A hierarchical mixed learning resource coordination scheduling method

By building a global prediction model in the cloud and deploying real-time decision-making strategies at the edge, and combining asynchronous federated learning and graph neural networks, the problem of insufficient global prediction and edge response capabilities of scheduling methods in cloud-edge collaborative environments is solved, achieving efficient collaborative optimization and stability of resources.

CN122173283APending Publication Date: 2026-06-09南宁桂电电子科技研究院有限公司 +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
南宁桂电电子科技研究院有限公司
Filing Date
2026-03-05
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing scheduling methods struggle to simultaneously balance global prediction capabilities and real-time edge response capabilities in cloud-edge collaborative environments. Furthermore, they are difficult to maintain a stable scheduling scheme under asynchronous operation of multiple domain nodes and changes in network conditions, leading to resource waste and untimely response issues.

Method used

A hierarchical hybrid learning-based resource collaborative scheduling method is adopted. This method involves building a global prediction and optimization model in the cloud, deploying a lightweight real-time decision-making strategy at the edge, and using asynchronous federated learning to continuously update multi-domain models. Finally, it combines graph neural networks and reinforcement learning for task allocation and resource optimization.

Benefits of technology

It achieves collaborative optimization of tasks and resources in complex distributed computing environments, improves task processing efficiency and resource utilization, maintains system stability and generalizability, and solves the robustness problem of cross-domain scheduling.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122173283A_ABST
    Figure CN122173283A_ABST
Patent Text Reader

Abstract

This invention discloses a hierarchical hybrid learning-based resource collaborative scheduling method, relating to the fields of distributed computing and resource management technology. The method includes the following steps: constructing a multi-domain heterogeneous resource topology graph composed of cloud nodes and edge nodes; extracting the resource status of each node using a graph neural network; building a global prediction and optimization model in the cloud domain; allocating resources to the task set within future time windows; designing an optimization objective function for constraint optimization and refining it using a graph neural network; distributing the global resource allocation scheme to each edge domain; deploying lightweight real-time decision-making strategies in the edge domains; and independently running lightweight reinforcement learning agents in each edge domain. An asynchronous federated learning mechanism dynamically couples the global layer with each edge layer to achieve continuous updates to the multi-domain model. This invention integrates the global resource budget in the cloud with the real-time policy control in the edge domain, enabling collaborative optimization of tasks and resources in complex distributed computing environments.
Need to check novelty before this filing date? Find Prior Art