Predicted Physical–SLA State Marginal-Drift Target Selection for Forced Handover in LEO Satellite Networks

Abstract

Forced handovers in low-Earth-orbit (LEO) satellite networks require a replacement satellite when the incumbent becomes invisible. The replacement choice affects shared access and backhaul congestion, handover interruption, and heterogeneous service-level-agreement (SLA) debt. This paper presents TPMD–PS, a target-selection method that evaluates each shortlisted replacement through a causal finite-horizon rollout of physical traffic queues and class-level SLA virtual queues. A network-aggregate rollout captures shared-resource coupling, while a fine-grid focal-user model resolves handover interruption and local queue evolution. The score combines endpoint Lyapunov-potential differences with discounted utility, load, and visibility terms. We show that rate stability of the SLA virtual queues implies their time-average budgets and derive an $H$ -step sufficient stability condition under comparison-policy drift slack, bounded non-drift cost, and bounded rollout-score error. The fluid-queue evaluation reports a workload-delay proxy and uses 30 paired realizations at each of four traffic loads. TPMD–PS substantially reduces the delay proxy and endpoint backlog relative to instantaneous-SINR selection under congestion, while differences from predictive variants are smaller. Paired same-state validation further shows that SLA-aware target changes reduce the endpoint virtual-debt potential on average across all tested loads, with realization-level intervals below zero from nominal through high load, and class-specific trajectories identify eMBB delay debt as the dominant active SLA state. The results demonstrate the value of jointly predicting physical congestion and heterogeneous SLA debt in forced-handover target selection.