Towards sustainable synchronization of digital twins in next-generation IoT networks

The rising number of connected devices in nextgeneration Internet of Things (IoT) networks demands more efficient ways to synchronize physical systems and their digital twins. Although frequent updates keep twins accurate, they also consume large amounts of bandwidth and energy, while less frequent updates increase the risk of model drift. This paper introduces doctoral research comprising a three-phase strategy that leverages digital twin interactions to reduce network overhead while preserving real-time insights. First, we identify a subset of correlated parameters for multi-output regression, while the remaining parameters update on a fixed schedule. A deep reinforcement learning model then adjusts the full synchronization intervals based on predictive confidence. Second, we introduce a graph-based learning approach that enables multiple digital twins to share data and collaboratively impute missing information. Finally, we use generative AI to handle multi-modal inputs and support collaboration among heterogeneous twins, further cutting down on communication needs. Initial experiments with the multi-output regression approach demonstrate high accuracy, minimal inference latency, and low memory and CPU usage in our runtime environment for digital twins, underscoring its promising potential for efficient synchronization and real-time application deployment. By continually refining these techniques, we pave the way for scalable, low-overhead digital twin synchronization in massive IoT deployments.