Satellite dag task offloading decision method and system based on graph neural network

By using a graph neural network-based approach, combining lightweight graph convolutional networks and graph attention networks, the systemic defects in the DAG mission offloading of LEO satellites were addressed, achieving low-latency and highly reliable mission offloading, and improving the processing efficiency and reliability of satellite edge computing.

CN122173284APending Publication Date: 2026-06-09CHONGQING UNIV OF POSTS & TELECOMM
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHONGQING UNIV OF POSTS & TELECOMM
Filing Date
2026-03-05
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing technologies have systemic defects in LEO satellite DAG mission offloading, including the disconnect between DAG mission dependencies and satellite topology, insufficient decision adaptability, insufficient resource constraint adaptability, and weak dynamic topology adaptability, resulting in problems such as high latency, numerous conflicts, poor stability, and low resource utilization.

Method used

A graph neural network-based approach is adopted, which couples DAG task dependencies with satellite topology features by modeling graph structures. Lightweight graph convolutional networks and graph attention networks are combined to extract key correlation information. Lightweight dynamic graph convolutional networks and graph attention networks are designed to extract features. A reinforcement learning framework is optimized by combining multi-objective reward functions and lightweight proximal policies to achieve centralized decision-making and two-way data interaction between satellite and ground, adapting to onboard resource constraints and dynamic topology environments.

Benefits of technology

It achieves low-latency, conflict-free, and highly reliable DAG task offloading, significantly improving task processing efficiency and reliability, adapting to the optimal offloading effect in complex scenarios, and promoting the development of satellite edge computing technology.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122173284A_ABST
    Figure CN122173284A_ABST
Patent Text Reader

Abstract

The present application relates to a kind of satellite DAG task unloading decision method and system based on graph neural network, belong to on-board computing unloading technical field.The method includes problem modeling and optimization target definition, task dependence and satellite topological graph modeling, graph neural feature extraction, access satellite centralized decision mechanism is constructed, satellite-ground collaborative evolution and closed loop update etc.Process, DAG task dependence is coupled with satellite topological characteristics by graph structure modeling, relevant information is extracted by means of graph neural, adapt to on-board resource constraint and dynamic topological environment.The system designs task input and modeling layer, satellite network modeling and environment layer, core processing and decision layer and satellite-ground collaborative and update layer to execute above-mentioned method, in the rigid constraint of on-board computing power, energy, bandwidth and LEO satellite dynamic topological environment, the low delay, conflict-free, high reliable unloading of DAG task containing dependence relationship is realized, and provides core algorithm support for the adaptive optimization of complex task satellite edge computing.
Need to check novelty before this filing date? Find Prior Art