CN-121981271-A - Abnormality attribution method suitable for multiple intelligent agents

CN121981271ACN 121981271 ACN121981271 ACN 121981271ACN-121981271-A

Abstract

The application discloses an anomaly attribution method suitable for a multi-agent, which belongs to the field of anomaly detection of the multi-agent, and comprises the steps of carrying out priori construction on physical communication relation data and logic constraint relation data of the multi-agent system to obtain priori constraint data of causal search space states of the multi-agent system, carrying out causal relation solving based on the priori constraint data to obtain causal relation data of the multi-agent system, and screening root causes of the causal relation data to obtain attribution information used for representing abnormal root causes of operation of the multi-agent system. According to the application, the causal relation solving and root cause screening are limited to the prior constraint, so that the accuracy of abnormal root cause screening is improved, the misjudgment condition caused by causal inference deviation is reduced, and the defect of insufficient accuracy of abnormal root cause screening is overcome.

Inventors

WANG PINGHUI
GUO WEI

Assignees

西交网络空间安全研究院
西安交通大学

Dates

Publication Date: 20260505
Application Date: 20260203

Claims (10)

1. An anomaly attribution method applicable to multiple agents, comprising: The method comprises the steps of carrying out priori construction on physical communication relation data and logic constraint relation data of a multi-agent system to obtain priori constraint data of a causal search space state of the multi-agent system, wherein the physical communication relation data are used for representing connection relations among agents in the multi-agent system, the logic constraint relation data are used for representing dependency relations among agents in the multi-agent system, and the priori constraint data are used for representing a feasibility limiting state of the causal search space; Carrying out causal relation solving based on the priori constraint data to obtain causal relation data of the multi-agent system, wherein the causal relation data is used for representing causal relation strength formed between a plurality of element events in the causal search space and a plurality of abnormal effects respectively; based on root cause screening of the causal relationship data, attribution information used for representing abnormal root cause of the multi-agent system is obtained.
2. The anomaly attribution method for multi-agent according to claim 1, wherein the a priori constructing the physical connectivity relationship data and the logical constraint relationship data of the multi-agent system to obtain a priori constraint data of causal search space states of the multi-agent system comprises: determining structural relationship data characterizing a connectivity relationship between each of the plurality of agents in the multi-agent system based on the physical connectivity relationship data; Determining, based on the logical constraint relationship data, logical relationship data that characterizes event transfer relationships between individual agents in the multi-agent system; Based on the structural relationship data and the logical relationship data, a priori constraint data characterizing the causal search space feasibility defining state is generated.
3. The anomaly attribution method for a multi-agent system according to claim 1, wherein the performing a causal relationship solution based on the a priori constraint data to obtain causal relationship data for the multi-agent system comprises: The prior constraint data is used as a limiting condition for the feasibility of causal relation among element events in the causal search space so as to determine a solving search range for the causal relation among the element events in the multi-agent system; And solving the causal relationship intensity among each element event in the multi-agent system based on the determined solving search range so as to obtain the causal relationship data.
4. The anomaly attribution method applicable to multiple agents of claim 1, wherein the root cause screening based on the causal relationship data to obtain attribution information for characterizing the root cause of the multiple agent system operation anomaly comprises: determining target causal relation intensities in the causal relation data according to a preset root cause screening threshold, and determining element events and abnormal effects corresponding to each target causal relation intensity as a group of first target element events; The attribution information is generated based on all of the first target element events.
5. The anomaly attribution method for multi-agent according to claim 1, the method is characterized in that the abnormality attribution method applicable to multiple intelligent agents further comprises the following steps: And adjusting the causal relationship data according to a preset fact condition set to obtain the adjusted causal relationship data.
6. The anomaly attribution method for multi-agent according to claim 5, wherein the adjusting the causal relationship data according to a preset set of facts conditions to obtain adjusted causal relationship data comprises: Determining each set of element events and abnormal effects that conflict with the fact conditions in the fact condition set as a set of second target element events respectively; and adjusting the causal relation intensity between the element events and the abnormal effects in the causal relation data in a one-to-one correspondence manner based on the determined second target element event, and determining the causal relation data with all the causal relation intensities adjusted as adjusted causal relation data.
7. An anomaly-based system for multiple agents, comprising: the system comprises a priori constraint module, a causal search space state judgment module and a causal search space state judgment module, wherein the priori constraint module is used for performing priori construction on physical communication relation data and logical constraint relation data of a multi-agent system to obtain priori constraint data of the causal search space state of the multi-agent system, the physical communication relation data are used for representing connection relations among the agents in the multi-agent system, the logical constraint relation data are used for representing dependency relations among the agents in the multi-agent system, and the priori constraint data are used for representing the feasibility limiting state of the causal search space; The causal relation module is used for carrying out causal relation solving based on the priori constraint data to obtain causal relation data of the multi-agent system, wherein the causal relation data is used for representing causal relation strength formed between a plurality of element events in the causal search space and a plurality of abnormal effects respectively; And the root cause screening module is used for screening the root causes of the causal relationship data to obtain attribution information used for representing the abnormal root cause of the multi-agent system operation.
8. A computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the anomaly attribution method for multi-agent according to any one of claims 1 to 6.
9. An electronic device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, which when executed by the processor performs the steps of the anomaly attribution method for multi-agent, as claimed in any one of claims 1 to 6.
10. A computer program product, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the steps of the anomaly attribution method for multi-agent according to any of claims 1 to 6.

Description

Abnormality attribution method suitable for multiple intelligent agents Technical Field The application belongs to the field of intelligent agent anomaly detection, and particularly relates to an anomaly attribution method, system, equipment, storage medium and computer program product applicable to multiple intelligent agents. Background With the wide application of multi-agent systems in the fields of industrial production scheduling, intelligent manufacturing, energy management, traffic cooperative control, large-scale distributed computing and the like, the overall operation state of the system is more and more dependent on the cooperative relationship among a plurality of agents. The multi-agent system generally has the characteristics of complex structure, frequent interaction, dynamic change of running state and the like, and once an agent or an interaction link in the system is abnormal, chain reaction is often caused, so that the system performance is reduced and even the whole system fails. Existing multi-agent anomaly detection and attribution techniques typically perform statistical analysis, pattern recognition, or machine learning based anomaly recognition based on system operational data. Such as by monitoring local state changes of the agents, identifying abnormal events using threshold detection, cluster analysis, or time series modeling, or attempting to build an interaction model between agents, locating potential sources of abnormalities through graph structure analysis, dependency modeling, or causal inference based correlation analysis. However, the prior art generally performs causal analysis based only on observed data, and often has difficulty in effectively distinguishing real causal links between a plurality of element events and a plurality of abnormal effects in a complex system, which results in inaccurate screening of abnormal root causes and failure to reliably output attribution information capable of characterizing the abnormal root causes of system operation. Disclosure of Invention The application aims to provide an anomaly attribution method, system, equipment, storage medium and computer program product applicable to multiple intelligent agents, which at least solve the problem of inaccurate screening of anomaly root causes. In a first aspect, an embodiment of the present application discloses an anomaly attribution method applicable to multiple agents, including: The method comprises the steps of carrying out priori construction on physical communication relation data and logic constraint relation data of a multi-agent system to obtain priori constraint data of a causal search space state of the multi-agent system, wherein the physical communication relation data are used for representing connection relations among agents in the multi-agent system, the logic constraint relation data are used for representing dependency relations among agents in the multi-agent system, and the priori constraint data are used for representing a feasibility limiting state of the causal search space; Carrying out causal relation solving based on the priori constraint data to obtain causal relation data of the multi-agent system, wherein the causal relation data is used for representing causal relation strength formed between a plurality of element events in the causal search space and a plurality of abnormal effects respectively; based on root cause screening of the causal relationship data, attribution information used for representing abnormal root cause of the multi-agent system is obtained. In a second aspect, an embodiment of the present application further discloses an anomaly attribution system applicable to multiple agents, including: the system comprises a priori constraint module, a causal search space state judgment module and a causal search space state judgment module, wherein the priori constraint module is used for performing priori construction on physical communication relation data and logical constraint relation data of a multi-agent system to obtain priori constraint data of the causal search space state of the multi-agent system, the physical communication relation data are used for representing connection relations among the agents in the multi-agent system, the logical constraint relation data are used for representing dependency relations among the agents in the multi-agent system, and the priori constraint data are used for representing the feasibility limiting state of the causal search space; The causal relation module is used for carrying out causal relation solving based on the priori constraint data to obtain causal relation data of the multi-agent system, wherein the causal relation data is used for representing causal relation strength formed between a plurality of element events in the causal search space and a plurality of abnormal effects respectively; And the root cause screening module is used for screening the root causes of the causal relationship dat