Traffic Data Imputation Based on Self-Supervised Learning

doi:10.12141/j.issn.1000-565X.220237

Abstract

Abstract:

In the regional highway network, there are numerous toll stations generating massive amounts of data on a daily basis. However, due to equipment and network issues, there may be delays in data transmission for some stations. In such cases, the transmitted data may not be sufficient to meet the requirements for real-time traffic flow prediction. To achieve real-time traffic data imputation and dynamic traffic flow prediction, this paper firstly proposed a method for data imputation of highway traffic flow data based on self-supervised learning, which adopts time series model based on attention mechanism (Seq2Seq-Att). Then the self-supervised learning method was used to train the model. Finally, the reliability of the method was verified by taking 80 toll stations in the highway network of Guangdong province as an example. The results show that the method in this paper can flexibly capture the missing pattern in traffic data and give a reasonable value according to the internal correlation of the data. This method is generally superior to other methods and has good performance under different missing rates. The overall MAPE is about 17.7% and the WMAPE is 12.8%. In the case of high missing rate, this method has obvious advantages over other methods. The results of traffic volume prediction indicate that the prediction accuracy of traffic flow prediction using the data completed by this method is close to the situation of using complete data.

Key words: data imputation, self-supervised learning, traffic flow prediction, machine learning, highway

CLC Number:

U491

ZHOU Chuhao, LIN Peiqun, YAN Mingyue. Traffic Data Imputation Based on Self-Supervised Learning[J]. Journal of South China University of Technology(Natural Science Edition), 2023, 51(4): 101-114.

Figures/Tables 12

Fig.1

Fig.2

Fig.3

Table 1

Parameter setting of the experiment of the traffic volume data imputation"

参数	取值	说明
l	12	数据会发生缺失的最大时段数
r	12	用于提取历史规律的时段数
n	80	入口站点数量
$p m a x$	0.8	最大缺失率
$d t i m e$	82	时间标签one-hot编码后的向量大小
$d m o d e l$	192	LSTM的隐层大小
$p 1, p 2, ⋯, p 8$	$0.1,0.2, ⋯, 0.8$	用于计算验证集损失的缺失率
$μ 1$	0.9	缺失数据的误差权重
Batch size	512	每次训练多少个样本
Epoch	5 000	完整训练的最大轮次
Stop Condition	200	连续200轮的验证集误差不再降低则停止训练

Table 1

Fig.4

Fig.5

Fig.6

Fig.7

Fig.8

Table 2

Table 3

Fig.9

References 23

1	NEWMAN D A ．Missing data：five practical guidelines［J］．Organizational Research Methods，2014，17（4）：372-411.
2	孟鸿程，陈淑燕．交通流缺失数据处理方法比较分析［J］．交通信息与安全，2018，36（2）：61-67.
	MENG Hong-cheng， CHEN Shu-yan ．A comparative analysis of data imputation methods for missing traffic flow data［J］．Journal of Transport Information and Safety，2018，36（2）：61-67.
3	黎镭，陈蔼祥，姚赞杰．EM算法对缺失数据极大似然估计的实证研究［J］．统计学与应用，2018，7（2）：210-220.
	LI Lei， CHEN Ai-xiang， YAO Zan-jie ．An empirical study on maximum likelihood estimation of missing data by EM algorithm［J］．Statistics and Application，2018，7（2）：210-220.
4	韩卫国，王劲峰，胡建军．交通流量数据缺失值的插补方法［J］．交通信息与安全，2005，23（1）：39-42.
	HAN Wei-guo， WANG Jin-feng， HU Jian-jun ．Imputation methods for missing values in traffic flow data［J］．Journal of Transport Information and Safety，2005，23（1）：39-42.
5	ROYSTON P ．Multiple imputation of missing values［J］．The Stata Journal，2004，4（3）：227-241.
6	LI L， ZHANG J， WANG Y，et al ．Multiple imputation for incomplete traffic accident data using chained equations［C］∥Proceedings of the 2017 IEEE 20th International Conference on Intelligent Transportation Systems （ITSC）．New York：IEEE，2017：1-5.
7	陆化普，屈闻聪，孙智源．基于S-G滤波的交通流故障数据识别与修复算法［J］．土木工程学报，2015，48（5）：123-128.
	LU Hua-pu， QU Wen-cong， SUN Zhi-yuan ．Detection and repair algorithm of traffic erroneous data based on S-G filtering［J］．China Civil Engineering Journal，2015，48（5）：123-128.
8	秦一菲，马明辉，王岩松，等．基于改进KNN算法的交通流异常数据修复方法［J］．计算机测量与控制，2018，26（12）：180-184.
	QIN Yi-fei， MA Ming-hui， WANG Yan-song，et al ．A recovery method for abnormal traffic flow data based on improved KNN algorithm［J］．Computer Measurement & Control，2018，26（12）：180-184.
9	KAUR M， SARBJEET S， NAVEEN A ．Missing traffic data imputation using a dual-stage error-corrected boosting regressor with uncertainty estimation［J］．Information Sciences，2022，586：344-373.
10	ZHANG Y， ZHOU B， CAI X，et al ．Missing value imputation in multivariate time series with end-to-end generative adversarial networks［J］．Information Sciences，2021，551：67-82.
11	CAO W， WANG D， LI J，et al ．BRITS：bidirectional recurrent imputation for time series［C］∥Proceedings of the 32nd Conference on Neural Information Processing Systems （NIPS）．California：NIPS，2018：10.48550/arXiv.1805.10572.
12	ZHANG Z， LIN X， LI M，et al ．A customized deep learning approach to integrate network-scale online traffic data imputation and prediction［J］．Transportation Research Part C，2021，132：103372.
13	王力，李敏，闫佳庆，等．基于生成式对抗网络的路网交通流数据补全方法［J］．交通运输系统工程与信息，2018，18（6）：63-71.
	WANG Li， LI Min， YAN Jia-qing，et al ．Urban traffic flow data recovery method based on generative adversarial network［J］．Journal of Transportation Systems Engineering and Information Technology，2018，18（6）：63-71.
14	LIANG Y， ZHAN Z， SUN L ．Memory-augmented dynamic graph convolution networks for traffic data imputation with diverse missing patterns［J］．Transportation Research Part C，2022，143：103826.
15	LEE H， HWANG S J， SHIN J ．Self-supervised label augmentation via input transformations［C］∥Proceedings of the International Conference on Machine Learning （ICML）．San Diego：JMLR，2020：10.48550/arXiv.1910.05872.
16	HE K， CHEN X， XIE S，et al ．Masked autoencoders are scalable vision learners［C］∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）．Los Alamitos：IEEE Computer Society，2022：15979-15988.
17	DEVLIN J， CHANG M， LEE K，et al ．BERT：pre-training of deep bidirectional transformers for language understanding［C］∥Proceedings of the 17th Annual Conference of the North American Chapter of the Association for Computational Linguistics：Human Language Technologies （NAACL-HLT）．Stroudsburg：ACL，2019：4171-4186.
18	李山海，吴艳雄，刘玉龙，等．基于编解码深度神经网络的交通流预测模型［J］．电子技术与软件工程，2021（14）：138-141.
	LI Shan-hai， WU Yan-xiong， LIU Yu-long，et al ．Traffic flow prediction model based on encoder-decoder deep neural network［J］．Electronic Technology & Software Engineering，2021（14）：138-141.
19	GRAVES A， SCHMIDHUBER J ．Framewise phoneme classification with bidirectional LSTM and other neural network architectures［J］．Neural Networks，2005，18（5/6）：602-610.
20	VASWANI A， SHAZEER N， PARMAR N，et al ．Attention is all you need［C］∥Proceedings of the 31st Annual Conference on Neural Information Processing Systems （NIPS）．California：NIPS，2017：10.48550/arXiv.1706.03762.
21	RUDER S ．An overview of gradient descent optimization algorithms［J］．arXiv preprint arXiv：.
22	郑智泉，王孟孟，田维琦．基于加权K近邻算法的缺失数据填补研究［J］．智能计算机与应用，2021，11（11）：31-33.
	ZHENG Zhi-quan， WANG Meng-meng， TIAN Wei-qi ．Research on missing data filling based on weighted k-nearest neighbor algorithm［J］．Intelligent Computer and Applications，2021，11（11）：31-33.
23	黄子敬．基于时空注意力机制的高速公路多收费站多时段出口流量预测方法研究［D］．广州：华南理工大学，2020.

方法	MAPE					WMAPE
方法	缺失率20%	缺失率40%	缺失率60%	缺失率80%	平均值	缺失率20%	缺失率40%	缺失率60%	缺失率80%	平均值
Proposed	0.172	0.174	0.177	0.184	0.177	0.125	0.126	0.128	0.133	0.128
Proposed （无时间）	0.179	0.183	0.188	0.205	0.189	0.129	0.131	0.135	0.142	0.134
LSTM-AutoEncoder	0.182	0.187	0.193	0.208	0.193	0.131	0.133	0.135	0.144	0.136
GAN	0.213	0.229	0.242	0.258	0.235	0.132	0.139	0.146	0.157	0.144
LSTM	0.205	0.225	0.242	0.262	0.234	0.129	0.136	0.145	0.156	0.142
Interpolate	0.166	0.174	0.193	0.25	0.196	0.112	0.118	0.13	0.168	0.132
KNN	0.174	0.185	0.209	0.277	0.211	0.12	0.129	0.145	0.189	0.146
Ffill	0.202	0.218	0.245	0.306	0.243	0.141	0.152	0.172	0.213	0.17
Mean	0.511	0.551	0.602	0.67	0.583	0.287	0.309	0.338	0.377	0.328
Median	0.48	0.54	0.605	0.675	0.575	0.3	0.331	0.364	0.399	0.349
Mode	0.449	0.488	0.531	0.575	0.51	0.373	0.395	0.417	0.436	0.405
方法	RMSE					MAE
方法	缺失率20%	缺失率40%	缺失率60%	缺失率80%	平均值	缺失率20%	缺失率40%	缺失率60%	缺失率80%	平均值
Proposed	20.37	20.6	21.06	21.84	20.97	13.61	13.73	14.00	14.56	13.97
Proposed （无时间）	21.07	21.32	21.92	22.93	21.81	14.09	14.28	14.68	15.45	14.63
LSTM-AutoEncoder	21.33	21.52	21.99	23.35	22.05	14.26	14.44	14.77	15.74	14.80
GAN	20.97	21.83	22.96	24.52	22.57	14.41	15.11	15.97	17.11	15.65
LSTM	20.61	21.51	22.70	24.40	22.31	14.08	14.87	15.79	17.06	15.45
Interpolate	18.79	19.82	22.15	28.88	22.41	12.34	12.95	14.32	18.48	14.52
KNN	20.09	21.53	24.29	31.68	24.4	13.23	14.19	15.96	20.78	16.04
Ffill	23.49	25.4	28.8	35.33	28.25	15.53	16.72	18.88	23.35	18.62
Mean	42.99	46.27	50.51	56.13	48.98	31.51	33.96	37.16	41.44	36.02
Median	47.41	51.8	56.3	60.84	54.09	32.92	36.35	40.01	43.86	38.28
Mode	64.19	66.72	68.94	70.73	67.65	40.98	43.42	45.77	47.85	44.5

预测方法	数据类型	MAPE					MAE					RMSE
预测方法	数据类型	缺失率0	缺失率20%	缺失率40%	缺失率60%	缺失率80%	缺失率0	缺失率20%	缺失率40%	缺失率60%	缺失率80%	缺失率0	缺失率20%	缺失率40%	缺失率60%	缺失率80%
KNN	缺失		0.512	0.713	1.203	1.699		134.6	197.6	256.5	308.7		178.4	251.9	325.0	417.6
	补全		0.092	0.098	0.097	0.097		51.68	53.19	53.82	54.48		73.66	75.19	76.70	77.89
	完整	0.086					46.78					68.10
GBRT	缺失		0.214	0.308	0.449	0.737		86.4	107.3	143.5	202.6		113.4	136.2	176.2	244.5
	补全		0.092	0.099	0.101	0.104		51.41	54.10	56.17	56.63		71.64	73.46	79.58	78.84
	完整	0.128					64.53					88.54
RF	缺失		0.225	0.412	0.631	0.956		88.7	125.7	176.5	238.4		115.7	159.2	213.7	279.5
	补全		0.088	0.093	0.094	0.101		50.46	52.39	52.60	56.43		69.09	73.05	72.01	79.61
	完整	0.111					58.35					80.67

[1]	YANG Yonghong, TANG Zude, WANG Chun, ZHU Guanru. Influence of Alignment Indexes of Highway Turning Section on Safety Performance of Concrete Guardrail [J]. Journal of South China University of Technology(Natural Science Edition), 2024, 52(1): 90-99.
[2]	HUANG Yan, FU Xinsha, ZENG Yanjie, et al. 3D Modeling Method of Highway Based on Lidar Odometer [J]. Journal of South China University of Technology(Natural Science Edition), 2023, 51(7): 129-138.
[3]	DONG Ping, WEI Shuyang, LIU Mingbo. Scheduling Strategies for Electric Vehicle Participation in Electricity Markets Under Multi-Network Collaboration [J]. Journal of South China University of Technology(Natural Science Edition), 2023, 51(12): 83-94.
[4]	WU Yanxia, ZHOU Tong, HUANG Shuai, et al. Research on Roadside Environment Safety Based on Driver’s Attention Distribution Model [J]. Journal of South China University of Technology(Natural Science Edition), 2022, 50(9): 49-57.
[5]	WANG Xiaofei, LI Siyu, CHEN Mi, et al. Influence of the Combination Equilibrium of Horizontal and Crest Vertical Curves on Highway Safety [J]. Journal of South China University of Technology(Natural Science Edition), 2022, 50(7): 76-84.
[6]	SONG Jian, WANG Wenlong, LI Dong, et al. Injection molding part size prediction method based on Stacking integration learning [J]. Journal of South China University of Technology(Natural Science Edition), 2022, 50(6): 19-26.
[7]	FU Xinsha, ZENG Yanjie, MA Li, et al. Weather Recognition of Highway Surveillance Scenes Based on Light-Weight Deep Neural Network [J]. Journal of South China University of Technology(Natural Science Edition), 2022, 50(3): 1-8.
[8]	HE Yongming, FENG Jia, QUAN Cong, et al. Braking Collision Avoidance System for Vehicles Driving on Superhighway Based on Co-simulation [J]. Journal of South China University of Technology(Natural Science Edition), 2022, 50(10): 19-28.
[9]	LIN Peiqun XIA Yu ZHOU Chuhao. Freeway Travel Time Prediction Based on Spatial and Temporal Characteristics of Road Networks [J]. Journal of South China University of Technology(Natural Science Edition), 2021, 49(8): 1-11.
[10]	JIA Ruo, DAI Shenghong, HUANG Ni, et al. Literature Review on Traffic Congestion Identification Methods [J]. Journal of South China University of Technology (Natural Science Edition), 2021, 49(4): 124-139.
[11]	ZHAO Jing, WANG Xuancang, FAN Zhenyang, et al. Evaluation on Performance of Asphalt Pavement Based on Support Vector Machine [J]. Journal of South China University of Technology (Natural Science Edition), 2020, 48(9): 116-123.
[12]	CHANG Xin, LI Haijian, RONG Jian, et al. Analysis of Capacity for Mixed Traffic Flow with Connected Vehicle Platoon on Freeway [J]. Journal of South China University of Technology (Natural Science Edition), 2020, 48(4): 142-148.
[13]	ZHANG Ziye, LI Mingchang, LIANG Lingrui, et al. Improved Transfer Learning Algorithm Based on Cross-domain in Recommendation System [J]. Journal of South China University of Technology (Natural Science Edition), 2020, 48(11): 99-106.
[14]	YUAN Hua CHEN Zehao. Short-term Traffic Flow Prediction Based on Temporal Convolutional Networks [J]. Journal of South China University of Technology (Natural Science Edition), 2020, 48(11): 107-113,122.
[15]	. Elastic Property Prediction of Materials Based on Machine Learning and Feature Selection [J]. Journal of South China University of Technology (Natural Science Edition), 2019, 47(5): 48-55.