False information detection method in multi-agent system

Abstract

The invention relates to the field of information detection, in particular to a false information detection method in a multi-agent system, and the method comprises the steps: calling an agent cluster to execute interaction data for a plurality of historical tasks of real information, and carrying out the analysis of the agent cluster; comparing the numerical value of the specific dimension feature corresponding to each path segment with a specific dimension feature standard range, and determining the offset of the task execution path topological graph; evaluating a deviation tendency characterization coefficient of the received information according to the offset and the information receiving frequency of the agent cluster; judging whether the corresponding rule execution path topological graph is consistent with the task execution path topological graph of the agent cluster or not, and detecting and analyzing received information of the agent cluster; according to the method and the system, the false information is detected more comprehensively and reliably by adopting a multi-dimensional evaluation mode, the received information is marked, the sender corresponding to the received information is identified, and the receiver of the information sent by the agent cluster is tracked.

Description

Technical Field

[0001] The present invention relates to the field of information detection, and particularly to a method for detecting false information in a multi-agent system. Background Art

[0002] An agent is a terminal that includes a server capable of performing corresponding computing tasks and capable of data interaction. An agent cluster can process relatively complex tasks sent by a sender and send the relevant result information after processing to a receiver. Agent clusters are widely used in various fields. Therefore, it is crucial to perform security detection during the interaction process of agent clusters, such as identifying false information, forged information, etc.

[0003] Chinese Patent Application Publication No.: CN117155616A, discloses a method for predicting deception attack intention in a multi-agent system based on inverse reinforcement learning, belonging to the field of information processing technology. This invention considers the existence of deception attack phenomena in a multi-agent system and proposes an effective method for detecting false information injection. A method of multi-sensor data fusion is used for positioning, making the positioning more accurate. A Kalman filter is added for distributed state estimation, which can solve the problems of sensor noise and system instability. Assisting state estimation through neighbor node information makes the estimation more accurate and also makes the impact of external attacks on system performance smaller. Based on the traditional false data injection attack detection method, inverse reinforcement learning technology is added to infer the attacker's attack intention, solving the problems of complex data encryption and decryption algorithms and attack hiding in traditional technologies, and can effectively improve the effectiveness and accuracy of attack detection.

[0004] However, the following problems still exist in the prior art. In the prior art, the identity of the sender may be disguised, and thus it is not easy to identify the authenticity of the sent information by using a single method. The interaction rules of each node in the agent cluster are not considered, and the reliability of identifying false information is poor. Summary of the Invention

[0005] Therefore, the present invention provides a method for detecting false information in a multi-agent system to overcome the problems in the prior art that the identity of the sender may be disguised, and thus it is not easy to identify the authenticity of the sent information by using a single method. The interaction rules of each node in the agent cluster are not considered, and the reliability of identifying false information is poor.

[0006] To achieve the above object, the present invention provides a method for detecting false information in a multi-agent system, which includes: Invoke the interaction data of the historical tasks of the intelligent agent cluster for several pieces of real information to analyze the intelligent agent cluster, including determining the fluctuation value ranking of the corresponding dimensional features of each path segment of the task execution path topology graph after the intelligent agent cluster receives several pieces of real information, so as to determine the specific dimensional features corresponding to each path segment and the standard range of the specific dimensional features; In response to the intelligent agent cluster receiving information, compare the specific dimensional features corresponding to each path segment with the standard range of the specific dimensional features to determine the offset of the task execution path topology graph; According to the offset and the information reception frequency of the intelligent agent cluster, evaluate the deviation tendency characterization coefficient of the received information; Based on the deviation tendency characterization coefficient, determine whether to invoke the operation rule file of the intelligent agent cluster to determine the corresponding rule execution path topology graph, and determine whether the rule execution path topology graph is consistent with the task execution path topology graph of the intelligent agent cluster, so as to detect and analyze the received information of the intelligent agent cluster, including, Identify several nodes corresponding to the task execution path topology graph of the intelligent agent cluster to determine unexpected nodes, determine the task execution impact degree on the intelligent agent cluster based on the unexpected nodes, calculate the topology deviation characterization value in combination with the number of associated nodes of the unexpected nodes, and determine whether it meets the topology deviation standard; In response to determining that it does not meet the topology deviation standard, mark the received information, identify the sender corresponding to the received information, and track the receiver of the information sent by the intelligent agent cluster.

[0007] Furthermore, the process of determining the specific dimensional features corresponding to each path segment and the standard range of the specific dimensional features includes, Determine the dimensional features corresponding to the path segments of the task execution path topology graph, including the average load at both ends, the data transmission volume, and the transmission frequency; Determine the fluctuation values of the dimensional features of each path segment and sort them respectively; Determine the dimensional feature corresponding to the minimum fluctuation value, and determine the dimensional feature as the specific dimensional feature; Record the maximum value and the minimum value of the specific dimensional feature in the path segment, and determine the standard range of the specific dimensional feature based on the maximum value and the minimum value; Among them, the fluctuation value is the absolute variance of the dimensional feature at several time nodes.

[0008] Furthermore, the process of determining the offset of the task execution path topology graph includes, Determine the specific dimensional features corresponding to each path segment respectively; Determine the proportion of the specific dimensional features corresponding to each path segment exceeding the corresponding standard range of the specific dimensional features; Determine the mean value of each of the said ratios as the said offset.

[0009] Furthermore, the process of evaluating the deviation tendency characterization coefficient of the received information includes, Taking the ratio of the offset of the task execution path topology graph to the offset threshold as the first deviation feature; Taking the ratio of the information reception frequency to the information reception frequency threshold as the second deviation feature; Taking the sum of the first deviation feature and the second deviation feature as the deviation tendency characterization coefficient.

[0010] Furthermore, determining whether to call the operation rule file of the intelligent agent cluster based on the deviation tendency characterization coefficient includes, If the deviation tendency characterization coefficient is greater than or equal to the deviation tendency characterization coefficient threshold, then call the operation rule file of the intelligent agent cluster; Among them, the operation rule file records a rule execution path topology graph.

[0011] Furthermore, the process of determining whether the rule execution path topology graph is consistent with the task execution path topology graph of the intelligent agent cluster includes, Extract the rule execution path topology graph from the operation rule file; Compare the task execution path topology graph with the rule execution path topology graph and calculate the coincidence degree; If the coincidence degree is greater than or equal to the preset coincidence degree threshold, then determine that they are consistent; If the coincidence degree is less than the preset coincidence degree threshold, then determine that they are not consistent, and detect and analyze the received information of the intelligent agent cluster.

[0012] Furthermore, the process of determining the task execution influence degree for the intelligent agent cluster based on the unanticipated node includes, Determine the unanticipated nodes according to the task execution path topology graph and the rule execution path topology graph; Determine the average value of the actual in-degree and out-degree of each unanticipated node to obtain the task execution influence degree; Among them, if a node appears in the task execution path topology graph and does not appear in the rule execution path topology graph, then determine that the node is an unanticipated node.

[0013] Furthermore, the process of calculating the topology deviation characterization value includes, Determine the associated nodes of each unanticipated node and determine the average value of the number of associated nodes; Taking the ratio of the task execution influence degree to the task execution influence degree threshold as the first topology deviation feature; Use the ratio of the average number of associated nodes to the threshold of the number of associated nodes as the second topological deviation feature; Perform weighted summation on the first topological deviation feature and the second topological deviation feature as the topological deviation representation value.

[0014] Furthermore, determining whether it meets the topological deviation standard includes, If the topological deviation representation value is greater than or equal to the topological deviation representation threshold, it is determined that it does not meet the topological deviation standard.

[0015] Furthermore, it also includes, in response to not meeting the topological deviation standard, controlling the agent cluster to stop receiving information and stop sending information.

[0016] Compared with the prior art, the present invention analyzes the agent cluster by invoking the historical task execution interaction data of the agent cluster for several real information; in response to the agent cluster receiving information, compares the specific dimension features corresponding to each path segment with the standard range of the specific dimension features to determine the offset of the task execution path topology graph; evaluates the deviation tendency representation coefficient of the received information according to the offset and the information reception frequency of the agent cluster; determines whether to invoke the operation rule file of the agent cluster based on the deviation tendency representation coefficient to determine the corresponding rule execution path topology graph, and determines whether the rule execution path topology graph is consistent with the task execution path topology graph of the agent cluster to detect and analyze the received information of the agent cluster; in response to determining that it does not meet the topological deviation standard, marks the received information, identifies the sender corresponding to the received information, and traces the receiver of the information sent by the agent cluster. The present invention adopts a multi-dimensional evaluation method to detect false information more comprehensively and reliably, ensuring the reliability of detecting false information.

[0017] In particular, the present invention determines the specific dimension features corresponding to each path segment and the standard range of the specific dimension features. In actual situations, the computing capabilities or functions of the nodes corresponding to different path segments are not exactly the same. Therefore, there are differences in information transmission between the corresponding path segments when performing corresponding tasks. Furthermore, they perform differently in terms of different dimension features. Therefore, considering and identifying the specific dimension features corresponding to the path segments, for the dimension features with large fluctuation values, their regularity is poor, it is difficult to reflect the daily transmission law, and they do not have the value of observation. Therefore, identifying the specific dimension features with strong data representativeness of the path segment data and determining the standard range of the specific dimension features provide support for subsequent determination of the deviation tendency representation coefficient, improving the analysis efficiency and reliability.

[0018] In particular, the present invention considers calculating a deviation tendency characterization coefficient. In actual situations, for a given sender, when an agent executes a task for received information, the nodes involved in the operation and the information interaction between the nodes are all related to the corresponding task, and there are usually certain historical laws. Based on this, the present invention determines the offset of the task execution path topology graph, considering the offset of the relative specific dimension features of different path segments from the standard range of specific dimension features, that is, the situation of deviation from the historical law. At the same time, by combining the information reception frequency of the agent cluster, the offset of the operation of the agent cluster relative to the historical law is comprehensively considered, which characterizes the potential risk that the information received by the agent cluster is tampered with or there is false information. Furthermore, it provides data support for subsequent determination of the consistency of the agent's execution situation. The present invention adopts a multi-dimensional evaluation method to detect false information more comprehensively and reliably, ensuring the reliability of false information detection.

[0019] In particular, in the case where the task execution path topology graph of the agent cluster does not conform to the consistency with the rule execution path topology graph, the present invention triggers further detection and analysis of the information received by the agent cluster, saving computing power and avoiding waste of computing power caused by traversal in-depth analysis. When performing detection and analysis, an unexpected node means that the node does not belong to the rule execution path topology graph. Subsequently, the influence degree of the task execution is calculated. The in-degree and out-degree of the unexpected node characterize the proportion of such nodes participating in the operation. The more the proportion, the more it indicates that it is not an accidental event, and the more important the unexpected node is, which characterizes that more operations involved in the tasks executed by the current agent cluster do not conform to the historical law. By calculating the topological deviation characterization value and considering the number of associated nodes of the unexpected node, the importance degree of the unexpected node participating in the operation is further corroborated. Furthermore, it comprehensively reflects the situation of the task execution deviating from the historical law from the inside of the agent cluster, providing data support for subsequent determination of whether it conforms to the topological deviation standard. The present invention adopts a multi-dimensional evaluation method to detect false information more comprehensively and reliably, ensuring the reliability of false information detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the steps of the false information detection method in the multi-agent system according to the embodiment of the invention; Figure 2 It is a logical decision diagram for determining whether to call the operation rule file of the agent cluster according to the embodiment of the invention; Figure 3 It is a logical decision diagram for determining whether the rule execution path topology graph conforms to the consistency with the task execution path topology graph of the agent cluster according to the embodiment of the invention; Figure 4 It is a logical decision diagram for determining whether it conforms to the topological deviation standard according to the embodiment of the invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To make the objectives and advantages of the present invention more clearly understood, the present invention will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0022] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and do not limit the protection scope of the present invention.

[0023] Please refer to Figure 1 As shown, it is a schematic diagram of the steps of the false information detection method in the multi-agent system of the embodiment of the present invention. The false information detection method in the multi-agent system of the embodiment of the present invention includes: Step S1, invoking the interaction data of the agent cluster for the historical tasks of a number of real information to analyze the agent cluster, including determining the fluctuation value ranking of the corresponding dimensionality features of each path segment of the task execution path topology graph after the agent cluster receives a number of real information, so as to determine the specific dimensionality features corresponding to each path segment and the standard range of the specific dimensionality features; Step S2, in response to the agent cluster receiving information, comparing the specific dimensionality features corresponding to each path segment with the standard range of the specific dimensionality features to determine the offset of the task execution path topology graph; Step S3, evaluating the deviation tendency characterization coefficient of the received information according to the offset and the information reception frequency of the agent cluster; Step S4, based on the deviation tendency characterization coefficient, determining whether to invoke the operation rule file of the agent cluster to determine the corresponding rule execution path topology graph, and determining whether the rule execution path topology graph is consistent with the task execution path topology graph of the agent cluster to detect and analyze the received information of the agent cluster, including, Identifying a number of nodes corresponding to the task execution path topology graph of the agent cluster to determine unexpected nodes, determining the task execution influence degree on the agent cluster based on the unexpected nodes, calculating the topology deviation characterization value in combination with the number of associated nodes of the unexpected nodes, and determining whether it meets the topology deviation standard; Step S5, in response to determining that the agent cluster does not meet the topology deviation standard, marking the received information, identifying the sender corresponding to the received information, and tracking the receivers of the information sent by the agent cluster.

[0024] Specifically, an agent is a terminal that includes a server capable of executing corresponding computing tasks and capable of data interaction, and the agent cluster includes a number of agents.

[0025] Specifically, in implementation, the agent cluster establishes connections with the sender and the receiver. The information sent by the sender contains the tasks to be executed. After the agent cluster performs operations, the information containing the operation results is sent to the receiver.

[0026] Specifically, the real information received by the agent can be pre-verified and marked by those skilled in the art as real information.

[0027] Specifically, the historical task execution interaction data includes the operation records of the agent cluster within a historical period. To ensure that the historical task execution interaction data has data representativeness, the historical period should not be less than 10 days.

[0028] Specifically, the task execution path topology graph is determined through the operation records of the agent cluster. The agents participating in the operation in the agent cluster are obtained, and the agents participating in the operation are used as nodes, and the connection of the agents generating interactions is used as edges to obtain the task execution path topology graph.

[0029] Specifically, the path segment of the task execution path topology graph includes two agents generating interactions and the corresponding edges.

[0030] Specifically, the process of determining the specific dimension features and the standard range of the specific dimension features corresponding to each of the path segments includes determining the dimension features corresponding to the path segment of the task execution path topology graph, including the average load at both ends, the data transmission volume, and the transmission frequency; respectively determining the fluctuation values of the dimension features of the path segment and sorting them; determining the dimension feature corresponding to the minimum fluctuation value, and determining the dimension feature as the specific dimension feature; recording the maximum value and the minimum value of the specific dimension feature in the path segment, and determining the standard range of the specific dimension feature based on the maximum value and the minimum value. In implementation, the standard range of the specific dimension feature is a closed interval, the lower limit of the interval is the minimum value, and the upper limit of the interval is the maximum value.

[0031] Among them, the fluctuation value is the absolute variance of the dimension feature at several time nodes.

[0032] In implementation, the average load at both ends is the average of the loads corresponding to the two agents corresponding to the path segment, the data transmission volume is the average of the data transmission volumes of the two agents corresponding to the path segment during several data interaction processes, and the transmission frequency is the transmission frequency between the two agents corresponding to the path segment.

[0033] It can be understood that the larger the fluctuation value, the worse the regularity of the corresponding dimension feature, and it itself has strong volatility, and a relatively standard range cannot be determined.

[0034] Specifically, the present invention determines the specific dimensional features corresponding to each path segment and the standard range of the specific dimensional features. In actual situations, the computing capabilities or functions of the nodes corresponding to different path segments are not exactly the same. Therefore, there are differences in the information transmission between the corresponding path segments when performing corresponding tasks. Furthermore, the performance in different dimensional features is different. Therefore, considering and identifying the specific dimensional features corresponding to the path segments, for the dimensional features with large fluctuation values, their regularity is poor, it is difficult to reflect the daily transmission regularity, and they do not have the value of observation. Therefore, identifying the specific dimensional features with strong data representativeness of the path segments and determining the standard range of the specific dimensional features provide support for determining the deviation tendency characterization coefficient later, improving the analysis efficiency and reliability.

[0035] Specifically, the process of determining the offset of the task execution path topology diagram includes, Determining the specific dimensional features corresponding to each path segment respectively; Determining the proportion of the specific dimensional features corresponding to each of the path segments exceeding the standard range of the corresponding specific dimensional features; Determining the mean value of each of the proportions as the offset.

[0036] It can be understood that exceeding the standard range of the corresponding specific dimensional feature means not being within the corresponding interval of the standard range of the specific dimensional feature, and the proportion is the ratio of the excess amount of the specific dimensional feature exceeding the standard range of the corresponding specific dimensional feature to the length of the corresponding interval of the standard range of the specific dimensional feature.

[0037] In implementation, if there is any value of the specific dimensional feature corresponding to a path segment that does not exceed the standard range of the specific dimensional feature, the corresponding proportion is determined to be 0.

[0038] Specifically, the process of evaluating the deviation tendency characterization coefficient of the received information includes, Taking the ratio of the offset of the task execution path topology diagram to the offset threshold as the first deviation feature; Taking the ratio of the information reception frequency to the information reception frequency threshold as the second deviation feature; Taking the sum of the first deviation feature and the second deviation feature as the deviation tendency characterization coefficient.

[0039] In this embodiment, the purpose of setting the offset threshold is to characterize the situation where the specific dimensional features corresponding to each path segment deviate relatively more from the standard range of the specific dimensional features. Therefore, the offset threshold should not be set too small. In implementation, it is selected within the interval [30%, 50%].

[0040] The information reception frequency threshold is determined based on the maximum value of the historical information reception frequency between the sender and the intelligent agent cluster, and is set as the product of the maximum value and the error coefficient. The error coefficient is selected within the interval [0.85, 0.95].

[0041] Specifically, the present invention considers the calculation deviation tendency characterization coefficient. In actual situations, for a given sender, when an agent executes a task for received information, the nodes participating in the operation and the information interaction between the nodes are all associated with the corresponding task, and there are usually certain historical laws. Based on this, the present invention determines the offset of the task execution path topology graph, considering the offset of the specific dimension features corresponding to different path segments relative to the standard range of specific dimension features, that is, the situation of deviation from the historical law. At the same time, in combination with the information reception frequency of the agent cluster, the offset of the operation situation of the agent cluster relative to the historical law is comprehensively considered to characterize the potential risk that the information received by the agent cluster is tampered with or there is false information, thereby providing data support for subsequent determination of the consistency of the agent execution situation. The present invention adopts a multi-dimensional evaluation method to detect false information more comprehensively and reliably, ensuring the reliability of false information detection.

[0042] Specifically, please refer to Figure 2 As shown, it is a logical decision diagram for determining whether to call the operation rule file of the agent cluster in an embodiment of the present invention. Based on the deviation tendency characterization coefficient, it is determined whether to call the operation rule file of the agent cluster, including If the deviation tendency characterization coefficient is greater than or equal to the deviation tendency characterization coefficient threshold, then call the operation rule file of the agent cluster; Among them, the rule execution path topology graph is recorded in the operation rule file.

[0043] In implementation, the deviation tendency characterization coefficient threshold is selected within the interval [2.24, 2.35].

[0044] Specifically, the rule execution path topology graph is pre-generated, and the operation records of the agent cluster after a number of real information are sent by the same sender are pre-recorded. The agents participating in the operation are determined as nodes, and the agents generating interactions are connected as edges to obtain the rule execution path topology graph.

[0045] Specifically, please refer to Figure 3 As shown, it is a logical decision diagram for determining whether the task execution path topology graph of the agent cluster in an embodiment of the present invention conforms to consistency. The process of determining whether the rule execution path topology graph conforms to the task execution path topology graph of the agent cluster includes Extract the rule execution path topology graph from the operation rule file; Compare the task execution path topology graph with the rule execution path topology graph and calculate the coincidence degree; If the coincidence degree is greater than or equal to the preset coincidence degree threshold, then it is determined to conform to consistency; If the degree of coincidence is less than a preset degree-of-coincidence threshold, it is determined that the consistency is not met, and the received information of the agent cluster is detected and analyzed.

[0046] Specifically, the ratio of the number of coincident nodes to the total number of nodes in the rule execution path topology graph is used as the degree of coincidence.

[0047] The degree-of-coincidence threshold is preset. Among them, after statistically obtaining several task execution path topology graphs corresponding to the agent cluster after several real messages are sent by the same sender in advance, the degrees of coincidence between each task execution path topology graph and the rule execution path topology graph are solved respectively, the average value of the degrees of coincidence is calculated, and the degree-of-coincidence threshold is set as the product of the average value of the degrees of coincidence and the precision coefficient, and the precision coefficient is selected within the interval [1.15, 1.3].

[0048] Specifically, the process of determining the task execution influence degree for the agent cluster based on the unexpected nodes includes, Determine the unexpected nodes according to the task execution path topology graph and the rule execution path topology graph; Determine the average value of the actual in-degree and out-degree of each unexpected node to obtain the task execution influence degree; Among them, if a node appears in the task execution path topology graph and does not appear in the rule execution path topology graph, then this node is determined as an unexpected node.

[0049] It can be understood that the in-degree and out-degree of a node are the average value of the data reception frequency and the data transmission frequency of this node.

[0050] Specifically, the process of calculating the topological deviation characterization value includes, Determine the associated nodes of each unexpected node and determine the average value of the number of associated nodes; Take the ratio of the task execution influence degree to the task execution influence degree threshold as the first topological deviation feature; Take the ratio of the average value of the number of associated nodes to the associated node number threshold as the second topological deviation feature; Perform weighted summation on the first topological deviation feature and the second topological deviation feature as the topological deviation characterization value.

[0051] In implementation, the task execution influence degree threshold is determined in advance. Among them, the average value of the in-degree and out-degree corresponding to each node in the rule execution path topology graph is used as the task execution influence degree threshold.

[0052] In implementation, the average value of the number of nodes associated with each node in the rule execution path topology graph is determined as the associated node number threshold.

[0053] Specifically, during the actual operation of the agent cluster, the deviation in the evaluation operation of the task execution impact determined according to the actual operation trajectory is more representative. Therefore, in implementation, the task execution impact is considered first, and a slightly higher weight is assigned to the first topological deviation feature calculated based on the task execution impact. Therefore, when performing weighted summation, the weight of the first topological deviation feature is set to 0.6, and the weight of the second topological deviation feature is set to 0.4.

[0054] Specifically, in the case where the topological graph of the task execution path of the agent cluster does not conform to the consistency of the topological graph of the rule execution path, the present invention triggers further detection and analysis of the information received by the agent cluster, saving computing power and avoiding waste of computing power in traversal-depth analysis. When performing detection and analysis, an unexpected node means that the node does not belong to the topological graph of the rule execution path. Subsequently, the task execution impact is calculated. The in-degree and out-degree of the unexpected node represent the proportion of such nodes participating in the operation. The larger the proportion, the more it indicates that it is not an accidental event, and the more important the unexpected node is, which represents that more operations involved in the tasks executed by the current agent cluster do not conform to historical laws. By calculating the topological deviation representation value and considering the number of associated nodes of the unexpected node, the importance of the unexpected node participating in the operation is further corroborated, and then comprehensively reflects the situation of the task execution deviation historical law from the inside of the agent cluster, providing data support for subsequent determination of whether it conforms to the topological deviation standard. The present invention adopts a multi-dimensional evaluation method to detect false information more comprehensively and reliably, ensuring the reliability of false information detection.

[0055] Specifically, please refer to Figure 4 as shown, which is the logical decision diagram for determining whether the present invention embodiment conforms to the topological deviation standard. Determining whether it conforms to the topological deviation standard includes, If the topological deviation representation value is greater than or equal to the topological deviation representation threshold, it is determined that it does not conform to the topological deviation standard; If the topological deviation representation value is less than the topological deviation representation threshold, it is determined that it conforms to the topological deviation standard.

[0056] It can be understood that the purpose of setting the topological deviation representation threshold is to represent that the abnormal deviation degree presented by the current agent cluster during the operation process is relatively large, that is, in the case where the topological deviation representation value is relatively large. Due to the calculation method of the topological deviation representation value, its ideal state should be around 1. The larger the topological deviation representation value, the more serious the topological deviation. Therefore, the topological deviation representation threshold is set to the product of 1 and the deviation coefficient, and the deviation coefficient is in the interval [1.3, 1.6].

[0057] Specifically, it further includes, in response to not conforming to the topological deviation standard, controlling the agent cluster to stop receiving information and stop sending information.

[0058] If the false information detection method in the multi-agent system of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0059] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of the present invention.

Claims

1. A method for detecting false information in a multi-agent system, characterized in that: include: Calling the intelligent agent cluster for the historical task execution interaction data of the real information to analyze the intelligent agent cluster, including determining the order of fluctuation values ​​of each dimensional feature corresponding to each path segment of the task execution path topology diagram after the intelligent agent cluster receives the real information, so as to determine the specific dimensional feature corresponding to each path segment and the standard range of the specific dimensional feature; In response to the agent cluster receiving the information, the specific dimensional features corresponding to each path segment are compared with the standard range of the specific dimensional features to determine the offset of the task execution path topology graph; According to the offset and the information receiving frequency of the intelligent agent cluster, evaluating the deviation tendency characterization coefficient of the received information; Based on the deviation tendency characterization coefficient, it is determined whether to call the operation rule file of the intelligent agent cluster to determine the corresponding rule execution path topology map, and whether the rule execution path topology map is consistent with the task execution path topology map of the intelligent agent cluster, so as to detect and analyze the received information of the intelligent agent cluster. include, Identify several nodes corresponding to the task execution path topology graph of the agent cluster to determine unexpected nodes, determine the impact of the task execution on the agent cluster based on the unexpected nodes, calculate the topology deviation characterization value in combination with the number of associated nodes of the unexpected nodes, and determine whether the topology deviation standard is met; In response to determining that the topology deviation standard is not met, the received information is marked, a sender corresponding to the received information is identified, and a receiver of the information sent by the intelligent agent cluster is tracked.

2. The method for detecting false information in a multi-agent system according to claim 1, characterized in that: The process of determining the specific dimensional features corresponding to each of the path segments and the standard range of the specific dimensional features includes: Determine the dimensional features corresponding to the path segments of the task execution path topology graph, including the load mean at both ends, data transmission volume, and transmission frequency; Determine and sort the fluctuation values ​​of the features of each dimension of the path segment respectively; Determine the dimensional feature corresponding to the minimum fluctuation value, and determine the dimensional feature as a specific dimensional feature; Recording the maximum value and the minimum value of the specific dimensional feature in the path segment, and determining the standard range of the specific dimensional feature based on the maximum value and the minimum value; Among them, the fluctuation value is the absolute variance of the dimension feature at several time nodes.

3. The method for detecting false information in a multi-agent system according to claim 2, characterized in that: The process of determining the offset of the task execution path topology graph includes: Determine the specific dimensional features corresponding to each path segment respectively; Determine the proportion of the specific dimensional features corresponding to each of the path segments that exceeds the standard range of the corresponding specific dimensional features; The mean of the ratios is determined as the offset.

4. The method for detecting false information in a multi-agent system according to claim 3, characterized in that: The process of evaluating the bias tendency characterization coefficient of the received information includes: The ratio of the offset of the task execution path topology graph to the offset threshold is used as the first deviation feature; The ratio of the information receiving frequency to the information receiving frequency threshold is used as the second deviation feature; The sum of the first deviation feature and the second deviation feature is used as the deviation tendency characterization coefficient.

5. The method for detecting false information in a multi-agent system according to claim 4, characterized in that: Determining whether to call the operation rule file of the agent cluster based on the deviation tendency characterization coefficient includes: If the deviation tendency characterization coefficient is greater than or equal to the deviation tendency characterization coefficient threshold, calling the operation rule file of the agent cluster; The operation rule file records a rule execution path topology diagram.

6. The method for detecting false information in a multi-agent system according to claim 5, characterized in that: The process of determining whether the rule execution path topology diagram is consistent with the task execution path topology diagram of the agent cluster includes: Extracting a rule execution path topology diagram from the operation rule file; Compare the task execution path topology map with the rule execution path topology map to calculate the overlap; If the overlap is greater than or equal to a preset overlap threshold, it is determined to be consistent; If the overlap is less than a preset overlap threshold, it is determined that the consistency is not met, and the received information of the intelligent agent cluster is detected and analyzed.

7. The method for detecting false information in a multi-agent system according to claim 1, characterized in that: The process of determining the task execution impact on the agent cluster based on the unexpected node includes: Determine unexpected nodes based on the task execution path topology map and the rule execution path topology map; Determine the actual in-and-out degree average of each unexpected node to obtain the task execution impact; If a node appears in the task execution path topology diagram but does not appear in the rule execution path topology diagram, the node is determined to be an unexpected node.

8. The method for detecting false information in a multi-agent system according to claim 7, characterized in that: The process of calculating the topological deviation characterization value includes: Determine the associated nodes of each unexpected node and determine the average number of associated nodes; The ratio of the task execution influence to the task execution influence threshold is used as the first topology deviation feature; The ratio of the mean value of the number of associated nodes to the threshold value of the number of associated nodes is used as the second topological deviation feature; A weighted sum of the first topology deviation feature and the second topology deviation feature is taken as the topology deviation characterization value.

9. The method for detecting false information in a multi-agent system according to claim 8, characterized in that: Determine whether the topological deviation criteria are met, including, If the topology deviation characterization value is greater than or equal to the topology deviation characterization threshold, it is determined that the topology deviation standard is not met.

10. The method for detecting false information in a multi-agent system according to claim 9, characterized in that: It also includes controlling the intelligent agent cluster to stop receiving information and stop sending information in response to not meeting the topology deviation standard.

Images