Dynamic Caching-Assisted Inter-Satellite Cooperative Computation Offloading in LEO Satellite Edge Computing

Low earth orbit (LEO) satellite edge computing network (LSECN) can complement the advantages of terrestrial networks and provide users with ubiquitous services. Currently, the majority of research on task software caching primarily concentrates on statically caching data at the network edge. However, in reality, user requests are time-varying, which makes it difficult to maintain high reusability. Therefore, this paper considers dynamic caching of task software in the mobile edge computing (MEC) servers to assist users in processing tasks. Tasks can be processed locally or offloaded to the LSECN to achieve on-board load balancing through inter-satellite cooperation. Specifically, considering the limited caching size and computing capabilities of MEC servers and the time-varying task requirements of users, a joint task software caching update and computation offloading problem is constructed to minimize the system cost, defined as the weighted sum of system delay and energy consumption, while ensuring resource constraints. To solve the above problem, we propose a computation offloading strategy and caching scheme based on twin delayed deep deterministic policy gradient (TD3) algorithm, which significantly reduces the system cost and has excellent convergence performance.