RL‐GA: An Optimization Strategy for Spatial Information Network Topology

Spatial information networks (SINs) are network systems characterized by their complex structure, high‐speed operation, and extreme dynamics. Their intricate topology, involving a vast number of nodes and links, coupled with a harsh operating environment, makes them susceptible to both natural and man‐made interference. This poses significant challenges to network stability. Therefore, the primary objective of topology optimization for SIN is to rapidly construct a network topology that is stable, highly connected, robust, and efficient. Previous optimization algorithms, including exact solutions based on dynamic programming and approximate solutions based on heuristics, struggle with limitations in escaping local optima and achieving rapid convergence. To address these drawbacks, this paper introduces a novel approach: a reinforcement learning–enhanced adaptive genetic algorithm (RL‐GA). This framework integrates reinforcement learning to dynamically adjust the key parameters of the genetic algorithm, enhancing its ability to escape local optima and accelerate convergence. Experimental results, based on simulations of the Iridium and Qianfan satellite constellations, demonstrate that the proposed RL‐GA algorithm outperforms existing methods in both solution quality (optimization results) and convergence speed. This advancement offers a more effective solution for SIN topology optimization, thereby enhancing the stability and efficiency of space‐based communication networks.