Authors - Maharajpet Sheela, Roy Ratnakirti, Thakur Manish Kumar Abstract - The swift expansion of networked vehicles and city traffic has presented major challenges to the management of traffic in smart cities and therefore solutions that are intelligent and privacy-protecting are needed. In this paper, a Drift-Aware Edge-Federated Spatio-Temporal Intelligence (EF-STI) model that utilizes Long Short-Term Memory (LSTM) networks to predict traffic flowing predictively and accurately is offered. Instead of using a traditional centralized or cloud-based model, EF-STI allows individual vehicle or roadside edge units to locally-train a lightweight LSTM model, which is only encrypted model parameters are shared with an aggregator located globally. In order to deal with the non-static and dynamic traffic, a drift-aware federated optimization plan is implemented, which enables the system to adjust to the sudden change and different traffic patterns. The framework uses Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) communication to predict traffic density and flow with minimal latency enabling proactive interventions to traffic management problems including dynamic signal control, route recommendations, and congestion warnings. It is proved by experimental analysis that EF-STI has better prediction accuracy, lesser communication overhead, and better adaptability than traditional methodology. The article demonstrates a special intersection of edge computing, privacy-sensitive federated learning, spatio-temporal LSTM modeling, and vehicular networking, building intelligent transportation systems to be scalable, secure, and autonomous in traffic management.
Open Zoom