A data resource scheduling method based on multi-agent game cooperation

By combining hierarchical agent modeling and multi-agent reinforcement learning, the problem of unbalanced and unstable resource scheduling in data centers and cloud computing platforms is solved, and efficient and stable data resource scheduling is achieved.

CN121277640BActive Publication Date: 2026-06-26XINGCHENSHUMENG (HANGZHOU) TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN ยท China
Patent Type
Patents(China)
Current Assignee / Owner
XINGCHENSHUMENG (HANGZHOU) TECH CO LTD
Filing Date
2025-10-13
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

In existing data centers and cloud computing platforms, traditional data resource scheduling methods suffer from slow response speed, rigid scheduling results, and uneven resource allocation when faced with multi-task concurrency, resource heterogeneity, and dynamic environmental changes. Furthermore, existing multi-agent game theory methods lack hierarchical modeling and refined game analysis, resulting in high computational complexity and unstable scheduling strategies.

Method used

By employing hierarchical agent modeling, multi-agent reinforcement learning, and local Nash response policy derivation, a hierarchical agent structure is constructed, comprising task execution, resource allocation, and coordination control classes. Through distributed reinforcement learning and local Nash response policy derivation, collaborative decision-making and adaptive scheduling of multiple agents are achieved.

Benefits of technology

It achieves globally optimal scheduling in a large-scale, multi-source, heterogeneous data environment, improving scheduling accuracy, resource utilization, and system stability, while reducing scheduling conflict rate and system energy consumption.

โœฆ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121277640B_ABST
    Figure CN121277640B_ABST
Patent Text Reader

Abstract

The application discloses a data resource scheduling method based on multi-agent game cooperation, which comprises the following steps: collecting running state data to generate scheduling state data set; constructing a hierarchical agent structure based on the scheduling state data set and generating an initial parameter set; obtaining an initial strategy network set based on the hierarchical agent structure and the initial parameter set; freezing the current strategy parameters based on the initial strategy network set to obtain a local Nash response strategy combination; constructing a joint optimization objective function based on the local Nash response strategy combination and the initial strategy network set; outputting an updated strategy network set by using a multi-agent reinforcement learning algorithm; collecting feedback information to form scheduling execution feedback data; and repeatedly executing an iteration process based on the scheduling execution feedback data and the updated strategy network set. The hierarchical agent game reinforcement learning realizes the cooperative scheduling of data resources.
Need to check novelty before this filing date? Find Prior Art