Fault determination method, device, equipment and computer storage medium

Abstract

The application discloses a kind of determination method, device, equipment and computer storage medium of failure. The method comprises: obtaining the running data of multiple devices in a preset period, constructing the first time lag matrix and multiple second time lag matrices of running data using sliding window, calculating the first Cauchy-Schwarz divergence of the first time lag matrix and the target second time lag matrix, the target second time lag matrix is any one in multiple second time lag matrices, if the first Cauchy-Schwarz divergence is greater than the first threshold value, determine that the running data corresponding to the target second time lag matrix is abnormal, determine the failure of multiple devices. Thus, by calculating the Cauchy-Schwarz divergence of the first time lag matrix and the second time lag matrix, according to the difference degree of the probability distribution between the normal state data and the test data, the failure problem is detected, the failure is identified earlier and more accurately, and the detection rate of minor failure is improved.

Description

Technical Field

[0001] The present application belongs to the technical field of fault diagnosis, and in particular relates to a fault determination method, apparatus, device, and computer storage medium. Background Art

[0002] With the growing demand for safe chemical process equipment and high-quality products, modern chemical industry is moving towards digitalization and intelligence, but it is also becoming more sophisticated and complex. Process monitoring is a key task in maintaining process safety. In the context of interconnected chemical industry, if a failure occurs at any point in the production process, it will spread and evolve through the interconnected system, causing economic losses such as equipment downtime and substandard product quality, or even safety accidents and casualties.

[0003] Intelligent process control in process industries, led by abnormality monitoring technology, has played an important role in ensuring the safe operation of complex industrial processes. In particular, data-driven fault detection and abnormality diagnosis technology based on multivariate statistical analysis methods has become one of the most active areas in process control research, with many achievements and widespread application.

[0004] However, the current detection effect of minor faults in chemical processes is poor, and the detection rate of minor faults is low. Summary of the Invention

[0005] The embodiments of the present application provide a method, apparatus, device and computer storage medium for determining faults, which can detect fault problems based on the difference in probability distribution between normal state data and test data, identify faults more accurately, and improve the detection rate of minor faults.

[0006] In a first aspect, an embodiment of the present application provides a method for determining a fault, the method comprising:

[0007] Obtain operating data of multiple devices within a preset time period;

[0008] constructing a first time-lag matrix and a plurality of second time-lag matrices of the operating data using a sliding window;

[0009] respectively calculating first Cauchy-Schwarz divergences of the first time-delay matrix and the target second time-delay matrix, where the first Cauchy-Schwarz divergences represent the difference between the probability distribution function of the first time-delay matrix and the probability distribution function of the target second time-delay matrix, and the target second time-delay matrix is ​​any one of the plurality of second time-delay matrices;

[0010] When the first Cauchy-Schwarz divergence is greater than the first threshold, it is determined that the operating data corresponding to the target second time-delay matrix is ​​abnormal, and multiple device failures are determined.

[0011] In one possible embodiment, respectively calculating the first Cauchy-Schwarz divergence of the first time-delay matrix and the target second time-delay matrix includes:

[0012] determining a probability distribution function of a first time-delay matrix and a probability distribution function of a target second time-delay matrix;

[0013] Determine whether the first Cauchy-Schwarz divergence of the first time-delay matrix and the target second time-delay matrix satisfies the following conditions:

[0014]

[0015] Among them, X test,k represents the target second time-delay matrix, X test,ref represents the first time-delay matrix, p(X test,k ) represents the probability distribution function of the target second time-delay matrix, p(X test,ref ) represents the probability distribution function of the first time-lag matrix.

[0016] In one possible embodiment, before obtaining the operating data of the plurality of devices within a preset time period, the method further includes:

[0017] Obtain historical operating data of multiple devices within a preset time period and use the historical operating data as a training set;

[0018] Using a sliding window to construct a third time-lag matrix and multiple fourth time-lag matrices of historical operating data in a training set;

[0019] respectively calculating the second Cauchy-Schwarz divergence of the third time-delay matrix and the target fourth time-delay matrix, where the second Cauchy-Schwarz divergence represents the difference between the probability distribution function of the third time-delay matrix and the probability distribution function of the target fourth time-delay matrix, and the target fourth time-delay matrix is ​​any one of the plurality of fourth time-delay matrices;

[0020] Based on a predetermined confidence level, the control limit corresponding to the second Cauchy-Schwarz divergence is calculated using the kernel density function, and the control limit is used as the first threshold;

[0021] The second Cauchy-Schwarz divergence of the third time-delay matrix and the target fourth time-delay matrix is ​​calculated respectively, and the following conditions are satisfied:

[0022]

[0023] Among them, p(X train,k ) represents the probability distribution function of the target fourth time-delay matrix, p(X train,ref ) represents the probability distribution function of the third time-delay matrix.

[0024] In one possible embodiment, the present invention further includes:

[0025] Converting the operating data of multiple devices in the target second time-lag matrix into a first variable corresponding to each device, where the first variable includes the operating data of the devices in the target second time-lag matrix;

[0026] Calculate the first conditional Cauchy-Schwarz divergence of the first variable under two adjacent second time-lag matrices. The first conditional Cauchy-Schwarz divergence represents the difference in the conditional probability distribution function of the first variable under two adjacent second time-lag matrices.

[0027] Normalizing the first conditional Cauchy-Schwarz divergence using a predetermined mean and a predetermined standard deviation to obtain a standardized first conditional Cauchy-Schwarz divergence;

[0028] Normalize the first conditional Cauchy-Schwarz divergence after normalization to obtain the target conditional Cauchy-Schwarz divergence;

[0029] Determine the first variable corresponding to the maximum objective condition Cauchy-Schwarz divergence as the fault isolation variable;

[0030] Determine the fault device information corresponding to the fault isolation variable;

[0031] The calculation of the first conditional Cauchy-Schwarz divergence of the first variable under two adjacent second time-lag matrices satisfies the following conditions:

[0032]

[0033] in, Represents the first variable at the k-1th moment The conditional probability distribution function under the second time lag matrix at the k-1th moment, Represents the first variable at the kth moment The conditional probability distribution function under the condition of the second time delay matrix at the kth moment, l represents the lth row of the time delay matrix, represents the lth row vector of the second time-delay matrix, represents the first variable except the l-th row vector in the second time-lag matrix;

[0034] The first conditional Cauchy-Schwarz divergence is standardized using a predetermined mean and a predetermined standard deviation, and the standardized first conditional Cauchy-Schwarz divergence satisfies the following conditions:

[0035]

[0036] in, represents the predetermined mean, represents the predetermined standard deviation;

[0037] The first conditional Cauchy-Schwarz divergence after normalization is normalized to obtain the target conditional Cauchy-Schwarz divergence that satisfies the following conditions:

[0038]

[0039] Where m represents the mth column of the delay matrix.

[0040] In one possible embodiment, before normalizing the first conditional Cauchy-Schwarz divergence using a predetermined mean and a predetermined standard deviation to obtain the normalized first conditional Cauchy-Schwarz divergence, the method further includes:

[0041] converting the operating data of the plurality of devices in the target fourth time-lag matrix into a second variable corresponding to each device, wherein the second variable includes the operating data of the devices in the target fourth time-lag matrix;

[0042] Calculate the second conditional Cauchy-Schwarz divergence of the second variable under two adjacent fourth time-lag matrices. The second Cauchy-Schwarz divergence represents the difference in the conditional probability distribution function of the second variable under two adjacent fourth time-lag matrices.

[0043] Calculate the mean and standard deviation of the second conditional Cauchy-Schwarz divergence, and use the mean as the predetermined mean and the standard deviation as the predetermined standard deviation;

[0044] The second conditional Cauchy-Schwarz divergence of the second variable under two adjacent fourth lag matrices satisfies the following conditions:

[0045]

[0046] in, Represents the second variable at the k-1th moment The conditional probability distribution function under the fourth time lag matrix at the k-1th moment is: Represents the second variable at the kth moment The conditional probability distribution function under the condition of the fourth time delay matrix at the kth time, l represents the lth row of the time delay matrix, represents the lth row vector of the fourth time-delay matrix, Represents the first variable including the l-th row vector in the second time-lag matrix.

[0047] In one possible embodiment, the present invention further includes:

[0048] Calculate the third conditional Cauchy-Schwarz divergence of the first variable under the second time lag matrix at every first target time, where the third conditional Cauchy-Schwarz divergence represents the difference between the conditional probability distribution functions of the first variable under the second time lag matrix at every first target time;

[0049] Calculate the average of the third conditional Cauchy-Schwarz divergence during the second target time period, where the second target time period is longer than the first target time period;

[0050] When the target third condition Cauchy-Schwarz divergence is greater than the average value, it is determined that a process abnormality occurs in the equipment corresponding to the first variable.

[0051] In one possible embodiment, calculating the third conditional Cauchy-Schwarz divergence of the first variable under the second time lag matrix at every first target time length includes:

[0052] The third conditional Cauchy-Schwarz divergence of the first variable under the second time lag matrix is ​​calculated every first target time, and the following conditions are met:

[0053]

[0054] in, Represents the first variable at the k-1th moment The conditional probability distribution function under the second time lag matrix at the k-1th moment, represents the k+wth c Second variable at time -1 At k+w c The conditional probability distribution function under the second time lag matrix condition at time -1, w c Indicates the first target duration, Represents the first variable including the l-th row vector in the second time-lag matrix.

[0055] In a second aspect, an embodiment of the present application provides a fault determination device, including:

[0056] An acquisition module is used to obtain the operating data of multiple devices within a preset time period;

[0057] A construction module, configured to construct a first time-lag matrix and a plurality of second time-lag matrices of the operating data using a sliding window;

[0058] a calculation module, configured to respectively calculate a first Cauchy-Schwarz divergence of the first time-delay matrix and a target second time-delay matrix, wherein the first Cauchy-Schwarz divergence represents a difference between a probability distribution function of the first time-delay matrix and a probability distribution function of the target second time-delay matrix, and the target second time-delay matrix is ​​any one of a plurality of second time-delay matrices;

[0059] The determination module is configured to determine that the operating data corresponding to the target second time-delay matrix is ​​abnormal and that multiple equipment failures are detected when the first Cauchy-Schwarz divergence is greater than a first threshold.

[0060] In a third aspect, an embodiment of the present application provides an electronic device, the device comprising:

[0061] a processor and a memory storing computer program instructions;

[0062] When the processor executes the computer program instructions, any one of the above methods for determining a fault is implemented.

[0063] In a fourth aspect, an embodiment of the present application provides a computer storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method for determining any one of the above-mentioned faults is implemented.

[0064] In a fifth aspect, an embodiment of the present application provides a computer program product, characterized in that when the instructions in the computer program product are executed by a processor of an electronic device, the electronic device is enabled to execute any one of the above-mentioned fault determination methods.

[0065] The fault determination method, apparatus, device, and computer storage medium of the embodiments of the present application include: obtaining operating data of multiple devices within a preset time period, constructing a first time-delay matrix and multiple second time-delay matrices of the operating data using a sliding window, and calculating first Cauchy-Schwarz divergences of the first time-delay matrix and a target second time-delay matrix, respectively. The first Cauchy-Schwarz divergence represents the difference between the probability distribution function of the first time-delay matrix and the probability distribution function of the target second time-delay matrix, where the target second time-delay matrix is ​​any one of the multiple second time-delay matrices. When the first Cauchy-Schwarz divergence is greater than a first threshold, determining that the operating data corresponding to the target second time-delay matrix is ​​abnormal, and determining multiple device faults. In this way, by calculating the Cauchy-Schwarz divergence of the first time-delay matrix and the second time-delay matrix, that is, based on the difference in probability distribution between normal state data and test data, fault problems are detected, faults are more accurately identified, and the detection rate of minor faults is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0067] Figure 1 This is a flowchart of a method for determining a fault provided by an embodiment of the present application;

[0068] Figure 2 is the time delay matrix X involved in another embodiment of the present application k Schematic diagram of the structure;

[0069] Figure 3 is a flowchart of a method for determining a fault provided by another embodiment of the present application;

[0070] Figure 4 is the time delay matrix X involved in another embodiment of the present application k Schematic diagram of rewriting variable index;

[0071] Figure 5 is a schematic diagram of a fault determination result of a numerical simulation process according to yet another embodiment of the present application;

[0072] Figure 6 is a schematic diagram of a result of determining a fault isolation variable based on conditional Cauchy-Schwarz divergence in a numerical simulation process according to yet another embodiment of the present application;

[0073] Figure 7 is a schematic diagram of a numerical simulation process according to another embodiment of the present application based on the conditional Cauchy-Schwarz divergence change point detection result;

[0074] Figure 8 Schematic diagram of a CSTR reaction process according to another embodiment of the present application;

[0075] Figure 9 is a schematic diagram of a result of determining a fault isolation variable based on conditional Cauchy-Schwarz divergence in a CSTR reaction process according to yet another embodiment of the present application;

[0076] Figure 10 Schematic diagram of a CSTR reaction process according to another embodiment of the present application based on the conditional Cauchy-Schwarz divergence change point detection result;

[0077] Figure 11 Schematic diagram of the CSTH reaction process involved in yet another embodiment of the present application;

[0078] Figure 12This is a schematic diagram of the changes of various variables over time in normal and fault conditions during a CSTH reaction process according to another embodiment of the present application;

[0079] Figure 13 FIG1 is a schematic diagram of fault detection performance under different sliding window lengths w according to another embodiment of the present application;

[0080] Figure 14 1 is a schematic diagram of change point detection results of a CSTH reaction process in two fault conditions, f1 and f2, according to another embodiment of the present application;

[0081] Figure 15 is a schematic diagram of a result of determining a fault isolation variable based on conditional Cauchy-Schwarz divergence in a CSTH reaction process according to yet another embodiment of the present application;

[0082] Figure 16 1 is a schematic diagram of a CSTH reaction process according to another embodiment of the present application, based on a conditional Cauchy-Schwarz divergence change point detection result under fault f1;

[0083] Figure 17 1 is a schematic diagram of a CSTH reaction process according to another embodiment of the present application, based on a conditional Cauchy-Schwarz divergence change point detection result under fault f2;

[0084] Figure 18 is a structural diagram of a fault determination device provided in yet another embodiment of the present application;

[0085] Figure 19 This is a structural diagram of an electronic device provided in yet another embodiment of the present application. DETAILED DESCRIPTION

[0086] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present application by illustrating the examples of the present application.

[0087] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, the elements defined by the phrase "comprising..." do not exclude the presence of other identical elements in the process, method, article, or device comprising the elements.

[0088] With the growing demand for safe chemical process equipment and high-quality products, modern chemical industry is moving towards digitalization and intelligence, but it is also becoming more sophisticated and complex. Process monitoring is a key task in maintaining process safety. In the context of interconnected chemical industry, if a failure occurs at any point in the production process, it will spread and evolve through the interconnected system, causing economic losses such as equipment downtime and substandard product quality, or even safety accidents and casualties.

[0089] Intelligent process control in process industries, led by abnormality monitoring technology, has played an important role in ensuring the safe operation of complex industrial processes. In particular, data-driven fault detection and abnormality diagnosis technology based on multivariate statistical analysis methods has become one of the most active areas in process control research, with many achievements and widespread application.

[0090] However, the current detection effect of minor faults in chemical processes is poor, and the detection rate of minor faults is low.

[0091] In order to solve the problems in the prior art, the embodiments of the present application provide a method, apparatus, device, and computer storage medium for determining a fault. The following first introduces the method for determining a fault provided by the embodiments of the present application.

[0092] Figure 1 A flow chart of a method for determining a fault provided in one embodiment of the present application is shown.

[0093] like Figure 1 As shown, the fault determination method provided in the embodiment of the present application includes S110 to S140.

[0094] S110: Obtain operating data of multiple devices within a preset time period.

[0095] Here, the operation data is online operation data in the chemical process. In some embodiments, the online operation data in the chemical process can be collected by a data acquisition device, and the online operation data in the chemical process can be collected in real time by the data acquisition device.

[0096] In some embodiments, a test set is constructed using the operating data of multiple devices within a preset time period. The running data is standardized using the mean and standard deviation of the running data in the test set to obtain the standardized running data. The standardized calculation formula (1) is as follows:

[0097]

[0098] in, It is the operating data of multiple devices within a preset time period. is the mean of the running data in the test set, is the standard deviation of the running data in the test set, x test,i The normalized operating data.

[0099] S120: Construct a first time-lag matrix and multiple second time-lag matrices of the operating data using a sliding window.

[0100] In some embodiments, a first time-lag matrix X of the operating data of the previous w moments is constructed using a sliding window of length w. test,w , and use the sliding window of length w to construct multiple second time delay matrices X test,k .

[0101] Among them, X test,w Expressed as:

[0102]

[0103] X test,k Expressed as:

[0104]

[0105] In some embodiments, for a chemical process under normal operation, the monitoring data stream contains dynamic operating rules that change over time. If the current state of the system deviates from the normal state and exceeds the acceptable range, the current operation process has a fault. Since the probability density function of the time series can better reflect the actual state of the operation process, in order to obtain a real-time dynamic estimate of the system operation, a sliding window method is used to capture the dynamic evolution of the chemical data stream. The length of the sliding window is selected as w, and at the monitoring time k, it is determined by x k and its previous w-1 running data {x test,k-w+1 ,…,x test,k-2 ,x test,k-1}Construct the delay matrix Xk , where x k =[x k,1 ,x k,2 ,…,x k,m ] T is the kth sample, i.e. the running data at time k, and the time delay matrix X k The structure is as Figure 2 shown.

[0106] S130. Calculate the first Cauchy-Schwarz divergence of the first time-delay matrix and the target second time-delay matrix respectively, where the first Cauchy-Schwarz divergence represents the difference between the probability distribution function of the first time-delay matrix and the probability distribution function of the target second time-delay matrix, where the target second time-delay matrix is ​​any one of the multiple second time-delay matrices.

[0107] Here, the Cauchy-Schwarz divergence is used to characterize the difference between two probability distribution functions.

[0108] In some embodiments, the first time delay matrix X before w time points is selected test,w (a time-delay matrix) is the normal reference state X test,ref The second delay matrix has n test -w. Calculate the normal reference state X respectively test,ref and the second time-delay matrix X test,k The first Cauchy-Schwarz divergence of , where k∈[w,n test It should be noted that the first Cauchy-Schwarz divergence characterizes the difference between the probability distribution function of the first time-delay matrix and the probability distribution function of the target second time-delay matrix, that is, the difference between the probability distribution function of the time-delay matrix at each moment and the time-delay matrix under the normal reference state is calculated.

[0109] S140 : When the first Cauchy-Schwarz divergence is greater than a first threshold, determining that the operating data corresponding to the target second time-delay matrix is ​​abnormal, and determining that multiple equipment failures are present.

[0110] In some embodiments, when the first Cauchy-Schwarz divergence is greater than a first threshold, that is, when the difference between the probability distribution function of the first time-lag matrix and the probability distribution function of the target second time-lag matrix is ​​greater than the first threshold, the operating data corresponding to the target second time-lag matrix is ​​determined to be abnormal, that is, multiple equipment failures are determined, and the faulty equipment information is output.

[0111] In some embodiments, in order to quantify the deviation of the operating state, the online monitoring process uses a sliding window to determine the statistical difference between the normal state and the abnormal state.test,ref =[x test,1 ,…,x test,w-1 ,x test,w ], then the time delay matrix X at the kth moment test,k Compared with the normal reference state X test,ref The consistency between them can be analyzed by the following hypothesis test (Formula (2)):

[0112]

[0113] Based on this hypothesis test, the stable normal operation state is that the change of the probability density function value of the expected operation data is below the first threshold.

[0114] In this way, by calculating the Cauchy-Schwarz divergence of the first time-delay matrix and the second time-delay matrix, that is, according to the difference in probability distribution between normal state data and test data, the fault problem is detected, the fault is identified more accurately, and the detection rate of minor faults is improved.

[0115] Based on this, in some embodiments, the above S130 may specifically include:

[0116] determining a probability distribution function of a first time-delay matrix and a probability distribution function of a target second time-delay matrix;

[0117] The first Cauchy-Schwarz divergence of the first time-delay matrix and the target second time-delay matrix is ​​calculated using formula (3), which is as follows:

[0118]

[0119] Among them, X test,k represents the target second time-delay matrix, X test,ref represents the first time-delay matrix, p(X test,k ) represents the probability distribution function of the target second time-delay matrix, p(X test,ref ) represents the probability distribution function of the first time-lag matrix.

[0120] In some embodiments, for process industries under normal operating conditions, the monitoring data should maintain the same distribution. Then the fault detection problem becomes analyzing whether the current sample and the normal reference state belong to the same distribution, that is, the two probability distributions X test,ref and X test,k The difference between CS (p(X test,k ); p(X test,ref )).

[0121] In this way, the formula can be used to accurately calculate the difference between the currently analyzed sample and the normal reference state.

[0122] In some embodiments, as Figure 3 As shown, before the above S110, the method further includes S210 to S240.

[0123] S210: Obtain historical operation data of multiple devices within a preset time period, and use the historical operation data as a training set.

[0124] Here, the historical operation data is the original operation data (offline operation data) in the chemical process.

[0125] In some embodiments, an input of historical operating data of multiple devices within a preset time period may be received. The historical operating data of multiple devices within the preset time period may be operating data stored in a database.

[0126] In some embodiments, a training set is constructed using historical operating data of multiple devices within a preset time period. Use the mean and standard deviation of the historical operation data in the training set to standardize the historical operation data to obtain the standardized training set Among them, the standardized calculation formula (4) is as follows:

[0127]

[0128] in, It is the historical operation data of multiple devices within a preset time period. is the mean of the historical running data in the training set, is the standard deviation of the running data in the test set, x train,i The normalized operating data.

[0129] S220 , constructing a third time-lag matrix and multiple fourth time-lag matrices for the historical operation data in the training set using a sliding window.

[0130] In some embodiments, a sliding window of length w is used to construct a third time lag matrix X of the historical operation data of the previous w moments. train,w , and use the sliding window of length w to construct multiple fourth time-delay matrices X train,k .

[0131] Among them, X train,w Expressed as:

[0132]

[0133] X train,k Expressed as:

[0134]

[0135] In some embodiments, for a chemical process under normal operation, the monitoring data stream contains dynamic operating rules that change over time. If the current state of the system deviates from the normal state and exceeds the acceptable range, the current operation process has a fault. Since the probability density function of the time series can better reflect the actual state of the operation process, in order to obtain a real-time dynamic estimate of the system operation, a sliding window method is used to capture the dynamic evolution of the chemical data stream. The length of the sliding window is selected as w, and at the monitoring time k, it is determined by x k and its previous w-1 running data {x train,k-w+1 ,…,x train,k-2 ,x train,k-1}Construct the delay matrix X k , where x k =[x k,1 ,x k,2 ,…,x k,m ] is the kth sample, i.e. the running data at time k, and the time delay matrix X k The structure is as Figure 2 shown.

[0136] S230. Calculate the second Cauchy-Schwarz divergence of the third time-delay matrix and the target fourth time-delay matrix respectively, where the second Cauchy-Schwarz divergence represents the difference between the probability distribution function of the third time-delay matrix and the probability distribution function of the target fourth time-delay matrix, where the target fourth time-delay matrix is ​​any one of the multiple fourth time-delay matrices.

[0137] Here, the Cauchy-Schwarz divergence is used to characterize the difference between two probability distribution functions.

[0138] In some embodiments, the third time delay matrix X before w moments is selected train,w (a time-delay matrix) is the normal reference state X train,ref The fourth time delay matrix has n train -w. Calculate the normal reference state X respectively train,ref and the fourth time-delay matrix X train,k The second Cauchy-Schwarz divergence of , where k∈[w,n train It should be noted that the second Cauchy-Schwarz divergence characterizes the difference between the probability distribution function of the third time-delay matrix and the probability distribution function of the target fourth time-delay matrix, that is, the difference between the probability distribution function of the time-delay matrix at each moment and the time-delay matrix under the normal reference state.

[0139] S240 . Based on a predetermined confidence level, calculate a control limit corresponding to the second Cauchy-Schwarz divergence using a kernel density function, and use the control limit as a first threshold.

[0140] When using D CS (p(X train,k ); p(X train,ref When monitoring the process, the normal state operation data stream in the offline phase is used to obtain the control threshold, and the control limit D under the predetermined confidence level η can be estimated by kernel density estimation. cl .

[0141] In some embodiments, at the first Cauchy-Schwarz divergence D CS (p(X test,k ); p(X test,ref )) is greater than the control limit D cl In this case, the operating data corresponding to the target second time-delay matrix is ​​determined to be abnormal, and multiple equipment failures are determined.

[0142] The second Cauchy-Schwarz divergence of the third time-delay matrix and the target fourth time-delay matrix is ​​calculated using formula (5), which is as follows:

[0143]

[0144] Among them, p(X train,k ) represents the probability distribution function of the target fourth time-delay matrix, p(X train,ref ) represents the probability distribution function of the third time-delay matrix.

[0145] In this way, the control limits are calculated with reference to historical data, and the control limits are used to alarm the fault and determine the equipment failure.

[0146] Based on this, in some embodiments, the method may further include:

[0147] Converting the operating data of multiple devices in the target second time-lag matrix into a first variable corresponding to each device, where the first variable includes the operating data of the devices in the target second time-lag matrix;

[0148] Calculate the first conditional Cauchy-Schwarz divergence of the first variable under two adjacent second time-lag matrices. The first conditional Cauchy-Schwarz divergence represents the difference in the conditional probability distribution function of the first variable under two adjacent second time-lag matrices.

[0149] Normalizing the first conditional Cauchy-Schwarz divergence using a predetermined mean and a predetermined standard deviation to obtain a standardized first conditional Cauchy-Schwarz divergence;

[0150] Normalize the first conditional Cauchy-Schwarz divergence after normalization to obtain the target conditional Cauchy-Schwarz divergence;

[0151] Determine the first variable corresponding to the maximum objective condition Cauchy-Schwarz divergence as the fault isolation variable;

[0152] Determine the fault device information corresponding to the fault isolation variable;

[0153] The first conditional Cauchy-Schwarz divergence of the first variable under two adjacent second time-lag matrices is calculated using formula (6), which is as follows:

[0154]

[0155] in, Represents the first variable at the k-1th moment The conditional probability distribution function under the second time lag matrix at the k-1th moment, Represents the first variable at the kth moment The conditional probability distribution function under the condition of the second time delay matrix at the kth moment, l represents the lth row of the time delay matrix, represents the lth row vector of the second time-delay matrix, represents the first variable except the l-th row vector in the second time-lag matrix;

[0156] The first conditional Cauchy-Schwarz divergence is standardized using a predetermined mean and a predetermined standard deviation to obtain a standardized first conditional Cauchy-Schwarz divergence, which is then standardized using formula (7). Formula (7) is as follows:

[0157]

[0158] in, represents the predetermined mean, represents the predetermined standard deviation;

[0159] The normalized first conditional Cauchy-Schwarz divergence is normalized to obtain the target conditional Cauchy-Schwarz divergence, which is normalized using formula (8). Formula (8) is as follows:

[0160]

[0161] Where m represents the mth column of the delay matrix, is the sum of the first conditional Cauchy-Schwarz divergences.

[0162] In some embodiments, when a device failure is detected, it is crucial to determine which device caused the failure for process control. Fault isolation refers to identifying specific device variables that cause state changes or contribute the most to the failure. During normal operation, each device should be in a stable state; after a failure occurs, due to the complex physical or information connections in the interconnected chemical industry, the fault inducement will evolve and affect other fault-related variables beyond the acceptable range, resulting in conditional correlations between variables that are different from the conditional correlations under normal conditions. Therefore, it is feasible to reveal the causal relationship of variables based on the anomalies of the conditional distribution and identify fault-related variables. Due to the evolution and propagation of faults, state offsets or oscillations will cause the conditional distribution values ​​corresponding to the time-lag matrix to change. In order to measure the causal correlation of variables, the operating data of multiple devices in the target second time-lag matrix are converted into the first variables corresponding to each device. The first variables include the operating data of the devices in the target second time-lag matrix. For example, the second time-lag matrix X at the kth moment is converted into the first variable corresponding to each device. test,k Rewrite as variable index x l (1≤l≤m) represents the l-th dimension variable, and the rewriting process is as follows Figure 4 As shown, where X k Represents X test,k .

[0163] It should be noted that X -l Represents all dimensions except the l-th dimension variable, that is, p k (x l |X -l ) represents the kth moment x l The conditional probability of .

[0164] In some embodiments, during the online monitoring process, the time window X k-1 and X k When judging whether the fault or abnormality is caused by the lth variable or sensor, the hypothesis test (Formula (9)) is as follows:

[0165]

[0166] This hypothesis test depends on the variable x l and X -l The dependence between them provides the explainability of fault isolation from the perspective of conditional distribution changes.

[0167] It should be noted that, in order to eliminate the unit or magnitude effects among multiple variables during online fault isolation, the first conditional Cauchy-Schwarz divergence of each first variable is standardized and scaled to a specific interval using formula (7).

[0168] In this way, the conditional Cauchy-Schwarz divergence of each variable in its current state is independently calculated within the sliding window, resulting in the normalized fully conditional Cauchy-Schwarz divergence of all primary variables. The primary variable with the largest T(l) value is considered the primary causative variable of the fault. This is a real-time online fault isolation method that facilitates the timely identification of the primary variable closely related to the current fault state.

[0169] Based on this, in some embodiments, before normalizing the first conditional Cauchy-Schwarz divergence using a predetermined mean and a predetermined standard deviation to obtain the normalized first conditional Cauchy-Schwarz divergence, the method may further include:

[0170] converting the operating data of the plurality of devices in the target fourth time-lag matrix into a second variable corresponding to each device, wherein the second variable includes the operating data of the devices in the target fourth time-lag matrix;

[0171] Calculate the second conditional Cauchy-Schwarz divergence of the second variable under two adjacent fourth time-lag matrices. The second Cauchy-Schwarz divergence represents the difference in the conditional probability distribution function of the second variable under two adjacent fourth time-lag matrices.

[0172] Calculate the mean and standard deviation of the second conditional Cauchy-Schwarz divergence, and use the mean as the predetermined mean and the standard deviation as the predetermined standard deviation;

[0173] The second conditional Cauchy-Schwarz divergence of the second variable under two adjacent fourth time-lag matrices is calculated using formula (10), which is as follows:

[0174]

[0175] in, Represents the second variable at the k-1th moment The conditional probability distribution function under the fourth time lag matrix at the k-1th moment is: Represents the second variable at the kth moment The conditional probability distribution function under the condition of the fourth time delay matrix at the kth time, l represents the lth row of the time delay matrix, represents the lth row vector of the fourth time-delay matrix, The matrix includes second variables corresponding to multiple devices.

[0176] In this way, the standard deviation and mean are set according to the historical operating data, the first conditional Cauchy-Schwarz divergence is standardized, and the unit or magnitude effects among multiple variables are eliminated.

[0177] Based on this, in some embodiments, it further includes:

[0178] Calculate the third conditional Cauchy-Schwarz divergence of the first variable under the second time lag matrix at every first target time, where the third conditional Cauchy-Schwarz divergence represents the difference between the conditional probability distribution functions of the first variable under the second time lag matrix at every first target time;

[0179] Calculate the average of the third conditional Cauchy-Schwarz divergence during the second target time period, where the second target time period is longer than the first target time period;

[0180] When the target third condition Cauchy-Schwarz divergence is greater than the average value, it is determined that a process abnormality occurs in the equipment corresponding to the first variable.

[0181] In some embodiments, dynamic process anomaly monitoring based on conditional Cauchy-Schwarz divergence change point detection includes: judging whether a first variable has an abnormality based on the average value of the third conditional Cauchy-Schwarz divergence of all site operation data, and determining that the equipment corresponding to the first variable has a fault when the target third conditional Cauchy-Schwarz divergence is greater than the average value.

[0182] This approach considers operating point deviations and process dynamic anomalies in actual chemical processes, allowing for phased segmentation of the current operating state and variable isolation. This allows for timely regulation of variables influencing the system state to prevent excursions that could lead to failures, helping to transform post-accident emergency response into pre-accident prevention. Unsupervised online monitoring, applicable to actual industrial processes, can detect dynamic anomalies caused by operating conditions, control loops, and other factors. This allows for identification of the cause and process control to ensure the system remains stable within normal operating conditions.

[0183] Based on this, in some embodiments, calculating the third conditional Cauchy-Schwarz divergence of the first variable under the second time lag matrix every first target time length includes:

[0184] Using formula (11), the third conditional Cauchy-Schwarz divergence of the first variable under the second time lag matrix is ​​calculated every first target time. Formula (11) is as follows:

[0185]

[0186] in, Represents the first variable at the k-1th moment The conditional probability distribution function under the second time lag matrix at the k-1th moment, represents the k+wth c Second variable at time -1 At k+w c The conditional probability distribution function under the second time lag matrix condition at time -1, w c Indicates the first target duration, The matrix includes first variables corresponding to multiple devices.

[0187] In some embodiments, in the interval length l c Under this condition, calculate the reliability interval [kl c ,k+l c The average value of the conditional Cauchy-Schwarz divergence of all first variables in the second lag matrix If the kth moment is in the observation interval [k-1,k+w c -1] third conditional Cauchy-Schwarz divergence Greater than During the operation, the device corresponding to the first variable at the kth moment fails.

[0188] In some embodiments, the length of the second target time period is 1. c .

[0189] In this way, compared with identifying the faulty device after the fault is detected, if the dynamic process abnormality can be monitored in time and the variables that cause the system state change can be identified, it will help to prevent accidents before they occur.

[0190] In the embodiments provided in this application, when the system state changes or a fault occurs, fault diagnosis based on conditional Cauchy-Schwarz divergence can quantitatively describe the correlation between devices, and even variable transmission relationships, to monitor and maintain normal operation of the system.

[0191] Traditional multivariate statistical analysis methods, such as principal component analysis, partial least squares, and independent component analysis, aim to project high-dimensional observational data onto a low-dimensional subspace based on mathematical statistics. This allows the detection and analysis of fault generation and evolution based on the correlation of variables within that subspace. Considering the dynamic and nonlinear nature of actual chemical production processes, constructing a time-lag matrix based on a sliding window to capture the temporal correlations between consecutive samples and the cumulative correlations within the production process is an effective approach to improving the performance of multivariate statistical analysis-based fault monitoring techniques.

[0192] Due to the complexity of chemical processes, faults are often considered deviations from the normal state, requiring human intervention to identify the fault and restore normal operation. This hinders timely diagnosis, repair, and system compensation, among other effective control measures. If divergence calculations can accurately estimate the degree of deviation between the current system state and the normal state, providing real-time information on whether the process is operating outside acceptable limits, timely diagnosis, repair, and system compensation can be implemented within controllable limits. Monitoring data streams in process industries are discretized representations of continuous processes, and the resulting probability distribution of time series better reflects the actual state of the process. However, real-time probability density estimation for continuous processes is a challenging problem in data-driven applications. To address this, a sliding window-based approach is employed to calculate the probability density function of the stream data within the current sliding window using nonparametric kernel density estimation, effectively capturing the dynamic evolution of the data stream. Furthermore, the more effective Cauchy-Schwarz divergence is employed to measure the difference between the probability distributions of the monitored state and the normal state. This approach enables early detection of minor faults based on deviations from the production process's operating state. Furthermore, conditional divergence is combined with the ability to identify specific sensor variables that trigger faults, helping decision makers understand and adopt the inference results of diagnostic models, thereby accurately identifying equipment faults without requiring human intervention.

[0193] In particular, compared to isolating variables after a fault is detected, timely monitoring of dynamic process anomalies and identifying the specific variables that cause system state changes will help prevent accidents before they occur. The embodiments provided in this application address the above-mentioned issues and propose a micro-fault diagnosis technology more suitable for process industries. This is of great significance for effectively ensuring the operational safety and reliability of complex chemical production systems.

[0194] Based on the fault determination method provided in the above embodiment, the effect is verified and analyzed as follows.

[0195] 1. Verification and analysis based on numerical simulation

[0196] Construct a multivariate numerical simulation process as shown below:

[0197] x=As+e,

[0198] Where x = [x1, x2, x3, x4, x5, x6] T is the process variable; s=[s1,s2,s3,s4] T is an independent Gaussian distribution data source, satisfying μ=[1.2,0.7,2.3,1.7] T ; e is Gaussian white noise, ∑ e =diag{0.031,0.023,0.192,0.173,0.150,0.113}; A is the coefficient matrix

[0199] Consider failures caused by equipment deviation:

[0200]

[0201] in, is the measured value of the process variable under fault conditions, x represents the normal process variable value without fault, and f represents the fault. A training set of 700 samples is generated, of which 300 are fault samples. The fault used here is f = 0.01 (k-700), and a slow-varying fault is introduced on the variable x4. k=701,...,1000.

[0202] It should be noted that one sample corresponds to a time-delay matrix at a certain moment, and k represents the sample. One variable corresponds to one device. The fault detection process is the fault identification process.

[0203] The Cauchy-Schwarz divergence (CS) method proposed in the embodiment of the present application is compared with the principal component analysis method (PCA), the statistical Mahalanobis distance method (SMD), and the Wasserstein distance method (WD). The sliding window length is set to 100 and the significance level of fault detection is set to 0.01. In the fault detection model of principal component analysis and Wasserstein distance, the number of principal components is 3 to capture more than 95% of the variance. The fault determination results of the numerical simulation process are shown in Figure 2. Figure 5 As shown, Figure 5 (a1) in the figure is the PAC-based fault detection (principal element space) process. Figure 5 (a2) in is the PAC-based fault detection (residual space) process, Figure 5 (b1) in is the fault detection (principal element space) process based on Wasserstein distance, Figure 5 (b2) in is the fault detection (residual space) process based on Wasserstein distance, Figure 5 (c) in the figure is the fault detection process based on Mahalanobis distance. Figure 5 (d) in the figure is the fault detection process based on Cauchy-Schwarz divergence. The dotted line indicated by the arrow is the location of the abnormal operation data determined for the first time, and the solid line represents the location where the abnormal operation data occurs. Figure 5It can be seen that the PCA-based fault determination method cannot detect faults in a timely manner and has low diagnostic accuracy. The Mahalanobis distance and Wasserstein distance statistics both successfully detect faults, but the detection delay is long and they do not identify faults in their early stages. However, the monitoring statistic based on the Cauchy-Schwarz divergence can detect faults earlier, is suitable for minor fault monitoring, and has high sensitivity.

[0204] In order to evaluate the stability of the fault determination method provided in the embodiment of the present application, the input was randomly interfered with 100 times under the above-mentioned fault scenario. Table 1 compares the average detection performance of different methods and shows the fault detection performance on the numerical simulation data set. As can be seen from Table 1, the fault detection rate (FDR) of the monitoring statistic based on Cauchy-Schwarz divergence is the highest (85.0%), and the fault false alarm rate (FAR) is the lowest (0.11%). Fault detection delay (FDD) is crucial for the early detection of faults. The fault determination method provided in the embodiment of the present application is 38 minutes and 74 minutes earlier than the method based on Wasserstein distance and the method based on statistical Mahalanobis distance, respectively. Earlier detection of faults is conducive to taking effective control measures to restore the process to normal.

[0205] Table 1

[0206]

[0207] Among them, T 2 Indicates Hotelling's T 2 Statistics; SPE stands for squared prediction error; W z and W e The mean and variance are used as statistics for the principal component space and residual space, respectively, based on the Wasserstein distance. The Mahalanobis distance method uses the mean and variance as statistics. Here, the sliding window length is 100, and the confidence level is set to 1%. FDR is the fault detection rate, FAR is the false alarm rate, and FDD is the fault detection delay, corresponding to the number of samples in the fault detection delay.

[0208] The change of the operating state is mostly caused by the individual evolution or overall deviation of the variables (for example, the first variable). The main variables most relevant to the current state are determined based on the conditional distribution values ​​of the normalized variables. The results of determining the fault isolation variables based on the conditional Cauchy-Schwarz divergence are as follows: Figure 6 As shown, from Figure 6The curve on the left side of the figure shows that the variable x4 deviates significantly from the steady state after the fault occurs, and the deviation is the largest. Figure 6 From the curve on the right side of the figure, it can be seen that the mean and median of the variable x4 have changed significantly, and it can be considered as the main variable most relevant to the current fault state. This is consistent with the actual cause of the slow-changing fault, verifying the effectiveness and interpretability of the fault determination method provided in the embodiment of the present application.

[0209] As an example, further dynamic process anomaly monitoring based on conditional Cauchy-Schwarz divergence change point detection is shown in Figure 7 It should be noted that the change point represents the variable whose target third condition Cauchy-Schwarz divergence is greater than the average value (the average value of all third condition Cauchy-Schwarz divergences in the second target time period). Figure 7 As can be seen from (a) in the figure, the location of the change point is [324, 427, 732]. The monitoring process of 1000 sample points is divided into 4 stages. The distribution of the third stage [427, 732] is shifted upward, and the fourth stage [732, 1000] is shifted upward slowly. The statistical distribution of the variables in each stage is shown in Figure 2. Figure 7 In (b), the variable distribution is relatively stable in the first and second stages. Although it is right-skewed in the first stage and left-skewed in the second stage, there is no obvious variable shift or abnormality, and both are in a normal state. Starting from the third stage, the upper quartile of variable x4 increases, causing other variables to return to right skewness. Variable x4 deviates significantly from the normal state in the fourth stage, reflecting that variable x4 is the main cause of the fault, verifying the effectiveness of change point detection based on conditional Cauchy-Schwarz divergence for dynamic process anomaly monitoring.

[0210] 2. Experimental verification and analysis based on the continuous stirred tank reactor (CSTR) process

[0211] CSTR is a closed-loop reactor widely used in polymerization chemical reactions. It is a typical highly nonlinear chemical reaction system in the process industry. Its characteristics can be expressed by the following continuous-time nonlinear differential equations:

[0212]

[0213] Where, u=[C i ,T i ,T ci ] T is the input signal, C i is the reactor inlet concentration, T i is the reactor inlet temperature, T ciis the cooling water temperature at the reactor inlet, y=[C,T,T c ,Q c ] T is the output signal, C is the outlet concentration, T is the reactor temperature, T c is the cooling water temperature, Q c is the coolant flow rate, v i is the system noise, is the Arrhenius rate constant, a and b are the reaction coefficients. The remaining parameters are described in Table 2.

[0214] The disturbance of the input signal will cause the system dynamics to change. Due to the nonlinearity of the reaction process, the measured value is a time-dependent, non-Gaussian distributed and noisy data stream. Figure 8 The reactor temperature T is controlled by the coolant flow Q c To maintain, the specific reaction parameters are shown in Table 2. For the six measurement variables x = [C, T, T c ,Q,C i ,T i ] T , 2000 sample points under normal conditions are collected as training sets, and the sampling interval is 1 minute.

[0215] Table 2

[0216]

[0217]

[0218] To verify the capability of the fault identification method provided in this embodiment, 100 fault data points were generated for each of 10 types of minor faults, with different random seeds for process noise, measurement noise, and input interference. The minor fault types (parameter settings) for the CSTR process are shown in Table 3.

[0219] Table 3

[0220]

[0221] All test sets contain 1600 samples, where the fault is introduced at sample point 201. The fault determination method provided in the embodiments of this application is a general framework. To ensure fairness in comparing different methods, the subspace projection technique is extended to the fault determination method and statistical Mahalanobis distance method provided in the embodiments of this application, similar to the principal component analysis method and Wasserstein distance method, which perform detection after spatial projection. The results are compared in Table 4.

[0222] Table 4

[0223]

[0224]

[0225] Among them, MD z , MD e , W z , W e and CS z , CS e They represent the fault detection statistical indicators in the principal component space and residual space of the fault determination methods based on the statistical Mahalanobis distance method, the Wasserstein distance method, and the Cauchy-Schwarz divergence, respectively.

[0226] When subspace projection is not introduced, the average fault detection rate of the fault determination method provided in the embodiment of the present application is 89.76%, which is higher than the average fault detection rate of 82.52% of the statistical Mahalanobis distance method and the average fault detection rate of 85.74% of the Wasserstein distance method; the average fault detection delay is 130 samples, which is lower than the average fault detection delay of 102 samples of the statistical Mahalanobis distance method, but the fault false alarm rate is better than the statistical Mahalanobis distance method; when subspace projection is introduced, the fault detection rate is improved in the residual space, reaching a fault detection rate of 89.93%, and the fault detection delay is significantly shortened. The average fault detection delay is only 60 samples, which is 34 and 104 samples earlier than the statistical Mahalanobis distance method and the Wasserstein distance method respectively. The fault false alarm rate is 0 on all data sets, verifying the effectiveness of the fault determination method provided in the embodiment of the present application in distinguishing between normal state data distribution and fault state data distribution, and has the ability to identify minor faults in the early stage and the detection stability of 10 types of faults. In the actual chemical production process, timely detection of minor faults can eliminate early hidden dangers and prevent faults from worsening.

[0227] In order to verify the accuracy of the fault isolation variable determination method provided in the embodiment of the present application, taking fault type f1 as an example: a sensor deviation of 5L / min is introduced on Q at the 201st sample point, the result of determining the fault isolation variable based on the conditional Cauchy-Schwarz divergence of fault f1 is as follows: Figure 9 As shown in Figure 2. Since the deviation value 5L is smaller than the initial value, f1 can be regarded as a minor fault. After the fault is introduced, the variables Q, C i , T iThere have been abnormal changes of varying degrees, step changes, and obvious deviations from the steady state, resulting in a large number of abnormal values ​​in the statistical distribution of the variables in the right figure, and they are concentrated on the side with larger values, showing a right skewness (such as Figure 9 As shown in the figure on the right, the value moves up). In fact, since the fault occurred early, with the system regulation, the CSTR process showed a steady state under the fault for a period of time, which is manifested in Figure 9 In the variable statistical distribution diagram on the right, there is no obvious mean shift for each variable.

[0228] For fault f1, the dynamic anomaly monitoring results based on conditional Cauchy-Schwarz divergence change point detection are as follows: Figure 10 As shown, Figure 10 As shown in (a), the change point divides the entire monitoring interval into 8 sub-state intervals, among which the first and second stages have obvious offsets relative to other monitoring stages, indicating that a fault has occurred here; further, corresponding to Figure 10 In (b), variables Q, C i ,T i In both the first and second stages, mean shifts and outliers appeared to varying degrees, indicating that these three variables were closely related to the fault. If these three variables could be adjusted promptly in the first stage, the impact of the fault on the system would be reduced. Starting from the third stage, the statistical distribution of each variable remained within a small range, indicating that the process resumed stable operation after the fault occurred. In the fifth stage, the numerical range of the variable increased, but this indicates that the process in this stage has undergone dynamic changes. Users can determine whether the variable needs to be adjusted based on actual conditions.

[0229] 3. Experimental verification and analysis based on the Continuous Stirred Tank Heater (CSTH) process

[0230] The process principle of CSTH is as follows Figure 11 As shown. For the verification of the fault determination method provided in the embodiment of the present application, a total of 5 variables x=[C L ,C F ,L,T,F],C L is the cold water flow rate, C Fis the cold water valve, L is the liquid level in the tank, F is the cold water flow rate, and T is the outlet water temperature. All measurements are electrical signals within the 420mA range. The two fault scenarios used to verify the fault determination method provided in the embodiment of the present application are shown in Table 5, denoted as f1 and f2: Fault f1 is that the hot water temperature rises from 50°C to 70°C within 300s, the outlet temperature T rises, and the control loop will cause the steam valve to close slowly; Fault f2 is that the cold water valve is stuck instantly, resulting in abnormal changes in the liquid level and cold water flow rate F. Both test sets contain 2000 samples, and the fault is introduced in the 601st sample. Figure 12 In the figure, (a) shows the changes of each variable over time under normal conditions, and (b) shows the changes of each variable over time under two fault conditions, f1 and f2. Under normal conditions, each variable fluctuates stably only within a small range near its own steady state, but under fault conditions, each variable fluctuates greatly.

[0231] Table 5

[0232]

[0233]

[0234] In the process of fault determination, if the sliding window length is too large, the local stationarity or smoothness assumption may be violated; if the sliding window length is too small, the distribution estimation may be unreliable due to insufficient monitoring samples. In order to verify the relationship between the fault determination method provided by the embodiment of the application and the sliding window length, the fault detection performance is compared under different sliding window lengths w. Figure 13 As shown, Figure 13 (a) in the figure represents the fault detection rate of fault f1 under different sliding window lengths w. Figure 13 (b) in the figure represents the false alarm rate of fault f1 under different sliding window lengths w. Figure 13 (c) in the figure represents the fault detection rate of fault f2 under different sliding window lengths w. Figure 13 (d) in the figure represents the false alarm rate of fault f2 under different sliding window lengths w. Figure 13 In (b) and (d), all methods have the same fault false alarm rate for faults f1 and f2, because the first 600 normal samples of the two types of fault test sets are generated with the same noise and interference settings. Figure 13 As shown in (a) of the figure, when w∈[10,110], the fault detection rate of fault f1 increases with the increase of the window length of the sliding window. Figure 13In (c), when w∈[40,80], the fault detection rate of fault f2 also increases with the increase of the sliding window length, but when w∈[80,120], the fault false alarm rate increases with the increase of the sliding window length. Therefore, for the fault determination method and the detection method based on the statistical Mahalanobis distance provided in the embodiment of the present application, w=70 is a good compromise between the fault detection rate and the fault false alarm rate, and w=60 is more suitable for the detection method based on the Wasserstein distance.

[0235] The detection performance of faults f1 and f2 is compared under the optimal sliding window length of each method. As shown in Table 6, the fault determination method provided by the embodiment of the present application has a higher fault detection rate than the method based on the statistical Mahalanobis distance and the method based on the Wasserstein distance, and even exceeds the fault detection method based on the deep network: variational autoencoder (VAE) and orthogonal self-attentive variational autoencoder (OSAVA). This is because the number of variables in the CSTH process is small and the deep architecture advantage of the deep network is not brought into play. In the case of fault f1, the fault determination method provided by the embodiment of the present application has the best detection performance, the highest fault detection rate (97.79%), the lowest fault false alarm rate (0) and the smallest fault detection delay (30 samples). Although the fault detection rate is relatively low in the residual space in the case of fault f2, it also reflects to a certain extent the effectiveness of change point detection based on conditional Cauchy-Schwarz divergence for dynamic process anomaly monitoring. The fault determination method provided in the embodiment of the present application detects the change points of the two fault conditions f1 and f2 respectively. Figure 14 As shown, Figure 14 (a) in the figure shows the change point detection in the case of f1 failure. Figure 14 (b) in FIG. 5 shows the change point detection in the case of an f2 fault, which can effectively detect the fault in a short time.

[0236] Table 6

[0237]

[0238] The sliding window size based on the statistical Mahalanobis distance method and the Cauchy-Schwarz divergence method is set to 70, the sliding window size based on the Wasserstein distance method is set to 60, and the confidence level is set to 5%.

[0239] The fault determination method provided in the embodiment of the present application calculates the variable dependency between two adjacent sliding windows by conditional Cauchy-Schwarz divergence. The result of determining the fault isolation variable based on the conditional Cauchy-Schwarz divergence is as follows: Figure 15 As shown. Under fault f1, Figure 15 As can be seen from the left side of (a), the outlet water temperature T and the cold water flow F deviate significantly from the steady state, and Figure 15 As can be seen in the figure on the right side of (a), the statistical distribution of the outlet water temperature T and the cold water flow F show abnormal values, which can be considered as the main variables most related to fault f1. In fact, the outlet water temperature T is the causal variable of fault f1. The cold water flow F, as the control variable, shows almost the same step change as the outlet water temperature T, and T gradually returns to a stable state under the control of F. Under fault f2, Figure 15 As can be seen from the left side of (b) in the figure, the liquid level L in the tank deviates significantly from the steady state and Figure 15 As can be seen in the right-hand graph of (b), the statistical distribution of the tank liquid level L exhibits an outlier, making it the primary variable most relevant to fault f2. In reality, fault f2 is caused by the cold water valve shutoff causing the control loop to become stuck. However, the cold water valve position and cold water flow rate remain within normal ranges, so only the tank liquid level L is identified as the variable most relevant to the current state.

[0240] Under faults f1 and f2, the dynamic anomaly monitoring results based on conditional Cauchy-Schwarz divergence change point detection are as follows: Figure 16 and Figure 17 As shown, from Figure 16 As can be seen from (a) in the figure, the fault f1 process is divided into 5 stages, and there are obvious step anomalies in the middle stages 2, 3, and 4; specifically, the corresponding Figure 16 In (b), from the statistical distribution, the outlet water temperature T and the cold water flow F have a large number of outliers in the first stage of segmentation, and mean drift occurs in the second and third stages. However, in the fourth stage, the control variable F returns to normal, and the numerical range of the outlet water temperature T is reduced. Until in the fifth stage, the outlet water temperature T has only outliers, and the system gradually tends to be stable and controlled under regulation. Figure 17 As can be seen from (a) in the figure, the fault f2 process is divided into 11 stages, corresponding to Figure 17In (b), from the statistical distribution point of view, a large number of outliers appeared in the liquid level L in the second stage of the segmentation, and mean drift appeared in the third stage. If the variable L can be regulated in the second stage, it is possible to avoid the mean drift in the third stage. Although the variables fluctuate in the subsequent stages, they are generally within the controllable range, reflecting the dynamic nature of the system. Table 7 is the fault isolation variable identification result of the top three most relevant variables for faults f1 and f2. The importance of the variables and the faults decreases row by row. On this basis, the user can regulate the relevant variables in combination with process knowledge to ensure that they operate in a steady state. In summary, the fault determination method provided in the embodiment of the present application is interpretable for dynamic monitoring results, and provides a reference for how to adjust the corresponding variables at each stage, which helps to transform post-accident emergency treatment into pre-accident prevention.

[0241] Table 7

[0242]

[0243]

[0244] Among them, C L is the cold water flow rate, C F is the cold water valve, L is the liquid level in the tank, F is the cold water flow rate, and T is the outlet water temperature. The importance of variables decreases row by row.

[0245] Based on the fault determination method provided in the above embodiment, the present application also provides a specific implementation of a fault determination device. Please refer to the following embodiment.

[0246] See first Figure 18 The fault determination device 300 provided in the embodiment of the present application includes:

[0247] An acquisition module 310 is used to acquire operation data of multiple devices within a preset time period;

[0248] A construction module 320, configured to construct a first time-lag matrix and a plurality of second time-lag matrices of the operating data using a sliding window;

[0249] a calculation module 330 for respectively calculating a first Cauchy-Schwarz divergence of the first time-delay matrix and a target second time-delay matrix, wherein the first Cauchy-Schwarz divergence represents a difference between a probability distribution function of the first time-delay matrix and a probability distribution function of the target second time-delay matrix, where the target second time-delay matrix is ​​any one of the plurality of second time-delay matrices;

[0250] The determination module 340 is configured to determine that the operating data corresponding to the target second time-delay matrix is ​​abnormal and to determine that multiple equipment failures are occurring when the first Cauchy-Schwarz divergence is greater than a first threshold.

[0251] Based on this, in some embodiments, the calculation module 330 is specifically configured to:

[0252] determining a probability distribution function of a first time-delay matrix and a probability distribution function of a target second time-delay matrix;

[0253] Determine whether the first Cauchy-Schwarz divergence of the first time-delay matrix and the target second time-delay matrix satisfies the following conditions:

[0254]

[0255] Among them, X tes,t k represents the target second time-delay matrix, X tes,t ref represents the first time-delay matrix, p(X test,k ) represents the probability distribution function of the target second time-delay matrix, p(X test,ref ) represents the probability distribution function of the first time-lag matrix.

[0256] Based on this, in some embodiments, the apparatus 300 may further include:

[0257] The acquisition module 310 is further configured to acquire historical operation data of the plurality of devices within a preset time period before acquiring the operation data of the plurality of devices within a preset time period, and use the historical operation data as a training set;

[0258] The construction module 320 is further configured to construct a third time-lag matrix and a plurality of fourth time-lag matrices of the historical operation data in the training set using a sliding window;

[0259] The calculation module 330 is further configured to respectively calculate a second Cauchy-Schwarz divergence of the third time-delay matrix and a target fourth time-delay matrix, where the second Cauchy-Schwarz divergence represents a difference between a probability distribution function of the third time-delay matrix and a probability distribution function of the target fourth time-delay matrix, where the target fourth time-delay matrix is ​​any one of the plurality of fourth time-delay matrices.

[0260] The calculation module 330 is further configured to calculate a control limit corresponding to the second Cauchy-Schwarz divergence using a kernel density function based on a predetermined confidence level, and use the control limit as a first threshold;

[0261] The second Cauchy-Schwarz divergence of the third time-delay matrix and the target fourth time-delay matrix is ​​calculated respectively, and the following conditions are satisfied:

[0262]

[0263] Among them, p(Xtrain,k ) represents the probability distribution function of the target fourth time-delay matrix, p(X train,ref ) represents the probability distribution function of the third time-delay matrix.

[0264] Based on this, in some embodiments, the apparatus 300 may further include:

[0265] A conversion module, configured to convert the operating data of the plurality of devices in the target second time-lag matrix into a first variable corresponding to each device, wherein the first variable includes the operating data of the devices in the target second time-lag matrix;

[0266] The calculation module 330 is further configured to calculate a first conditional Cauchy-Schwarz divergence of the first variable under two adjacent second time-lag matrices, where the first conditional Cauchy-Schwarz divergence represents a difference in the conditional probability distribution function of the first variable under two adjacent second time-lag matrices.

[0267] The calculation module 330 is further configured to normalize the first conditional Cauchy-Schwarz divergence using a predetermined mean and a predetermined standard deviation to obtain a normalized first conditional Cauchy-Schwarz divergence;

[0268] The calculation module 330 is further configured to normalize the normalized first conditional Cauchy-Schwarz divergence to obtain a target conditional Cauchy-Schwarz divergence;

[0269] The determination module 340 is further configured to determine that the first variable corresponding to the maximum target condition Cauchy-Schwarz divergence is a fault isolation variable;

[0270] The determination module 340 is further configured to determine the fault device information corresponding to the fault isolation variable;

[0271] The calculation of the first conditional Cauchy-Schwarz divergence of the first variable under two adjacent second time-lag matrices satisfies the following conditions:

[0272]

[0273] in, Represents the first variable at the k-1th moment The conditional probability distribution function under the second time lag matrix at the k-1th moment, Represents the first variable at the kth moment The conditional probability distribution function under the condition of the second time delay matrix at the kth moment, l represents the lth row of the time delay matrix, represents the lth row vector of the second time-delay matrix, represents the first variable except the l-th row vector in the second time-lag matrix;

[0274] The first conditional Cauchy-Schwarz divergence is standardized using a predetermined mean and a predetermined standard deviation, and the standardized first conditional Cauchy-Schwarz divergence satisfies the following conditions:

[0275]

[0276] in, represents the predetermined mean, represents the predetermined standard deviation;

[0277] The first conditional Cauchy-Schwarz divergence after normalization is normalized to obtain the target conditional Cauchy-Schwarz divergence that satisfies the following conditions:

[0278]

[0279] Where m represents the mth column of the delay matrix.

[0280] Based on this, in some embodiments, the apparatus 300 may further include:

[0281] The conversion module is further configured to convert the operating data of the plurality of devices in the target fourth time-lag matrix into a second variable corresponding to each device before normalizing the first conditional Cauchy-Schwarz divergence using a predetermined mean and a predetermined standard deviation to obtain the standardized first conditional Cauchy-Schwarz divergence, the second variable including the operating data of the device in the target fourth time-lag matrix;

[0282] The calculation module 330 is further configured to calculate a second conditional Cauchy-Schwarz divergence of the second variable under two adjacent fourth time-lag matrices, where the second Cauchy-Schwarz divergence represents a difference in the conditional probability distribution function of the second variable under two adjacent fourth time-lag matrices.

[0283] The calculation module 330 is further configured to calculate a mean and a standard deviation of the second conditional Cauchy-Schwarz divergence, and use the mean as the predetermined mean and the standard deviation as the predetermined standard deviation;

[0284] The second conditional Cauchy-Schwarz divergence of the second variable under two adjacent fourth lag matrices satisfies the following conditions:

[0285]

[0286] in, Represents the second variable at the k-1th moment The conditional probability distribution function under the fourth time lag matrix at the k-1th moment is: Represents the second variable at the kth moment The conditional probability distribution function under the condition of the fourth time delay matrix at the kth time, l represents the lth row of the time delay matrix, represents the lth row vector of the fourth time-delay matrix, Represents the first variable including the l-th row vector in the second time-lag matrix.

[0287] Based on this, in some embodiments, the apparatus 300 may further include:

[0288] The calculation module 330 is further configured to calculate a third conditional Cauchy-Schwarz divergence of the first variable under the second time lag matrix at every first target duration, where the third conditional Cauchy-Schwarz divergence represents a degree of difference between conditional probability distribution functions of the first variable under the second time lag matrix at every first target duration;

[0289] The calculation module 330 is further configured to calculate an average value of the third conditional Cauchy-Schwarz divergence within a second target time period, where the second target time period is longer than the first target time period;

[0290] The determination module 340 is further configured to determine that a process abnormality occurs in the equipment corresponding to the first variable when the target third condition Cauchy-Schwarz divergence is greater than the average value.

[0291] Based on this, in some embodiments, the calculation module 330 may be specifically used to:

[0292] The third conditional Cauchy-Schwarz divergence of the first variable under the second time lag matrix is ​​calculated every first target time, and the following conditions are met:

[0293]

[0294] in, Represents the first variable at the k-1th moment The conditional probability distribution function under the second time lag matrix at the k-1th moment, represents the k+wth c Second variable at time -1 At k+w c The conditional probability distribution function under the second time lag matrix condition at time -1, w c Indicates the first target duration, Represents the first variable including the l-th row vector in the second time-lag matrix.

[0295] The various modules of the fault determination device provided in the embodiment of the present application can realize the functions of the various steps of the above-mentioned fault determination method and achieve the corresponding technical effects. For the sake of brevity, they will not be repeated here.

[0296] Based on the same inventive concept, an embodiment of the present application also provides an electronic device.

[0297] Figure 19 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application is shown.

[0298] The electronic device may include a processor 401 and a memory 402 storing computer program instructions.

[0299] Specifically, the processor 401 may include a central processing unit (CPU) or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0300] Memory 402 may include a large capacity memory for data or instructions. By way of example and not limitation, memory 402 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 402 may include removable or non-removable (or fixed) media. Where appropriate, memory 402 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, memory 402 is a non-volatile solid-state memory.

[0301] The memory may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Thus, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to an aspect of the present disclosure.

[0302] The processor 401 reads and executes computer program instructions stored in the memory 402 to implement any one of the fault determination methods in the above embodiments.

[0303] In one example, the electronic device may further include a communication interface 403 and a bus 410. Figure 19 As shown, the processor 401 , the memory 402 , and the communication interface 403 are connected via a bus 410 and communicate with each other.

[0304] The communication interface 403 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0305] The bus 410 includes hardware, software, or both, and couples the components of the electronic device to each other. By way of example and not limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Linear Predictive Coding (LPC) bus, a memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (Peripheral Component Interconnect-X, PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VESA Local Bus, VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, the bus 410 may include one or more buses. Although the embodiments of the present application describe and illustrate a specific bus, the present application contemplates any suitable bus or interconnect. The electronic device can execute the fault determination method of the embodiments of the present invention, thereby implementing the above-mentioned fault determination method.

[0306] In addition, in conjunction with the fault determination method in the above embodiments, embodiments of the present application may provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any of the fault determination methods in the above embodiments is implemented.

[0307] The present application also provides a computer program product. When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes each process of implementing any one of the above-mentioned fault determination method embodiments.

[0308] It should be understood that the present application is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present application is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present application.

[0309] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of the present application are programs or code segments used to perform the required tasks. The programs or code segments can be stored in a machine-readable medium, or transmitted on a transmission medium or communication link via a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memories (ROMs), flash memories, erasable read-only memories (EROMs), floppy disks, compact disc read-only memories (CD-ROMs), optical discs, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segments can be downloaded via computer networks such as the Internet and intranets.

[0310] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps. In other words, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0311] Aspects of the present disclosure have been described above with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit. It is also understood that each box in the block diagram and / or flowchart and the combination of the boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0312] The above is only a specific implementation method of the present application. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the scope of protection of the present application is not limited to this. Any technician familiar with this technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the scope of protection of this application.

Claims

1. A method for determining a fault, characterized in that: include: Obtain operating data of multiple devices within a preset time period; constructing a first time-lag matrix and a plurality of second time-lag matrices of the operating data using a sliding window; respectively calculating first Cauchy-Schwarz divergences of the first time-delay matrix and a target second time-delay matrix, where the first Cauchy-Schwarz divergences represent a degree of difference between a probability distribution function of the first time-delay matrix and a probability distribution function of the target second time-delay matrix, and the target second time-delay matrix is ​​any one of the plurality of second time-delay matrices; When the first Cauchy-Schwarz divergence is greater than a first threshold, determining that the operating data corresponding to the target second time-delay matrix is ​​abnormal, and determining that the multiple devices are faulty; The method of constructing the first time-lag matrix and the plurality of second time-lag matrices of the operating data by using a sliding window includes: Using a sliding window of length w, the first time-lag matrix of the running data of the previous w moments is constructed, and multiple k and x k The second time-lag matrix is ​​composed of the running data of the previous w-1 moments, where x k =[x k,1 ,x k,2 ,…,x k,m ] is the operating data at the kth moment; The respectively calculating first Cauchy-Schwarz divergences of the first time-lag matrix and the target second time-lag matrix includes: Determining a probability distribution function of the first time-delay matrix and a probability distribution function of the target second time-delay matrix; Determine that the first Cauchy-Schwarz divergence of the first time-delay matrix and the target second time-delay matrix satisfies the following condition: Among them, X test,k represents the target second time-delay matrix, X test,ref represents the first time-delay matrix, p(X test,k ) represents the probability distribution function of the target second time-delay matrix, p(X test,ref ) represents the probability distribution function of the first time-lag matrix.

2. The method for determining a fault according to claim 1, wherein: Before obtaining the operating data of the plurality of devices within a preset time period, the method further includes: Acquire historical operation data of the plurality of devices within the preset time period, and use the historical operation data as a training set; Constructing a third time-lag matrix and a plurality of fourth time-lag matrices of the historical operating data in the training set using a sliding window; respectively calculating a second Cauchy-Schwarz divergence of the third time-delay matrix and a target fourth time-delay matrix, wherein the second Cauchy-Schwarz divergence represents a difference between a probability distribution function of the third time-delay matrix and a probability distribution function of the target fourth time-delay matrix, where the target fourth time-delay matrix is ​​any one of the plurality of fourth time-delay matrices; Based on a predetermined confidence level, calculating a control limit corresponding to the second Cauchy-Schwarz divergence using a kernel density function, and using the control limit as a first threshold; The second Cauchy-Schwarz divergence of the third time-delay matrix and the target fourth time-delay matrix is ​​calculated respectively, and the following conditions are satisfied: Among them, p(X train,k ) represents the probability distribution function of the target fourth time-delay matrix, p(X train,ref ) represents the probability distribution function of the third time-delay matrix.

3. The method for determining a fault according to claim 2, wherein: Also includes: Converting the operating data of multiple devices in the target second time-lag matrix into a first variable corresponding to each device, wherein the first variable includes the operating data of the devices in the target second time-lag matrix; calculating a first conditional Cauchy-Schwarz divergence of the first variable under two adjacent second time-lag matrices, where the first conditional Cauchy-Schwarz divergence represents a difference in a conditional probability distribution function of the first variable under two adjacent second time-lag matrices; Normalizing the first conditional Cauchy-Schwarz divergence using a predetermined mean and a predetermined standard deviation to obtain a standardized first conditional Cauchy-Schwarz divergence; Normalizing the standardized first conditional Cauchy-Schwarz divergence to obtain a target conditional Cauchy-Schwarz divergence; Determine the first variable corresponding to the maximum objective condition Cauchy-Schwarz divergence as the fault isolation variable; Determining fault device information corresponding to the fault isolation variable; The first conditional Cauchy-Schwarz divergence of the first variable under two adjacent second time-lag matrices is calculated to satisfy the following conditions: in, Represents the first variable at the k-1th moment The conditional probability distribution function under the second time lag matrix at the k-1th moment, Represents the first variable at the kth moment The conditional probability distribution function under the condition of the second time delay matrix at the kth moment, l represents the lth row of the time delay matrix, represents the lth row vector of the second time-delay matrix, represents the first variable except the l-th row vector in the second time-lag matrix; The first conditional Cauchy-Schwarz divergence is normalized by using a predetermined mean and a predetermined standard deviation, so that the normalized first conditional Cauchy-Schwarz divergence satisfies the following condition: in, represents the predetermined mean, represents the predetermined standard deviation; The first conditional Cauchy-Schwarz divergence after normalization is normalized to obtain a target conditional Cauchy-Schwarz divergence that satisfies the following conditions: Where m represents the mth column of the delay matrix.

4. The method for determining a fault according to claim 3, wherein: Before normalizing the first conditional Cauchy-Schwarz divergence using a predetermined mean and a predetermined standard deviation to obtain the normalized first conditional Cauchy-Schwarz divergence, the method further includes: converting the operating data of the plurality of devices in the target fourth time-lag matrix into a second variable corresponding to each device, wherein the second variable includes the operating data of the devices in the target fourth time-lag matrix; calculating a second conditional Cauchy-Schwarz divergence of the second variable under two adjacent fourth time-lag matrices, where the second Cauchy-Schwarz divergence represents a difference in conditional probability distribution functions of the second variable under two adjacent fourth time-lag matrices; Calculating a mean and a standard deviation of the second conditional Cauchy-Schwarz divergence, and using the mean as a predetermined mean and the standard deviation as a predetermined standard deviation; The second conditional Cauchy-Schwarz divergence of the second variable under two adjacent fourth time-lag matrices is calculated to satisfy the following conditions: in, Represents the second variable at the k-1th moment The conditional probability distribution function under the fourth time lag matrix at the k-1th moment is: Represents the second variable at the kth moment The conditional probability distribution function under the condition of the fourth time delay matrix at the kth time, l represents the lth row of the time delay matrix, represents the lth row vector of the fourth time-delay matrix, Represents the first variable including the l-th row vector in the second time-lag matrix.

5. The method for determining a fault according to claim 3 or 4, characterized in that: Also includes: calculating a third conditional Cauchy-Schwarz divergence of the first variable under the second time-lag matrix at every first target duration, where the third conditional Cauchy-Schwarz divergence represents a degree of difference between conditional probability distribution functions of the first variable under the second time-lag matrix at every first target duration; calculating an average of the third conditional Cauchy-Schwarz divergence within a second target time period, where the second target time period is longer than the first target time period; When the target third condition Cauchy-Schwarz divergence is greater than the average value, it is determined that a process abnormality occurs in the equipment corresponding to the first variable.

6. The method for determining a fault according to claim 5, wherein: The calculating the third conditional Cauchy-Schwarz divergence of the first variable under the second time-lag matrix at every first target time length includes: The third conditional Cauchy-Schwarz divergence of the first variable under the second time-lag matrix is ​​calculated every first target time period to meet the following conditions: in, Represents the first variable at the k-1th moment The conditional probability distribution function under the second time lag matrix at the k-1th moment, represents the k+wth c Second variable at time -1 At k+w c -1 time under the condition of the second time lag matrix condition, the w c Indicates the first target duration, Represents the first variable including the l-th row vector in the second time-lag matrix.

7. A fault determination device, characterized in that: The device comprises: An acquisition module is used to obtain the operating data of multiple devices within a preset time period; A construction module, configured to construct a first time-lag matrix and a plurality of second time-lag matrices of the operating data using a sliding window; a calculation module, configured to respectively calculate a first Cauchy-Schwarz divergence of the first time-delay matrix and a target second time-delay matrix, wherein the first Cauchy-Schwarz divergence represents a difference between a probability distribution function of the first time-delay matrix and a probability distribution function of the target second time-delay matrix, and the target second time-delay matrix is ​​any one of the plurality of second time-delay matrices; A determination module, configured to determine that the operating data corresponding to the target second time-delay matrix is ​​abnormal and determine that the multiple devices have failed when the first Cauchy-Schwarz divergence is greater than a first threshold; The building blocks are specifically used for: Using a sliding window of length w, the first time-lag matrix of the running data of the previous w moments is constructed, and multiple k and x k The second time-lag matrix is ​​composed of the running data of the previous w-1 moments, where x k =[x k,1 ,x k,2 ,…,x k,m ] is the operating data at the kth moment; The computing module is specifically configured to: Determining a probability distribution function of the first time-delay matrix and a probability distribution function of the target second time-delay matrix; Determine that the first Cauchy-Schwarz divergence of the first time-delay matrix and the target second time-delay matrix satisfies the following condition: Among them, X test,k represents the target second time-delay matrix, X test,ref represents the first time-delay matrix, p(X test,k ) represents the probability distribution function of the target second time-delay matrix, p(X test,ref ) represents the probability distribution function of the first time-lag matrix.

8. An electronic device, characterized in that: The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the fault determination method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the fault determination method according to any one of claims 1 to 6.

10. A computer program product, characterized in that When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device is enabled to execute the fault determination method according to any one of claims 1 to 6.

Images