Task Scheduling for Low-Orbit Satellite Edge Computing Based on Deep Q Network

In this paper, a task scheduling model is constructed on the basis of the reinforcement learning and the task scheduling strategy is developed with a deep Q network algorithm, which is purposed to improve the efficiency of real-time multi-task processing for low-earth orbit satellites and reduce the delay of task processing. Under the DQN task scheduling strategy, a reward mechanism based on thread service time is applied to ensure the rapid convergence of deep neural networks. At the same time, the centralized training of multiple is performed to reduce computational dimensions and shorten the time taken to train deep neural networks. Therefore, it is suitable for the real-time task processing of low-earth orbit satellite internet. Experimental results show that the DQN task scheduling algorithm proposed in this paper outperforms the other task scheduling algorithms in the delay of task processing and the chance of success in task processing.