A method for unmanned aerial vehicle cooperative unloading based on double-flow space-time graph reinforcement learning

By constructing a dual-stream heterogeneous graph and a gating fusion mechanism, the problem of topological feature confusion in UAV cooperative networks is solved, and adaptive and efficient decision-making of UAV cooperative offloading strategies is achieved.

CN122114049APending Publication Date: 2026-05-29TIANJIN POLYTECHNIC UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-01-09
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

In existing drone collaborative networks, current technologies forcibly use a single physical topology to aggregate features, which makes it impossible to distinguish between situations with good communication quality but saturated computing power and situations with poor communication quality but idle computing power. This causes the offloading strategy to oscillate between transmission failures and queuing timeouts, reducing the system's convergence speed and collaborative efficiency.

Method used

A dual-stream heterogeneous graph is constructed to represent the physical topology of communication and the logical topology of computation, respectively. A gating fusion mechanism is designed to adaptively weight and fuse communication and collaboration features. Temporal dependencies are captured through a dual-stream graph convolutional network and a gating recurrent unit to generate the UAV's mission offloading decision.

Benefits of technology

The adaptive capability of the UAV collaborative unloading strategy has been realized, which improves the convergence speed and collaborative efficiency of the system and avoids unloading failure caused by topological feature confusion.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122114049A_ABST
    Figure CN122114049A_ABST
Patent Text Reader

Abstract

The application discloses a kind of based on double-flow space-time graph reinforcement learning's unmanned plane cooperation unloading method, belong to wireless communication and mobile edge computing field.The method includes: collecting unmanned plane cluster state, constructs the physical topology graph of characterizing communication quality and the logical topology graph of characterizing cooperation demand;Communication features and cooperation features are respectively extracted using double-flow graph convolution network;Through gate fusion unit, adaptive generation fusion coefficient is carried out to the weighting fusion of two kinds of features;The fusion features are input into the gate cycle unit to extract timing information, and generate unloading action through strategy network;Finally, multi-agent deep deterministic policy gradient algorithm is used to update network parameters.The application decouples communication constraint and cooperation demand by double-flow heterogeneous graph, realizes environment adaptive feature fusion using gating mechanism, effectively improves the cooperation unloading efficiency and system robustness of unmanned plane cluster in dynamic environment.
Need to check novelty before this filing date? Find Prior Art