Mobile-Edge Computing in SAGINs: A Hybrid Action Space P-DDQN Algorithm for Joint Offloading and Resource Allocation
The flexible deployment of uncrewed aerial vehicles (UAVs) and the wide-area coverage of low Earth orbit (LEO) satellites make their integration in space–air–ground integrated networks (SAGINs) a promising solution for communication in resource-constrained remote areas. This paper proposes a SAGIN framework supporting mobile edge computing (MEC) with a three-layer architecture, which provides heterogeneous computing resources for ground Internet of Things (IoT) devices and enables users in remote and underdeveloped regions to access computational services. Our objective is to minimize the weighted sum of energy consumption and latency in the SAGIN subject to satellite coverage time constraints and partial task offloading requirements. The optimization problem is formulated as a mixed-integer nonlinear programming (MINLP) challenge that jointly optimizes the UAV’s three-dimensional trajectory, IoT device association, transmit power, and task assignment. The coupled optimization variables form a hybrid action space with both discrete and continuous actions. To address this challenge, a parameterized double deep Q-network (P-DDQN) algorithm based on deep reinforcement learning (DRL) is proposed. The proposed method employs the DDQN algorithm to handle discrete actions and the deep deterministic policy gradient (DDPG) algorithm to generate continuous actions. Simulation results show that the proposed algorithm outperforms several baseline schemes in terms of system cost, providing an efficient solution for highly coupled hybrid decision optimization problems in SAGINs.