A fault root cause tracing method based on dynamic temporal causal network

Abstract

The present invention discloses a fault root cause tracing method based on a dynamic temporal causal network. The method extracts sub-time series by sliding window from time series data of normal mode and fault mode; calculates the causal strength fluctuation range of each edge by using the causal network constructed by the sub-time series of the normal mode; finds out the abnormal causal relationship in the network by judging whether the causal strength of each edge exceeds the allowable fluctuation range for each causal network constructed by the sub-time series of the fault mode, and calculates the reliability distribution of a fault root cause; finally, fuses all the reliability distributions of the fault root causes obtained, obtains the final reliability distribution of the fault root cause, and determines the fault root cause according to the distribution; the present invention can more accurately capture the dynamic changes of the causal relationship between variables and realize the accurate positioning of the fault root cause.

Description

Technical Field

[0001] The present invention relates to the technical field of causal analysis and fault diagnosis, and in particular to a fault root cause tracing method based on a dynamic temporal causal network. Background Art

[0002] With the continuous advancement of science and technology and the rapid development of industrialization, the complexity of various systems continues to increase. The relationship between variables in the system is no longer a single linear or static relationship, but a dynamic relationship with strong nonlinearity, time-varying and multi-causal characteristics. Once an abnormality or failure occurs in the system, quickly and accurately tracing the root cause of the failure is crucial to ensuring the safety, reliability and efficient operation of the system.

[0003] Dynamic temporal causal networks can more accurately capture the dynamic changes of causal relationships between variables and effectively make up for the shortcomings of traditional methods. Especially in scenarios where causal relationships are complex and dynamically changing, dynamic temporal causal networks show significant advantages and can accurately locate the root cause of faults. They can also provide scientific and reliable support for operation monitoring, fault diagnosis and optimization decisions of complex systems. However, traditional root cause tracing methods based on causal analysis focus more on the modeling of static causal relationships, but ignore the dynamic temporal causal characteristics between variables. This makes it difficult to fully utilize key dynamic feature information, which affects the accuracy of root cause tracing. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a fault root cause tracing method based on a dynamic temporal causal network, which shows better performance in tracing the fault root cause of complex systems with strong nonlinearity, time-varying and multi-causal characteristics;

[0005] The technical solution is as follows: A fault root cause tracing method based on a dynamic temporal causal network includes the following steps:

[0006] S1: Use the time sliding window algorithm to divide the time series data of normal mode and fault mode into sub-time series of the same width;

[0007] Preferably, the time series data of the normal mode and the fault mode are extracted by sliding window, and the total length of the data is recorded as , take the sliding window size as , the step length is , divide the time series data into N groups of sub-time series with the same width, The calculation formula is as follows:

[0008] +1

[0009] Where [x] represents the largest integer not exceeding x.

[0010] S2: Connect the variables of the sub-time series in pairs to obtain a completely undirected network, and gradually delete the edges through the conditional independence test method to obtain the causal skeleton network skG;

[0011] Preferably, the method for obtaining the causal skeleton network skG further comprises:

[0012] S201: Use sub-time series to construct a causal skeleton network, which is represented by ,in, , represents n fault variables, , represents the edge set between variables;

[0013] S202: Initialize the causal skeleton network skG at the beginning of the algorithm, connect the variables in the variable set to each other, and obtain a completely undirected network;

[0014] S203: When the first layer is looped, condition sets of different lengths are traversed from small to large to test the independence of conditions. The length of the condition set L starts from 0, that is, the initial condition set is an empty set, and the loop is continued until all the condition sets in the The edge ;

[0015] S204: When the second layer loops, traverse all the nodes that meet the given conditions. Edge , for the edge , whose condition set only appears in Among the neighbors, ;

[0016] S205: When the third layer loops, traverse all sets with a set length of L ,like and If conditionally independent under a given condition Z, then delete the edge in skG , loop until and The edges between them have been deleted, or all Z have been traversed;

[0017] S206: Through the above steps S201 to S205, the final causal skeleton network skG is obtained.

[0018] S3: Use the sub-network generation and generation of intermediate variables to build forward and reverse models, determine the direction and strength of the causal relationship in the causal skeleton network, and obtain the final causal network G;

[0019] Preferably, the method for obtaining the final causal network G further comprises the following steps:

[0020] S301: Decompose the causal skeleton network skG into multiple sub-networks , for the subnetwork , there is a central variable ,have Non-central variables ,Right now It is expressed as follows: ;

[0021] S302: Traverse all sub-networks, determine the direction and weight of the edges in the sub-networks, and , using the variational autoencoder as a generative model, Non-central variables Generate an intermediate variable M, representing all non-central variables;

[0022] S303: In the forward model, for M and Regression analysis yields: , test noise Independence from M, and get the independence index ;

[0023] S304: In the reverse model, and each non-central variable Regression analysis yields: , test noise and Independence of ;

[0024] S305: Determine the direction of the edge in the sub-network according to the causal identification standard of ANM. If , then and The edges between are oriented → On the contrary, if , then the orientation is → ;

[0025] S306: Measuring causal strength by the degree of asymmetry between forward and reverse models , The calculation formula is as follows:

[0026]

[0027] S307: The processed sub-network is restored to the causal skeleton network, and the conflicting edges are processed by accepting the result with the greater weight, so as to obtain the final causal network G.

[0028] S4: Construct a corresponding causal network for each group of sub-time series in the normal mode, record the weight of each edge in the network, and obtain the fluctuation range of the causal strength of each edge ;

[0029] Preferably, the causal strength fluctuation range of each edge is obtained The specific method is: construct the corresponding causal network of N groups of sub-time series of normal mode , for causal networks , represented as a weighted directed network ,in ,express The set of directed edges of ,express The weight of the directed edge of the network, the weight of each edge in the network, The set of weights in the N networks It is expressed as follows:

[0030]

[0031] By the side The weight set Calculate the allowable fluctuation range of the causal strength of the edge , the calculation formula is as follows:

[0032]

[0033] in, Represents edge The weight set The minimum and maximum values ​​of .

[0034] S5: Construct a corresponding causal network for each group of sub-time series of the fault mode, and judge whether the causal strength of each edge exceeds the allowable fluctuation range. , confirm the abnormal causal relationship in the network and calculate the confidence distribution of a fault root cause;

[0035] Preferably, calculating a reliability distribution of a fault root cause further includes:

[0036] S501: Construct corresponding causal networks from N groups of sub-time series of failure modes , for causal networks , represented as a weighted directed network ,in ,express The set of directed edges of express The weights of the directed edges of ;

[0037] S502: Causal Network for Failure Modes , traverse all the edges in the network , determine the weight of the edge , that is, whether the causal strength exceeds the allowable fluctuation range of the causal strength , determine the network abnormal causality in;

[0038] S503: When side It is an abnormal causal relationship. As an abnormal point, it caused changes, yes The count is increased by one. express is the number of outliers, calculated by Get a set of confidence distributions P of root causes of faults, Representation variables is the confidence of the root cause of the failure, The calculation formula is as follows:

[0039]

[0040] in, express is the number of outliers.

[0041] S6: performing a fusion process on all obtained reliability distributions of the root causes of the fault to obtain a final reliability distribution of the root cause of the fault, and determining the root cause of the fault according to the reliability distribution;

[0042] Preferably, determining the root cause of the fault specifically includes: for the obtained N groups of root causes of the fault, the reliability distribution , and perform fusion processing to obtain the final reliability distribution of the root cause of the fault ,according to To determine the root cause of the fault, the fusion formula is as follows:

[0043]

[0044] in, Variables in the reliability distribution of the mth group of root causes of failure Reliability Represents the variable in the reliability distribution of the mth group of root causes of failure reliability.

[0045] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention realizes accurate positioning of the root cause of system faults through the steps of sliding window division of time series, causal network construction, calculation of the allowable fluctuation range of the strength of causal relationships, and calculation and fusion of the credibility distribution of the root causes of faults. At the same time, compared with the traditional root cause tracing method of faults based on causal analysis, the root cause of the fault is determined by comparing the changes in the causal relationship before and after the fault. In fact, the causal relationship that is not affected by the fault often originates from the normal process operation mechanism rather than the fault propagation. The method based on the dynamic temporal causal network can eliminate such redundant information, so that the model pays more attention to the causal relationship related to the fault. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0047] Figure 1 A flow chart of fault root cause tracing based on a dynamic temporal causal network and its use method provided by an embodiment of the present invention;

[0048] Figure 2 A causal skeleton network constructed for a first group of sub-time series of a normal mode provided by an embodiment of the present invention;

[0049] Figure 3 A causal network constructed for the first group of sub-time series of the normal mode provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] In this embodiment, combined Figure 1 ,like Figure 1 As shown, through the steps of sliding window division of time series, causal network construction, calculation of allowable fluctuation range of causal relationship strength, and calculation and fusion of reliability distribution of root cause of fault, the accurate location of root cause of system fault is achieved. The causal network construction here means: constructing causal networks for sub-time series after sliding window division of time series data of normal mode and fault mode respectively; in order to facilitate understanding of the above technical solution of the present invention, the above technical solution of the present invention is described in detail below through specific embodiments;

[0052] In this embodiment, the present invention provides a fault root cause tracing method based on a dynamic temporal causal network, comprising the following steps:

[0053] S1: Use the time sliding window algorithm to divide the time series data of normal mode and fault mode into sub-time series of the same width;

[0054] In this embodiment, the data set of this embodiment is composed of time series data of 6 characteristic variables, expressed as 1_x, 2_x, ..., 6_x, which respectively represent the vibration data of the 6 positions of the elevator, namely, the No. 1 traction wheel, the No. 2 guide wheel, the No. 3 car top wheel, the No. 4 counterweight wheel, the No. 5 car three-axis, and the No. 6 counterweight three-axis;

[0055] Preferably, by constructing 1 normal mode data set and 3 fault mode data sets for fault root cause tracing, namely, guide wheel wear data set, car top wheel wear data set, and guide wheel bearing damage data set, the data volume is 50,000, and each fault corresponds to a fault root cause. For example, the fault root cause corresponding to the guide wheel wear fault is guide wheel No. 2, that is, the fault root cause variable is 2_x. Through the method studied, 3 groups of results can be obtained for the 3 fault categories respectively, and finally the effectiveness of the method is verified;

[0056] Preferably, the time series data of the normal mode and the fault mode are extracted by sliding window, and the total length of the data is recorded as , take the sliding window size as , the step length is , divide the time series data into N groups of sub-time series with the same width. For example, in this embodiment, the total length of the data is defined as , sliding window size , step length , the time series data of normal mode and failure mode are divided into Sub-time series with the same group width, The calculation formula is shown in formula (1):

[0057] +1 (1)

[0058] Where [x] represents the largest integer not exceeding x.

[0059] S2: Normal mode and fault mode The sub-time series with the same group width all construct causal networks. This step is the first step in constructing a causal network. The variables of the sub-time series are connected in pairs to obtain a completely undirected network. The edges are gradually deleted through the conditional independence test method to obtain the causal skeleton network skG;

[0060] In this embodiment, combined Figure 2, the causal skeleton network constructed by the first group of sub-time series of the normal mode is as follows Figure 2 As shown, by using the sub-time series to construct a causal skeleton network, the causal skeleton network is expressed as ,in, , represents n fault variables, , represents the edge set between variables;

[0061] Preferably, at the beginning of the algorithm, the causal skeleton network skG is initialized, and the variables in the variable set are connected in pairs to obtain a completely undirected network;

[0062] In the first loop, condition sets of different lengths are traversed from small to large to test conditional independence. The condition set length L starts from 0, that is, the initial condition set is an empty set, and the loop continues until all the condition sets are in the same state. The edge ;

[0063] In the second loop, traverse all the Edge , for the edge For example, its condition set should only appear in Among the neighbors, ;

[0064] In the third loop, traverse all sets with a set length of L ,like and If conditionally independent under a given condition Z, then delete the edge in skG , loop until and The edges between have been deleted, or all Z have been traversed; through the above steps, the final causal skeleton network skG is obtained.

[0065] S3: This step is the second step of constructing the causal network. It determines the direction and strength of the causal relationship in the causal skeleton network obtained by S2, and converts the many-to-one causal relationship structure into an approximate one-to-one causal relationship structure in a way that a sub-network is generated and an intermediate variable is generated. Based on the principle of nonlinear additive noise ANM, forward and reverse models are constructed in the sub-network for identification;

[0066] In this embodiment, combined with Figure 3 , the causal network constructed by the first group of sub-time series of the normal mode is as follows Figure 3 As shown, the first step is to decompose the causal skeleton network skG into multiple sub-networks. , for the subnetwork , there is a central variable ,have Non-central variables ;

[0067] The second step is to traverse all sub-networks and determine the direction and weight of the edges in the sub-networks. , using the variational autoencoder as a generative model, Non-central variables Generate an intermediate variable M, representing all non-central variables. In the forward model, M and Regression analysis , test noise Independence from M, and get the independence index In the reverse model, and each non-central variable Regression analysis , test noise and Independence of ; Finally, the direction of the edge in the subnetwork is determined according to the causal identification standard of ANM. , then and The edges between are oriented → On the contrary, if , then the orientation is → The ability to identify causal structures is due to the asymmetry between the cause variable and the result variable. Therefore, the degree of asymmetry between the forward model and the reverse model is used to measure the causal strength. , The calculation formula is shown in (2):

[0068] (2)

[0069] The third step is to restore the processed sub-network to the causal skeleton network. The way to deal with conflicting edges is to accept the result with heavy weight and obtain the final causal network G.

[0070] S4: Construct the corresponding causal network for the 50 groups of sub-time series of the normal mode, and count the weight of each edge in the network to calculate the fluctuation range of the causal strength of each edge , the calculated causal strength fluctuation range of each edge ;

[0071] For example, construct the corresponding causal network of N groups of sub-time series of normal mode , for causal networks , represented as a weighted directed network ,in ,express The set of directed edges of ,express The weight of the directed edge of the network, the weight of each edge in the network, The set of weights in the N networks It is expressed as follows:

[0072] (3)

[0073] By the side The weight set Calculate the allowable fluctuation range of the causal strength of the edge , the calculation formula is as follows:

[0074] (4)

[0075] Preferably, only the causal strength fluctuation range is listed below The edges that are not 0 are shown in Table 1;

[0076] Table 1: The fluctuation range of causal strength of each edge

[0077]

[0078] S5: Construct corresponding causal networks for the 50 groups of sub-time series of the fault mode. For each causal network, determine whether the causal strength of each edge exceeds the allowable fluctuation range. , find out the abnormal causal relationship in the network and calculate the confidence distribution of the root cause of the fault.

[0079] In this embodiment, the N groups of sub-time series of the fault mode are used to construct the corresponding causal network , for causal networks , represented as a weighted directed network ,in ,express The set of directed edges of ,express The weights of the directed edges of ;

[0080] For example, the causal network of failure modes , traverse all the edges in the network , determine the weight of the edge , that is, whether the causal strength exceeds the allowable fluctuation range of the causal strength , find the network The abnormal causal relationship in side It is an abnormal causal relationship. As an abnormal point, it caused changes, yes The count is increased by one. express is the number of abnormal points, and the confidence distribution P of a set of fault root causes is calculated by formula (5). Representation variables is the reliability of the root cause of the fault, and formula (5) is as follows:

[0081] (5)

[0082] For example, 50 groups of sub-time series of failure modes can obtain the confidence distribution of 50 root causes of failures. , for example, guide wheel wear failure As shown in Table 2.

[0083] Table 2: Reliability distribution

[0084]

[0085] S6: Fusion processing is performed on the reliability distribution of the 50 root causes of faults in S5 to obtain the final reliability distribution of the root cause of faults. , the root cause of the fault can be determined based on the distribution;

[0086] In this embodiment, for the obtained N groups of fault root causes, , and perform fusion processing to obtain the final reliability distribution of the root cause of the fault ,according to To determine the root cause of the fault, the fusion formula is as follows:

[0087] (6)

[0088] Preferably, for the guide wheel wear fault, the final reliability distribution is obtained As shown in Table 3;

[0089] Table 3: Confidence distribution of root causes of guide wheel wear failure

[0090]

[0091] For example, the root cause of the guide wheel wear fault is guide wheel No. 2, that is, the root cause variable is 2_x. The highest confidence level is 2_x, which can accurately locate the root cause of the guide wheel wear failure;

[0092] Preferably, for the car top wheel wear fault, the final reliability distribution is obtained As shown in Table 4;

[0093] Table 4: Confidence distribution of root causes of car top wheel wear failure

[0094]

[0095] For example, the root cause of the car top wheel wear fault is the No. 3 car top wheel, that is, the root cause variable is 3_x, and the reliability distribution is The highest reliability is 3_x, which realizes the accurate location of the root cause of the car top wheel wear failure; preferably, for the guide wheel bearing damage failure, the final reliability distribution is obtained As shown in Table 5;

[0096] Table 5: Confidence distribution of root causes of guide wheel bearing damage failure

[0097]

[0098] For example, the root cause of the guide wheel bearing damage fault is the guide wheel No. 2, that is, the root cause variable is 2_x, and the reliability distribution is The one with the highest reliability is 2_x, which can accurately locate the root cause of the guide wheel bearing damage failure.

[0099] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0100] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A fault root cause tracing method based on a dynamic temporal causal network, characterized by: The following steps are involved: S1: Use the time sliding window algorithm to divide the time series data of normal mode and fault mode into sub-time series of the same width; S2: Connect the variables of the sub-time series in pairs to obtain a completely undirected network, and gradually delete the edges through the conditional independence test method to obtain the causal skeleton network skG; S3: Use the sub-network generation and generation of intermediate variables to build forward and reverse models, determine the direction and strength of the causal relationship in the causal skeleton network, and obtain the final causal network G; The method for obtaining the final causal network G comprises the following steps: S301: Decompose the causal skeleton network skG into multiple sub-networks S i = {V i ,E i }, i = 1, 2, ..., n, for sub-network S i , there is a central variable There is k i Non-central variables Right now It is expressed as follows: S302: Traverse all sub-networks and determine the direction and weight of the edges in the sub-networks. i , using the variational autoencoder as a generative model, k i Non-central variables Generate an intermediate variable M, representing all non-central variables; S303: In the forward model, for M and Regression analysis yields: Test noise Independence from M, and get the independence index Ind M ; S304: In the reverse model, and each non-central variable Regression analysis yields: Test noise and Independence of S305: Determine the direction of the edge in the sub-network according to the causal identification standard of ANM. If Then and The edges between are oriented On the contrary, if The orientation is S306: Measuring causal strength w by the degree of asymmetry between the forward model and the reverse model ij , w ij The calculation formula is as follows: S307: restore the processed sub-network to the causal skeleton network, and handle the conflicting edges by accepting the result with the greater weight, so as to obtain the final causal network G; S4: Construct a corresponding causal network for each group of sub-time series of the normal mode, record the weight of each edge in the network, and obtain the causal strength fluctuation range FR of each edge; S5: Construct a corresponding causal network for each group of sub-time series of the fault mode, determine whether the causal strength of each edge exceeds the allowable fluctuation range FR, confirm the abnormal causal relationship in the network, and calculate a confidence distribution of the root cause of the fault; S6: performing fusion processing on all obtained reliability distributions of the root causes of the fault to obtain a final reliability distribution of the root cause of the fault, and determining the root cause of the fault according to the reliability distribution.

2. The fault root cause tracing method based on a dynamic temporal causal network according to claim 1 is characterized in that: The specific process of step S1 further includes: performing sliding window extraction on the time series data of the normal mode and the fault mode, recording the total length of the data as Len, taking the sliding window size as ξ, and the step length as step, dividing the time series data into N groups of sub-time series with the same width, and the calculation formula of N is as follows: Where [x] represents the largest integer not exceeding x.

3. The fault root cause tracing method based on a dynamic temporal causal network according to claim 2 is characterized in that: The method for obtaining the causal skeleton network skG in step S2 further comprises: S201: Use the sub-time series to construct a causal skeleton network, which is represented by skG = {V, E}, where V = {v i |i=1,2…,n}, represents n fault variables, represents the set of edges between variables; S202: Initialize the causal skeleton network skG at the beginning of the algorithm, connect the variables in the variable set to each other, and obtain a completely undirected network; S203: When the first layer is looped, condition sets of different lengths are traversed from small to large to test conditional independence. The condition set length L starts from 0, that is, the initial condition set is an empty set, and the loop is continued until all edges (v i ,v j ):|adj(skG,v i )\{v j }| <L; S204: When the second layer loops, traverse all the nodes that meet the given condition |adj(skG,v i )\{v j }|≥L edge (v i ,v j ), for edge (v i ,v j ), whose conditional set only appears in v i Among the neighbors, adj(skG,v i )\{v j }; S205: When the third layer loops, traverse all sets with a set length of L If v i and v j Given the condition Z, we can delete the edge (v i ,v j ), loop until v i and v j The edges between them have been deleted, or all Z have been traversed; S206: Through the above steps S201 to S205, the final causal skeleton network skG is obtained.

4. The fault root cause tracing method based on a dynamic temporal causal network according to claim 3 is characterized by: The method for obtaining the final causal network G in step S3 also includes the following prerequisites: the direction and strength of the causal relationship in the causal skeleton network obtained in step S2, converting the many-to-one causal relationship structure into an approximate one-to-one causal relationship structure in a way of generating a sub-network and generating intermediate variables, and constructing forward and reverse models in the sub-network for identification based on the principle of nonlinear additive noise ANM.

5. The fault root cause tracing method based on a dynamic temporal causal network according to claim 4 is characterized in that: The specific method for obtaining the causal strength fluctuation range FR of each edge in step S4 is: construct the corresponding causal network G1, G2, ..., G N , for the causal network G m ,m=1,2,…,N, which means a weighted directed network in Represents G m The set of directed edges of Represents G m The weight of the directed edge of the network, the weight of each edge in the network, the edge (v i ,v j ) The weight set W in N networks ij It is expressed as follows: By the edge (v i ,v j )’s weight set W ij Calculate the allowable fluctuation range FR of the causal strength of the edge ij , the calculation formula is as follows: Among them, Min(W ij ),Max(W ij ) represents the edge (v i ,v j )’s weight set W ij The minimum and maximum values ​​of .

6. The fault root cause tracing method based on a dynamic temporal causal network according to claim 5 is characterized in that: In step S5, calculating a reliability distribution of a fault root cause also includes: S501: Construct corresponding causal networks G'1, G'2, ... G' from N groups of sub-time series of failure modes N , for the causal network G' m ,m=1,2,…,N, expressed as a weighted directed network G' m =(V,E' m ,W' m ),in G' m The set of directed edges of G' m The weights of the directed edges of ; S502: Causal network G' for failure mode m , traverse all the edges in the network (v i ,v j )∈E' m , determine the weight of the edge That is, whether the causal strength exceeds the allowable fluctuation range of the causal strength FR ij , determine the network G' m abnormal causality in; S503: When Edge (v i ,v j ) is an abnormal causal relationship, v i is an abnormal point, triggering v j changes in a i Count plus one, a i Indicates v i is the number of outliers, by calculating P(v i ) to obtain a set of confidence distributions of the root causes of faults P, P(v i ) represents the variable v i is the confidence of the root cause of the fault, P(v i ) is calculated as follows: Among them, a j Indicates v j is the number of outliers.

7. The fault root cause tracing method based on dynamic temporal causal network according to claim 6 is characterized by: Determining the root cause of the fault in step S6 specifically includes: for the obtained N groups of root causes of faults, the reliability distribution P m ,m=1,2,…,N, and perform fusion processing to obtain the final reliability distribution P of the root cause of the fault result , according to P result To determine the root cause of the fault, the fusion formula is as follows: Among them, P m (v i ) represents the variable v in the reliability distribution of the mth group of fault root causes i The reliability, P m (v j ) represents the variable v in the reliability distribution of the mth group of fault root causes j reliability.

Images