Predict first, sync when needed: A risk-aware approach to adaptive digital twin synchronization under bandwidth constraints

Maintaining real-time synchronization between physical assets and their Digital Twins over bandwidth-limited wireless links is a complex challenge. Existing approaches either transmit all sensed data, resulting in unnecessary network overhead, or rely solely on forecasting, which cannot guarantee fidelity when operating conditions change. In response, we present a risk-aware, event-driven synchronization framework

that leverages predictive modeling to minimize network traffic while maintaining the fidelity of digital twins. Our approach utilizes a Gated Recurrent Unit-based quantile forecaster to predict key states with calibrated uncertainty. Additionally, it employs

online quantile coverage and Kullback-Leibler divergence tests to monitor reliability and detect changes in distribution. The system operates in a predictive mode, fetching only selected states, and shifts to full synchronization and model retraining when deviations are detected. We conducted experiments using public LTE and 5G datasets, achieving R2 > 0.80 in most cases, with a peak of 0.99. Real-world evaluations using a commercial smartphone showed network-traffic reductions of 5.4–7.0% versus

continuous synchronization, along with mean absolute error (MAE) reductions of about 64% on average relative to simpler predictive methods. These findings demonstrate that predictive synchronization, coupled with awareness of drift, enables highfidelity

digital twinning even under strict bandwidth constraints.