CN-122022346-A - Multi-agent mixing and synergy method

CN122022346ACN 122022346 ACN122022346 ACN 122022346ACN-122022346-A

Abstract

The application provides a multi-agent hybrid cooperation method, which comprises the steps of firstly obtaining task targets of a target service scene and agent cooperation requirement description, generating service task cooperation element topology through multi-dimensional semantic deconstruction and cooperation element association modeling, distributing tasks to an agent capability domain in a layering manner based on the topology, pre-judging conflicts through an agent capability-task adaptation conflict pre-judging matrix, determining conflict-free pre-distributing task tensors, constructing a cross-agent real-time cooperation-conflict resolution protocol cluster based on the tensors, generating a cooperation interaction rule protocol cluster, and finally carrying out dynamic cooperation scheduling and execution deviation correction on multi-agent execution state real-time feedback data through the protocol cluster to generate a multi-agent cooperation execution dynamic tensor instruction map so as to realize full-flow closed loop control of cooperation tasks. And the smoothness, reliability and task completion quality of multi-agent cooperation are improved.

Inventors

WANG JIAYING
LI HUAWEI
LI HAIYANG
WU XIAOYUAN
WU GUANGPENG
Zhang Ninglu

Assignees

北京甲板智慧科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. The multi-agent mixing and cooperating method is characterized by comprising the following steps: step 1, acquiring task targets and agent cooperative demand description in a target service scene, and performing multidimensional semantic deconstructment and cooperative element association modeling on the task targets and the agent cooperative demand description to generate service task cooperative element topology; Step 2, carrying out hierarchical task allocation on the capability domains of various intelligent agents based on the service task collaborative element topology to obtain a hierarchical intelligent agent task allocation scheme, and carrying out conflict prejudgment through an intelligent agent capability-task adaptation conflict prejudgment matrix to determine a conflict-free preassigned task tensor; Step 3, constructing a cross-agent real-time collaboration-conflict resolution protocol cluster based on the conflict-free pre-allocation task tensor to generate a collaboration interaction rule protocol cluster; And 4, carrying out dynamic collaborative scheduling and execution deviation correction on the real-time feedback data of the execution state of the multi-agent through the collaborative interaction rule protocol cluster to generate a multi-agent collaborative execution dynamic tensor instruction map so as to manage and control collaborative tasks in a target service scene based on the multi-agent collaborative execution dynamic tensor instruction map.
2. The multi-agent hybrid synergy method of claim 1, wherein step 1 comprises: step 11, carrying out service dimension disassembly and core index extraction on a task target in a target service scene to generate a task core index tensor; Step 12, carrying out collaborative element identification and association relation labeling on the intelligent agent collaborative demand description so as to generate a collaborative demand element conformation; And 13, fusing the task core index tensor and the element dimension of the cooperative demand element conformation to perform topological modeling so as to generate a service task cooperative element topology, wherein nodes of the service task cooperative element topology are formed by the index dimension of the task core index tensor and the element dimension of the cooperative demand element conformation together, and the edge weight is determined by the association matching degree of the task core index tensor and the cooperative demand element conformation.
3. The multi-agent hybrid synergy method of claim 2, wherein step 11 comprises: step 111, performing business process node splitting and index dimension definition on a task target to generate task process node characteristics; step 112, core index screening and quantization rule formulation are carried out on the node characteristics of the task flow so as to generate an initial index tensor; And 113, performing redundancy elimination and logic association integration on the initial index tensor to generate a task core index tensor, wherein the task core index tensor performs tensor dimension mapping on each core index according to the dimension of the business process node.
4. The multi-agent hybrid synergy method of claim 2, wherein step 13 comprises: Step 131, constructing a task-collaborative element association mapping rule, and performing feature matching on a task core index tensor and a collaborative demand element conformation to generate an element association pair, wherein the matching basis of the element association pair is the business association of the index dimension of the task core index tensor and the element dimension of the collaborative demand element conformation; Step 132, performing topological node definition and edge weight assignment on the element association pairs to generate an initial collaborative topology; And 133, carrying out integrity check and structure optimization on the initial collaborative topology to generate a business task collaborative element topology, wherein the business task collaborative element topology carries out linkage binding on index nodes of a task core index tensor and element nodes of a collaborative demand element conformation.
5. The multi-agent hybrid synergy method of claim 1, wherein step 2 comprises: step 21, collecting capability parameters and business adaptation records of each type of intelligent agent, and constructing intelligent agent capability domain feature topology; step 22, carrying out hierarchical disassembly on the tasks based on the business task collaborative element topology and the intelligent agent capability domain feature topology and matching the tasks to corresponding intelligent agents so as to generate a hierarchical intelligent agent task allocation scheme; and 23, constructing an intelligent body capability-task adaptation conflict pre-judging matrix, and carrying out conflict detection and resolution on the hierarchical intelligent body task allocation scheme to determine a conflict-free pre-allocation task tensor, wherein the conflict-free pre-allocation task tensor carries out Zhang Liangchong construction on tasks in the hierarchical intelligent body task allocation scheme according to the intelligent body dimension.
6. The multi-agent hybrid synergy method of claim 5, wherein step 21 comprises: step 211, carrying out real-time acquisition and standardization processing on capability parameters such as computing power, communication bandwidth, task processing expertise and the like of each intelligent agent so as to generate basic capability characteristics of the intelligent agent; step 212, the history service cooperative completion rate and the adaptation type record of each intelligent agent are called to generate intelligent agent service adaptation characteristics; and 213, fusing the basic capability features of the intelligent agents with the service adaptation features to construct an intelligent capability domain feature topology, wherein the intelligent capability domain feature topology constructs a hierarchical topology structure according to the capability features of the intelligent agents.
7. The multi-agent hybrid synergy method of claim 1, wherein step 3 comprises: Step 31, defining a communication protocol and a data interaction format of cross-agent cooperative interaction based on a task association relation of a conflict-free pre-allocation task tensor; Step 32, constructing a multi-agent hierarchical conflict tracing-shunting resolution strategy library, and formulating response mechanisms of different types of conflicts; And step 33, fusing the cooperative interaction communication protocol and the multi-agent hierarchical conflict tracing-splitting resolution strategy library, and constructing a cross-agent real-time cooperative-conflict resolution protocol cluster to generate a cooperative interaction rule protocol cluster, wherein the cooperative interaction rule protocol cluster carries out protocol binding on the communication protocol and the resolution strategy according to conflict types.
8. The multi-agent hybrid synergy method of claim 7, wherein step 31 comprises: step 311, determining a real-time communication link and a data transmission interface which are cooperated across the intelligent agents, and defining field specifications and check rules of interface data; step 312, aiming at the cooperative demands of different types of tasks, formulating a synchronous/asynchronous interactive communication protocol, wherein the synchronous interactive protocol is adapted to the high-timeliness cooperative demands of a core task layer; Step 313, integrating the communication link configuration and the interaction protocol to generate a cooperative interaction communication protocol module, wherein the cooperative interaction communication protocol module embeds the communication link parameters and the interaction protocol rules into task nodes of the conflict-free pre-allocation task tensor.
9. The multi-agent hybrid synergy method of claim 1, wherein step 4 comprises: step 41, deploying a multi-agent execution state acquisition node, and acquiring feedback data such as task progress, resource occupation, abnormal alarm and the like of each agent in real time; step 42, inputting feedback data into the collaborative interaction rule protocol cluster, and performing execution deviation recognition and collaborative scheduling instruction generation, wherein the collaborative scheduling instruction generation needs to call a resolution strategy module in the collaborative interaction rule protocol cluster; And 43, integrating the scheduling instruction and the task execution baseline, generating a multi-agent cooperative execution dynamic tensor instruction map, performing full-flow closed loop control on the cooperative task based on the multi-agent cooperative execution dynamic tensor instruction map, and dynamically updating the scheduling instruction according to the agent dimension and the task dimension by the multi-agent cooperative execution dynamic tensor instruction map.
10. The multi-agent hybrid co-process of claim 9, wherein step 41 comprises: step 411, embedding a state acquisition interface in a task execution module of each agent, and setting a data acquisition period and reporting rules; Step 412, performing format normalization and anomaly filtering on the collected original state data to generate standardized feedback data, wherein the standardized feedback data needs to be matched with the data dimension of the conflict-free pre-allocation task tensor; And 413, constructing time sequence storage and associated indexes of feedback data to form a multi-agent execution state real-time feedback data pool, wherein the index dimension of the data pool is consistent with the node dimension of the agent capability domain feature topology.

Description

Multi-agent mixing and synergy method Technical Field The application relates to the technical field of multi-agent coordination, in particular to a multi-agent mixing coordination method. Background In complex business scenes such as intelligent manufacturing, intelligent transportation, distributed AI decision and the like, the application of the multi-agent system is more and more extensive, and the core requirement is that complex business tasks are efficiently completed through the cooperative cooperation of a plurality of agents, so that the overall working efficiency and quality are improved. Along with the improvement of the complexity of the service scene, higher requirements are provided for the real-time performance, compatibility and stability of the cooperation of multiple intelligent agents. In the prior art, a multi-agent cooperation scheme exists, wherein the scheme firstly carries out simple splitting on service tasks, then distributes subtasks according to static function parameters of agents, realizes data interaction among the agents through a preset fixed communication protocol, and adopts a unified processing mode to deal with the conflict. The general flow is that after the service task is defined, the service task is split into a plurality of subtasks according to the functional module, the tasks are distributed according to the static attributes of the agent such as calculation power, task processing expertise and the like, the agent transmits data according to a fixed format, and when the resource competition or data interaction problem occurs, a unified conflict resolution flow is started. However, the existing scheme has obvious technical defects that dynamic evaluation and element association analysis on the coordination capability of the intelligent agents are lacking, task allocation is only based on static attributes, the coordination suitability among the intelligent agents and potential conflict in task execution are not fully considered, the conflict occurrence rate is high, the conflict processing is lacking in pertinence, the coordination efficiency is low, the task execution deviation is large, and the high-quality requirement on the coordination of multiple intelligent agents in a complex service scene cannot be met. Disclosure of Invention In order to solve the above technical problems, the present application provides a multi-agent hybrid synergistic method to at least alleviate the above technical problems. The technical scheme provided by the embodiment of the application is as follows: A multi-agent hybrid cooperation method includes the steps of obtaining task targets and agent cooperation requirement description under a target service scene, conducting multi-dimensional semantic deconstruction and cooperation element association modeling on the task targets and the agent cooperation requirement description to generate service task cooperation element topology, conducting hierarchical task distribution on capability domains of various types of agents based on the service task cooperation element topology to obtain a hierarchical agent task distribution scheme, conducting conflict prejudgment through an agent capability-task adaptation conflict prejudgment matrix to determine a conflict-free pre-distribution task tensor, constructing a cross-agent real-time cooperation-conflict resolution protocol cluster based on the conflict-free pre-distribution task tensor to generate a cooperation interaction rule protocol cluster, conducting dynamic cooperation scheduling and execution deviation correction on multi-agent execution state real-time feedback data through the cooperation interaction rule protocol cluster to generate a multi-agent cooperation execution dynamic tensor instruction map, and conducting control on cooperation tasks under the target service scene based on the multi-agent cooperation execution dynamic tensor instruction map. The multi-agent mixing and synergy method provided by the application has the technical advantages that: In the task and demand analysis link, the existing scheme only carries out simple splitting on the task, and does not consider the association relation between the task core index and the cooperative element, so that the task allocation lacks pertinence. According to the scheme, through carrying out service dimension disassembly and core index extraction on the task targets, element identification and association labeling are carried out on the cooperative demands, and service task cooperative element topology is generated in a fusion mode, so that the task and the demands are more comprehensive in deconstructment and clearer in association. The multi-dimensional semantic deconstructing and topological modeling mode can accurately capture key nodes for task execution and core elements of cooperative requirements, provides a basis for follow-up task allocation to be more fit with an actual scene, and solves the problem that the ta