A data-driven parallel adaptive large neighborhood search algorithm for a large-scale inter-satellite link scheduling problem

Abstract

Large-scale low-orbit communication satellite constellations play an indispensable role in the satellite Internet of Things, remote sensing and other fields, while the inter-satellite link scheduling problem is a crucial problem that significantly affects constellation communication performance. To address the inter-satellite link scheduling problem of large-scale low-orbit communication satellite constellations, this paper proposes a time-discrete network multi-commodity flow model in which satellites are viewed as network nodes, inter-satellite links are viewed as transmission paths and virtual intermediate nodes and virtual endpoints are set. Based on this model, a data-driven parallel adaptive large neighborhood search (DP-ALNS) algorithm, an extension of the ALNS, is proposed for solving the problem. In this algorithm, a probability prediction model is trained based on extreme gradient boosting to predict the probabilities that a satellite is connected to its visible satellites. An initial solution generation strategy is further adopted for constructing a high-quality initial solution. Meanwhile, an adaptive mechanism including a rule adaptive layer and an operator adaptive layer is used to improve the search performance of the algorithm, in which the rule adaptive layer affects solution selection and the operator adaptive layer is used in the neighborhood search. Computational experiments indicate that the DP-ALNS algorithm can optimize the average transmission time delay of all on-board data from approximately 8 units to only 4 units when the ratio of the number of gateway satellites to entry satellites is set to 1:8. Simultaneously, the DP-ALNS algorithm presents better overall performance than other state-of-the-art algorithms.