LightGBM-based link provisioning framework for low ATP delay in optical satellite networks

Optical satellite networks (OSNs) have emerged as a critical infrastructure for space-terrestrial integrated communications, enabled by high-capacity laser-based inter-satellite links (ISLs). While laser ISL establishment requires a precise acquisition, tracking, and pointing (ATP) procedure, it inevitably introduces additional setup delays. As user services are transmitted over multiple ISLs, ATP delays are incurred on each traversed link and accumulate along multi-hop paths in serial or parallel ways when services span multiple satellites, resulting in non-negligible end-to-end latency. Existing routing studies in OSNs have considered this ATP-induced delay less. To address this problem, this paper formulates the problem as an integer linear programming (ILP) model with the objective of minimizing the total ATP delay under both serial and parallel ISL establishment, serving as an optimizing performance benchmark. To improve service responsiveness, we propose a prediction-based link provisioning (LP) framework that predicts service requests and proactively provisions candidate ISLs. The LP framework consists of request prediction, graph construction, and resource scheduling modules. Specifically, a LightGBM-based prediction module is designed to forecast service requests, and both correctly and incorrectly predicted requests are considered in their corresponding topology graphs. Corresponding ISLs were reserved in advance via the ATP-aware path selection (APS) algorithm to overlap ATP delays. Simulation results verify that the proposed framework achieves near-optimal performance with only a 4.9% gap compared to the ILP benchmark, while reducing the total ATP waiting time by up to 90.83% and 89.81% under serial and parallel dynamic service arrival scenarios, respectively.