Abstract
This paper presents a traffic-driven scaling framework for a digital twin proxy pool (DTPP) in vehicular edge computing (VEC), designed to eliminate the latency and synchronization issues inherent in conventional digital twin (DT) migration approaches. The core innovation lies in replacing the migration of vehicle DTs between edge servers (ESs) with instantaneous switching within a pre-allocated pool of DT proxies, thereby achieving zero migration latency and continuous synchronization. The proposed architecture differentiates between short-term DTs (SDTs) hosted in edge-side in-memory databases for real-time, low-latency services, and long-term DTs (LDTs) in the cloud for historical data aggregation. A queuing-theoretic model formulates the DTPP as an M/M/c system, deriving a closed-form lower bound for the minimum number of proxies required to satisfy a predefined queuing-delay constraint, thus transforming quality-of-service targets into analytically computable resource allocations. The scaling mechanism operates on a cloud–edge collaborative principle: a cloud-based predictor, employing a TCN-Transformer fusion model, forecasts hourly traffic arrival rates to set a baseline proxy count, while edge-side managers perform monotonic, 5 min scale-ups based on real-time monitoring to absorb sudden traffic bursts without causing service jitter. Extensive evaluations were conducted using the PeMS dataset. The TCN-Transformer predictor significantly outperforms single-model baselines, achieving a mean absolute percentage error (MAPE) of 17.83%. More importantly, dynamic scaling at the ES reduces delay violation rates substantially—for instance, from 13.57% under static provisioning to just 1.35% when the minimum proxy count is 2—confirming the system’s ability to maintain service quality under highly dynamic conditions. These findings shows that the DTPP framework provides a robust solution for resource-efficient and latency-guaranteed DT services in VEC.