An Optimal Power Allocation Algorithm for Multi-Beam Satellites Based on Reinforcement Learning

This paper presents an innovative approach to multi-beam power allocation for Geostationary Earth Orbit (GEO) satellites utilizing reinforcement learning techniques. The proposed method entails the development of a reinforcement learning model tailored for the optimization of power allocation across multiple beams of GEO satellites. This model delineates the pertinent components of reinforcement learning, including the definition of state space, action space, and reward function. Specifically, the reward function is intricately designed to encapsulate the dual objectives of enhancing communication coverage area and communication quality, with an explicit emphasis on the continuous improvement of these metrics. Leveraging the reinforcement learning framework, the constructed multi-beam power allocation model for GEO satellites undergoes comprehensive training to achieve optimal power distribution. Notably, the algorithm incorporates a hybrid approach, blending random exploration with optimal evolution strategies to circumvent convergence towards a fixed trajectory during optimization processes. Through rigorous simulation-based comparison experiments, the efficacy and robustness of the proposed optimization algorithm are rigorously evaluated and corroborated.