L, et al. Forecasting the subway passenger flow under event occurrences with multivariate disturbances[J]. Expert Systems With Applications, 2022, 188: 116057. DOI: 10.1016/j.eswa.2021.116057.
[8]	赵建东, 朱丹, 刘佳欣. 基于时间序列分解与门控循环单元的地铁换乘客流预测[J]. 华南理工大学学报(自然科学版), 2022, 50(5): 22-31. DOI: 10.12141/j.issn.1000-565X.210559.
[9]	惠阳, 王永岗, 彭辉, 等. 基于优化PSO-BP算法的耦合时空特征下地铁客流预测[J]. 交通运输工程学报, 2021, 21(4): 210-222. DOI: 10.19818/j.cnki.1671-1637.2021.04.016.
[10]	LI C, BAI L, LIU W, et al. A multi-task memory network with knowledge adaptation for multimodal demand forecasting[J]. Transportation Research Part C: Emerging Technologies, 2021, 131: 103352. DOI: 10.1016/j.trc.2021.103352.
[11]	李德奎, 杜书波, 张鹏. 基于ARIMA和LSTM的城市轨道交通延时客流预测方法比较[J]. 青岛理工大学学报, 2021, 42(4): 135-142. DOI: 10.3969/j.issn.1673-4602.2021.04.020.
[12]	邵必林, 饶媛, 何欣. 基于SARIMA-SVM组合模型的地铁客流量预测研究[J]. 软件导刊, 2022, 21(11): 24-30. DOI: 10.11907/rjdk.212729.
[13]	韩皓, 徐圣安, 赵蒙. 考虑线网结构的LightGBM轨道交通短时客流预测模型[J]. 铁道运输与经济, 2021, 43(10): 109-117. DOI: 10.16668/j.cnki.issn.1003-1421.2021.10.17.
[14]	ROOS J, GAVIN G, BONNEVAY S. A dynamic Bayesian network approach to forecast short-term urban rail passenger flows with incomplete data[J]. Transportation Research Procedia, 2017, 26: 53-61. DOI: 10.1016/j.trpro.2017.07.008.
[15]	CHEN P F, FU X D, WANG X. A graph convolutional stacked bidirectional unidirectional-LSTM neural network for metro ridership prediction[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(7): 6950-6962. DOI: 10.1109/TITS.2021.3065404.
[16]	沈永发, 郑煜, 刘辉冉, 等. 基于历史数据的轨道交通站内客流预测模型[J]. 武汉理工大学学报(信息与管理工程版), 2022, 44(6): 887-893. DOI: 10.3963/j.issn.2095-3852.2022.06.003.
[17]	王金水, 欧雪雯, 陈俊岩, 等. 基于组合模型的城市轨道站点短时客流分类预测[J]. 铁道科学与工程学报, 2023, 20(6): 2004-2012. DOI: 10.19713/j.cnki.43-1423/u.t20221280.
[18]	张金雷, 陈奕洁, Panchamy Krishnakumari, 等. 基于注意力机制的城市轨道交通网络级多步短时客流时空综合预测模型[J]. 地球信息科学学报, 2023, 25(4): 698-713. doi: 10.12082/dqxxkx.2023.220817