Abstract
Real-time traffic management in smart cities demands low-latency decision-making that centralized cloud architectures cannot guarantee. Edge-cloud collaborative architectures promise to address this by distributing computation between edge nodes and cloud data centers. However, existing frameworks often lack systematic evaluation of orchestration strategies, data fusion techniques, and scalability under realistic traffic conditions. This paper proposes a unified edge-cloud collaborative architecture integrating vehicular fog computing, federated learning, and digital twin models for real-time traffic signal control and congestion prediction. The architecture is evaluated through a hybrid simulation combining SUMO (Simulation of Urban MObility) with a custom edge-cloud simulator. Performance metrics include end-to-end latency, throughput, prediction accuracy, and resource utilization. Results demonstrate that the proposed architecture reduces average response time by 38% compared to cloud-only approaches and achieves 91% prediction accuracy for traffic flow. Key contributions include a novel task offloading algorithm based on fuzzy logic, a digital twin synchronization protocol, and a lightweight federated learning aggregation scheme. Findings highlight the importance of dynamic edge-cloud orchestration and collaborative filtering in adapting to traffic variability. The study provides design guidelines for smart city infrastructure planners and underscores the potential of hybrid architectures for scalable, resilient traffic management.
Keywords
edge-cloud collaboration, real-time traffic management, smart cities, latency optimization, vehicular fog computing, federated learning, digital twin, task offloading