AI-Empowered Trajectory Anomaly Detection for Intelligent Transportation Systems: A Hierarchical Federated Learning Approach
Xiaoding Wang,Wenxin Liu,Hui Lin,Jia Hu,Kuljeet Kaur,M. Shamim Hossain
DOI: https://doi.org/10.1109/tits.2022.3209903
IF: 8.5
2023-04-01
IEEE Transactions on Intelligent Transportation Systems
Abstract:The vigorous development of positioning technology and ubiquitous computing has spawned trajectory big data. By analyzing and processing the trajectory big data in the form of data streams in a timely and effective manner, anomalies hidden in the trajectory data can be found, thus serving urban planning, traffic management, safety control and other applications. Limited by the inherent uncertainty, infinity, time-varying evolution, sparsity and skewed distribution of trajectory big data, traditional anomaly detection techniques cannot be directly applied to anomaly detection in trajectory big data. To solve this problem, we propose a hierarchical trajectory anomaly detection scheme for Intelligent Transportation Systems (ITS) using both machine learning and blockchain technologies. To be specific, a hierarchical federated learning strategy is proposed to improve the generalization ability of the global trajectory anomaly detection model by secondary fusion of the multi-area trajectory anomaly detection model. Then, by integrating blockchain and federated learning, the iterative exchange and fusion of the global trajectory anomaly detection model can be realized by means of on-chain and off-chain coordinated data access. Experiments show that the proposed scheme can improve the generalization ability of the trajectory anomaly detection model in different areas, while ensuring its reliability.
engineering, electrical & electronic,transportation science & technology, civil