A distributed multi-agent collaborative decision-making method based on an attention mechanism

Abstract

The invention discloses a distributed multi-agent collaborative decision-making method based on an attention mechanism. the distributed multi-agent observation information is fused based on an attention mechanism; The method comprises a local observation information coding step, an attention mechanism-based weight calculation step, an attention coefficient calculation step, a local observation information weighted summation step, an attention information and position information fusion step and a decision probability solving step. On the basis of a data driving mode, the mutual relation of local information is discovered autonomously, global decision information is constructed for a single agent, position information is processed independently, and a multi-agent system of which the information is richer and the spatial position relation is processed in a targeted manner can be realized.

Description

technical field [0001] The invention relates to the field of intelligent decision-making, specifically, a distributed multi-agent collaborative decision-making method based on an attention mechanism, which fuses the observation information of distributed multi-agents and combines it with the id of a single agent Coding fusion forms decision-making features, provides rich and effective information for final decision-making, and provides decision-making level. Background technique [0002] A multi-agent system is a system composed of computing units—agents that cooperate with each other. Agent collaboration is to organize multiple single-agents into a group and make each agent coordinate and cooperate effectively, so as to generate the overall problem-solving ability. There are two types of multi-agent systems: centralized and distributed. A centralized system consists of a core Agent and multiple independent cooperative Agents that are structurally dispersed. The core Agen...

AI Technical Summary

Problems solved by technology

However, BiCNet's organizational structure idea still does not get rid of the chain topology or ring topology, and does not have dynamic programming capabilities, which will cause great problems in actual practice

In the context of real tactics of mutual destruction, the destruction of a point in the network without dynamic programming capabilities will cause the complete termination of all information exchanges passing through that point

Under the premise of no recovery, the chain topology and ring topology rely too much on each endpoint in the network, resulting in the destruction of a few points, which will have a devastating impact on the agent communication network, and the group will be completely split and lose the ability to communicate and coordinate.

[0005] Therefore, the traditional multi-agent decision-making network is too cumbersome to mine the correlation of local information itself, and the traditional decision-making network does not pay enough attention to the location information

Images