.png){kind=icon}
.png){kind=icon}
Return
A Deep Reinforcement Learning-Based Service Caching and Computation Offloading Algorithm for Multi-UAV-Enabled MEC Systems
.png){kind=icon}
PRE
AI
DOI：10.1109/TCCN.2026.3670152.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper investigates the joint service caching and computation offloading optimization problem in a multi-UAV-enabled mobile edge computing (MEC) system. The problem is formulated as a non-linear integer programming problem with an objective to minimize the average service delay of all computing tasks generated by ground users in the system. To solve the formulated problem, the original problem is first decomposed into two coupled subproblems, i.e., service caching and computation offloading, and a two-layer learning framework is then introduced to solve the subproblems. The learning framework consists of an upper layer and a lower layer, which are used for service caching policy learning and computation offloading policy learning, respectively. Based on the learning framework, a deep reinforcement learning (DRL)-based joint optimization algorithm, which consists of a lower-layer algorithm and an upper-layer algorithm, is further proposed to perform service caching and computation offloading. The lower-layer algorithm uses a graph neural network (GNN) and an attention mechanism to extract relevance information on the states of ground users in the system and uses a deep Q-network (DQN) to make UAV assignment decisions based on the extracted information. The upper-layer algorithm uses an actor-critic structure to make optimal service caching decisions. Simulation results demonstrate that the proposed algorithm can significantly reduce the average service delay of all computing tasks in the system compared with benchmark algorithms.
Keywords:
UAV
MEC
deep reinforcement learning
service caching
computation offloading
Journal
I
IF:
7
Papers:
1.5K
Citations:
5.5K