A fault tracing method and device based on causal analysis, and electronic equipment.

Abstract

This disclosure relates to a fault tracing method, apparatus, and electronic device based on causal analysis, comprising: acquiring a dataset to be processed, wherein the dataset includes multiple related variables in the target system corresponding to the target fault when the target fault occurs, and a data sequence of each related variable within a preset time period; performing causal analysis on the dataset to be processed to determine the dependent variable that has a causal relationship with the target fault among the multiple related variables; determining the intervention probability corresponding to each dependent variable based on a Bayesian network model, wherein the intervention probability corresponding to each dependent variable is used to represent the probability value that the dependent variable is the cause of the fault; and determining the fault cause corresponding to the target fault based on the intervention probability corresponding to each dependent variable. The embodiments of this disclosure can eliminate the influence of confounding factors and mediating variables, improve the accuracy of fault tracing results, and avoid instability in fault tracing results caused by noise.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a fault tracing method and apparatus, and electronic equipment based on causal analysis. Background Technology

[0002] Faults occurring during industrial production can spread rapidly within industrial systems, impacting production safety and product quality. Therefore, timely fault tracing is crucial to identify the root cause. However, due to the interplay between production variables, multiple alarm paths may exist from the root cause to the fault detection point, potentially including false alarms, making accurate fault tracing difficult. Summary of the Invention

[0003] In view of this, this disclosure proposes a fault tracing method and device based on causal analysis, as well as a technical solution for electronic equipment.

[0004] According to one aspect of this disclosure, a fault tracing method based on causal analysis is provided, comprising: acquiring a dataset to be processed, wherein the dataset includes multiple related variables in the target system corresponding to the target fault when a target fault occurs in the target system, and a data sequence of each related variable within a preset time period; performing causal analysis on the dataset to be processed to determine a dependent variable that has a causal relationship with the target fault among the multiple related variables; determining an intervention probability corresponding to each dependent variable based on a Bayesian network model, wherein the intervention probability corresponding to each dependent variable is used to represent the probability value that the dependent variable is the cause of the fault; and determining the fault cause corresponding to the target fault based on the intervention probability corresponding to each dependent variable.

[0005] In one possible implementation, the step of performing causal analysis on the dataset to be processed to determine the dependent variable among the plurality of relevant variables includes: grouping the dataset to be processed to determine a processing data group and a control data group; determining the causal regression coefficient corresponding to each of the relevant variables based on the processing data group and the control data group, using a stable learning algorithm, wherein the causal regression coefficient corresponding to each of the relevant variables is used to represent the correlation between the relevant variable and the target fault; and determining the dependent variable among the plurality of relevant variables based on the causal regression coefficient corresponding to each of the relevant variables.

[0006] In one possible implementation, grouping the dataset to be processed to determine the processing data group and the control data group includes: determining an index vector corresponding to each of the relevant variables based on the mean of normal samples corresponding to each of the relevant variables, wherein the mean of normal samples corresponding to each of the relevant variables is the average value of the relevant variable when the target system is in a normal state; and dividing the dataset to be processed into the processing data group and the control data group based on the index vector corresponding to each of the relevant variables.

[0007] In one possible implementation, determining the causal regression coefficient corresponding to each of the relevant variables based on a stable learning algorithm, according to the processed data set and the control data set, includes: determining a target optimization problem based on the processed data set and the control data set, wherein the target optimization problem is used to extract the dependent variable from the plurality of relevant variables; and solving the target optimization problem using the stable learning algorithm to determine the causal regression coefficient corresponding to each of the relevant variables.

[0008] In one possible implementation, determining the dependent variable from the plurality of related variables based on the causal regression coefficients corresponding to each of the related variables includes: sorting the causal regression coefficients corresponding to each of the related variables to determine a sorted coefficient sequence, wherein the coefficient sequence includes the absolute value of the causal regression coefficient corresponding to each of the related variables, and is sorted in descending order of absolute value; sequentially accumulating the absolute values ​​of the causal regression coefficients corresponding to each of the related variables included in the coefficient sequence, and comparing the cumulative percentage determined after each accumulation with a preset threshold; when the cumulative percentage corresponding to the first k related variables in the coefficient sequence is less than or equal to the preset threshold, and the cumulative percentage corresponding to the first k+1 related variables in the coefficient sequence is greater than the preset threshold, the first k related variables in the coefficient sequence are determined as the dependent variable.

[0009] In one possible implementation, determining the intervention probability corresponding to each dependent variable based on a Bayesian network model includes: determining a causal graph corresponding to the target fault based on the dependent variable and the label data sequence corresponding to the target system, wherein the label data sequence represents the working state of the target system within a preset time period; discretizing each dependent variable based on the data sequence of each dependent variable within the preset time period to obtain a discretized variable corresponding to each dependent variable; constructing a Bayesian network model based on the causal graph and the discretized variables corresponding to each dependent variable; determining a conditional probability table corresponding to the target fault and a marginal probability corresponding to each node in the causal graph based on the Bayesian network model, wherein the conditional probability table represents the probability of the target fault occurring when each dependent variable is abnormal, and the marginal probability represents the probability of each dependent variable being abnormal; and determining the intervention probability corresponding to each dependent variable based on the conditional probability table and the marginal probability.

[0010] In one possible implementation, the method further includes: when the target system experiences the target fault, obtaining the original dataset corresponding to the target system; performing data standardization processing on the original dataset to determine the dataset to be processed.

[0011] According to another aspect of this disclosure, a fault tracing device based on causal analysis is provided, comprising: a data acquisition module for acquiring a dataset to be processed, wherein the dataset to be processed includes multiple related variables in the target system corresponding to the target fault when a target fault occurs in the target system, and a data sequence of each related variable within a preset time period; a dependent variable extraction module for performing causal analysis on the dataset to be processed to determine the dependent variable that has a causal relationship with the target fault among the multiple related variables; an intervention probability determination module for determining the intervention probability corresponding to each dependent variable based on a Bayesian network model, wherein the intervention probability corresponding to each dependent variable is used to represent the probability value that the dependent variable is the cause of the fault; and a fault source determination module for determining the fault cause corresponding to the target fault based on the intervention probability corresponding to each dependent variable.

[0012] According to another aspect of this disclosure, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the above-described method when executing instructions stored in the memory.

[0013] According to another aspect of this disclosure, a non-volatile computer-readable storage medium is provided that stores computer program instructions thereon, wherein the computer program instructions, when executed by a processor, implement the above-described method.

[0014] According to another aspect of this disclosure, a computer program product is provided, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0015] In this embodiment, a dataset to be processed is acquired. This dataset includes multiple related variables in the target system corresponding to the target fault when the target system experiences a target fault, and a data sequence for each related variable within a preset time period. Causal analysis is performed on the dataset to determine the dependent variable with a causal relationship to the target fault among the multiple related variables. This identifies key variables related to the target fault, improving the accuracy and efficiency of fault cause localization. Based on a Bayesian network model, the intervention probability corresponding to each dependent variable is determined. Based on the intervention probability of each dependent variable, the fault cause corresponding to the target fault can be directly determined. Compared to traditional fault tracing methods based on conditional probability, this eliminates the influence of confounding factors and mediating variables, improves the accuracy of fault tracing results, and avoids instability caused by noise.

[0016] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0017] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0018] Figure 1 This illustrates a causal graph of the relationship between a processing variable X and an experimental result Y in a related technique.

[0019] Figure 2 This illustrates a causal graph of the relationship between a processing variable X and an experimental result Y in a related technique.

[0020] Figure 3 This illustrates a causal graph of the relationship between a processing variable X and an experimental result Y in a related technique.

[0021] Figure 4 A flowchart illustrating a fault tracing method based on causal analysis according to an embodiment of the present disclosure is shown.

[0022] Figure 5 A histogram showing the causal regression coefficients of a condenser cooling water inlet temperature fault according to an embodiment of the present disclosure is provided.

[0023] Figure 6 A histogram showing the squared prediction error of a condenser cooling water inlet temperature fault in the related art is shown.

[0024] Figure 7 This diagram illustrates a cause-effect graph corresponding to a feed ratio variation fault according to an embodiment of the present disclosure.

[0025] Figure 8 A cause-effect diagram is shown corresponding to a feed temperature fault according to an embodiment of the present disclosure.

[0026] Figure 9 A histogram showing the causal regression coefficients corresponding to a fault in a frame roll gap control actuator according to an embodiment of the present disclosure is provided.

[0027] Figure 10 A histogram showing the causal regression coefficients corresponding to a failure of a cooling water valve actuator between racks according to an embodiment of the present disclosure;

[0028] Figure 11 A cause-effect diagram is shown corresponding to a failure of a cooling water valve actuator between racks according to an embodiment of the present disclosure.

[0029] Figure 12 A block diagram of a fault tracing device based on causal analysis according to an embodiment of the present disclosure is shown;

[0030] Figure 13 A block diagram of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0031] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0032] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0033] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0034] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0035] When a fault occurs in an industrial system during production, it can spread rapidly throughout the entire system, severely impacting production safety and product quality. Therefore, it is crucial to promptly trace the fault and identify the root cause. However, due to the interplay between production variables, multiple alarm paths may exist between the root cause and the fault detection point, potentially including false alarms, making fault tracing difficult.

[0036] In existing technologies, Bayesian network-based inference methods are commonly used to analyze fault tracing problems, determining the propagation path and root cause of faults through conditional probability inference. However, the large number of components in complex industrial systems and the difficulty in acquiring relevant knowledge increase the complexity of building graph models based on prior knowledge. Furthermore, most data-driven graph model building methods are computationally intensive, further complicating the construction of Bayesian network models.

[0037] In the prior art, confounding factors refer to variables that simultaneously affect both the treatment variables and the experimental results in causal analysis.

[0038] Figure 1 This illustrates a causal graph showing the relationship between a processing variable X and an experimental result Y in a related technique. For example... Figure 1 As shown, there is no direct path between the treatment variable X and the experimental result Y, but both are influenced by variable Z. Therefore, when performing causal analysis based on relevant data to analyze the causal relationship between the treatment variable X and the experimental result Y, the presence of variable Z results in a conditional probability P(Y|X)≠0. If we take the conditional probability P(Y|X)≠0 as the result of the causal analysis, it will lead to a spurious causal relationship, or pseudo-correlation, between the treatment variable X and the experimental result Y. In fact, the conditional probability P(Y|X)≠0 here only indicates a correlation between the treatment variable X and the experimental result Y, not a causal relationship. In this analysis, variable Z is the confounding factor.

[0039] Figure 2 This illustrates a causal graph showing the relationship between a processing variable X and an experimental result Y in a related technique. For example... Figure 2As shown, the treatment variable X points to the experimental result Y, and the variable Z points to both the treatment variable X and the experimental result Z. In this case, during the causal analysis of the treatment variable X and the experimental result Y, due to the confounding factor Z, the conditional probability P(Y|X) obtained based on the relevant data cannot reflect the true causal relationship between the treatment variable X and the experimental result Y.

[0040] Traditional Bayesian network-based inference methods rely on conditional probabilities between variables. As discussed above, conditional probabilities in Bayesian networks essentially represent the correlation between two variables. However, according to the correlation theory of causal inference, the variable relationships obtained from correlation analysis may be affected by confounding factors, thus impacting the performance of the fault diagnosis system and making the fault attribution results lack interpretability. Furthermore, due to the influence of confounding factors, the causal graph constructed based on variable data may deviate from the true causal graph, containing spurious connections and leading to inaccurate fault attribution results.

[0041] Figure 3 This illustrates a causal graph showing the relationship between a processing variable X and an experimental result Y in a related technique. For example... Figure 2 As shown, there are two paths between the treatment variable X and the experimental result Y: X→Y and X→Z→Y, with a confounding factor C also present. In this case, variable Z is called the mediating variable. The first path, X→Y, between the treatment variable X and the experimental result Y is a direct causal path.

[0042] After a failure occurs, the propagation process from the root cause node to the fault detection node may be affected by the propagation from the mediator variable to the fault detection node. Therefore, in order to accurately determine the direct causal path from the root cause node to the fault detection node, it is also necessary to consider the influence of the mediator variable, which has not been considered in existing Bayesian network inference methods.

[0043] This disclosure provides a fault tracing method based on causal analysis. Based on the principles of causal analysis, it eliminates the influence of confounding factors and mediating variables on fault tracing, directly determining the true cause of the fault. The following provides a detailed description of the fault tracing method based on causal analysis provided in this disclosure.

[0044] Figure 4A flowchart illustrating a fault tracing method based on causal analysis according to an embodiment of this disclosure is shown. This fault tracing method based on causal analysis can be executed by an electronic device such as a terminal device or a server. The terminal device can be a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, in-vehicle device, wearable device, etc. This fault tracing method based on causal analysis can be implemented by a processor calling computer-readable instructions stored in memory. Alternatively, this fault tracing method based on causal analysis can be executed by a server. Figure 4 As shown, this fault tracing method based on causal analysis includes:

[0045] In step S41, a dataset to be processed is obtained, wherein the dataset to be processed includes multiple related variables in the target system corresponding to the target fault when the target system experiences the target fault, and the data sequence of each related variable within a preset time period.

[0046] The target system here can be an industrial system or any other system with multiple components. A target failure can be an abnormal phenomenon occurring during the operation and / or production of the target system, where its characteristics or parameters deviate significantly from acceptable ranges, leading to a decline in the system's performance. The relevant variables corresponding to the target failure are variables in the target system that may be associated with it. For example, when target system A experiences a target failure, the temperature, humidity, pressure, etc., corresponding to component a of target system A are all relevant variables for the target failure.

[0047] The dataset to be processed includes multiple relevant variables in the target system corresponding to the target fault, and the data sequence of each relevant variable within a preset time period. The data sequence of any relevant variable within the preset time period includes the data corresponding to that variable at each moment within that preset time period. For example, the temperature corresponding to component 'a' in target system A is a relevant variable corresponding to the target fault occurring in target system A. The preset time period includes moments T1 to T3. The data corresponding to this relevant variable at moment T1 is 35℃, at moment T2 it is 39℃, and at moment T3 it is 32℃. Therefore, the data sequence corresponding to this relevant variable within the preset time period can be determined as Y = [35, 39, 32]. The preset time period includes the time when the target fault occurs and can be set according to actual usage requirements; this disclosure does not impose specific limitations on it. The specific form of the data sequence of each relevant variable within the preset time period can be set according to actual usage requirements; this disclosure does not impose specific limitations on it.

[0048] For specific methods of obtaining the dataset to be processed, please refer to the implementation methods in related technologies. This disclosure does not limit the specific methods in this regard.

[0049] In step S42, a causal analysis is performed on the dataset to be processed to identify the dependent variable that has a causal relationship with the target fault among multiple related variables.

[0050] By performing causal analysis on the dataset to be processed, the dependent variable that has a causal relationship with the target fault can be identified from multiple related variables corresponding to the target fault. The dependent variable can be used to represent the variables corresponding to the fault causes that may lead to the target fault in the target system. Fault tracing based on the dependent variable can narrow down the candidate set of fault causes, improving the accuracy and efficiency of fault cause localization.

[0051] The following sections will describe in detail the process of causal analysis and the determination of the dependent variable that has a causal relationship with the target fault among multiple relevant variables, based on the possible implementation methods of this disclosure; these details will not be elaborated upon here.

[0052] In step S43, based on the Bayesian network model, the intervention probability corresponding to each dependent variable is determined, wherein the intervention probability corresponding to each dependent variable is used to represent the probability value that the dependent variable is the cause of the fault.

[0053] After determining the dependent variable corresponding to the target fault based on causal analysis, a Bayesian network model containing the dependent variable can be constructed, and the intervention probability corresponding to each dependent variable can be determined based on the Bayesian network model.

[0054] The Bayesian network model containing dependent variables determined in this disclosure embodiment, compared with the traditional Bayesian network model in the prior art, takes into more consideration the relationship between the dependent variable and the target fault, as well as the relationship between different dependent variables, and can eliminate the influence of confounding factors and mediating variables.

[0055] The following sections will describe in detail the process of determining the intervention probability corresponding to each dependent variable based on the Bayesian network model, taking into account the possible implementation methods of this disclosure. These details will not be elaborated upon here.

[0056] In step S44, the cause of the target fault is determined based on the intervention probability corresponding to each dependent variable.

[0057] After determining the intervention probability for each dependent variable, the dependent variable causing the target fault can be directly determined based on the intervention probability, thereby determining the cause of the target fault.

[0058] In this embodiment, a dataset to be processed is acquired. This dataset includes multiple related variables in the target system corresponding to the target fault when the target system experiences a target fault, and a data sequence for each related variable within a preset time period. Causal analysis is performed on the dataset to determine the dependent variable with a causal relationship to the target fault among the multiple related variables. This identifies key variables related to the target fault, improving the accuracy and efficiency of fault cause localization. Based on a Bayesian network model, the intervention probability corresponding to each dependent variable is determined. Based on the intervention probability of each dependent variable, the fault cause corresponding to the target fault can be directly determined. Compared to traditional fault tracing methods based on conditional probability, this eliminates the influence of confounding factors and mediating variables, improves the accuracy of fault tracing results, and avoids instability caused by noise.

[0059] In one possible implementation, the method further includes: when a target system fails, obtaining the original dataset corresponding to the target system; performing data standardization on the original dataset to determine the dataset to be processed.

[0060] When a target system experiences a target fault, the raw dataset of the target system within a preset time period can be obtained. The raw dataset may include multiple relevant variables in the target system corresponding to the target fault, as well as the initial data sequence of each relevant variable within the preset time period. Specifically, the initial data sequence for each relevant variable may include the unprocessed initial data collected at each moment within the preset time period for that relevant variable.

[0061] Since the initial data for different related variables in the original dataset may have different value ranges and units, after obtaining the original dataset, data standardization can be performed to determine the corresponding dataset to be processed, which facilitates subsequent fault tracing. Specific methods for data standardization can be found in related technical implementations, and this disclosure does not impose specific limitations on them.

[0062] In one example, the dataset to be processed can be represented by formula (1):

[0063]

[0064] Where X′ represents the dataset to be processed; X represents the original dataset, and X∈R n×p , where n represents the number of data sequences of the relevant variables in the original dataset within a preset time period, p represents the number of relevant variables in the original dataset; M represents the mean of the normal samples corresponding to each relevant variable, and M∈R 1×p ;e i Let represent the standard deviation of the i-th relevant variable in the original dataset X, where i = 1, 2, ..., p.

[0065] In one possible implementation, causal analysis is performed on the dataset to be processed to determine the dependent variable among multiple relevant variables, including: grouping the dataset to be processed to determine a processing data group and a control data group; determining the causal regression coefficient corresponding to each relevant variable based on a stable learning algorithm, according to the processing data group and the control data group, wherein the causal regression coefficient corresponding to each relevant variable is used to represent the correlation between the relevant variable and the target fault; and determining the dependent variable among multiple relevant variables based on the causal regression coefficient corresponding to each relevant variable.

[0066] In existing technologies, traditional machine learning typically assumes that the training and test datasets have the same distribution. When the dataset does not meet this assumption in real-world applications, it leads to unstable predictive performance of traditional machine learning methods. Stable learning is a causal analysis method based on a latent outcome model framework. When the actual causal structure is unknown, it can improve the stability of predictions and enhance the performance of traditional machine learning by balancing the influence of confounding factors.

[0067] Based on this, after obtaining the dataset to be processed, it can be grouped to determine the processing data group and the control data group. Using a stable learning algorithm, causal analysis is performed based on the processing and control data groups to determine the causal regression coefficients corresponding to each relevant variable.

[0068] Since the causal regression coefficient corresponding to each relevant variable can represent the correlation between that relevant variable and the target fault, the dependent variable corresponding to the target fault can be determined from multiple relevant variables based on the causal regression coefficient corresponding to each relevant variable.

[0069] In one possible implementation, the dataset to be processed is grouped to determine the processing data group and the control data group, including: determining the index vector corresponding to each relevant variable based on the normal sample mean corresponding to each relevant variable, wherein the normal sample mean corresponding to each relevant variable is the average value of the relevant variable when the target system is in a normal state; and dividing the dataset to be processed into the processing data group and the control data group based on the index vector corresponding to each relevant variable.

[0070] Based on the data of each relevant variable in the dataset to be processed under the normal state of the target system, the mean of the normal samples corresponding to each relevant variable can be determined. The mean of the normal samples corresponding to each relevant variable can represent the average value of that relevant variable under the normal state of the target system.

[0071] By using the mean of the normal samples corresponding to each relevant variable, the indicator vector corresponding to each relevant variable can be determined. The indicator vector for each relevant variable can be used to indicate whether the data corresponding to that relevant variable exceeds the mean of the normal samples. For the specific method of determining the indicator vector for each relevant variable, please refer to the implementation methods in related technologies; this disclosure does not impose specific limitations on it.

[0072] In one example, for the dataset to be processed, which is determined by data standardization using formula (1), the index vector corresponding to each relevant variable in the dataset can be represented by formula (2):

[0073]

[0074] Where X′(i,j) represents the i-th data point in the data sequence for the j-th related variable in the dataset X′ to be processed; I j (i) represents the index vector I j The i-th data, i = 1, 2, 3, ..., n; j = 1, 2, 3, ..., p.

[0075] For the dataset to be processed, which is determined by data standardization using formula (1), when X′(i,j)≥0, it indicates that the corresponding data in the original dataset exceeds the mean of the corresponding normal sample; when X′(i,j)<0, it indicates that the corresponding data in the original dataset does not exceed the mean of the corresponding normal sample. Therefore, the index vector determined by formula (2) can distinguish between the data that exceed the mean of the normal sample and the data that do not exceed the mean of the normal sample in all data of the j-th variable of the dataset to be processed, thereby dividing the dataset to be processed into a processing data group and a control data group.

[0076] In one possible implementation, based on the processing data set and the control data set, the causal regression coefficients corresponding to each relevant variable are determined using a stable learning algorithm. This includes: determining an objective optimization problem based on the processing data set and the control data set, wherein the objective optimization problem is used to extract the dependent variable from multiple relevant variables; and solving the objective optimization problem using a stable learning algorithm to determine the causal regression coefficients corresponding to each relevant variable.

[0077] A stable learning algorithm based on Causally Regularized Logistic Regression (CRLR) works by sequentially setting each relevant variable in the dataset as a treatment variable and the remaining relevant variables as confounding factors. It then estimates the causal impact of the treated variables on the target fault. During this process, the CRLR algorithm learns a sample weighting matrix to balance the moments of confounding factors between the treatment and control data groups. This sample weighting matrix ensures that all input variables set as treatment variables are independent, thus minimizing the influence of confounding factors.

[0078] Logistic regression analysis is used to study influence relationships, which can be described as the effect of variable A on variable B, where variable A is categorical data and variable B can be either quantitative or categorical data. The CRLR algorithm incorporates a sample weighting matrix into the loss function of the logistic regression model, and then determines the sample weighting matrix and causal regression coefficients through iterative optimization.

[0079] Based on the above principles, the objective optimization problem can be expressed as formulas (3) to (6):

[0080]

[0081]

[0082]

[0083]

[0084] In this context, formula (3) represents the weighted logistic regression loss function determined based on the CRLR algorithm; formula (4) can be used to balance the influence of confounding factors; formula (5) can be used to avoid model overfitting; and formula (6) can be used to ensure that the sample weighting matrix W is non-negative and not all zero. W represents the sample weighting matrix; Y represents the label data sequence, and Y∈R n×1 The labeled data sequence can be used to represent the working state of the target system within a preset time period; β represents the causal regression coefficient sequence; This indicates the result after setting the j-th column of the dataset X′ to be processed to 0 (i.e., removing the j-th variable of X′); γ1, γ2, γ3 and γ4 represent different preset thresholds, which can be set according to actual usage requirements, and this disclosure does not make specific limitations on them.

[0085] The CRLR algorithm can be used to iteratively optimize and solve the above objective optimization problem, determine the optimization hyperparameters, and determine the final causal regression coefficients based on the accuracy. Sample weighting matrix and hyperparameters and

[0086] In addition to the methods mentioned above, other stable learning algorithms can also be used to construct and solve the corresponding objective optimization problem, and determine the causal regression coefficients corresponding to each relevant variable. This disclosure does not make any specific limitations on this.

[0087] The causal regression coefficient corresponding to any relevant variable can reflect the correlation between that relevant variable and the target fault.

[0088] In one example, the target system is a condenser, and the target fault is a condenser cooling water inlet temperature failure. After the fault occurs, to maintain a constant system temperature, closed-loop control is used to change the condenser cooling water flow rate. A dataset corresponding to the target fault is obtained, which includes 52 relevant variables (variable 1 to variable 52) and the data sequence of each variable within a preset time period. Based on the aforementioned process, the causal regression coefficient corresponding to each relevant variable in the dataset can be determined. In comparison, the squared prediction error (SPE) corresponding to each relevant variable in the dataset can also be determined using the principal component analysis method in the prior art.

[0089] Figure 5 A histogram showing the causal regression coefficients of a condenser cooling water inlet temperature fault according to an embodiment of the present disclosure is presented. Figure 6 This paper presents a histogram of the squared prediction error of a condenser cooling water inlet temperature fault in the related art.

[0090] like Figure 5 As shown, the histogram of causal regression coefficients reveals that the proportion of relevant variable 52 exceeds 50%, indicating a causal relationship between relevant variable 52 and the target fault. Based on prior knowledge of the target system, it can be determined that relevant variable 52 represents the condenser cooling water flow rate, which aligns with the actual situation where, after the aforementioned fault occurs, closed-loop control is used to change the condenser cooling water flow rate to maintain a constant system temperature. Figure 6As shown, based on the histogram of the squared prediction error determined by principal component analysis of the condenser, the most significant contributing variable to the target fault is identified as variable 38, while the contribution of variable 52 is relatively small. A comparison reveals that, compared to the existing method of determining the correlation between each relevant variable and the target fault through principal component analysis, this embodiment uses a stable learning algorithm to perform causal analysis on multiple relevant variables corresponding to the target fault, determining the causal regression coefficient for each relevant variable. This reveals the causal relationship between the relevant variables and the target fault, eliminates the influence of confounding factors, and more accurately represents the correlation between the relevant variables and the target fault, thereby improving the accuracy and efficiency of fault cause localization.

[0091] In one possible implementation, the dependent variable is determined from multiple related variables based on the causal regression coefficients corresponding to each related variable. This includes: sorting the causal regression coefficients corresponding to each related variable to determine a sorted coefficient sequence, wherein the coefficient sequence includes the absolute value of the causal regression coefficient corresponding to each related variable, and sorted in descending order of absolute value; sequentially accumulating the absolute values ​​of the causal regression coefficients corresponding to each related variable included in the coefficient sequence, and comparing the cumulative percentage determined after each accumulation with a preset threshold; when the cumulative percentage corresponding to the first k related variables in the coefficient sequence is less than or equal to the preset threshold, and the cumulative percentage corresponding to the first k+1 related variables in the coefficient sequence is greater than the preset threshold, the first k related variables in the coefficient sequence are determined as the dependent variable.

[0092] Specifically, the obtained causal regression coefficient sequence can be... The causal regression coefficients corresponding to each relevant variable are sorted in ascending order of absolute value to obtain the sorted coefficient sequence.

[0093] The absolute values ​​of the causal regression coefficients corresponding to each relevant variable in the coefficient sequence are summed up sequentially, and the corresponding cumulative percentage can be obtained after each summation.

[0094] In one example, the cumulative percentage can be expressed as formula (7):

[0095]

[0096] Where η represents the cumulative percentage; k represents the current number of accumulations.

[0097] Preset thresholds can be set according to actual usage needs. When the cumulative percentage of the first k related variables in the coefficient sequence is less than or equal to the preset threshold, and the cumulative percentage of the first k+1 related variables in the coefficient sequence is greater than the preset threshold, the first k related variables in the coefficient sequence are determined as the dependent variables.

[0098] In one example, the preset threshold is 30%, the cumulative percentage corresponding to the first two related variables in the coefficient sequence is 20%, and the cumulative percentage corresponding to the first three related variables in the coefficient sequence is 80%. At this time, it can be determined that the number of dependent variables is 2, and the first two related variables in the coefficient sequence are the dependent variables.

[0099] In particular, when the absolute value of the causal regression coefficient corresponding to the first relevant variable in the coefficient sequence is greater than a preset threshold, the first relevant variable in the coefficient sequence can be directly identified as the dependent variable.

[0100] In one example, the preset threshold is 30%, and the absolute value of the causal regression coefficient corresponding to the first relevant variable in the coefficient sequence is 50%. In this case, the first relevant variable in the coefficient sequence can be directly identified as the dependent variable.

[0101] In one possible implementation, based on a Bayesian network model, the intervention probability corresponding to each dependent variable is determined, including: determining a causal graph corresponding to the target fault based on the dependent variable and the label data sequence corresponding to the target system, wherein the label data sequence is used to represent the working state of the target system within a preset time period; discretizing each dependent variable based on the data sequence of each dependent variable within the preset time period to obtain a discretized variable corresponding to each dependent variable; constructing a Bayesian network model based on the causal graph and the discretized variables corresponding to each dependent variable; determining a conditional probability table corresponding to the target fault and a marginal probability corresponding to each node in the causal graph based on the Bayesian network model, wherein the conditional probability table is used to represent the probability of the target fault occurring when each dependent variable is abnormal, and the marginal probability is used to represent the probability of each dependent variable being abnormal; and determining the intervention probability corresponding to each dependent variable based on the conditional probability table and the marginal probability.

[0102] When a target system experiences a failure, the tag data sequence of the target system within a preset time period can be obtained. The tag data sequence can be used to represent the operating status of the target system within the preset time period. The specific form of the tag data sequence can be found in related technical implementations, and this disclosure does not impose specific limitations on it.

[0103] In one example, the preset time includes times T1 to T4. The target system is in a normal state at times T1 and T2, and in a target fault state at times T3 and T4. When the target system is in a normal state, the corresponding tag data can be determined to be 0; when the target system is in a target fault state, the corresponding tag data can be determined to be 1. Therefore, the tag data sequence of the target system from time T1 to time T4 can be determined as Y = [0, 0, 1, 1].

[0104] For specific methods of obtaining tag data sequences, please refer to the implementation methods in related technologies; this disclosure does not impose any specific limitations on them.

[0105] Based on the label data sequences corresponding to the dependent variables and the target system, the causal graph corresponding to the target fault can be determined. The specific method for determining the causal graph can refer to implementation methods in related technologies. For example, a matrix can be formed by combining the label data sequences corresponding to the k dependent variables and the target system, and the causal graph can be obtained by performing Bayesian network structure learning using the DAG-NOTEARS structure learning algorithm. This disclosure does not specifically limit the specific implementation of this method.

[0106] Based on the data sequence of each dependent variable within a preset time period and the mean of the corresponding normal samples, each dependent variable can be discretized to obtain the discretized variable corresponding to each dependent variable. The specific method of discretization can be found in related technical implementations, and this disclosure does not impose specific limitations on it.

[0107] In one example, the discretized variable can be represented by formula (8):

[0108]

[0109] Where "1" indicates that the target system has experienced a target fault, and "0" indicates that the target system is in a normal state; t represents the preset discretization threshold. Using formula (8), each dependent variable can be discretized into "0" and "1".

[0110] Based on the causal graph and the discretized variables corresponding to each dependent variable, a Bayesian network model containing the dependent variable can be constructed. Compared to directly constructing a Bayesian network model based on all relevant variables, this reduces the computational cost of model construction and improves its efficiency.

[0111] By using Bayesian estimation methods, network parameters can be learned from a Bayesian network model to determine the conditional probability table corresponding to the target fault and the marginal probability corresponding to each node in the causal graph. The conditional probability table represents the probability of the target fault occurring when each dependent variable is abnormal, while the marginal probabilities represent the probability of each dependent variable being abnormal. Based on the conditional probability table and the marginal probabilities, the intervention probability corresponding to each dependent variable can be determined.

[0112] The specific method for determining the intervention probability corresponding to each dependent variable based on the conditional probability table and marginal probabilities can refer to the implementation methods in related technologies, such as using a backdoor adjustment formula to calculate the intervention probability, etc. This disclosure does not make specific limitations in this regard.

[0113] Based on the intervention probability corresponding to each dependent variable, the cause of the target fault can be directly determined. Compared with the reasoning method of inferring the cause of the fault based on conditional probability in traditional fault tracing, it can eliminate the influence of confounding factors and mediating variables, improve the accuracy of fault tracing results, and avoid noise causing instability in fault tracing results.

[0114] In one example, the target system is a feeding system, and the target fault is a feed ratio change fault during the feeding process, which is a step fault. Using a stable learning algorithm, the dependent variables corresponding to the feed ratio change fault during the feeding process are extracted as correlation variable 1, correlation variable 17, correlation variable 31, correlation variable 42, and correlation variable 44. Using the above dependent variables and the label data sequence corresponding to the target system, a cause-effect graph corresponding to the feed ratio change fault can be obtained.

[0115] Figure 7 A cause-effect diagram is shown corresponding to a feed ratio variation fault according to an embodiment of the present disclosure. For example... Figure 7 As shown, the target fault does not point to dependent variable 17, therefore there is no true causal relationship between dependent variable 17 and the target fault. In subsequent fault tracing, the influence of dependent variable 17 on the target fault can be disregarded. Based on the aforementioned method, the intervention probabilities corresponding to dependent variables 1, 31, 42, and 44 can be determined respectively.

[0116] Referring to Table 1, for the fault of feed ratio change, the corresponding conditional probability is determined for each dependent variable based on the traditional Bayesian network fault tracing method, and the corresponding intervention probability is determined for each dependent variable based on the fault tracing method based on causal analysis of the present disclosure embodiment.

[0117] Table 1

[0118]

[0119] As shown in Table 1, the conditional probability of dependent variable 1 is 95.1%, and the conditional probability of dependent variable 44 is 95.2%. Since their conditional probabilities are very close, the cause of the fault cannot be directly determined based on the conditional probabilities. However, the intervention probability of dependent variable 1 is 10.0%, and the intervention probability of dependent variable 44 is 95.5%. Therefore, the cause of the fault can be directly determined to be dependent variable 44 based on the intervention probability.

[0120] Based on the known information of the target system, it can be determined that dependent variable 44 is an operational variable representing the feed rate; dependent variable 1 is a process variable representing the feed flow rate. Since operational variables affect process variables in actual production, dependent variable 44 is the cause of the target system failure. The fault tracing results determined using the causal analysis-based fault tracing method of this disclosure are consistent with the actual situation.

[0121] In one example, the target system is a feeding system, and the target fault is a feed temperature fault, which is a random fault. Using a stable learning algorithm, the dependent variables corresponding to the feed temperature fault are identified as correlation variables 7, 11, 13, and 18. Using these dependent variables and the label data sequence corresponding to the target system, a cause-effect graph corresponding to the feed temperature fault can be obtained.

[0122] Figure 8 A cause-effect diagram corresponding to a feed temperature failure according to an embodiment of the present disclosure is shown. For example... Figure 8 As shown, the target faults point to dependent variables 7, 11, 13, and 18. Based on the aforementioned method, the intervention probabilities corresponding to dependent variables 7, 11, 13, and 18 can be determined respectively.

[0123] Referring to Table 2, for the feed temperature fault, the corresponding conditional probability is determined for each dependent variable based on the traditional Bayesian network fault tracing method, and the corresponding intervention probability is determined for each dependent variable based on the causal analysis-based fault tracing method of this disclosure embodiment.

[0124] Table 2

[0125]

[0126] As shown in Table 2, the conditional probability of dependent variable 7 is 53.2%, that of dependent variable 13 is 52.3%, and that of dependent variable 18 is 54.7%, all of which are almost equal. Therefore, the cause of the fault cannot be directly determined based on the conditional probabilities. However, the intervention probability of dependent variable 7 is 2.2%, that of dependent variable 13 is 32.3%, and that of dependent variable 18 is 76.4%. Therefore, the cause of the fault can be directly determined to be dependent variable 18 based on the intervention probability.

[0127] Based on the known information of the target system, the dependent variable 18 can be determined to represent the stripping tower system temperature. Since raw materials first enter the stripping tower system during actual production, abnormal sampling values ​​of the stripping tower system temperature can lead to target failure; therefore, dependent variable 18 is the cause of the target failure. The fault tracing results determined using the causal analysis-based fault tracing method of this disclosure are consistent with the actual situation.

[0128] Therefore, for target systems that require fault tracing, regardless of whether the target fault is a step fault or a random fault, the fault tracing method based on causal analysis disclosed herein can effectively improve the accuracy of the fault tracing results.

[0129] In one example, the target system is a complex industrial system used for hot rolling processes, comprising multiple components and involving several different production steps. Specifically, the hot rolling process is the process of processing slabs into steel plates, and the entire process can be divided into steps such as heating, rough rolling, flying shearing, finish rolling, laminar cooling, and coiling, among which the finish rolling process is the core part of the entire process.

[0130] Referring to Table 3, two types of faults in the finishing rolling process are identified as target faults. Target fault 1 is a fault in the stand roll gap control actuator, and target fault 2 is a fault in the cooling water valve actuator between stands.

[0131] Table 3

[0132]

[0133] The relevant variables in the dataset to be processed corresponding to the target fault are the roll gap, rolling force, and bending roll force for each stand. The bending roll force is sampled starting from the second stand and includes data for both the target system in normal and target fault states.

[0134] Using the aforementioned method, the causal regression coefficients corresponding to each relevant variable of target fault 1, as well as the dependent variable corresponding to target fault 1, can be determined.

[0135] Figure 9 A histogram showing the causal regression coefficients corresponding to a fault in a frame roll gap control actuator according to an embodiment of the present disclosure is shown. Figure 9 As shown, the dependent variable corresponding to target fault 1 is only related variable 4. Based on prior knowledge of the target system, dependent variable 4 represents the roll gap of the machine frame. Therefore, it can be determined that when the machine frame roll gap control actuator fails, the sampled value of the roll gap of the machine frame becomes abnormal, which is consistent with the actual situation in the production process.

[0136] Using the aforementioned method, the causal regression coefficients corresponding to each relevant variable of target fault 2, as well as the dependent variable corresponding to target fault 2, can be determined.

[0137] Figure 10 A histogram showing the causal regression coefficients corresponding to a cooling water valve actuator failure between racks according to an embodiment of the present disclosure is presented. Figure 10 As shown, the dependent variables corresponding to target fault 2 are related variable 3 and related variable 10. Using the aforementioned method, and utilizing the dependent variables and the corresponding label data sequence of the target system, a cause-effect graph can be drawn between target fault 2 and dependent variables 3 and 10.

[0138] Figure 11 A cause-effect diagram is shown corresponding to a cooling water valve actuator failure between racks according to an embodiment of the present disclosure. For example... Figure 11As shown, target fault 2 points to dependent variable 3 and dependent variable 10, and both dependent variable 3 and dependent variable 10 have a causal relationship with target fault 2.

[0139] Referring to Table 4, for target fault 2, the corresponding conditional probability is determined for each dependent variable based on the traditional Bayesian network fault tracing method, and the corresponding intervention probability is determined for each dependent variable based on the causal analysis-based fault tracing method of this disclosure embodiment.

[0140] Table 4

[0141]

[0142] As shown in Table 4, the conditional probability of dependent variable 3 is 96.6%, and the conditional probability of dependent variable 10 is 99.8%. Since their conditional probabilities are almost equal, the cause of the fault cannot be directly determined based on the conditional probabilities. However, the intervention probability of dependent variable 3 is 50%, and the intervention probability of dependent variable 10 is 99.8%. Therefore, the cause of the fault can be directly determined to be dependent variable 10 based on the intervention probability.

[0143] Based on prior knowledge of the target system, dependent variable 10 represents the rolling force of the mill stand. In actual production, when a fault occurs, it first causes an abnormal change in the rolling force of the mill stand, which in turn affects the roll gap of the mill stand, consistent with actual conditions. Therefore, dependent variable 10 is the cause of the target fault. The fault tracing results determined using the fault tracing method based on causal analysis disclosed in this invention are consistent with reality.

[0144] Therefore, for complex systems with many components and production processes, the fault tracing method based on causal analysis disclosed herein can effectively improve the accuracy and efficiency of fault tracing results.

[0145] In this embodiment, a dataset to be processed is acquired. This dataset includes multiple related variables in the target system corresponding to the target fault when the target system experiences a target fault, and the data sequence of each related variable within a preset time period. Based on a stable learning algorithm, causal analysis is performed on the dataset to be processed to determine the dependent variable with a causal relationship to the target fault among the multiple related variables. This identifies key variables related to the target fault, reduces the number of candidate related variables, and improves the accuracy and efficiency of fault cause localization. Based on a Bayesian network model, the intervention probability corresponding to each dependent variable is determined. Based on the intervention probability corresponding to each dependent variable, the fault cause corresponding to the target fault can be directly determined. Compared to the inference method based on conditional probability in traditional fault tracing, this method eliminates the influence of confounding factors and mediating variables, improves the accuracy of fault tracing results, and avoids instability caused by noise in the fault tracing results.

[0146] It is understood that the various method embodiments mentioned above in this disclosure can be combined with each other to form combined embodiments without violating the principle and logic. Due to space limitations, this disclosure will not elaborate further. Those skilled in the art will understand that in the above methods of specific implementation, the specific execution order of each step should be determined by its function and possible internal logic.

[0147] In addition, this disclosure also provides a fault tracing device, electronic device, computer-readable storage medium, and program based on causal analysis. All of the above can be used to implement any of the fault tracing methods based on causal analysis provided in this disclosure. The corresponding technical solutions and descriptions are described in the corresponding records in the method section and will not be repeated here.

[0148] Figure 12 A block diagram of a fault tracing device based on causal analysis according to an embodiment of the present disclosure is shown. Figure 12 As shown, the device 1200 includes:

[0149] The data acquisition module 1201 acquires the dataset to be processed, which includes multiple related variables in the target system corresponding to the target fault when the target system experiences the target fault, as well as the data sequence of each related variable within a preset time period.

[0150] The dependent variable extraction module 1202 is used to perform causal analysis on the dataset to be processed, and to identify the dependent variable that has a causal relationship with the target fault among multiple related variables;

[0151] The intervention probability determination module 1203 is used to determine the intervention probability corresponding to each dependent variable based on a Bayesian network model. The intervention probability corresponding to each dependent variable is used to represent the probability value that the dependent variable is the cause of the fault.

[0152] The fault source determination module 1204 is used to determine the fault cause corresponding to the target fault based on the intervention probability corresponding to each dependent variable.

[0153] In one possible implementation, the data acquisition module 1201 is further configured to: acquire the original dataset corresponding to the target system when a target system fails; perform data standardization processing on the original dataset to determine the dataset to be processed.

[0154] In one possible implementation, the dependent variable extraction module 1202 is further configured to: group the dataset to be processed to determine the processing data group and the control data group; determine the causal regression coefficient corresponding to each relevant variable based on the processing data group and the control data group and a stable learning algorithm, wherein the causal regression coefficient corresponding to each relevant variable is used to represent the correlation between the relevant variable and the target fault; and determine the dependent variable among multiple relevant variables based on the causal regression coefficient corresponding to each relevant variable.

[0155] In one possible implementation, the dependent variable extraction module 1202 is further configured to: determine the index vector corresponding to each relevant variable based on the mean of the normal samples corresponding to each relevant variable, wherein the mean of the normal samples corresponding to each relevant variable is the average value of the relevant variable when the target system is in a normal state; and divide the dataset to be processed into a processing data group and a control data group based on the index vector corresponding to each relevant variable.

[0156] In one possible implementation, the dependent variable extraction module 1202 is further configured to: determine the target optimization problem based on the processed data set and the control data set, wherein the target optimization problem is used to extract the dependent variable from multiple related variables; and solve the target optimization problem using a stable learning algorithm to determine the causal regression coefficient corresponding to each related variable.

[0157] In one possible implementation, the dependent variable extraction module 1202 is further configured to: sort the causal regression coefficients corresponding to each relevant variable to determine the sorted coefficient sequence, wherein the coefficient sequence includes the absolute value of the causal regression coefficient corresponding to each relevant variable, and is sorted in descending order of absolute value; sequentially accumulate the absolute values ​​of the causal regression coefficients corresponding to each relevant variable included in the coefficient sequence, and compare the cumulative percentage determined after each accumulation with a preset threshold; when the cumulative percentage corresponding to the first k relevant variables in the coefficient sequence is less than or equal to the preset threshold, and the cumulative percentage corresponding to the first k+1 relevant variables in the coefficient sequence is greater than the preset threshold, the first k relevant variables in the coefficient sequence are determined as dependent variables.

[0158] In one possible implementation, the intervention probability determination module 1203 is further configured to: determine the causal graph corresponding to the target fault based on the dependent variable and the label data sequence corresponding to the target system, wherein the label data sequence represents the working state of the target system within a preset time period; discretize each dependent variable based on the data sequence of each dependent variable within the preset time period to obtain the discretized variable corresponding to each dependent variable; construct a Bayesian network model based on the causal graph and the discretized variable corresponding to each dependent variable; determine the conditional probability table corresponding to the target fault and the marginal probability corresponding to each node in the causal graph based on the Bayesian network model, wherein the conditional probability table represents the probability of the target fault occurring when each dependent variable is abnormal, and the marginal probability represents the probability of each dependent variable being abnormal; and determine the intervention probability corresponding to each dependent variable based on the conditional probability table and the marginal probability.

[0159] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0160] This disclosure also proposes a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the above-described method. The computer-readable storage medium can be volatile or non-volatile.

[0161] This disclosure also proposes an electronic device, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0162] This disclosure also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0163] Figure 13 A block diagram of an electronic device according to an embodiment of the present disclosure is shown. For example, electronic device 1900 may be provided as a server or terminal device. (Refer to...) Figure 13 The electronic device 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0164] Device 1900 may also include a power supply component 1926 configured to perform power management of device 1900, a wired or wireless network interface 1950 configured to connect device 1900 to a network, and an input / output (I / O) interface 1958. Device 1900 can operate on an operating system stored in memory 1932, such as Windows Server™, MacOS X™, Unix™, Linux™, FreeBSD™, or similar.

[0165] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of the device 1900 to perform the above-described method.

[0166] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0167] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0168] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0169] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0170] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0171] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0172] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0173] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0174] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A fault tracing method based on causal analysis, characterized in that, include: Obtain a dataset to be processed, wherein the dataset to be processed includes multiple related variables in the target system corresponding to the target failure when the target system experiences a target failure, and a data sequence of each related variable within a preset time period; A causal analysis is performed on the dataset to be processed to determine the dependent variable that has a causal relationship with the target fault among the multiple related variables; Based on the Bayesian network model, the intervention probability corresponding to each dependent variable is determined, wherein the intervention probability corresponding to each dependent variable is used to represent the probability value that the dependent variable is the cause of the fault. Based on the intervention probability corresponding to each dependent variable, determine the cause of the target fault; The step of determining the intervention probability corresponding to each dependent variable based on a Bayesian network model includes: Based on the dependent variable and the tag data sequence corresponding to the target system, a cause-effect graph corresponding to the target fault is determined, wherein the tag data sequence is used to represent the working state of the target system within the preset time period; Based on the data sequence of each dependent variable within a preset time period, each dependent variable is discretized to obtain the discretized variable corresponding to each dependent variable; Based on the causal graph and the discretized variables corresponding to each dependent variable, a Bayesian network model is constructed. Based on the Bayesian network model, a conditional probability table corresponding to the target fault and a marginal probability corresponding to each node in the causal graph are determined. The conditional probability table is used to represent the probability of the target fault occurring when each dependent variable is abnormal, and the marginal probability is used to represent the probability of each dependent variable being abnormal. Based on the conditional probability table and the marginal probabilities, the intervention probability corresponding to each dependent variable is determined.

2. The method according to claim 1, characterized in that, The step of performing causal analysis on the dataset to be processed, and determining the dependent variable from the plurality of relevant variables, includes: The dataset to be processed is grouped to determine the processing data group and the control data group; Based on the processed data set and the control data set, and using a stable learning algorithm, the causal regression coefficient corresponding to each of the relevant variables is determined, wherein the causal regression coefficient corresponding to each of the relevant variables is used to represent the correlation between the relevant variable and the target fault; The dependent variable is determined from the plurality of related variables based on the causal regression coefficients corresponding to each of the related variables.

3. The method according to claim 2, characterized in that, The step of grouping the dataset to be processed to determine the processing data group and the control data group includes: Based on the mean of normal samples corresponding to each of the relevant variables, an index vector corresponding to each of the relevant variables is determined, wherein the mean of normal samples corresponding to each of the relevant variables is the average value of the relevant variable when the target system is in a normal state; Based on the index vector corresponding to each of the relevant variables, the dataset to be processed is divided into the processed data group and the control data group.

4. The method according to claim 2 or 3, characterized in that, The step of determining the causal regression coefficient corresponding to each of the relevant variables based on the processed data set and the control data set, using a stable learning algorithm, includes: Based on the processed data set and the control data set, a target optimization problem is determined, wherein the target optimization problem is used to extract the dependent variable from the plurality of related variables; The stable learning algorithm is used to solve the objective optimization problem and determine the causal regression coefficients corresponding to each of the relevant variables.

5. The method according to claim 2 or 3, characterized in that, The step of determining the dependent variable from the plurality of related variables based on the causal regression coefficient corresponding to each of the related variables includes: The causal regression coefficients corresponding to each of the relevant variables are sorted to determine the sorted coefficient sequence, wherein the coefficient sequence includes the absolute value of the causal regression coefficients corresponding to each of the relevant variables, and the coefficients are sorted in descending order of absolute value; The absolute values ​​of the causal regression coefficients corresponding to each of the relevant variables included in the coefficient sequence are accumulated sequentially, and the cumulative percentage determined after each accumulation is compared with a preset threshold. When the cumulative percentage corresponding to the first k related variables in the coefficient sequence is less than or equal to the preset threshold, and the cumulative percentage corresponding to the first k+1 related variables in the coefficient sequence is greater than the preset threshold, the first k related variables in the coefficient sequence are determined as the dependent variable.

6. The method according to claim 1, characterized in that, The method further includes: When the target system experiences the target failure, the original dataset corresponding to the target system is obtained; The original dataset is subjected to data standardization to determine the dataset to be processed.

7. A fault tracing device based on causal analysis, characterized in that, include: The data acquisition module acquires a dataset to be processed, wherein the dataset to be processed includes multiple related variables in the target system corresponding to the target fault when the target system experiences a target fault, and a data sequence of each related variable within a preset time period; The dependent variable extraction module is used to perform causal analysis on the dataset to be processed and to determine the dependent variable that has a causal relationship with the target fault among the multiple related variables. The intervention probability determination module is used to determine the intervention probability corresponding to each dependent variable based on a Bayesian network model, wherein the intervention probability corresponding to each dependent variable is used to represent the probability value that the dependent variable is the cause of the fault. The fault source determination module is used to determine the fault cause corresponding to the target fault based on the intervention probability corresponding to each dependent variable. The intervention probability determination module is specifically used for: Based on the dependent variable and the tag data sequence corresponding to the target system, a cause-effect graph corresponding to the target fault is determined, wherein the tag data sequence is used to represent the working state of the target system within the preset time period; Based on the data sequence of each dependent variable within a preset time period, each dependent variable is discretized to obtain the discretized variable corresponding to each dependent variable; Based on the causal graph and the discretized variables corresponding to each dependent variable, a Bayesian network model is constructed. Based on the Bayesian network model, a conditional probability table corresponding to the target fault and a marginal probability corresponding to each node in the causal graph are determined. The conditional probability table is used to represent the probability of the target fault occurring when each dependent variable is abnormal, and the marginal probability is used to represent the probability of each dependent variable being abnormal. Based on the conditional probability table and the marginal probabilities, the intervention probability corresponding to each dependent variable is determined.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images