Multi-agent reinforcement learning method

By generating the final hyper-edge and correlation matrix of the multi-agent system, dynamically adjusting the communication between agents, the problem of message understanding and fusion in heterogeneous agent collaboration is solved, and efficient collaboration strategy learning and system adaptability are achieved.

CN120337975AInactive Publication Date: 2025-07-18BEIJING JUNDE INTELLIGENT COMPUTING TECHNOLOGY CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202510409796.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-04-02
Publication Date
2025-07-18
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In the prior art, the message understanding and fusion between isomeric agents has not yet fully considered the dynamic nature of communication and the correlation between agent connections, making it difficult for agents to learn useful collaboration strategies, and are low in flexibility, making it impossible for agents to adapt to complex and changeable scenarios.

Method used

By generating multiple final hyper-edges, based on the observation information and complex strategy mechanism of the multi-agent system, combining the multi-dimensional feature vector and the correlation matrix, the adaptation mode is dynamically adjusted to optimize communication and cooperation between the agents.

Benefits of technology

It realizes accurate capture and efficient integration of complex relationships between different agents, improving the accuracy of correlation calculation and system adaptability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120337975A_ABST
    Figure CN120337975A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-agent reinforcement learning method, and relates to the technical field of multi-agent reinforcement learning, and the method comprises the steps: generating a plurality of final hyperedges based on the observation information of each agent in a multi-agent system and a complex variable strategy mechanism; obtaining a multi-dimensional feature vector of each agent, and generating an association degree matrix corresponding to the multi-agent system based on the communication structure and the multi-dimensional feature vector; fusing the communication information of each agent based on the correlation degree matrix to obtain fused communication information; the converged communication information is used for guiding actions of corresponding agents; by setting a complex variable strategy mechanism, hyperedges are adjusted according to task state values and environment state values of the intelligent agents, an adaptive mode is dynamically adjusted, communication cooperation between the intelligent agents is optimized, and the effect of accurately capturing and efficiently fusing complex relations between different intelligent agents is achieved. Multi-dimensional feature vectors are introduced to generate an association degree matrix, and the effect of improving the accuracy of association degree calculation is achieved.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Multi-agent reinforcement learning method and device, electronic equipment and storage medium

    CN118052272A