REFOL: Resource-Efficient Federated Online Learning for Traffic Flow Forecasting

被引:0
作者
Liu, Qingxiang [1 ,2 ]
Sun, Sheng [1 ]
Liang, Yuxuan [3 ,4 ]
Xu, Xiaolong [5 ,6 ]
Liu, Min [1 ,7 ]
Bilal, Muhammad [8 ]
Wang, Yuwei [1 ]
Li, Xujing [1 ,2 ]
Zheng, Yu [9 ,10 ]
机构
[1] Chinese Acad Sci, Inst Comp Technol, Beijing 100190, Peoples R China
[2] Univ Chinese Acad Sci, Beijing 100190, Peoples R China
[3] Hong Kong Univ Sci & Technol Guangzhou, INTR Thrust, Guangzhou 510000, Peoples R China
[4] Hong Kong Univ Sci & Technol Guangzhou, DSA Thrust, Guangzhou 510000, Peoples R China
[5] Nanjing Univ Informat Sci & Technol, Sch Software, Nanjing 210044, Peoples R China
[6] Nanjing Univ Informat Sci & Technol, Jiangsu Collaborat Innovat Ctr Atmospher Environm, Nanjing 210044, Peoples R China
[7] Zhongguancun Lab, Beijing 100094, Peoples R China
[8] Univ Lancaster, Sch Comp & Commun, Lancaster LA1 4WA, England
[9] JD Intelligent Cities Res, Beijing 102300, Peoples R China
[10] JD Technol, JD iCity, Beijing 102300, Peoples R China
基金
中国国家自然科学基金;
关键词
Predictive models; Concept drift; Correlation; Forecasting; Computational modeling; Data models; Optimization; Adaptation models; Urban areas; Training; Traffic flow forecasting; federated learning; concept drift; online learning; graph convolution; PREDICTION;
D O I
10.1109/TITS.2024.3510913
中图分类号
TU [建筑科学];
学科分类号
0813 ;
摘要
Multiple federated learning (FL) methods are proposed for traffic flow forecasting (TFF) to avoid heavy-transmission and privacy-leaking concerns resulting from the disclosure of raw data in centralized methods. However, these FL methods adopt offline learning which may yield subpar performance, when concept drift occurs, i.e., distributions of historical and future data vary. Online learning can detect concept drift during model training, thus more applicable to TFF. Nevertheless, the existing federated online learning method for TFF fails to efficiently solve the concept drift problem and causes tremendous computing and communication overhead. Therefore, we propose a novel method named Resource-Efficient Federated Online Learning (REFOL) for TFF, which guarantees prediction performance in a communication-lightweight and computation-efficient way. Specifically, we design a data-driven client participation mechanism to detect the occurrence of concept drift and determine clients' participation necessity. Subsequently, we propose an adaptive online optimization strategy, which guarantees prediction performance and meanwhile avoids meaningless model updates. Then, a graph convolution-based model aggregation mechanism is designed, aiming to assess participants' contribution based on spatial correlation without importing extra communication and computing consumption on clients. Finally, we conduct extensive experiments on real-world datasets to demonstrate the superiority of REFOL in terms of prediction improvement and resource economization.
引用
收藏
页码:2777 / 2792
页数:16
相关论文
共 45 条
[1]   A Survey of Road Traffic Congestion Measures towards a Sustainable and Resilient Transportation System [J].
Afrin, Tanzina ;
Yodo, Nita .
SUSTAINABILITY, 2020, 12 (11)
[2]  
Arcas Y. B. A., 2017, PMLR, V54, P1273, DOI DOI 10.48550/ARXIV.1602.05629
[3]  
Baena-Garcia M, 2006, 4 INT WORKSH KNOWL D
[4]   An Incremental Change Detection Test Based on Density Difference Estimation [J].
Bu, Li ;
Zhao, Dongbin ;
Alippi, Cesare .
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS, 2017, 47 (10) :2714-2726
[5]   Short-Term Traffic Flow Prediction Method for Urban Road Sections Based on SpaceTime Analysis and GRU [J].
Dai, Guowen ;
Ma, Changxi ;
Xu, Xuecai .
IEEE ACCESS, 2019, 7 :143025-143035
[6]  
Feng HF, 2006, IEEE ICC, P597
[7]  
Gama J, 2004, LECT NOTES ARTIF INT, V3171, P286
[8]   Strictly proper scoring rules, prediction, and estimation [J].
Gneiting, Tilmann ;
Raftery, Adrian E. .
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION, 2007, 102 (477) :359-378
[9]  
Gu F, 2016, IEEE IJCNN, P24, DOI 10.1109/IJCNN.2016.7727176
[10]  
Guo SN, 2019, AAAI CONF ARTIF INTE, P922