Managing urban traffic dynamics is critical in Intelligent Transportation Systems (ITS), where short-term traffic prediction is vital for effective congestion management and vehicle routing. While existing centralized deep learning (DL) models have achieved high prediction accuracy, their applicability is limited in decentralized ITS environments. The increasing use of connected vehicles and mobile sensors has led to decentralized data generation in ITS, presenting an opportunity to improve traffic prediction through collaborative machine learning. Recently, blockchain technology has shown promise in improving ITS efficiency, security, and reliability. In conjunction with blockchain, Federated Learning (FL) is a suitable approach to leverage online data streams in ITS; however, most research on FL for traffic prediction focuses on offline learning scenarios. This paper researches a blockchain-enhanced architecture for training online traffic prediction models using FL. The proposed approach enables decentralized model training at the edge of the ITS network, and extensive experiments used dynamically collected arterial traffic data shards as a case study to evaluate online learning performance. The results demonstrate that our online FL approach outperforms the per-device, non-federated baseline models for most sensors while maintaining a suitable execution time and latency for real-world deployment.
Adaptive Traffic Prediction at the ITS Edge With Online Models and Blockchain-Based Federated Learning
IEEE Transactions on Intelligent Transportation Systems ; 25 , 9 ; 10725-10740
01.09.2024
4835760 byte
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Multi-period traffic prediction method based on federated learning
Europäisches Patentamt | 2025
|Cluster type federated learning-based urban traffic flow prediction method
Europäisches Patentamt | 2024
|