Abstract
Urban digital twins (UDTs) are increasingly recognized as powerful tools for simulating, monitoring, and managing city systems in real time. However, the massive data volumes generated by Internet of Things (IoT) sensors and the stringent latency requirements of real-time applications pose significant computational challenges. This paper proposes a novel edge-computing architecture tailored for real-time urban digital twins, enabling distributed data processing at the network edge to reduce latency and bandwidth consumption. The framework integrates a hierarchical edge layer with lightweight digital twin models that synchronize with cloud-based counterparts. We evaluate the system through a simulated smart city scenario involving traffic monitoring and environmental sensing. Results demonstrate that the edge-based approach achieves sub-100 millisecond latency for 95% of data streams, while reducing cloud data transfer by 68% compared to a cloud-only baseline. Furthermore, we analyze the trade-offs between model fidelity and computational efficiency, showing that selective model compression can maintain prediction accuracy above 92% while reducing inference time by 40%. The findings highlight the viability of edge computing for enabling scalable, real-time urban digital twins and provide design guidelines for practitioners deploying such systems in smart city contexts.