Minimize Average Tasks Processing Time in Satellite MobileEdge Computing Systems via a Deep Reinforcement Learning Method

In recent years, the development of Low Earth Orbit (LEO) satellites and the advancement of the Mobile Edge Computing (MEC) paradigm have addressed the requirement for Satellite Mobile Edge Computing (Sat-MEC) servers to support remote Internet of Things (IoT) Mobile devices (IMDs) in the absence of terrestrial networks. However, optimizing the execution time of ground-based tasks in this scenario is still a huge challenge due to the heterogeneity of tasks and Sat-MEC servers.In this paper, we proposed an algorithm in the context of Sat-MEC architecture aimed at minimizing average tasks processing time. In a significant portion of other papers that have considered the access of Sat-MEC to terrestrial network architectures, the interplay between task characteristics and server characteristics is often not considered simultaneously, and basically only single tasks are considered in sequence for offloading decisions, which is difficult to seek the optimal average processing time for multiple tasks, therefore, we proposed a algorithm which considers several factors, including the collaboration among LEO satellites, the concurrency and heterogeneity of tasks, the dynamics of LEO satellites, the heterogeneity of computation capability of Sat-MEC servers, and the heterogeneity of the initial queue for task computation. Due to the stochastic nature of the optimization problem and the high-dimensional nature of the solution space, we apply a deep reinforcement learning method to navigate through this high-dimensional space. And use a self-attentive mechanism in the Q-network to extract relational features between tasks and Sat-MEC servers. The decision process of task offloading solutions is determined by the similarity values between tasks and Sat-MEC servers. Simulation results confirm the effectiveness of our proposed algorithm, showing that the average processing time of tasks is lower compared to other baseline solutions.