SmartChargeNet: An Edge-Embedded Multi-Agent Sensor Intelligence Architecture for Real-Time Electric Vehicle Routing and Charging Infrastructure Coordination in Smart Urban Grids

The proliferation of electric vehicles (EVs) in urban environments generates a dense, continuous stream of sensor data from onboard battery-management sensors, GPS localization sensors, charging-station occupancy sensors, and road-infrastructure sensors, creating unprecedented demands on distributed sensing and coordination infrastructure. This paper introduces SmartChargeNet, an edge-embedded multi-agent sensor intelligence architecture whose primary contribution is a hierarchical sensor data processing pipeline that fuses heterogeneous, high-velocity sensor streams at the network edge to enable real-time EV routing and charging-station coordination as its application. The proposed system employs graph neural networks (GNNs) for multi-modal sensor data fusion, proximal policy optimization (PPO)-based reinforcement learning for sensor-driven scheduling decisions, and a hierarchical consensus protocol over a three-tier edge sensor-processing topology to address the multi-objective problem of minimizing travel time, reducing charging wait time, maximizing energy efficiency, and balancing grid load. SmartChargeNet is evaluated on the NYC-EV-2024 synthetic-enhanced dataset (real NYC mobility traces augmented with physics-model-generated BMS and occupancy sensor variables), comprising 50,000 EV sensor trajectories and 1200 charging stations over six months, and on SUMO-Urban-Grid, a sensor-rich simulation with 25,000 EVs and 800 charging stations under varied urban conditions. The architecture outperforms eight competitive baselines with improvements of 14.9% in average travel time, 20.9% in average charging wait time, 6.6% in energy efficiency, and 31.5% in grid load variance over the best-performing baseline, all measured exclusively on SUMO-Urban-Grid as the primary closed-loop evaluation environment; NYC-EV-2024 is used solely for GNN fine-tuning and offline sensor-state prediction validation, maintaining a mean system response time of 47 ms. The edge-embedded sensor-processing pipeline achieves O(N log N) routing complexity under a bounded per-cluster vehicle load constraint of at most 500 simultaneously active vehicles per cluster, enabling real-time processing of high-velocity, high-volume sensor streams without GPU acceleration, and validating SmartChargeNet as a scalable sensor intelligence platform for next-generation smart city deployments.