Deep Reinforcement Learning for Safe Satellite Operational Mode Switching

Space weather phenomena can degrade satellite mission efficacy, particularly during space weather events where their magnitude intensifies significantly and unexpectedly. This research proposes a smart Satellite Control System (SCS) operational mode selection framework using a deep Dyna-Q algorithm, designed to mitigate the effects of space weather events and also enhance overall mission efficiency in real-time. The reinforcement learning agent selects the optimal operational mode between three modes: orbit transfer, orbit correction, and the safe hold mode. The decision-making process is carried out to maximize the reward function, which is defined to incorporate space weather health risks due to radiation in conjunction with mission performance metrics, such as tracking precision, promptness, and fuel consumption, among others.