Multi-Agent Deep Reinforcement Learning-Based Computation Offloading in LEO Satellite Edge Computing System
Jian Wu, Min Jia, Ningtao Zhang, Qing Guo
Abstract
Efficient computation offloading is crucial for resource-constrained users in low earth orbit (LEO) satellite edge computing system. The proposed Dueling Double Deep Q Network (D3QN)-based computation offloading algorithm considers LEO satellite mobility, dynamic load levels, queuing theory, and jointly optimizes system delay and energy consumption. Simulation results show that the proposed algorithm has better system cost than other comparison algorithms.
Topics & Concepts
Computer scienceReinforcement learningComputation offloadingSatelliteComputationEnhanced Data Rates for GSM EvolutionEdge computingArtificial intelligenceDistributed computingComputer networkAlgorithmEngineeringAerospace engineeringSatellite Communication SystemsIoT and Edge/Fog Computing