Failure path determination method and apparatus, computer device, and storage medium

Abstract

The application relates to a failure path determination method and device, a computer device and a storage medium. The method comprises the following steps: constructing a directed failure hypergraph corresponding to a failure process of a multi-agent task according to a task execution track of the multi-agent task; determining a candidate failure path corresponding to a terminal failure node in the directed failure hypergraph; the terminal failure node is a node corresponding to a final state of the multi-agent task in the directed failure hypergraph; determining a target failure path from the candidate failure paths according to the influence degree of each candidate failure path on the final state, so that the process of jointly causing failure of multiple upstream abnormalities in the multi-agent task can be modeled, the influence of path ambiguity on the identification of a dominant failure path can be effectively reduced, and the identification accuracy of the dominant failure path is improved.

Description

Technical Field

[0001] This application relates to the field of multi-agent system technology, and in particular to a method, apparatus, computer device, and storage medium for determining a failed path. Background Technology

[0002] With the rapid development of cloud-native architecture and large-scale language model technology, multi-agent systems are widely used in complex scenarios such as software development, task planning, collaborative question answering, and operation and maintenance analysis. These multi-agent systems consist of multiple agents with autonomous decision-making, interaction, and collaboration capabilities. These agents work collaboratively around a common goal to complete complex tasks, which has also led to the need to analyze the failure behavior of multi-agent system tasks.

[0003] In related technologies, the analysis of failure behavior in multi-agent system tasks mainly includes: identifying failure types, determining the responsible agent, or locating key failure steps.

[0004] However, the failure of a multi-agent system task may be caused by multiple upstream anomalies, resulting in multiple candidate failure paths. Existing methods struggle to accurately pinpoint the critical path that leads to the failure of a multi-agent system task when faced with multiple candidate failure paths. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, computer device, and storage medium for determining the failure path in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a method for determining a failed path, including:

[0007] Based on the task execution trajectory of a multi-agent task, a directed failure hypergraph corresponding to the failure process of the multi-agent task is constructed; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events;

[0008] In the directed failure hypergraph, candidate failure paths corresponding to the final failure nodes are determined; the final failure nodes are the nodes in the directed failure hypergraph that correspond to the final state of the multi-agent task.

[0009] Based on the degree of influence of each candidate failure path on the final state, the target failure path is determined from the candidate failure paths.

[0010] In one embodiment, the task execution trajectory consists of multiple trajectory events. Based on the task execution trajectory of the multi-agent task, a directed failure hypergraph corresponding to the failure process of the multi-agent task is constructed, including:

[0011] For each trajectory event in the task execution trajectory, construct a state snapshot of the trajectory event based on the task state after the trajectory event ends;

[0012] The fuzzy failure phenomenon corresponding to the state snapshot is discretized to obtain the set of failure atoms; the set of failure atoms includes multiple minimum failure units;

[0013] A directed failure hypergraph is constructed using each failed atom in the set of failed atoms as a node; the hyperedges of the directed failure hypergraph are used to characterize the dependencies between each failed atom.

[0014] In one embodiment, determining the target failure path from the candidate failure paths based on the propagation weights of the hyperedges in each candidate failure path includes:

[0015] For each candidate failure path, intervention processing is performed on the key nodes and / or key hyperedges in the candidate failure path, and local counterfactual replay is performed on the candidate failure path after intervention processing to obtain the counterfactual replay trajectory; the key nodes and / or key hyperedges are determined based on the propagation weights of the hyperedges in the candidate failure path.

[0016] The comprehensive score of the candidate failure path is determined based on the counterfactual replay trajectory and the propagation weight of the superedges in the candidate failure path.

[0017] Based on the comprehensive score of each candidate failure path, the target failure path is determined from among them.

[0018] In one embodiment, a comprehensive score for a candidate failure path is determined based on the counterfactual replay trajectory and the propagation weights of the hyperedges in the candidate failure path, including:

[0019] The counterfactual contribution value of the candidate failure path is determined based on the difference between the failure probability of the counterfactual replay trajectory and the failure probability of the candidate failure path.

[0020] The path strength of the candidate failure path is obtained by weighted summation of the propagation weights of each hyperedge in the candidate failure path based on a preset attenuation coefficient.

[0021] A comprehensive score for candidate failed paths is determined based on counterfactual contribution value and path strength.

[0022] In one embodiment, a comprehensive score for candidate failed paths is determined based on counterfactual contribution value and path strength, including:

[0023] Based on the path strength of each candidate failure path corresponding to the final failure node, the path strength of each candidate failure path is normalized to obtain the path weight of each candidate failure path.

[0024] The path ambiguity entropy corresponding to the final failure node is determined based on the path weight of each candidate failed path; the path ambiguity entropy is used to characterize the degree of competition among multiple candidate failed paths in explaining the same final failure.

[0025] The identifiability of candidate failure paths is determined based on their counterfactual contribution values ​​and path ambiguity entropy.

[0026] The path strength, counterfactual contribution value, and identifiability of the candidate failed paths are weighted and summed to obtain a comprehensive score for the candidate failed paths.

[0027] In one embodiment, the method further includes:

[0028] Based on the key nodes and / or key hyperedges in the target failure path, construct a candidate correction set; the candidate correction set includes at least one of key nodes and key hyperedges.

[0029] The minimum failure set is determined from each candidate correction set based on the objective function.

[0030] In one embodiment, determining the minimum losing set from each candidate correction set based on the objective function includes:

[0031] Solve for the objective function to obtain the minimum losing set; the objective function aims to minimize the sum of the number of elements in the candidate correction set and the correction cost.

[0032] Secondly, this application also provides a failed path determination apparatus, comprising:

[0033] The hypergraph construction module is used to construct a directed failure hypergraph corresponding to the failure process of a multi-agent task based on the task execution trajectory of the multi-agent task; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events;

[0034] The first determining module is used to determine the candidate failure path corresponding to the final failure node in the directed failure hypergraph; the final failure node is the node in the directed failure hypergraph that corresponds to the final state of the multi-agent task.

[0035] The second determination module is used to determine the target failure path from among the candidate failure paths based on the degree of influence of each candidate failure path on the final state.

[0036] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0037] Based on the task execution trajectory of a multi-agent task, a directed failure hypergraph corresponding to the failure process of the multi-agent task is constructed; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events;

[0038] In the directed failure hypergraph, candidate failure paths corresponding to the final failure nodes are determined; the final failure nodes are the nodes in the directed failure hypergraph that correspond to the final state of the multi-agent task.

[0039] Based on the degree of influence of each candidate failure path on the final state, the target failure path is determined from the candidate failure paths.

[0040] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0041] Based on the task execution trajectory of a multi-agent task, a directed failure hypergraph corresponding to the failure process of the multi-agent task is constructed; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events;

[0042] In the directed failure hypergraph, candidate failure paths corresponding to the final failure nodes are determined; the final failure nodes are the nodes in the directed failure hypergraph that correspond to the final state of the multi-agent task.

[0043] Based on the degree of influence of each candidate failure path on the final state, the target failure path is determined from the candidate failure paths.

[0044] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0045] Based on the task execution trajectory of a multi-agent task, a directed failure hypergraph corresponding to the failure process of the multi-agent task is constructed; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events;

[0046] In the directed failure hypergraph, candidate failure paths corresponding to the final failure nodes are determined; the final failure nodes are the nodes in the directed failure hypergraph that correspond to the final state of the multi-agent task.

[0047] Based on the degree of influence of each candidate failure path on the final state, the target failure path is determined from the candidate failure paths.

[0048] The aforementioned failure path determination method, apparatus, computer device, and storage medium construct a directed failure hypergraph corresponding to the failure process of a multi-agent task based on the task execution trajectory of the multi-agent task. The directed failure hypergraph represents the dependencies between multiple failure events. Candidate failure paths corresponding to the final failure node are determined within the directed failure hypergraph. The final failure node is the node in the directed failure hypergraph corresponding to the final state of the multi-agent task. The target failure path is determined from the candidate failure paths based on the degree of influence of each candidate failure path on the final state. In this embodiment, by constructing a directed failure hypergraph representing the dependencies between failure events and searching for candidate failure paths corresponding to the final failure node within it, and then selecting the target failure path based on the degree of influence of each candidate failure path on the final state, the process of multiple upstream anomalies jointly causing failure in a multi-agent task can be modeled. This effectively reduces the impact of path ambiguity on the identification of the dominant failure path, thereby improving the accuracy of identifying the dominant failure path. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is an internal structural diagram of a computer device provided in an embodiment of this application;

[0051] Figure 2 This is a flowchart illustrating a method for determining a failed path provided in an embodiment of this application;

[0052] Figure 3 This is a flowchart illustrating a method for constructing a directed failure hypergraph provided in an embodiment of this application;

[0053] Figure 4 This is a flowchart illustrating a method for determining a target failure path provided in an embodiment of this application;

[0054] Figure 5 This is a flowchart illustrating a comprehensive scoring method provided in an embodiment of this application;

[0055] Figure 6 This is a flowchart illustrating another comprehensive scoring method provided in an embodiment of this application;

[0056] Figure 7 This is a flowchart illustrating a method for determining the minimum losing set provided in an embodiment of this application;

[0057] Figure 8 This is a structural block diagram of a failure path determination device provided in an embodiment of this application. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0060] With the rapid development of cloud-native architecture and large-scale language model technology, multi-agent systems are widely used in complex scenarios such as software development, task planning, collaborative question answering, and operation and maintenance analysis. These multi-agent systems consist of multiple agents with autonomous decision-making, interaction, and collaboration capabilities. These agents work collaboratively around a common goal to complete complex tasks, which has also led to the need to analyze the failure behavior of multi-agent system tasks.

[0061] In related technologies, the analysis of failure behavior in multi-agent system tasks mainly includes: identifying failure types, determining the responsible agent, or locating key failure steps.

[0062] However, the failure of a multi-agent system task may be caused by multiple upstream anomalies. Therefore, the same final failure can be explained by multiple seemingly reasonable propagation paths. Existing methods are unable to accurately locate the critical path that dominates the failure of a multi-agent system task. In other words, existing methods are unable to eliminate the problem of inaccurate root cause localization caused by path ambiguity.

[0063] The failure path determination method provided in this application embodiment can be applied to, for example, Figure 1 The application environment shown. Figure 1 This is an internal structure diagram of a computer device provided in an embodiment of this application. The computer device may be a server, and its internal structure diagram may be as follows: Figure 1As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements a failure path determination method.

[0064] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0065] In one embodiment, such as Figure 2 As shown, Figure 2 This is a flowchart illustrating a method for determining a failed path provided in an embodiment of this application. This method can be applied to... Figure 1 The method, using a computer device, includes the following steps:

[0066] S201, Based on the task execution trajectory of the multi-agent task, construct a directed failure hypergraph corresponding to the failure process of the multi-agent task.

[0067] Among them, the directed failure hypergraph is used to characterize the dependencies between multiple failure events.

[0068] In this embodiment, the full trajectory data of a single multi-agent task execution process can be collected and integrated into the same computable trajectory to obtain the task execution trajectory of the aforementioned multi-agent task.

[0069] Optional full trajectory data may include user input, role messages, inference output, tool call parameters, tool return results, shared memory updates, verification logs, final results, as well as cloud-native runtime information such as container logs, event bus messages, and service orchestration logs.

[0070] In one embodiment, the task execution trajectory may consist of multiple trajectory events. For example, the task execution trajectory may be represented as follows:

[0071]

[0072] in, For the task execution trajectory, For the first A trajectory event. Further, a trajectory event is defined as:

[0073]

[0074] in, Indicates execution of the first Role identifiers for agents in a trajectory event. Indicates the type of behavior. Indicates the input context. Indicates the output result. Indicates a timestamp or sequence number. This indicates the execution of metadata.

[0075] In this embodiment, the final failure of the multi-agent task is caused by the accumulation of multiple failure events. Among them, a certain downstream failure event may be triggered by multiple upstream failure events. That is, there is a certain dependency relationship between these failure events. Therefore, a directed failure hypergraph corresponding to the failure process of the multi-agent task can be constructed based on the dependency relationship between multiple failure events, so as to intuitively represent the joint failure relationship of multiple upstream events through the directed failure hypergraph.

[0076] S202, determine the candidate failure path corresponding to the final failure node in the directed failure hypergraph.

[0077] In this context, the final failure node is the node in the directed failure hypergraph that corresponds to the final state of the multi-agent task.

[0078] Optionally, in a directed failure hypergraph, a reverse search can be performed starting from the final failure node to generate multiple candidate failure paths.

[0079] For example, any candidate failed path can be represented as an ordered sequence of alternating nodes and hyperedges:

[0080]

[0081] in, Indicates the first Candidate failed paths; This indicates the number of nodes in the candidate failed path; Indicates the first candidate failure path Strip of edge; Indicates the first candidate failure path There are nodes, and they satisfy... , for The downstream node.

[0082] S203, Based on the degree of influence of each candidate failure path on the final state, determine the target failure path from each candidate failure path.

[0083] Optionally, the impact of each candidate failure path on the final state of the multi-agent task can be evaluated sequentially, and the candidate failure path most likely to cause the multi-agent task to fail can be identified as the target failure path.

[0084] In this embodiment of the application, a directed failure hypergraph is constructed to characterize the dependencies between failure events, and candidate failure paths corresponding to the final failure nodes are searched within it. Then, the target failure path is selected based on the degree of influence of each candidate failure path on the final state. This can model the process of multiple upstream anomalies jointly causing failure in multi-agent tasks, effectively reduce the impact of path ambiguity on the identification of the dominant failure path, and thus improve the accuracy of the identification of the dominant failure path.

[0085] Reference Figure 3 , Figure 3 This is a flowchart illustrating a directed failure hypergraph construction method provided in an embodiment of this application. This embodiment relates to a possible implementation of how to construct a directed failure hypergraph corresponding to the failure process of a multi-agent task based on the task execution trajectory of the multi-agent task. Based on the above embodiment, S201 includes the following steps:

[0086] S301: For each trajectory event in the task execution trajectory, construct a state snapshot of the trajectory event based on the task state after the trajectory event ends.

[0087] For example, the task state after a certain trajectory event ends can be obtained, and a state snapshot can be constructed based on that state:

[0088]

[0089] in, Indicates the first A snapshot of the state after a trajectory event ends. Represents the state convergence function, used to determine the state convergence function based on the first... The historical task execution trajectory generation task status representation after the completion of each trajectory event.

[0090] Optionally, the state snapshot may include, for example, the task target state, the role local state, the shared memory state, satisfied constraints, unsatisfied constraints, tool result summary, and verification result markers.

[0091] S302, discretize the fuzzy failure phenomenon corresponding to the state snapshot to obtain the set of failed atoms.

[0092] The set of failed atoms includes multiple smallest failed units.

[0093] In this embodiment, for each trajectory event in the task execution trajectory, the fuzzy failure phenomenon corresponding to the state snapshot of the trajectory event can be discretized into the smallest propagable, computable, and localizable failure unit based on local anomalies, generating a set of failure atoms:

[0094]

[0095] in, For the set of failed atoms, For the first One failed atom, This represents the total number of invalid atoms.

[0096] Further define the failed atom:

[0097]

[0098] in, The source anchor point of the failed atom is used to identify its corresponding trajectory event or state snapshot; Indicates the failure category; This indicates the confidence level in identifying the failed atom, and .

[0099] It should be noted that the failed atom is the smallest failed unit.

[0100] For example, failure categories may include: task specification deviation, role responsibility deviation, information omission, context loss, incorrect tool selection, incorrect parameter construction, lack of verification, premature termination, and inconsistency between results and objectives.

[0101] S303 constructs a directed failure hypergraph using each failed atom in the set of failed atoms as a node.

[0102] In this context, the hyperedges of the directed failure hypergraph are used to characterize the dependencies between each failure atom.

[0103] In this embodiment, a directed failure hypergraph can be constructed using each failed atom as a node and the dependencies between failed atoms as hyperedges:

[0104]

[0105] in, For a set of nodes, For a set of superedges, This is for mapping hyperedge attributes.

[0106] Any hyperedge is defined as:

[0107]

[0108] in, Indicates the first Strip of edge; Indicates the first The set of upstream nodes corresponding to each superedge; Indicates the first The downstream node corresponding to the superedge; Indicates the dependency type; This indicates the propagation weight.

[0109] Optional, dependency type It includes at least one of semantic dependency, state dependency, temporal dependency, validation dependency, and resource dependency.

[0110] The propagation weight is defined as:

[0111]

[0112] in, Indicates the semantic causal consistency score. Indicates state-dependent scores. Indicates the score of historical inheritance. Indicates the verification gap score; For the corresponding weight coefficients, the following conditions must be met:

[0113]

[0114] in addition, ,thereby .

[0115] In this embodiment, by dividing the task execution trajectory into multiple trajectory events and constructing a state snapshot for each trajectory event, the fuzzy failure phenomenon is discretized into the smallest independently processable failure unit. Then, a directed failure hypergraph is constructed with failure atoms as nodes. The dependency relationship between failure events is displayed intuitively through the directed failure hypergraph, which effectively reduces the impact of path ambiguity on the identification of the dominant failure path and improves the identification accuracy of the dominant failure path.

[0116] Reference Figure 4 , Figure 4 This is a flowchart illustrating a target failure path determination method provided in an embodiment of this application. This embodiment relates to a possible implementation of determining a target failure path from among candidate failure paths based on the degree of influence of each candidate failure path on the final state. Based on the above embodiment, S203 includes the following steps:

[0117] S401, for each candidate failure path, perform intervention processing on the key nodes and / or key superedges in the candidate failure path, and perform local counterfactual replay on the candidate failure path after intervention processing to obtain the counterfactual replay trajectory.

[0118] Among them, key nodes and / or key hyperedges are determined based on the propagation weights of hyperedges in candidate failure paths.

[0119] In one possible implementation, the hyperedges in the candidate failure path can be sorted according to their propagation weights, with hyperedges having propagation weights higher than a preset threshold being identified as critical hyperedges. Furthermore, for each node in the candidate failure path, the nodes can be sorted according to the cumulative propagation weights of the hyperedges associated with them, with nodes having cumulative propagation weights higher than a preset threshold being identified as critical nodes.

[0120] In this embodiment, intervention processing can be performed on key nodes and / or key hyperedges in the candidate failed paths to construct path-level intervention parameters:

[0121]

[0122] in, Indicates the candidate failure path Intervention procedures.

[0123] Based on the above intervention, a partial counterfactual replay of the original trajectory is performed to obtain the counterfactual trajectory:

[0124]

[0125] in, Indicates the candidate failed path Counterfactual replay trajectory after intervention, This represents a local playback function.

[0126] Optionally, the intervention process may include at least one of deletion, replacement, supplementary verification, and partial rewriting.

[0127] S402. Determine the comprehensive score of the candidate failure path based on the counterfactual replay trajectory and the propagation weight of the superedge in the candidate failure path.

[0128] For example, the failure results corresponding to the counterfactual replay trajectory can be compared with the failure results corresponding to the candidate failure paths to determine the contribution of the candidate failure paths to the final failure. Furthermore, the path strength and identifiability of the candidate failure paths can be calculated based on the propagation weights of the hyperedges in the candidate failure paths. Thus, the path strength, identifiability, and contribution of the candidate failure paths to the final failure are used as three evaluation dimensions, and a comprehensive score for the candidate failure paths is determined based on the evaluation results of these three dimensions.

[0129] S403, based on the comprehensive score of each candidate failure path, determine the target failure path from each candidate failure path.

[0130] For example, the candidate failure path corresponding to the highest comprehensive score can be determined as the target failure path:

[0131]

[0132] in, The target failure path is the dominant failure path. Candidate failure paths The overall score.

[0133] In this embodiment, for each candidate failure path, intervention processing is performed on key nodes and / or key hyperedges in the candidate failure path, and local counterfactual replay is performed on the intervened candidate failure path to obtain the counterfactual replay trajectory; based on the counterfactual replay trajectory and the propagation weights of the hyperedges in the candidate failure path, a comprehensive score for the candidate failure path is determined; based on the comprehensive scores of each candidate failure path, the target failure path is determined from all candidate failure paths. By performing intervention processing on key nodes or hyperedges in the candidate failure path and performing local counterfactual replay, the contribution of the candidate failure path to the final failure can be quantified, and the comprehensive score can be determined by combining the propagation weights of the hyperedges, which improves the accuracy and reliability of the dominant failure path identification and reduces the computational cost.

[0134] Reference Figure 5 , Figure 5 This is a flowchart illustrating a comprehensive scoring method provided in an embodiment of this application. This embodiment relates to a possible implementation of determining the comprehensive score of a candidate failure path based on the counterfactual replay trajectory and the propagation weights of the hyperedges in the candidate failure path. Based on the above embodiment, S402 includes the following steps:

[0135] S501, Based on the difference between the failure probability of the counterfactual replay trajectory and the failure probability of the candidate failure path, determine the counterfactual contribution value of the candidate failure path.

[0136] For example, the counterfactual contribution value of a candidate failure path can be defined:

[0137]

[0138] in, This is the failure probability estimation function; This indicates when considering candidate failed paths The extent to which the probability of final failure decreased after intervention. The larger the value, the greater the contribution of the candidate failure path to the final failure.

[0139] S502, the propagation weights of each hyperedge in the candidate failure path are weighted and summed based on the preset attenuation coefficient to obtain the path strength of the candidate failure path.

[0140] For example, the path strength of a candidate failed path can be defined as:

[0141]

[0142] in, Indicate candidate failure paths Path strength, The preset attenuation coefficient satisfies , Indicates the first candidate failure path The propagation weight of a superedge.

[0143] S503, based on counterfactual contribution value and path strength, determines the comprehensive score of candidate failed paths.

[0144] Furthermore, the identifiability of candidate failure paths can be calculated based on their path strength. Thus, the path strength, identifiability, and contribution of candidate failure paths to the final failure can be used as three evaluation dimensions. The comprehensive score of candidate failure paths can be determined based on path strength, counterfactual contribution value, and identifiability.

[0145] In this embodiment, the counterfactual contribution value of the candidate failure path is determined based on the difference between the failure probability of the counterfactual replay trajectory and the failure probability of the candidate failure path; the propagation weights of each hyperedge in the candidate failure path are weighted and summed based on a preset attenuation coefficient to obtain the path strength of the candidate failure path; and the comprehensive score of the candidate failure path is determined based on the counterfactual contribution value and the path strength, thereby conducting a comprehensive evaluation of the candidate failure path from multiple dimensions and further improving the accuracy of the dominant failure path evaluation.

[0146] Reference Figure 6 , Figure 6 This is a flowchart illustrating another comprehensive scoring method provided in this application embodiment. This embodiment relates to a possible implementation of determining the comprehensive score of a candidate failed path based on counterfactual contribution value and path strength. Based on the above embodiment, S503 includes the following steps:

[0147] S601, based on the path strength of each candidate failure path corresponding to the final failure node, normalize the path strength of each candidate failure path to obtain the path weight of each candidate failure path.

[0148] For example, to quantify the uncertainty when the same endgame failure can be explained by multiple candidate paths, normalized path weights are defined:

[0149]

[0150] in, This represents the total number of candidate failed paths. For temperature coefficient, ,and .

[0151] S602, determine the path ambiguity entropy corresponding to the final failed node based on the path weight of each candidate failed path.

[0152] Among them, path ambiguity entropy is used to characterize the degree of competition among multiple candidate failed paths in explaining the same final failure.

[0153] For example, path ambiguity entropy can be defined as:

[0154]

[0155] in, The larger the path ambiguity entropy value, the stronger the competition and the greater the ambiguity among candidate propagation paths for final failure; the smaller the path ambiguity entropy value, the more concentrated the dominant failure paths.

[0156] S603. Based on the counterfactual contribution value and path ambiguity entropy of the candidate failure path, determine the identifiability of the candidate failure path.

[0157] For example, the identifiability of a candidate failure path can be defined as:

[0158]

[0159] in, Indicate candidate failure paths Identifiability in the current path contention environment.

[0160] S604 calculates a weighted sum of the path strength, counterfactual contribution value, and identifiability of the candidate failed paths to obtain a comprehensive score for the candidate failed paths.

[0161] In this embodiment, the path strength of candidate failed paths can be evaluated. and counterfactual contribution value Normalize them separately, denoted as and And define the comprehensive score as:

[0162]

[0163] in, The path strength weight represents the proportion of propagation structure strength in the overall score; Counterfactual contribution weight represents the proportion of the decrease in failure probability after counterfactual intervention in the overall score; The path distinguishability weight represents the proportion of the path's distinguishability / dominance among multiple candidate paths in the overall score. And satisfy:

[0164]

[0165] In this embodiment, by introducing path ambiguity entropy on top of normalized path strength, the degree of competition among multiple candidate failure paths in explaining the same final failure is quantified. Furthermore, the counterfactual contribution value is combined with the path ambiguity entropy to construct identifiability, placing each path in a global path competition environment for evaluation. A weighted sum of path strength, counterfactual contribution value, and identifiability yields a comprehensive score, effectively mitigating the root cause localization instability problem caused by the coexistence of multiple seemingly reasonable paths in traditional methods, and further improving the accuracy of selecting the dominant failure path.

[0166] Reference Figure 7 , Figure 7 This is a flowchart illustrating a method for determining the minimum losing set provided in an embodiment of this application. Based on the above embodiment, the method further includes the following steps:

[0167] S701, construct a candidate correction set based on the key nodes and / or key superedges in the target failure path.

[0168] The candidate correction set includes at least one of key nodes and key superedges.

[0169] In this embodiment, key nodes and / or key hyperedges can be selected from the target failure path as candidate correction objects to construct a candidate correction set, and the update operator after correcting the failure hypergraph is defined as follows:

[0170]

[0171] in, Represents a set The modified hypergraph is obtained by performing deletion, replacement, correction, or verification reinforcement on nodes or hyperedges in the graph.

[0172] S702 determines the minimum losing set from each candidate correction set based on the objective function.

[0173] For example, the objective function can be defined as:

[0174]

[0175] in, Denotes the minimum losing set. Represents a set Number of elements in the middle Represents a set Implement the modified cost function, This represents the cost weight.

[0176] Solving the objective function requires satisfying the following constraints:

[0177]

[0178] in, This represents the execution result obtained by replaying the original trajectory under the modified hypergraph constraints. This is the acceptable failure threshold.

[0179] In one implementation, the target failure path and the minimum failure set can be solved sequentially based on the above method, and the calculation results are output after the solution is completed. The output calculation results include: the target failure path; a ranked list of candidate failure paths; the category, source anchor point, and identification confidence of each failure atom; the dependency type and propagation weight of each hyperedge in the directed failure hypergraph; path ambiguity entropy; the minimum failure set; a counterfactual replay evidence log and the corresponding failure probability change results.

[0180] Optionally, the output format of the calculation results may include, for example, a graph structure report, a JSON report, a visual audit page, a diagnostic manual, etc.

[0181] In this embodiment, a candidate correction set is constructed by extracting key nodes and key superedges from the target failure path, and the minimum failure set is solved using an objective function that minimizes the sum of the number of correction elements and the correction cost. This yields the minimum set of intervention objects required to reduce the probability of final failure, and the task execution trajectory can be corrected with the minimum correction cost based on the minimum failure set, effectively reducing the failure probability of multi-agent tasks.

[0182] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0183] Based on the same inventive concept, this application also provides a failure path determination apparatus for implementing the failure path determination method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more failure path determination apparatus embodiments provided below can be found in the limitations of the failure path determination method described above, and will not be repeated here.

[0184] In one embodiment, such as Figure 8 As shown, Figure 8 This is a structural block diagram of a failed path determination device provided in an embodiment of this application. The device 800 includes:

[0185] The hypergraph construction module 801 is used to construct a directed failure hypergraph corresponding to the failure process of a multi-agent task based on the task execution trajectory of the multi-agent task; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events.

[0186] The first determining module 802 is used to determine the candidate failure path corresponding to the final failure node in the directed failure hypergraph; the final failure node is the node in the directed failure hypergraph that corresponds to the final state of the multi-agent task.

[0187] The second determining module 803 is used to determine the target failure path from among the candidate failure paths based on the degree of influence of each candidate failure path on the final state.

[0188] In one embodiment, the task execution trajectory consists of multiple trajectory events, and the hypergraph construction module 801 includes:

[0189] The state snapshot construction unit is used to construct a state snapshot of each trajectory event in the task execution trajectory based on the task state after the trajectory event ends.

[0190] The failure atom discrete unit is used to discretize the fuzzy failure phenomenon corresponding to the state snapshot to obtain the failure atom set; the failure atom set includes multiple minimum failure units.

[0191] The hypergraph construction unit is used to construct a directed failure hypergraph using each failure atom in the set of failure atoms as a node; the hyperedges of the directed failure hypergraph are used to characterize the dependencies between each failure atom.

[0192] In one embodiment, the second determining module 803 includes:

[0193] The intervention unit is used to perform intervention processing on key nodes and / or key hyperedges in each candidate failure path, and to perform local counterfactual replay on the candidate failure path after intervention processing to obtain the counterfactual replay trajectory; key nodes and / or key hyperedges are determined based on the propagation weights of hyperedges in the candidate failure path.

[0194] The comprehensive scoring calculation unit is used to determine the comprehensive score of the candidate failure path based on the counterfactual replay trajectory and the propagation weight of the hyperedge in the candidate failure path;

[0195] The target failure path determination unit is used to determine the target failure path from among the candidate failure paths based on the comprehensive score of each candidate failure path.

[0196] In one embodiment, the comprehensive scoring calculation unit includes:

[0197] The first determining subunit is used to determine the counterfactual contribution value of the candidate failure path based on the difference between the failure probability of the counterfactual replay trajectory and the failure probability of the candidate failure path.

[0198] The weighted sub-unit is used to perform a weighted summation with the propagation weights of each hyperedge in the candidate failure path based on a preset attenuation coefficient to obtain the path strength of the candidate failure path.

[0199] The second determining subunit is used to determine the comprehensive score of candidate failed paths based on counterfactual contribution value and path strength.

[0200] In one embodiment, the second determining subunit is specifically configured to: normalize the path strength of each candidate failure path based on the path strength of each candidate failure path corresponding to the final failure node to obtain the path weight of each candidate failure path; determine the path ambiguity entropy corresponding to the final failure node based on the path weight of each candidate failure path; the path ambiguity entropy is used to characterize the degree of competition among multiple candidate failure paths in explaining the same final failure; determine the identifiability of the candidate failure path based on the counterfactual contribution value and the path ambiguity entropy; and perform a weighted summation of the path strength, counterfactual contribution value, and identifiability of the candidate failure path to obtain a comprehensive score for the candidate failure path.

[0201] In one embodiment, the device 800 further includes:

[0202] A candidate correction set construction module is used to construct a candidate correction set based on key nodes and / or key hyperedges in the target failure path; the candidate correction set includes at least one of key nodes and key hyperedges.

[0203] The third determination module is used to determine the minimum failure set from each candidate correction set based on the objective function.

[0204] In one embodiment, the third determining module is specifically used to solve the objective function to obtain the minimum losing set; the objective function aims to minimize the sum of the number of elements in the candidate correction set and the correction cost.

[0205] Each module in the aforementioned failure path determination device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0206] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0207] Based on the task execution trajectory of a multi-agent task, a directed failure hypergraph corresponding to the failure process of the multi-agent task is constructed; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events;

[0208] In the directed failure hypergraph, candidate failure paths corresponding to the final failure nodes are determined; the final failure nodes are the nodes in the directed failure hypergraph that correspond to the final state of the multi-agent task.

[0209] Based on the degree of influence of each candidate failure path on the final state, the target failure path is determined from the candidate failure paths.

[0210] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0211] For each trajectory event in the task execution trajectory, construct a state snapshot of the trajectory event based on the task state after the trajectory event ends;

[0212] The fuzzy failure phenomenon corresponding to the state snapshot is discretized to obtain the set of failure atoms; the set of failure atoms includes multiple minimum failure units;

[0213] A directed failure hypergraph is constructed using each failed atom in the set of failed atoms as a node; the hyperedges of the directed failure hypergraph are used to characterize the dependencies between each failed atom.

[0214] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0215] For each candidate failure path, intervention processing is performed on the key nodes and / or key hyperedges in the candidate failure path, and local counterfactual replay is performed on the candidate failure path after intervention processing to obtain the counterfactual replay trajectory; the key nodes and / or key hyperedges are determined based on the propagation weights of the hyperedges in the candidate failure path.

[0216] The comprehensive score of the candidate failure path is determined based on the counterfactual replay trajectory and the propagation weight of the superedges in the candidate failure path.

[0217] Based on the comprehensive score of each candidate failure path, the target failure path is determined from among them.

[0218] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0219] The counterfactual contribution value of the candidate failure path is determined based on the difference between the failure probability of the counterfactual replay trajectory and the failure probability of the candidate failure path.

[0220] The path strength of the candidate failure path is obtained by weighted summation of the propagation weights of each hyperedge in the candidate failure path based on a preset attenuation coefficient.

[0221] A comprehensive score for candidate failed paths is determined based on counterfactual contribution value and path strength.

[0222] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0223] Based on the path strength of each candidate failure path corresponding to the final failure node, the path strength of each candidate failure path is normalized to obtain the path weight of each candidate failure path.

[0224] The path ambiguity entropy corresponding to the final failure node is determined based on the path weight of each candidate failed path; the path ambiguity entropy is used to characterize the degree of competition among multiple candidate failed paths in explaining the same final failure.

[0225] The identifiability of candidate failure paths is determined based on their counterfactual contribution values ​​and path ambiguity entropy.

[0226] The path strength, counterfactual contribution value, and identifiability of the candidate failed paths are weighted and summed to obtain a comprehensive score for the candidate failed paths.

[0227] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0228] Based on the key nodes and / or key hyperedges in the target failure path, construct a candidate correction set; the candidate correction set includes at least one of key nodes and key hyperedges.

[0229] The minimum failure set is determined from each candidate correction set based on the objective function.

[0230] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0231] Solve for the objective function to obtain the minimum losing set; the objective function aims to minimize the sum of the number of elements in the candidate correction set and the correction cost.

[0232] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0233] Based on the task execution trajectory of a multi-agent task, a directed failure hypergraph corresponding to the failure process of the multi-agent task is constructed; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events;

[0234] In the directed failure hypergraph, candidate failure paths corresponding to the final failure nodes are determined; the final failure nodes are the nodes in the directed failure hypergraph that correspond to the final state of the multi-agent task.

[0235] Based on the degree of influence of each candidate failure path on the final state, the target failure path is determined from the candidate failure paths.

[0236] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0237] For each trajectory event in the task execution trajectory, construct a state snapshot of the trajectory event based on the task state after the trajectory event ends;

[0238] The fuzzy failure phenomenon corresponding to the state snapshot is discretized to obtain the set of failure atoms; the set of failure atoms includes multiple minimum failure units;

[0239] A directed failure hypergraph is constructed using each failed atom in the set of failed atoms as a node; the hyperedges of the directed failure hypergraph are used to characterize the dependencies between each failed atom.

[0240] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0241] For each candidate failure path, intervention processing is performed on the key nodes and / or key hyperedges in the candidate failure path, and local counterfactual replay is performed on the candidate failure path after intervention processing to obtain the counterfactual replay trajectory; the key nodes and / or key hyperedges are determined based on the propagation weights of the hyperedges in the candidate failure path.

[0242] The comprehensive score of the candidate failure path is determined based on the counterfactual replay trajectory and the propagation weight of the superedges in the candidate failure path.

[0243] Based on the comprehensive score of each candidate failure path, the target failure path is determined from among them.

[0244] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0245] The counterfactual contribution value of the candidate failure path is determined based on the difference between the failure probability of the counterfactual replay trajectory and the failure probability of the candidate failure path.

[0246] The path strength of the candidate failure path is obtained by weighted summation of the propagation weights of each hyperedge in the candidate failure path based on a preset attenuation coefficient.

[0247] A comprehensive score for candidate failed paths is determined based on counterfactual contribution value and path strength.

[0248] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0249] Based on the path strength of each candidate failure path corresponding to the final failure node, the path strength of each candidate failure path is normalized to obtain the path weight of each candidate failure path.

[0250] The path ambiguity entropy corresponding to the final failure node is determined based on the path weight of each candidate failed path; the path ambiguity entropy is used to characterize the degree of competition among multiple candidate failed paths in explaining the same final failure.

[0251] The identifiability of candidate failure paths is determined based on their counterfactual contribution values ​​and path ambiguity entropy.

[0252] The path strength, counterfactual contribution value, and identifiability of the candidate failed paths are weighted and summed to obtain a comprehensive score for the candidate failed paths.

[0253] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0254] Based on the key nodes and / or key hyperedges in the target failure path, construct a candidate correction set; the candidate correction set includes at least one of key nodes and key hyperedges.

[0255] The minimum failure set is determined from each candidate correction set based on the objective function.

[0256] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0257] Solve for the objective function to obtain the minimum losing set; the objective function aims to minimize the sum of the number of elements in the candidate correction set and the correction cost.

[0258] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0259] Based on the task execution trajectory of a multi-agent task, a directed failure hypergraph corresponding to the failure process of the multi-agent task is constructed; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events;

[0260] In the directed failure hypergraph, candidate failure paths corresponding to the final failure nodes are determined; the final failure nodes are the nodes in the directed failure hypergraph that correspond to the final state of the multi-agent task.

[0261] Based on the degree of influence of each candidate failure path on the final state, the target failure path is determined from the candidate failure paths.

[0262] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0263] For each trajectory event in the task execution trajectory, construct a state snapshot of the trajectory event based on the task state after the trajectory event ends;

[0264] The fuzzy failure phenomenon corresponding to the state snapshot is discretized to obtain the set of failure atoms; the set of failure atoms includes multiple minimum failure units;

[0265] A directed failure hypergraph is constructed using each failed atom in the set of failed atoms as a node; the hyperedges of the directed failure hypergraph are used to characterize the dependencies between each failed atom.

[0266] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0267] For each candidate failure path, intervention processing is performed on the key nodes and / or key hyperedges in the candidate failure path, and local counterfactual replay is performed on the candidate failure path after intervention processing to obtain the counterfactual replay trajectory; the key nodes and / or key hyperedges are determined based on the propagation weights of the hyperedges in the candidate failure path.

[0268] The comprehensive score of the candidate failure path is determined based on the counterfactual replay trajectory and the propagation weight of the superedges in the candidate failure path.

[0269] Based on the comprehensive score of each candidate failure path, the target failure path is determined from among them.

[0270] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0271] The counterfactual contribution value of the candidate failure path is determined based on the difference between the failure probability of the counterfactual replay trajectory and the failure probability of the candidate failure path.

[0272] The path strength of the candidate failure path is obtained by weighted summation of the propagation weights of each hyperedge in the candidate failure path based on a preset attenuation coefficient.

[0273] A comprehensive score for candidate failed paths is determined based on counterfactual contribution value and path strength.

[0274] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0275] Based on the path strength of each candidate failure path corresponding to the final failure node, the path strength of each candidate failure path is normalized to obtain the path weight of each candidate failure path.

[0276] The path ambiguity entropy corresponding to the final failure node is determined based on the path weight of each candidate failed path; the path ambiguity entropy is used to characterize the degree of competition among multiple candidate failed paths in explaining the same final failure.

[0277] The identifiability of candidate failure paths is determined based on their counterfactual contribution values ​​and path ambiguity entropy.

[0278] The path strength, counterfactual contribution value, and identifiability of the candidate failed paths are weighted and summed to obtain a comprehensive score for the candidate failed paths.

[0279] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0280] Based on the key nodes and / or key hyperedges in the target failure path, construct a candidate correction set; the candidate correction set includes at least one of key nodes and key hyperedges.

[0281] The minimum failure set is determined from each candidate correction set based on the objective function.

[0282] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0283] Solve for the objective function to obtain the minimum losing set; the objective function aims to minimize the sum of the number of elements in the candidate correction set and the correction cost.

[0284] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0285] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0286] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for determining a failed path, characterized in that, The method includes: Based on the task execution trajectory of the multi-agent task, a directed failure hypergraph corresponding to the failure process of the multi-agent task is constructed; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events; In the directed failure hypergraph, candidate failure paths corresponding to the final failure nodes are determined; the final failure nodes are the nodes in the directed failure hypergraph that correspond to the final state of the multi-agent task. Based on the degree of influence of each of the candidate failure paths on the final state, a target failure path is determined from each of the candidate failure paths.

2. The method according to claim 1, characterized in that, The task execution trajectory consists of multiple trajectory events. The construction of a directed failure hypergraph corresponding to the failure process of the multi-agent task based on the task execution trajectory of the multi-agent task includes: For each trajectory event in the task execution trajectory, a state snapshot of the trajectory event is constructed based on the task state after the trajectory event ends; The fuzzy failure phenomenon corresponding to the state snapshot is discretized to obtain a set of failure atoms; the set of failure atoms includes multiple minimum failure units. A directed failure hypergraph is constructed using each failed atom in the set of failed atoms as a node; the hyperedges of the directed failure hypergraph are used to characterize the dependencies between each failed atom.

3. The method according to claim 1, characterized in that, The step of determining the target failure path from the candidate failure paths based on the degree of influence of each candidate failure path on the final state includes: For each of the candidate failure paths, intervention processing is performed on the key nodes and / or key hyperedges in the candidate failure paths, and local counterfactual replay is performed on the candidate failure paths after intervention processing to obtain the counterfactual replay trajectory; the key nodes and / or key hyperedges are determined based on the propagation weights of the hyperedges in the candidate failure paths. Based on the counterfactual replay trajectory and the propagation weights of the hyperedges in the candidate failure paths, a comprehensive score for the candidate failure paths is determined. The target failure path is determined from the candidate failure paths based on the comprehensive score of each candidate failure path.

4. The method according to claim 3, characterized in that, The step of determining the comprehensive score of the candidate failure path based on the counterfactual replay trajectory and the propagation weights of the hyperedges in the candidate failure path includes: The counterfactual contribution value of the candidate failure path is determined based on the difference between the failure probability of the counterfactual replay trajectory and the failure probability of the candidate failure path. The propagation weights of each hyperedge in the candidate failure path are weighted and summed based on a preset attenuation coefficient to obtain the path strength of the candidate failure path. Based on the counterfactual contribution value and the path strength, a comprehensive score is determined for the candidate failed path.

5. The method according to claim 4, characterized in that, The comprehensive score for determining the candidate failed path based on the counterfactual contribution value and the path strength includes: Based on the path strength of each candidate failure path corresponding to the final failure node, the path strength of each candidate failure path is normalized to obtain the path weight of each candidate failure path. The path ambiguity entropy corresponding to the final failure node is determined based on the path weight of each of the candidate failed paths; the path ambiguity entropy is used to characterize the degree of competition among multiple candidate failed paths in explaining the same final failure. The identifiability of the candidate failure path is determined based on the counterfactual contribution value of the candidate failure path and the path ambiguity entropy. The path strength, counterfactual contribution value, and identifiability of the candidate failed path are weighted and summed to obtain a comprehensive score for the candidate failed path.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: Based on the key nodes and / or key hyperedges in the target failure path, a candidate correction set is constructed; the candidate correction set includes at least one of the key nodes and key hyperedges. The minimum failure set is determined from each of the candidate correction sets based on the objective function.

7. The method according to claim 6, characterized in that, The determination of the minimum losing set from each of the candidate correction sets based on the objective function includes: Solve for the objective function to obtain the minimum failure set; the objective function aims to minimize the sum of the number of elements in the candidate correction set and the correction cost.

8. A failure path determination device, characterized in that, The device includes: The hypergraph construction module is used to construct a directed failure hypergraph corresponding to the failure process of the multi-agent task based on the task execution trajectory of the multi-agent task; wherein, the directed failure hypergraph is used to represent the dependency relationship between multiple failure events; The first determining module is used to determine the candidate failure path corresponding to the final failure node in the directed failure hypergraph; the final failure node is the node in the directed failure hypergraph that corresponds to the final state of the multi-agent task. The second determining module is used to determine the target failure path from the candidate failure paths based on the degree of influence of each candidate failure path on the final state.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images