Causal discovery-oriented fault root cause diagnosis method for time series data and related device

Abstract

The application discloses a time series data fault root cause diagnosis method for causal discovery and a related device, relates to the technical field of fault identification, and comprises the following steps: acquiring satellite telemetry time series data to be processed; calculating the Hotelling T 2 statistic of each data sample by using a principal component analysis algorithm, and determining a fault sample by comparing the Hotelling T 2 statistic with a comparison result of a control limit of the Hotelling T statistic; for each fault sample, screening out a fault candidate variable corresponding to the fault sample by using a fault isolation method based on LASSO; applying a channel attention sparse causal network model to the fault candidate variables of each fault sample to obtain a causal matrix between the fault candidate variables; and determining a causal topology between the fault candidate variables according to the causal matrix to realize identification of a fault propagation path and positioning of a fault root cause. The application can effectively identify the propagation path of a correlated fault and accurately position the root cause.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault identification, in particular to a time series data fault root cause diagnosis method for causal discovery and a related device. BACKGROUND

[0002] During the operation of a satellite, material flow, information flow and energy flow interact with each other. Once a fault occurs in a system unit, due to the characteristics of fault propagation, it is easy to cause the abnormality or fault of other system units. Therefore, root cause diagnosis (RCD) is an important task. RCD methods mainly include traditional causal inference algorithm-based methods and deep learning-based methods. Traditional causal inference algorithms face great challenges when dealing with highly complex and large-scale systems. With the deep integration of industrialization and informationization, multiple sensors are installed on each component of the satellite in orbit through a distributed measurement system to collect and store operation data as telemetry data. Through analysis of the telemetry data, the operation state of the satellite system can be understood. In recent years, the development of deep learning technology has made deep learning-based methods more and more popular in the field of RCD. Through RCD of a faulty satellite system, the propagation path of the associated fault can be effectively identified and the root cause can be accurately located, thereby helping to reduce the risk of fault propagation.

[0003] RCD mainly includes two steps of fault isolation and fault root cause location. Common fault isolation methods include contribution plot method and reconstructed contribution plot method. When a fault is detected, the fault isolation method is used to determine the related variables that cause the fault. The contribution plot method is the most commonly used method, which calculates the contribution value of each variable of the sample to the monitoring statistic, and the variable with a large contribution value is considered more likely to cause the occurrence of the fault. However, when using the contribution plot method for fault location, due to the influence of the fault variable, the contribution value of the non-fault variable will become larger, which can easily cause the problem of residual pollution and misdiagnosis. Although the reconstructed contribution plot method improves the traditional contribution plot method, the failure of the commonly used diagnostic indicators (T 2 and SPE statistics) and the problem of residual pollution cannot be perfectly solved. In addition, the contribution plot method can locate the variables related to the fault, but it cannot identify the root cause variable of the fault and identify the propagation path, because the contribution plot method always considers the correlation between fault variables, but does not consider the causality between them.

[0004] The currently used causal analysis methods mainly include transfer entropy (TE), convergent cross mapping (CCM) and Granger causality (GC). Among them, the TE method has a large amount of calculation and cannot distinguish direct and indirect causal relationships, and the CCM is only suitable for weak to moderate coupling systems, and for strong coupling systems, it will face the problem of generalized synchronization. Although the GC method can be used to analyze the causal relationship between variable data, there are still some problems that have not been well solved: (1) The selection of variables participating in the GC method analysis completely depends on the subjective experience of engineers, and the uncertainty of the number of selections will lead to the uncertainty of the results; (2) The traditional GC method cannot analyze nonlinear time series data, and is prone to false and redundant causal relationships; (3) The traditional GC method is designed to infer the causal relationship between two variables, and is not suitable for satellite systems with multiple variables.

[0005] Therefore, there is an urgent need for a technology that can effectively identify the propagation path of the associated fault and accurately locate the root cause. SUMMARY

[0006] The purpose of the present application is to provide a time series data fault root cause diagnosis method and related device for causal discovery, which can effectively identify the propagation path of the associated fault and accurately locate the root cause.

[0007] To achieve the above-mentioned purpose, the present application provides the following solutions:

[0008] In a first aspect, the present application provides a time series data fault root cause diagnosis method for causal discovery, comprising:

[0009] Obtaining satellite telemetry time series data to be processed; the satellite telemetry data at a time point in the satellite telemetry time series data to be processed is a data sample; a data sample includes a plurality of satellite telemetry variables;

[0010] Calculating the Hotelling T 2 statistic of each data sample by using a principal component analysis algorithm, and determining the data sample regarded as a fault sample according to the comparison result of the Hotelling T 2 statistic and the Hotelling T 2 statistic control limit;

[0011] For each fault sample, a fault candidate variable corresponding to the fault sample is screened out by using a fault isolation method based on LASSO; the fault candidate variable is part of the plurality of satellite telemetry variables;

[0012] Applying a channel attention sparse causal network model to the fault candidate variables of each fault sample to obtain a causal matrix between the fault candidate variables.

[0013] A causal topology between the fault candidate variables is determined according to the causal matrix to realize identification of a fault propagation path and positioning of a fault root cause.

[0014] In a second aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the above-mentioned fault root cause diagnosis method for time series data based on causal discovery.

[0015] In a third aspect, the present application provides a computer readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the above-mentioned fault root cause diagnosis method for time series data based on causal discovery.

[0016] In a fourth aspect, the present application provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the above-mentioned fault root cause diagnosis method for time series data based on causal discovery.

[0017] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0018] The present application provides a fault root cause diagnosis method for time series data based on causal discovery, and related devices, satellite telemetry time series data to be processed is acquired; satellite telemetry data at one time point is a data sample; a Hotelling T 2 statistic of each data sample is calculated by using a principal component analysis algorithm, and a data sample regarded as a fault sample is determined according to a comparison result of the Hotelling T 2 statistic and a Hotelling T 2 control limit; for each fault sample, a fault candidate variable corresponding to the fault sample is screened out by using a fault isolation method based on LASSO; a causal matrix between the fault candidate variables is obtained by applying a channel attention sparse causal network model to the fault candidate variables of each fault sample; and a causal topology between the fault candidate variables is determined according to the causal matrix to realize identification of a fault propagation path and positioning of a fault root cause. In the present application, the fault candidate variable is screened out by using the fault isolation method based on LASSO, the problems of invalidation of a diagnostic index of a traditional contribution graph and residual pollution are solved, and a fault variable candidate set is determined, and in view of the defect that a traditional Granger causal analysis method cannot analyze a multivariable causal relationship and a nonlinear causal relationship, a causal relationship analysis model of a channel attention sparse causal network is proposed, the nonlinear causal relationship between the fault variables can be analyzed by using the proposed channel attention sparse causal network model, a fault propagation causal topology is determined, and a fault root cause is positioned. Therefore, the method of the present application can not only accurately perform fault detection, but also accurately position a fault root cause. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0020] Figure 1 An application environment diagram of a fault root cause diagnosis method for causal discovery-oriented time series data according to an embodiment of the present application;

[0021] Figure 2 A flowchart of a fault root cause diagnosis method for causal discovery-oriented time series data provided by an embodiment of the present application;

[0022] Figure 3 A principle diagram of a fault root cause diagnosis method for causal discovery-oriented time series data provided by an embodiment of the present application;

[0023] Figure 4 A structure diagram of a channel attention sparse causal network model provided by an embodiment of the present application;

[0024] Figure 5 A structure diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0026] In order to make the above purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0027] The fault root cause diagnosis method for causal discovery-oriented time series data provided by the embodiments of the present application can be applied in an application environment as shown in Figure 1 The terminal communicates with the server through the network. The data storage system can store the data required to be processed by the server. The data storage system can be separately arranged, integrated on the server, placed on the cloud or other servers. The terminal can send the satellite telemetry time series data to be processed to the server. After receiving the satellite telemetry time series data to be processed, the server obtains the Hotelling T2 statistical quantities, and comparison with Hotelling T 2 The comparison result of the control limits of the statistical quantities determines the fault samples; for each fault sample, a fault candidate variable corresponding to the fault sample is screened out by using a LASSO-based fault isolation method; a causal matrix between the fault candidate variables is obtained by applying a channel attention sparse causal network model to the fault candidate variables of each fault sample; and a causal topology between the fault candidate variables is determined according to the causal matrix to realize identification of a fault propagation path and positioning of a fault root cause. The server can feed back the obtained causal topology between the fault candidate variables to the terminal. In addition, in some embodiments, the time series data fault root cause diagnosis method for causal discovery can also be implemented by the server or the terminal alone, for example, the terminal can directly perform fault detection and root cause positioning processing on the satellite telemetry time series data to be processed, or the server can obtain the satellite telemetry time series data to be processed from the data storage system and perform fault detection and root cause positioning processing.

[0028] The terminal can be, but is not limited to, various desktop computers, notebook computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things device can be a smart speaker, a smart television, a smart air conditioner, a smart vehicle-mounted device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0029] In an exemplary embodiment, as shown in Figure 2 and Figure 3 A time series data fault root cause diagnosis method for causal discovery is provided, which is executed by a computer device, specifically, can be executed by a terminal or a server, or can be executed by a terminal and a server. In the embodiments of the present application, the method is applied to a server in Figure 1 , which includes the following steps 101 to 105. Wherein:

[0030] Step 101, obtaining satellite telemetry time series data to be processed.

[0031] The to-be-processed telemetry data is stored in a table form of M rows and N columns; a position of a first row and a first column of the table is a blank; the to-be-processed telemetry data includes N-1 telemetry variables and is stored in positions of the first row of the table except the blank; the to-be-processed telemetry data includes M-1 time points and is stored in positions of the first column of the table except the blank; the to-be-processed telemetry data includes (M-1)*(N-1) data points and is stored in positions of the table except the first column and the first row; a first row of a marker column of the table is used to store a marker telemetry variable, and other rows of the marker column are used to store data points of the marker telemetry variable collected at different time points; the marker column is any column of the table except the first column, and the marker telemetry variable is any telemetry variable. Therefore, in the to-be-processed satellite telemetry time series data, satellite telemetry data of one time point is one data sample; one data sample includes a plurality of satellite telemetry variables.

[0032] In step 102, a Hotelling T 2 statistic of each data sample is calculated by using a principal component analysis algorithm, and a data sample regarded as a fault sample is determined according to a comparison result of the Hotelling T 2 statistic and a control limit of the Hotelling T 2 statistic.

[0033] In step 103, for each fault sample, a fault candidate variable corresponding to the fault sample is screened out by using a fault isolation method based on LASSO; the fault candidate variable is part of the plurality of satellite telemetry variables.

[0034] In step 104, a causal matrix between the fault candidate variables of each fault sample is obtained by applying a channel attention sparse causal network model to the fault candidate variables.

[0035] In step 105, a causal topology between the fault candidate variables is determined according to the causal matrix to realize identification of a fault propagation path and positioning of a fault root cause.

[0036] After the causal matrix between all fault variables in the fault candidate variable set is obtained, the causal topology structure between the fault variables can be determined. Based on the causal topology structure, a fault propagation path can be generated. A variable with no in-degree node is regarded as a root cause variable causing the fault.

[0037] During the implementation of steps 101 to 105, first, principal component analysis is used to detect faults. If a fault is detected, the subsequent root cause location steps are performed. Secondly, the present invention proposes a fault isolation method based on LASSO to determine an appropriate number of fault-related variables for subsequent causal analysis. Finally, a deep learning-based causal analysis method, called a channel-attention sparse causal network, is proposed to locate the root cause of the fault. This method combines causal analysis with fault root cause location, addressing the problem that most current fault identification methods focus solely on fault detection. The present invention introduces a LASSO-based fault isolation method to overcome the diagnostic indicator failure and residual contamination issues of traditional contribution graphs. It also introduces a channel-attention sparse causal network model to overcome the shortcomings of traditional Granger causality analysis methods, such as their inability to analyze causal relationships between multiple variables and nonlinear causal relationships. Using this method, ground engineers can promptly locate the root cause of in-orbit satellite faults, effectively avoiding the risk of fault propagation. Furthermore, the present method combines GC with neural networks, is completely data-driven, and does not require domain expertise. In addition, the present invention detects faults in industrial equipment through telemetry time series data from on-orbit satellites, identifies the subsequent propagation paths of the detected faults, and locates the root causes of the faults.

[0038] In another exemplary embodiment of the present application, for the fault detection in step 102, when the telemetry data to be processed is analyzed using the principal component analysis algorithm, it is necessary to first use the principal component analysis method to train the telemetry data when the satellite is in normal operation, and the standard T 2 Control limits (i.e. Hotelling T 2 Statistical control limit); the standard T 2 The control limits are used to compare the T of the telemetry data to be processed. 2 Therefore, in step 102, the Hotelling T of each data sample is calculated using the principal component analysis algorithm. 2 statistic, and according to the Hotelling T 2 Statistic and Hotelling's T 2 The comparison results of the statistical control limits determine the data samples that are considered as fault samples, including:

[0039] (a1) Obtain telemetry data on the normal operation of the satellite.

[0040] (a2) The principal component analysis method is used to train the telemetry data of the normal operation of the satellite to obtain the Hotelling T 2 Statistical control limits.

[0041] (a3) Calculate the Hotelling T of each data sample using the principal component analysis algorithm2 statistic.

[0042] (a4) determining whether the Hotelling T 2 statistic is greater than the Hotelling T 2 statistic control limit.

[0043] (a5) if yes, the corresponding data sample is regarded as a fault sample; if no, it is a normal sample, and no subsequent fault root cause localization is needed.

[0044] In another exemplary embodiment of the present application, for step 103, when a fault occurs, a fault candidate variable set is screened according to a LASSO-based fault isolation method; specifically, when a traditional contribution plot is used for fault localization, T 2 and SPE are usually used as indexes, but in some special cases, these two indexes will fail. The special cases are:

[0045] (1) when the number of principal components in the PCA model of the original data is 1, the RCP method based on T 2 cannot effectively perform fault localization;

[0046] (2) when the number of principal components is less than the number of variables by 1, the RCP method based on SPE will also fail to diagnose.

[0047] Therefore, the present application converts the fault reconstruction problem into a constraint combination optimization problem in the same form as LASSO. Based on solving this optimization problem, variables that contribute greatly to fault reconstruction are calculated. Therefore, for step 103, for each fault sample, a fault candidate variable corresponding to the fault sample is screened by using a LASSO-based fault isolation method, specifically including:

[0048] (b1) for each fault sample, a constraint optimization model for fault sample reconstruction is constructed.

[0049] The fault sample is reconstructed to eliminate the influence of the fault on the fault sample, thereby restoring the fault sample to a normal sample. Assuming that there is a fault sample x, fault reconstruction is represented by formula (1). Where x * represents the reconstructed normal sample; Ξ is a vector representing the fault direction, whose elements are 0 or 1, used to represent variables affected by the detected fault; e represents the amplitude of the fault, i.e., the strength of the influence of the fault on the variables; ° represents the Hadamard product. After reconstruction, x * should be close to the normal sample, which means that for x * the T 2 statistic will be lower than the control limit.

[0050]

[0051] Since the failure of the satellite telemetry process has sparsity, the failure usually only affects part of the variables, and the number of non-zero elements in the failure vector f must be limited, which can be represented as a constraint optimization problem shown in equation (2). Where, ||·||1 represents the L1 norm, and c is the maximum number of failure variables specified in advance.

[0052]

[0053] (b2) Convert the constraint optimization model of the failure sample reconstruction into a constraint combination optimization model in the form of LASSO.

[0054]

[0055] Where P is the load matrix, which refers to the load matrix in the PCA algorithm, A is a diagonal matrix composed of the first k eigenvalues, the eigenvalues are the eigenvalues obtained by the PCA algorithm, L is A -1 The lower triangular matrix obtained after Cholesky decomposition, λ is the weight. That is, the constraint optimization model of solving (2) is converted into y=(PL) T x, Z=(PL) T , β=f LASSO problem, that is, a constraint combination optimization model in the form of LASSO:

[0056] (b3) Solve the constraint combination optimization model in the form of LASSO to obtain the fault candidate variables corresponding to each failure sample.

[0057] The above failure reconstruction problem is converted into a constraint combination optimization problem in the form of LASSO, and the failure vector f is obtained by solving the optimization problem. The non-zero elements in f are the screened failure variables.

[0058] The solution of LASSO is a sparse variable set, that is, a subset of all variables. The variables in the sparse variable set are considered to be variables that contribute more to the failure, and the fault candidate variable set is determined based on this principle.

[0059] In another exemplary embodiment of the present application, corresponding to step 104, based on the proposed channel attention sparse causal network model and the above-mentioned fault candidate variable set, a causal matrix between the fault candidate variables is obtained. First, based on the proposed channel attention sparse causal network model, the prediction of the satellite telemetry variables (the prediction of the fault candidate variables) is obtained, as shown in equation (3). Where X represents the fault candidate variables in the satellite telemetry data, CASCN(·) represents the channel sparse causal network model, represents the prediction of the fault candidate variables.

[0060]

[0061] Secondly, the sparse weight is obtained by performing sparse penalty on the weight of the channel sparse causal network model. The weight is a causal matrix obtained based on the model, and the non-zero elements of the causal matrix represent that there is a causal relationship between the corresponding fault variables. The sparse weight is obtained by solving the optimization problem of formula (4), which is a loss calculation of the model. Wherein, W represents the weight of the model; λ GL represents a LASSO sparse penalty coefficient; λ Ridge represents a ridge regression coefficient, which is used to prevent overfitting of the model; ||·||2 represents an L2 norm.

[0062]

[0063] Therefore, in step 104, the channel attention sparse causal network model is applied to the fault candidate variables of each fault sample to obtain a causal matrix between the fault candidate variables, specifically including:

[0064] (c1) taking the fault candidate variables in the first preset time period as the model input, and using the channel attention sparse causal network model to output the fault candidate variable prediction data in the second preset time period; the time point in the second preset time period is greater than the time point in the first preset time period.

[0065] (c2) calculating the loss of the model according to the fault candidate variables in the first preset time period, the fault candidate variable prediction data in the second preset time period, and the related regularization applied to the weight value of the model, that is, formula (4).

[0066] (c3) the weight value of the model when the loss of the model is determined to be the minimum is the causal matrix between the fault candidate variables.

[0067] In another exemplary embodiment of the present application, as Figure 4 shown, the channel attention sparse causal network model includes a plurality of channel attention sparse causal networks; the input of each channel attention sparse causal network is the fault candidate variables of each fault sample, and each channel attention sparse causal network outputs prediction data of a fault candidate variable; the number of channel attention sparse causal networks is equal to the number of fault candidate variables.

[0068] Each of the channel attention sparse causal networks includes an attention module and a multi-channel time domain convolution module connected in sequence.

[0069] The attention module includes a time series data input layer, an expansion dimension and reshaping layer, an input embedding layer and a Transformer encoder layer connected in sequence; the input of the time series data input layer is the fault candidate variables of each fault sample.

[0070] The multi-channel time domain convolution module comprises a linear head layer, a reshaping shape layer, a multi-channel time sequence convolution network layer and a single variable prediction output layer connected in sequence; the input of the linear head layer is the output of the Transformer encoder layer; and the output of the single variable prediction output layer is prediction data of a fault candidate variable.

[0071] The expansion dimension and the reshaping shape layer are pre-operation of an input embedding layer; the input embedding layer is an indispensable operation before self-attention of the Transformer, including position embedding and token embedding.

[0072] The linear head layer is usually composed of linear layers and is mainly used for aligning the output dimension in the application; the reshaping shape layer is used for reshaping the shape of the output to meet the input requirement of the multi-channel convolution network; the time sequence convolution network is a popular network, and the multi-channel time sequence convolution network is an extension thereof; and the single variable prediction output layer is a linear layer in the application and is used for processing the multi-dimensional output of the multi-channel time sequence convolution network into a single-dimensional output.

[0073] The application is a deep learning-based method, and a time sequence data fault root cause diagnosis method for causal discovery is proposed. Unlike traditional fault diagnosis methods, the proposed method focuses more on identifying the propagation path after the fault occurs and the root cause of the fault. First, a fault isolation method based on LASSO is studied to solve the problem of diagnostic index failure in the fault positioning process, which avoids the diagnostic failure problem. The method is combined with PCA to realize fault detection and selection of a fault variable candidate set, which is used to select a proper number of variables to participate in the construction of a fault propagation causal topology. Finally, a channel attention sparse causal network causal analysis method is proposed to establish a fault propagation causal topology and locate the fault root cause. Satellite ground management personnel can timely locate the root cause of the on-orbit satellite fault by using the method, so that relevant measures can be taken to repair the fault and inhibit the risk of fault propagation, thereby ensuring the smooth execution of space tasks.

[0074] The application also provides an application scenario of the time sequence data fault root cause diagnosis method for causal discovery. Specifically, the time sequence data fault root cause diagnosis method for causal discovery provided in the embodiment can be applied in a satellite system fault diagnosis scenario, which includes a data acquisition link, a data processing link and a data result display link. The data acquisition link is used to acquire satellite telemetry time sequence data to be processed. The data processing link is used to calculate the Hotelling T 2The statistical quantity is determined, and based on this, a fault sample is determined. For each fault sample, a fault isolation method based on LASSO is used to screen out fault candidate variables corresponding to the fault sample. The channel attention sparse causal network model is applied to the fault candidate variables of each fault sample to obtain a causal matrix between the fault candidate variables, and then a causal topology between the fault candidate variables is determined to realize identification of a fault propagation path and positioning of a fault root cause. In the data display link, the results of the fault propagation path and the fault root cause are displayed. The time series data fault root cause diagnosis method provided in the embodiment belongs to the data processing link.

[0075] In an exemplary embodiment, a computer device is provided, which can be a server or a terminal, and an internal structure diagram thereof can be as shown in Figure 5 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store fault root cause diagnosis data. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement a time series data fault root cause diagnosis method based on causal discovery.

[0076] Those skilled in the art can understand that Figure 5 the structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0077] In an exemplary embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the steps in the above method embodiments.

[0078] In an example embodiment, a computer program product is provided, including a computer program which, when executed by a processor, implements the steps of any of the above method embodiments.

[0079] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0080] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to memory, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0081] The database involved in the embodiments provided by the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided by the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0082] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, it should be understood that the application encompasses all possible combinations of the technical features unless such a combination is not technically possible.

[0083] The principles and implementation manners of the present application are described herein by using specific examples, and the above embodiments are only used to help understand the method of the present application and its core idea; meanwhile, according to the idea of the present application, the specific implementation manners and application scopes will be changed by those skilled in the art. In conclusion, the content of the present specification should not be understood as a limitation of the present application.

Claims

1. A root cause diagnosis method for time series data faults oriented to causal discovery, characterized by: The time series data fault root cause diagnosis method for causal discovery includes: Obtaining satellite telemetry time series data to be processed; in the satellite telemetry time series data to be processed, satellite telemetry data at a time point is a data sample; and a data sample includes a plurality of satellite telemetry variables; The Hotelling T of each data sample is calculated using the principal component analysis algorithm. 2 statistic, and according to the Hotelling T 2 Statistic and Hotelling's T 2 The comparison results of statistical control limits determine the data samples that are regarded as fault samples; For each fault sample, a fault isolation method based on LASSO is used to screen out the fault candidate variables corresponding to the fault sample; the fault candidate variables are some variables among the several satellite telemetry variables; Applying a channel attention sparse causal network model to the fault candidate variables of each fault sample to obtain a causal matrix between the fault candidate variables; Determining the causal topology between the fault candidate variables according to the causal matrix to identify the fault propagation path and locate the root cause of the fault; For each fault sample, the fault isolation method based on LASSO is used to screen out the fault candidate variables corresponding to the fault sample, including: For each fault sample, a constrained optimization model for fault sample reconstruction is constructed; Converting the constrained optimization model for fault sample reconstruction into a constrained combination optimization model with the same form as LASSO; Solve the constrained combination optimization model with the same form as LASSO to obtain the fault candidate variables corresponding to each fault sample; Among them, the constrained combination optimization model with the same form as LASSO is: where y = (PL) T x, Z = (PL) T , β = f; Where P is the load matrix; A is a diagonal matrix consisting of the first k eigenvalues, and L is A -1 The lower triangular matrix obtained after Cholesky decomposition; λ is the weight; x is the fault sample; f is the fault vector; The channel attention sparse causal network model is applied to the fault candidate variables of each fault sample to obtain a causal matrix between the fault candidate variables, specifically including: Taking the fault candidate variables of the first preset time period as the model input, the channel attention sparse causal network model is used to output the fault candidate variable prediction data of the second preset time period; the time point of the second preset time period is greater than the time point of the first preset time period; Calculating the loss of the model based on the fault candidate variables for the first preset time period, the fault candidate variable prediction data for the second preset time period, and the relevant regularization applied to the model weight values; When determining the model with the minimum loss, the model weight value is the causal matrix between the fault candidate variables; Among them, the calculation expression of the model weight value when the model loss is the smallest is: Where W represents the model weight; λ GL represents the LASSO sparse penalty coefficient; λ Ridge represents the ridge regression coefficient; X represents the fault candidate variable of the first preset time period; represents the fault candidate variable prediction data for the second preset time period; ||·||2 represents the L2 norm.

2. The root cause diagnosis method for time series data faults oriented to causal discovery according to claim 1 is characterized in that: The Hotelling T of each data sample is calculated using the principal component analysis algorithm. 2 statistic, and according to the Hotelling T 2 Statistic and Hotelling's T 2 The comparison results of the statistical control limits determine the data samples that are considered as fault samples, including: Obtaining telemetry data on the normal operation of satellites; The principal component analysis method is used to train the telemetry data of the normal operation of the satellite to obtain the Hotelling T 2 Statistical control limits; The Hotelling T of each data sample is calculated using the principal component analysis algorithm. 2 Statistics; Determine the Hotelling T 2 Is the statistic greater than the Hotelling T 2 Statistical control limits; If so, the corresponding data sample is regarded as a fault sample.

3. The time series data fault root cause diagnosis method for causal discovery according to claim 1 is characterized in that: The channel attention sparse causal network model includes multiple channel attention sparse causal networks; the input of each channel attention sparse causal network is the fault candidate variable of each fault sample, and each channel attention sparse causal network outputs prediction data of a fault candidate variable; Each of the channel attention sparse causal networks includes an attention module and a multi-channel time domain convolution module connected in sequence; The attention module includes a sequentially connected time series data input layer, a dimension expansion and reshaping layer, an input embedding layer, and a Transformer encoder layer; the input of the time series data input layer is the fault candidate variable of each fault sample; The multi-channel time domain convolution module includes a linear head layer, a reshaping layer, a multi-channel temporal convolution network layer and a single variable prediction output layer connected in sequence; the input of the linear head layer is the output of the Transformer encoder layer; the output of the single variable prediction output layer is the predicted data of a fault candidate variable.

4. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the time series data fault root cause diagnosis method for causal discovery according to any one of claims 1 to 3.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for diagnosing the root cause of a time series data fault oriented to causal discovery according to any one of claims 1 to 3 is implemented.

6. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for diagnosing the root cause of a time series data fault oriented to causal discovery according to any one of claims 1 to 3 is implemented.

Images