Knowledge-assisted adaptive large neighbourhood search algorithm for the satellite-ground link scheduling problem

Abstract

Communications between satellites and the ground supported by satellite-ground links (SGLs) are indispensable to the daily management of large-scale low-orbit communication satellite constellations. The SGL scheduling problem (SGLSP) focuses on determining which nonconflicting SGLs should be activated to maximize the total linking time. To solve this problem, we build an integer programming (IP) model and simultaneously propose a knowledge-assisted adaptive large neighbourhood search algorithm (KA-ALNS) based on this model, unifying the strengths of IP and data mining within the ALNS framework. In this algorithm, IP, which is deeply embedded in the ALNS, is employed to further refine the solution improved by ALNS. Then, a data mining method named frequent pattern mining is employed for extracting knowledge of excellent solutions to construct new solutions. The KA-ALNS algorithm that combines data mining methods and IP modeling can effectively solve the SGLSP. A three-day experimental setting with 20 ground stations in the China region, 40 ground stations in the world region (each station is equipped with 2 antennas), and a constellation containing 168 satellites is conducted. Computational tests with 12 different scale instances indicate that KA-ALNS outperforms two state-of-the-art algorithms given both 10 and 60 min of scheduling time.