Throughput Optimization in Faulty Prone Scenarios in LEO-UAV-SG Network Based on Q-Learning

The advancement of low-Earth orbit satellite networks has increasingly drawn attention to the integration of smart grids with these satellites. This paper focuses on a lowEarth orbit satellite and drone-assisted smart grid network architecture. First, the communication relationship between multiple users and drone relays is analyzed. Then, the concept of ‘fault nodes,’ a category of users that necessitate the occupation of fixed resources for stable communication, is introduced. Moreover, this paper investigates the throughput optimization problem. We transform it into a resource-matching problem and propose a Q-learning approach based on improved reward function to efficiently solve it. Simulation results verify that our proposed scheme yields superior throughput when the number of faults is small, and can effectively avoid resources occupied by fault nodes when the number of faults is high.