Digital twin-guided service-aware graph attention ant colony optimization for reliable dynamic IPv6 routing in power data networks

Power data networks operating over IPv6-enabled communication infrastructures carry heterogeneous grid services with different latency, bandwidth, operational-risk, and recovery-priority requirements. Static routing metrics and manually weighted Quality of Service (QoS) objectives are insufficient to represent the dynamic coupling among traffic load, service priority, link risk, queueing delay, and fault states in dynamic IPv6 routing environments. This paper proposes a digital twin-guided service-aware graph attention ant colony optimization framework, termed SAG-RTACO, for reliability-oriented dynamic IPv6 routing in power data networks. A routing-oriented network digital twin is first constructed to synchronize topology, traffic, QoS, risk, service, and fault states into a time-varying graph for IPv6 service flows. A service-aware Graph Attention Network (GAT) then learns link suitability under heterogeneous service requests, and the learned edge scores are incorporated into a risk-time Ant Colony Optimization (ACO) search process. A priority-aware primary-backup routing and fault-recovery mechanism is further designed to improve the recovery priority of critical grid services carried by the IPv6-enabled network. Experiments on a 14-node power data network with ten representative service types show that SAG-RTACO achieves a mean path satisfaction of 0.538, reduces average risk cost to 28.7, improves backup availability to 88.9%, and lowers the service failure rate to 9.4%. Additional out-of-distribution IPv6 routing tests under priority-skewed traffic, bursty arrivals, clustered faults, and combined distribution shifts yield GAT AUC values of 0.761–0.816 and SAG-RTACO path satisfaction values of 0.478–0.522 without retraining, indicating graceful performance degradation under unseen operating conditions. Additional fidelity, ablation, robustness, and statistical analyses further support the effectiveness of the proposed framework under the tested routing scenarios.