| dc.description.abstract | The satellite-ground (SG) integrated networks are regarded as a promising network structure, which can provide ubiquitous intelligence and pervasive services for multiple ground users. Moreover, digital twin (DT) can drive real-time data mapping and wireless access from usual physical utilities to digital units. Therefore, the fusion of SG and DT can decrease the gap between real-time data analysis and physical system states, which can help boost SG-DT edge intelligence paradigms. Nevertheless, the unexpected task arrivals, time-varying channel gains, and distrust among ground devices cause the network service performance degradation. Hence, in this paper, we propose a privacy-assisted blockchain computation offloading model to shine upon original tasks to the corresponding aerial platforms, and then orchestrate the task scheduling, resource allocation, and privacy protection. Additionally, we envision a Lyapunov stability theory-based multi-agent federated reinforcement learning (LST-MAFRL) algorithm to further resolve the CPU cycle frequency, the size of each blockchain, the number of DTs, and related harvested solar energy to minimize the execution energy consumption and privacy time overhead. Finally, extensive simulation results indicate that the proposed LST-MAFRL algorithm framework outperforms some state-of-the-art benchmarks for the sake of execution energy efficiency, processed bit quantities, and privacy time overhead. | en_US |
