| In order to realize the vision of "ubiquitous intelligence" in the 6G era,space-air-ground integrated networks(SAGIN)are regarded as a powerful solution.By providing global seamless wide coverage and low-latency highperformance cloud computing services,Internet of Things(IoT)devices in remote areas can enjoy holographic computing offloading services that save local computing resources and energy by connecting with satellites and unmanned aerial vehicles.However,the SAGIN are highly vulnerable to eavesdropping attacks,which makes it a challenging problem to protect users’ data privacy in task scheduling.In response to this problem,this thesis introduces encryption service providers and studies the privacy-driven resource allocation for SAGIN based on machine learning.The main work includes the following two aspects.First,this thesis designs an encryption configuration level and encryption service pricing mechanism model for data transmission in the SatelliteTerrestrial Internet of Things.This model mainly considers that different encryption configurations correspond to different encryption costs,and both users and service providers hope to maximize their own benefits while reducing their own costs.Therefore,this thesis models the interaction process between users and service providers choosing encryption configurations and service prices as a Stackelberg game,and employ a distributed reinforcement learning algorithm to solve the game’s Nash equilibrium.Simulation results prove the effectiveness of the proposed algorithm.Second,this thesis proposes a privacy-driven security-aware SAGIN task scheduling mechanism based on multi-agent proximal policy optimization.Starting from the different privacy protection needs of users,this thesis jointly optimize the delay,energy consumption and security utility of computing of(loading.In this thesis,IoT devices in the SAGIN are regarded to be cooperative,and an algorithm combining convex optimization and multi-agent proximal strategy optimization is proposed to solve the strategies of task scheduling,computing resource allocation and encryption configuration.Simulation results demonstrate that the proposed algorithm outperforms other benchmarks in terms of convergence speed,offload overhead,and security. |
