Fault diagnosis method and device and model prototype acquisition method

Abstract

The disclosure provides a fault diagnosis method and device and a prototype acquisition method in a model, and relates to the technical field of fault diagnosis. The method comprises the following steps: acquiring business data; performing first processing on the business data to obtain first feature data and second feature data corresponding to each first index and a plurality of time windows; inputting the first feature data into a first group of network layers of a Gaussian prototype network model to obtain third feature data; inputting the second feature data into a second group of network layers of the Gaussian prototype network model to obtain fourth feature data; splicing the third feature data and the fourth feature data to obtain fifth feature data; inputting the fifth feature data into a third group of network layers of the Gaussian prototype network model to obtain sixth feature data; and performing fault diagnosis according to the sixth feature data and a prototype of the Gaussian prototype network model.

Description

Technical Field

[0001] This disclosure relates to the field of fault diagnosis technology, and in particular to a fault diagnosis method and apparatus, and a method for obtaining a prototype in a model. Background Technology

[0002] Fault diagnosis is a crucial aspect of intelligent operation and maintenance for services such as IPTV. Among related technologies, prototype networks can be used for fault diagnosis. Summary of the Invention

[0003] The inventors noticed that the accuracy of fault diagnosis in related technologies was low. Through analysis, they discovered that in related technologies, the prototype network only performs fault diagnosis based on the characteristics of business data in a single dimension, failing to fully extract the information contained within the business data, thus resulting in low accuracy of fault diagnosis.

[0004] To address the aforementioned problems, the present disclosure proposes the following solutions.

[0005] According to one aspect of the embodiments of this disclosure, a fault diagnosis method is provided, comprising: acquiring business data, the business data including data of multiple first indicators within each time window of multiple time windows; performing a first processing on the business data to obtain first feature data and second feature data corresponding to each first indicator and the multiple time windows; inputting the first feature data into a first group of network layers of a Gaussian prototype network model to obtain third feature data; inputting the second feature data into a second group of network layers of the Gaussian prototype network model to obtain fourth feature data, wherein at least one type of network layer in the first group of network layers does not belong to the second group of network layers; concatenating the third feature data and the fourth feature data to obtain fifth feature data; inputting the fifth feature data into a third group of network layers of the Gaussian prototype network model to obtain sixth feature data, the third group of network layers including at least one fully connected layer; and performing fault diagnosis based on the sixth feature data and the prototype of the Gaussian prototype network model.

[0006] In some embodiments, the sixth feature data includes an encoding vector and the covariance matrix of the encoding vector; fault diagnosis based on the sixth feature data and the prototype of the Gaussian prototype network model includes: performing fault diagnosis based on the encoding vector and the prototype of the Gaussian prototype network model.

[0007] In some embodiments, the first process includes: decomposing the data of each first indicator of the business data into multiple time windows through seasonal trend decomposition to obtain a first trend item, a first period item, and a first residual item corresponding to each time window; determining the first feature data based on the first trend item and the first period item; and determining the second feature data based on the first residual item.

[0008] In some embodiments, the second feature data includes one or more of the mean, standard deviation, kurtosis, and skewness of the first residual term.

[0009] In some embodiments, the prototype of the Gaussian prototype network model is obtained as follows: Multiple sets of first sample service data labeled with at least one fault category are obtained, each set of first sample service data including data of multiple second indicators within each of multiple first time windows; at least one set of first sample service data is augmented to obtain at least one set of second sample service data; each set of sample service data in each group of at least one set of sample service data undergoes a second processing to obtain first sample feature data and second sample feature data corresponding to each second indicator and the multiple first time windows, wherein the at least one set of sample service data includes a first set of sample service data, which belongs to a sample service dataset. The set consists of multiple sets of first sample service data and at least one set of second sample service data; the first sample feature data is input into the first group of network layers to obtain third sample feature data; the second sample feature data is input into the second group of network layers to obtain fourth sample feature data; the third sample feature data and the fourth sample feature data are concatenated to obtain fifth sample feature data; the fifth sample feature data is input into the third group of network layers to obtain sixth sample feature data, the sixth sample feature data including a sample encoding vector and the sample covariance matrix of the sample encoding vector; based on the sixth sample feature data of each set of sample service data in the first group of sample service data, the prototype of the Gaussian prototype network model corresponding to each fault category is determined.

[0010] According to another aspect of the embodiments of this disclosure, a method for obtaining a prototype in a Gaussian prototype network model is provided, comprising: obtaining multiple sets of first sample service data labeled with at least one fault category, each set of first sample service data including data of multiple second indicators within each of multiple first time windows; performing amplification processing on at least one set of first sample service data to obtain at least one set of second sample service data; performing a second processing on each set of sample service data in each set of at least one set of sample service data to obtain first sample feature data and second sample feature data corresponding to each second indicator and the multiple first time windows, wherein the at least one set of sample service data includes a first set of sample service data, the first set of sample service data belonging to a sample service dataset, and the sample service dataset consisting of the multiple sets of first sample service data and the at least one set of second sample service data; The first sample feature data is input into the first group of network layers of the Gaussian prototype network model to obtain the third sample feature data; the second sample feature data is input into the second group of network layers of the Gaussian prototype network model to obtain the fourth sample feature data, wherein at least one type of network layer in the first group of network layers does not belong to the second group of network layers; the third sample feature data and the fourth sample feature data are concatenated to obtain the fifth sample feature data; the fifth sample feature data is input into the third group of network layers of the Gaussian prototype network model to obtain the sixth sample feature data, wherein the sixth sample feature data includes a sample encoding vector and the sample covariance matrix of the sample encoding vector, and the third group of network layers includes at least one fully connected layer; based on the sixth sample feature data of each sample service data in the first group of sample service data, the prototype of the Gaussian prototype network model corresponding to each fault category is determined.

[0011] In some embodiments, the second process includes: decomposing the data of each second indicator of each sample business data in each group of sample business data into the plurality of first time windows by seasonal trend decomposition to obtain a second trend term, a second period term and a second residual term corresponding to each first time window; determining the first sample feature data based on the second trend term and the second period term; and determining the second sample feature data based on the second residual term.

[0012] In some embodiments, the second sample feature data includes one or more of the mean, standard deviation, kurtosis, and skewness of the second residual term.

[0013] In some embodiments, the number of network layers in the first group of network layers is greater than the number of network layers in the second group of network layers.

[0014] In some embodiments, the first set of network layers includes the second set of network layers.

[0015] In some embodiments, the first group of network layers includes a one-dimensional convolutional layer, a max pooling layer, a temporary de-escalation layer, a fully connected layer, and a linear rectifier layer; and / or the second group of network layers includes a fully connected layer and a linear rectifier layer.

[0016] In some embodiments, the third set of network layers includes three fully connected layers.

[0017] In some embodiments, the service data is IPTV-related service data.

[0018] In some embodiments, determining the Gaussian prototype network model and the prototype corresponding to each fault category includes: calculating the covariance inverse matrix of the sample encoding vector of the i-th sample service data in the first group of sample service data for the c-th fault category using formula (1).

[0019]

[0020] in, Let represent the sample covariance matrix of the i-th sample service data in the first group of sample service data for the c-th fault category, 1≤c≤M, where M is the number of fault categories, 1≤i≤N, where N is the number of sample service data in the first group of sample service data for the c-th fault category; use formula (2) to determine the Gaussian prototype network model corresponding to the c-th fault category.

[0021]

[0022] in, express A vector consisting of the diagonal elements. This represents the sample encoding vector of the i-th sample service data in the first group of sample service data for the c-th fault category; based on the Gaussian prototype network model and the c-th fault category... Determine the prototype of the Gaussian prototype network model corresponding to the c-th fault category.

[0023] In some embodiments, the at least one set of sample service data further includes a second set of sample service data, the second set of sample service data and the first set of sample service data constituting the sample service dataset; according to the Gaussian prototype network model and the c-th fault category Determining the prototype of the Gaussian prototype network model corresponding to the c-th fault category includes: using formula (3) to calculate the sample encoding vector of the i-th sample service data in the second group of sample service data for the c-th fault category and... distance

[0024]

[0025] Using formula (4), calculate the i-th sample service data in the second group of sample service data for the c-th fault category. probability

[0026]

[0027] Calculate the value of the c-th fault category using formula (5). The corresponding loss l:

[0028]

[0029] The Gaussian prototype network model is determined based on the loss to correspond to the prototype of the c-th fault category.

[0030] In some embodiments, amplifying at least one set of first sample business data to obtain at least one set of second sample business data includes: for any set of first sample business data, replacing the data of at least one set of multiple second indicators within a first time window in the set of first sample business data with the corresponding data of another set.

[0031] In some embodiments, any one set of first sample business data is data within a certain unit of time, and the other set of data is data of multiple second indicators within the at least one first time window within another unit of time before the certain unit of time.

[0032] In some embodiments, the other unit of time is the preceding unit of time adjacent to the first unit of time.

[0033] In some embodiments, the plurality of first time windows include the at least one first time window and other first time windows preceding the at least one first time window, wherein the other data is predicted based on data from a plurality of second indicators within the other first time windows.

[0034] In some embodiments, the other set of data is predicted using a differentially integrated moving average autoregressive model.

[0035] In some embodiments, the other set of data corresponding to each first time window in the at least one first time window is data of multiple second indicators within the corresponding other first time window.

[0036] In some embodiments, the other first time window is an adjacent first time window.

[0037] In some embodiments, the at least one first time window includes at least one pair of first time windows, and each pair of first time windows includes two adjacent first time windows.

[0038] In some embodiments, the data of multiple second indicators are replaced within two adjacent first time windows.

[0039] According to another aspect of the present disclosure, a fault diagnosis device is provided, comprising: an acquisition module configured to acquire business data, the business data including data of multiple first indicators within each time window of multiple time windows; a processing module configured to perform a first processing on the business data to obtain first feature data and second feature data corresponding to each first indicator and the multiple time windows; an input module configured to input the first feature data into a first group of network layers of a Gaussian prototype network model to obtain third feature data; input the second feature data into a second group of network layers of the Gaussian prototype network model to obtain fourth feature data, wherein at least one type of network layer in the first group of network layers does not belong to the second group of network layers; input the fifth feature data into a third group of network layers of the Gaussian prototype network model to obtain sixth feature data, wherein the third group of network layers includes at least one fully connected layer; a splicing module configured to splice the third feature data and the fourth feature data to obtain fifth feature data; and a diagnosis module configured to perform fault diagnosis based on the sixth feature data and the prototype of the Gaussian prototype network model.

[0040] According to another aspect of the present disclosure, an apparatus for acquiring a prototype in a Gaussian prototype network model is provided, comprising: an acquisition module configured to acquire multiple sets of first sample service data labeled with at least one fault category, each set of first sample service data including data of multiple second indicators within each of multiple first time windows; an amplification module configured to amplify at least one set of first sample service data to obtain at least one set of second sample service data; and a processing module configured to perform a second processing on each set of sample service data in each set of at least one set of sample service data to obtain first sample feature data and second sample feature data corresponding to each of the multiple first time windows for each second indicator, wherein the at least one set of sample service data includes a first set of sample service data, the first set of sample service data belongs to a sample service dataset, and the sample service dataset is composed of the multiple sets of first sample service data and the at least one set of second sample service data; and an output module. The input module is configured to input the first sample feature data into the first group of network layers of the Gaussian prototype network model to obtain the third sample feature data; input the second sample feature data into the second group of network layers of the Gaussian prototype network model to obtain the fourth sample feature data, wherein at least one type of network layer in the first group of network layers does not belong to the second group of network layers; input the fifth sample feature data into the third group of network layers of the Gaussian prototype network model to obtain the sixth sample feature data, wherein the sixth sample feature data includes a sample encoding vector and a sample covariance matrix of the sample encoding vector, and the third group of network layers includes at least one fully connected layer; the splicing module is configured to splice the third sample feature data and the fourth sample feature data to obtain the fifth sample feature data; and the determination module is configured to determine the prototype of the Gaussian prototype network model corresponding to each fault category based on the sixth sample feature data of each sample service data in the first group of sample service data.

[0041] According to another aspect of the present disclosure, an electronic device is provided, including: a memory; and a processor coupled to the memory, the processor being configured to execute the method described in any of the above embodiments based on instructions stored in the memory.

[0042] According to another aspect of the present disclosure, a computer-readable storage medium is provided, including computer program instructions, wherein the computer program instructions, when executed by a processor, implement the method described in any of the above embodiments.

[0043] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein the computer program, when executed by a processor, implements the method described in any of the above embodiments.

[0044] In the above embodiment, the first and second feature data corresponding to each first indicator are obtained by processing the business data. These first and second feature data are then input into different first and second network layers of a Gaussian prototype network model to obtain third and fourth feature data, respectively. The third and fourth feature data are then concatenated and input into the third network layer of the Gaussian prototype network model to obtain sixth feature data. Finally, fault diagnosis is performed based on the sixth feature data and the prototype of the Gaussian prototype network model. This approach, which uses first and second feature data from different dimensions for fault diagnosis, fully leverages the information within the business data and improves the accuracy of fault diagnosis.

[0045] The technical solutions of this disclosure will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic flowchart of a fault diagnosis method according to some embodiments of the present disclosure;

[0048] Figure 2 This is a flowchart illustrating a method for obtaining a prototype in a Gaussian prototype network model according to some embodiments of this disclosure.

[0049] Figure 3 This is a schematic diagram of the structure of a fault diagnosis device according to some embodiments of the present disclosure;

[0050] Figure 4 This is a schematic diagram of the structure of a prototype acquisition device in a Gaussian prototype network model according to some embodiments of the present disclosure;

[0051] Figure 5 This is a schematic diagram of the structure of an electronic device according to some embodiments of the present disclosure. Detailed Implementation

[0052] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0053] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this disclosure.

[0054] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0055] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0056] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0057] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0058] Figure 1 This is a schematic flowchart of a fault diagnosis method according to some embodiments of the present disclosure.

[0059] In step 102, business data is obtained, which includes data of multiple first indicators within each time window in multiple time windows.

[0060] For example, there are a total of 10 time windows and 10 primary indicators. The business data includes data within the 10 time windows, and the data within each time window includes data from the 10 primary indicators.

[0061] In some embodiments, the service data is IPTV-related service data. As one implementation, the service data is obtained from a user's set-top box.

[0062] As some implementation methods, several primary metrics include IPTV user perception metrics, such as the number of IPTV network jitters, the number of stutters, the length of time exceeding the preset time corresponding to the alarm, the media packet loss rate, the latency factor, and one or more of the number of handover failures.

[0063] In some embodiments, if the acquired business data is missing, data imputation is performed on the acquired business data. For example, if the continuous duration of missing data within a certain time window in a unit of time (e.g., a day) is greater than a threshold (e.g., one hour), data from the same time period in another unit of time (e.g., the previous unit of time) is imputed to the missing area; if the continuous duration of missing data within a certain time window in a unit of time (e.g., a day) is less than a threshold (e.g., one hour), linear imputation is performed at the missing data area.

[0064] In step 104, the business data is processed to obtain first feature data and second feature data corresponding to each first indicator and multiple time windows.

[0065] It should be understood that the first feature data and the second feature data can reflect the characteristics of data from multiple time windows for each first indicator.

[0066] In some embodiments, both the first feature data and the second feature data are vectors.

[0067] In step 106, the first feature data is input into the first set of network layers of the Gaussian prototype network model to obtain the third feature data.

[0068] In step 108, the second feature data is input into the second set of network layers of the Gaussian prototype network model to obtain the fourth feature data.

[0069] Here, both the first and second group of network layers include multiple network layers. Furthermore, at least one type of network layer in the first group does not belong to the second group. In other words, the second group does not include at least one type of network layer.

[0070] In step 110, the third feature data and the fourth feature data are concatenated to obtain the fifth feature data.

[0071] For example, the third feature data is vector X1[a1,a2], the fourth feature data is vector X2[b1,b2], and vectors X1 and X2 are concatenated to obtain vector X3[a1,a2,b1,b2] as the fifth feature data.

[0072] In step 112, the fifth feature data is input into the third group of network layers of the Gaussian prototype network model to obtain the sixth feature data, wherein the third group of network layers includes at least one fully connected layer.

[0073] For example, the third group of network layers may include only one fully connected layer; or, for another example, the third group of network layers may include multiple fully connected layers. In some embodiments, the third group of network layers may include three fully connected layers.

[0074] It should be understood that after steps 106-112, the first and second feature data of each first indicator can be used to obtain the corresponding sixth feature data. In other words, after step 112, the sixth feature data corresponding to each of the multiple first indicators can be obtained.

[0075] In step 114, fault diagnosis is performed based on the sixth feature data and the prototype of the Gaussian prototype network model.

[0076] It should be understood that fault diagnosis can be performed based on the sixth feature data corresponding to any first indicator and the prototype of the Gaussian prototype network model. For example, based on the sixth feature data corresponding to a certain first indicator and the prototype of the Gaussian prototype network model, the fault category corresponding to that business indicator can be obtained. In this way, the fault categories corresponding to different business indicators can be obtained.

[0077] In some embodiments, the sixth feature data includes an encoding vector and its covariance matrix. In this case, fault diagnosis is performed based on the encoding vector in the sixth feature data and the prototype of the Gaussian prototype network model. In other words, fault diagnosis is not required based on the covariance matrix of the encoding vector.

[0078] For example, the encoding vector is x∈R D The covariance matrix of the encoded vector is The sixth feature data is mathematically represented as: Where D is the dimension of the encoding vector, D S It is the dimension of the covariance matrix of the encoded vector, D = D S .

[0079] In the above embodiment, the first and second feature data corresponding to each first indicator are obtained by processing the business data. These first and second feature data are then input into different first and second network layers of a Gaussian prototype network model to obtain third and fourth feature data, respectively. The third and fourth feature data are then concatenated and input into the third network layer of the Gaussian prototype network model to obtain sixth feature data. Finally, fault diagnosis is performed based on the sixth feature data and the prototype of the Gaussian prototype network model. This approach, which uses first and second feature data from different dimensions for fault diagnosis, fully leverages the information within the business data and improves the accuracy of fault diagnosis.

[0080] The process of the first processing in step 104 is described below with reference to some embodiments.

[0081] First, decompose the data of multiple time windows of each first indicator of the business data through Seasonal-Trend-Loss to obtain the first trend term, the first periodic term, and the first residual term corresponding to each time window.

[0082] Then, determine the first feature data according to the first trend term and the first periodic term, and determine the second feature data according to the first residual term. The first feature data is also called the trend feature; the second feature data is also called the statistical feature.

[0083] As some implementation manners, the first feature data is the sum of the first trend term and the first periodic term.

[0084] As some other implementation manners, determine the first feature data according to the following manner.

[0085] First, add the corresponding first periodic term and the first trend term obtained by decomposing the data of multiple time windows of the j-th first indicator to obtain x j =(x j,1 ,…,x j,w ), where 1≤j≤J, J is the number of first indicators, 1≤w≤W, and W is the number of time windows.

[0086] Then, adopt formula (1), that is, the Piecewise Aggregate Approximation (PAA) method to perform dimensionality reduction on x j to obtain the first feature data

[0087]

[0088] where, 1<t≤r, 1≤q≤w, w is the dimension of x j before dimensionality reduction; r is the dimension of obtained after dimensionality reduction; r<w, and w is divisible by r; is the t-th data of obtained after dimensionality reduction; x j,q is the q-th data of x j before dimensionality reduction.

[0089] In the above embodiments, the business data is decomposed through Seasonal-Trend-Loss to obtain two feature data with different dimensions, which more fully mines the information contained in the business data and further improves the accuracy of fault diagnosis.

[0090] In some embodiments, the second feature data includes one of the mean, standard deviation, kurtosis, and skewness of the first residual term. <00​​​= [noise j,1 , …, noise j,w , calculate the mean mean(noise j,t )、standard deviation std(noise j ), kurtosis kurt(noise j ), skewness skew(noise j ) for noise(0 < t ≤ w) to obtain the second feature data x' j . x' j can include one of the mean, standard deviation, kurtosis, and skewness. For example, x′ j = [mean(noise j )].

[0092] In the above embodiment, through data processing of the first residual term, the obtained second feature data includes one of the mean, standard deviation, kurtosis, and skewness of the first residual term. In this way, the information contained in the business data is more fully mined, and the accuracy of fault diagnosis is further improved.

[0093] In some other embodiments, the second feature data includes multiple of the mean, standard deviation, kurtosis, and skewness of the first residual term, for example, two, three, or four. For example, the second feature data includes the mean and standard deviation, or includes kurtosis and skewness, or includes the mean, standard deviation, and kurtosis, which will not be enumerated here.

[0094] In the above embodiment, through data processing of the first residual term, the obtained second feature data includes multiple of the mean, standard deviation, kurtosis, and skewness of the first residual term. In this way, the information contained in the business data is more fully mined, and the accuracy of fault diagnosis is further improved.

[0095] Figure 2 is a schematic flowchart of a method for obtaining prototypes in a Gaussian prototype network model according to some embodiments of the present disclosure.

[0096] In step 202, obtain multiple first sample business data of at least one fault category that has been labeled. Each first sample business data includes data of multiple second indicators within each of the multiple first time windows.

[0097] For example, use different labels to label multiple first sample business data of each fault category. For example, the labels can include: no fault and having a fault. Having a fault includes fault categories such as user operation error, set-top box network cable loose, user terminal fault, broadband network fault, etc.

[0098] ​​In some embodiments, multiple first sample service data sets for each of the labeled fault categories may be obtained.

[0099] In some embodiments, the labels corresponding to multiple fault categories include "no fault" and "faulty", and the number of multiple first sample service data sets labeled "no fault" is less than the number of multiple first sample service data sets labeled "faulty". For example, the first sample service data set labeled "no fault" is downsampled to avoid excessive imbalance in the distribution of labeled sample service data, which could affect the accuracy of fault diagnosis.

[0100] In step 204, at least one first sample of business data is amplified to obtain at least one second sample of business data.

[0101] It should be understood that multiple sets of first sample business data can be amplified to obtain multiple sets of corresponding second sample business data.

[0102] The following sections will introduce some implementation methods of amplification processing in conjunction with different embodiments.

[0103] In step 206, each sample business data in each group of sample business data in at least one group of sample business data is subjected to a second processing to obtain the first sample feature data and the second sample feature data corresponding to each second indicator and multiple first time windows.

[0104] Here, at least one set of sample business data includes a first set of sample business data, which belongs to a sample business dataset. The sample business dataset consists of multiple sets of first sample business data and at least one set of second sample business data obtained in step 204.

[0105] It should be noted that when multiple sets of second sample business data are obtained in step 204, the sample business dataset consists of multiple sets of first sample business data and multiple sets of second sample business data obtained in step 204.

[0106] In some implementations, the second processing may take a different approach than the first processing. In other implementations, the second processing may take the same approach as the first processing.

[0107] In step 208, the first sample feature data is input into the first set of network layers of the Gaussian prototype network model to obtain the third sample feature data.

[0108] In step 210, the second sample feature data is input into the second group of network layers of the Gaussian prototype network model to obtain the fourth sample feature data. At least one type of network layer in the first group of network layers does not belong to the second group of network layers.

[0109] In step 212, the feature data of the third sample and the feature data of the fourth sample are concatenated to obtain the feature data of the fifth sample.

[0110] For example, the third sample feature data is vector X1[a1,a2], the fourth sample feature data is vector X2[b1,b2], and vectors X1 and X2 are concatenated to obtain vector X3[a1,a2,b1,b2] as the fifth feature data.

[0111] In step 214, the fifth sample feature data is input into the third group of network layers of the Gaussian prototype network model to obtain the sixth sample feature data. The sixth sample feature data includes the sample encoding vector and the sample covariance matrix of the sample encoding vector. The third group of network layers includes at least one fully connected layer.

[0112] In some embodiments, the third group of network layers includes three fully connected layers.

[0113] In step 216, based on the sixth sample feature data of each sample service data in the first group of sample service data, the prototype of the Gaussian prototype network model and the prototype corresponding to each fault category are determined.

[0114] In the above embodiments, on the one hand, the sample business data is processed to obtain the first sample feature data and the second sample feature data corresponding to each first indicator. The first and second sample feature data are then input into different first and second network layers of the Gaussian prototype network model to obtain the third and fourth sample feature data, respectively. The third and fourth sample feature data are then concatenated and input into the third network layer of the Gaussian prototype network model to obtain the sixth sample feature data. Finally, the prototype of the Gaussian prototype network model corresponding to each fault category is determined based on the sixth sample feature data. This approach, by performing fault diagnosis based on first and second sample feature data from different dimensions, fully leverages the information in the sample business data and improves the accuracy of the prototype in the Gaussian prototype network.

[0115] On the other hand, by amplifying the first sample service data to obtain the second sample service data, the number of sample service data that need to be labeled is reduced, avoiding the problems of low network failure frequency, difficulty in manual fault diagnosis, and high resource consumption in manual labeling in certain scenarios (such as IPTV service scenarios), and further improving the efficiency of obtaining prototypes in Gaussian prototype networks.

[0116] In some embodiments, the data of each first time window of each second metric in each set of sample service data is decomposed through seasonal trend decomposition to obtain a second trend term, a second periodic term, and a second residual term corresponding to each first time window; the first sample feature data is determined according to the second trend term and the second periodic term; the second sample feature data is determined according to the second residual term. Thus, through seasonal trend decomposition, the sample service data is decomposed to obtain two sample feature data of different dimensions, which more fully exploits the information contained in the sample service data and further improves the accuracy of the prototypes in the Gaussian prototype network.

[0117] In some embodiments, the second sample feature data includes one of the mean, standard deviation, kurtosis, and skewness of the second residual term.

[0118] For example, after decomposing the data of multiple time windows of the j-th second metric, the corresponding second residual term noise j =[noise j,1 ,…,noise j,w is obtained. The mean mean(noise j,t ), standard deviation std(noise j ), kurtosis kurt(noise j ), and skewness skew(noise j ) are respectively calculated for noise j (0 < t ≤ w) to obtain the second sample feature data y’ j . y’ j may include the mean, standard deviation, kurtosis, and skewness. For example, y′ j =[mean(noise j )].

[0119] In the above embodiments, through data processing of the second residual term, the obtained second feature data includes one of the mean, standard deviation, kurtosis, and skewness of the second residual term. Thus, the information contained in the service data is more fully exploited, and the accuracy of the prototypes in the Gaussian prototype network is further improved.

[0120] In some other embodiments, the second sample feature data includes multiple of the mean, standard deviation, kurtosis, and skewness of the second residual term, such as two, three, or four. For example, the second sample feature data includes the mean and standard deviation, or includes the mean, standard deviation, and kurtosis, or includes the mean, kurtosis, and skewness, which will not be enumerated here.

[0121] In the above embodiments, by processing the data of the second residual term, the obtained second sample feature data includes multiple values ​​of the second residual term, such as mean, standard deviation, kurtosis, and skewness. This more fully extracts the information contained in the business data and further improves the accuracy of the prototype in the Gaussian prototype network.

[0122] In some embodiments, at least one set of sample service data in step 206 further includes a second set of sample service data, and the second set of sample service data and the first set of sample service data constitute a sample service dataset.

[0123] For example, a set of sample service data corresponding to the c-th fault category contains 10 sample service data sets, of which 7 sample service data sets belong to the first set of sample service data, also known as the support set, and are used to construct the Gaussian prototype network corresponding to the c-th fault category. The other three data samples belong to the second group of sample business data, also known as the query set, and are used to... Determine the prototype of the Gaussian prototype network model corresponding to the c-th fault category.

[0124] In the above embodiment, the c-th fault category is constructed in the Gaussian prototype network using the first set of sample service data. And based on the sample encoding vector of the second set of sample business data and Determining the prototype corresponding to the c-th fault category in the Gaussian prototype network model further improves the accuracy of the prototype in the Gaussian prototype network.

[0125] In some embodiments, the number of network layers in the first group of network layers is greater than the number of network layers in the second group of network layers. For example, the first group of network layers contains 5 network layers, and the second group of network layers contains 2 network layers. Thus, by inputting feature data from two different dimensions into two different groups of network layers, the information contained in the business data is fully extracted during fault diagnosis, improving the accuracy of fault diagnosis. On the other hand, when obtaining the prototype of the Gaussian prototype network model, the information contained in the sample business data is fully extracted, improving the accuracy of the Gaussian prototype network model prototype.

[0126] In some embodiments, the first set of network layers includes the second set of network layers. This allows for a more thorough extraction of the information contained within the business data during fault diagnosis, further improving the accuracy of fault diagnosis. On the other hand, when obtaining the prototype of the Gaussian prototype network model, the information contained in the sample business data is more fully extracted, further improving the accuracy of the Gaussian prototype network model prototype.

[0127] In some embodiments, the first set of network layers includes a one-dimensional convolutional layer, a max pooling layer, a temporary de-escalation layer, a fully connected layer, and a linear rectifier layer; and / or the second set of network layers includes a fully connected layer and a linear rectifier layer.

[0128] In the above embodiments, in the first group of network layers, a one-dimensional convolutional layer extracts information from the first sample feature data, a max pooling layer samples this information, and a fully connected layer summarizes the convolutional feature data and maps it into two-dimensional feature data for output to facilitate subsequent concatenation. The temporary fallback layer and the linear rectifier layer prevent gradient explosion and gradient vanishing. The second sample feature data includes one or more of the mean, standard deviation, kurtosis, and skewness of the second residual term, suitable for a relatively simple network structure. The temporary fallback layer and the linear rectifier layer effectively prevent overfitting. Through the first, second, and third group of network layers, the information contained in the business data is more fully explored when performing fault diagnosis on the business data, further improving the accuracy of fault diagnosis. On the other hand, when obtaining the prototype of the Gaussian prototype network model, the information contained in the sample business data is more fully explored, improving the accuracy of the Gaussian prototype network model prototype.

[0129] The following examples illustrate different implementations of step 216.

[0130] In some embodiments, step 216 is implemented as follows.

[0131] First, the covariance inverse matrix of the sample encoding vector of the i-th sample service data in the first group of sample service data of the c-th fault category is calculated using formula (2).

[0132]

[0133] in, Let M be the sample covariance matrix of the i-th sample business data in the first group of sample business data for the c-th fault category, where 1 ≤ c ≤ M, M is the number of fault categories, and 1 ≤ i ≤ N, where N is the number of sample business data in the first group of sample business data for the c-th fault category.

[0134] Then, using formula (3), the Gaussian prototype network model corresponding to the c-th fault category is determined.

[0135]

[0136] in, express A vector consisting of the diagonal elements. This represents the sample encoding vector of the i-th sample service data in the first group of sample service data for the c-th fault category, where * indicates bitwise multiplication.

[0137] For example, vectors Then the diagonal vector of vector z is [a1a2].

[0138] Then, based on the Gaussian prototype network model and the c-th fault category... Determine the prototype of the Gaussian prototype network model corresponding to the c-th fault category.

[0139] In the above embodiments, the Gaussian prototype network model corresponding to the c-th fault category is obtained through formulas (2) and (3). Reuse Determining the prototype corresponding to the c-th fault category in the Gaussian prototype network model further improves the accuracy of the prototype in the Gaussian prototype network model.

[0140] As some implementation methods, This serves as the prototype for the Gaussian prototype network model and the c-th fault category.

[0141] As another implementation, based on the Gaussian prototype network model and the c-th fault category... Determine the prototype of the Gaussian prototype network model corresponding to the c-th fault category, including:

[0142] Using formula (4), calculate the sample encoding vector of the i-th sample service data in the second group of sample service data for the c-th fault category. distance

[0143]

[0144] Using formula (5), i.e., the softmax formula, the i-th sample service data in the second group of sample service data of the c-th fault category belongs to... probability

[0145]

[0146] Using formula (6), calculate the loss l corresponding to the c-th fault category:

[0147]

[0148] The Gaussian prototype network model is determined based on the loss, and the prototype corresponding to the c-th fault category is identified.

[0149] As one implementation, when constructing the loss function, a larger weight is given to fault categories with a lower probability of occurrence, in order to mitigate the adverse effects of uneven distribution on the prototype generation in the Gaussian prototype network model.

[0150] In the above embodiments, by calculating the i-th sample business data and... The distance is calculated, and the softmax formula is introduced to obtain the probability value of the i-th sample business data belonging to the c-th fault category. The loss function is constructed using formula (6) to determine the Gaussian prototype network model and the prototype corresponding to each fault category, which further improves the accuracy of the prototype in the Gaussian prototype network model.

[0151] The following describes different implementation methods of step 206 with reference to different embodiments.

[0152] In some embodiments, for any first sample business data, the data of at least one set of multiple second indicators within a first time window in any first sample business data is replaced with the corresponding data of another set.

[0153] In the above embodiments, the data of multiple second indicators within at least one first time window in any first sample business data is replaced with the corresponding data of another sample, thereby expanding the sample business data, reducing the amount of sample business data, further improving the efficiency of constructing the prototype in the Gaussian prototype network model, and reducing human resources costs.

[0154] The following section introduces several ways to implement another set of data.

[0155] As one implementation method, any first sample of business data is data within a certain unit of time, and another set of data is data of multiple second indicators within at least one first time window within another unit of time before a certain unit of time.

[0156] In some embodiments, the other unit of time is the preceding unit of time adjacent to a certain unit of time. For example, a certain unit of time can be a day, a week, or a month. Taking a day as an example, the data from 19:00 to 20:00 on May 21 is replaced with the data from 19:00 to 20:00 on May 20 to obtain the second sample data.

[0157] As some other implementations, multiple first time windows include at least one first time window and other first time windows preceding at least one first time window, with another set of data based on predictions of multiple second indicators within the other first time windows.

[0158] For example, the data from 0:00 to 24:00 on May 21st is randomly divided into two segments: 0:00 to 12:00 and 12:00 to 24:00. Based on the real data from 0:00 to 12:00, predicted data for 12:00 to 24:00 is generated. The predicted data is then used to replace the real data from 12:00 to 24:00 to obtain the second sample data.

[0159] In some embodiments, another set of data is predicted using a differentially integrated moving average autoregressive model. The differentially integrated moving average autoregressive model, also known as the ARIMA model, is a model used for time series forecasting.

[0160] In some other implementations, at least one set of data corresponding to each first time window within a first time window is data from multiple second indicators within another corresponding first time window. In other words, data from multiple second indicators within another first time window is used to replace data from multiple second indicators within a certain first time window.

[0161] In some embodiments, another first time window in any first sample of business data is an adjacent first time window.

[0162] For example, replace the data from 10:00 to 11:00 on May 21 with the data from 9:00 to 10:00 on May 21 to obtain the second sample of business data.

[0163] In some cases, at least one first time window in any first sample of business data includes at least one pair of first time windows, and each pair of first time windows includes two adjacent first time windows.

[0164] For example, if a pair of first time windows are May 21st 9:00-10:00 and May 21st 10:00-11:00, the data from May 21st 8:00-9:00 replaces the data from May 21st 9:00-10:00, and the data from May 21st 11:00-12:00 replaces the data from May 21st 10:00-11:00 to obtain the second sample business data.

[0165] In other cases, at least one first time window in any first sample of business data includes at least one pair of first time windows, each pair of first time windows includes two adjacent first time windows, and the data of multiple second indicators are interchanged within the two adjacent first time windows.

[0166] For example, the data from 8:00-9:00 on May 21st is swapped with the data from 9:00-10:00 on May 21st, and the data from 10:00-11:00 on May 21st is swapped with the data from 11:00-12:00 on May 21st to obtain the second sample business data.

[0167] In some embodiments, when performing fault diagnosis on business data, business data with fault diagnosis results greater than a threshold are added to the sample business dataset to obtain the prototype of the Gaussian prototype network model, thereby further improving the accuracy of the prototype in the Gaussian prototype network model.

[0168] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to mutually. For the apparatus embodiments, since they largely correspond to the method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0169] Figure 3 This is a schematic diagram of the structure of a fault diagnosis device according to some embodiments of the present disclosure.

[0170] like Figure 3 As shown, the fault diagnosis device includes an acquisition module 301, a processing module 302, an input module 303, a splicing module 304, and a diagnosis module 305.

[0171] The acquisition module 301 is configured to acquire business data, which includes data of multiple first indicators within each time window.

[0172] The processing module 302 is configured to perform a first processing on the business data to obtain first feature data and second feature data corresponding to each first indicator and multiple time windows.

[0173] The input module 303 is configured to input first feature data into the first group of network layers of the Gaussian prototype network model to obtain third feature data; input second feature data into the second group of network layers of the Gaussian prototype network model to obtain fourth feature data, wherein at least one type of network layer in the first group of network layers does not belong to the second group of network layers; and input fifth feature data into the third group of network layers of the Gaussian prototype network model to obtain sixth feature data, wherein the third group of network layers includes at least one fully connected layer.

[0174] The splicing module 304 is configured to splice the third feature data and the fourth feature data to obtain the fifth feature data.

[0175] The diagnostic module 305 is configured to perform fault diagnosis based on the sixth feature data and the prototype of the Gaussian prototype network model.

[0176] In some embodiments, the fault diagnosis apparatus further includes other modules to perform the fault diagnosis methods of the other embodiments described above.

[0177] Figure 4 This is a schematic diagram of the structure of a prototype acquisition device in a Gaussian prototype network model according to some embodiments of the present disclosure.

[0178] like Figure 4 As shown, the prototype acquisition device in the Gaussian prototype network model includes an acquisition module 401, an amplification module 402, a processing module 403, an input module 404, a splicing module 405, and a determination module 406.

[0179] The acquisition module 401 is configured to acquire multiple first sample service data sets labeled with at least one fault category, each first sample service data set including data of multiple second indicators within each first time window.

[0180] The amplification module 402 is configured to amplify at least one first sample of service data to obtain at least one second sample of service data.

[0181] The processing module 403 is configured to perform a second processing on each sample business data in each of the at least one set of sample business data to obtain first sample feature data and second sample feature data corresponding to each second indicator and multiple first time windows. The at least one set of sample business data includes a first set of sample business data, which belongs to a sample business dataset. The sample business dataset consists of multiple first sample business data and at least one second sample business data.

[0182] The input module 404 is configured to input the first sample feature data into the first group of network layers of the Gaussian prototype network model to obtain the third sample feature data; input the second sample feature data into the second group of network layers of the Gaussian prototype network model to obtain the fourth sample feature data, wherein at least one type of network layer in the first group of network layers does not belong to the second group of network layers; and input the fifth sample feature data into the third group of network layers of the Gaussian prototype network model to obtain the sixth sample feature data, wherein the sixth sample feature data includes the sample encoding vector and the sample covariance matrix of the sample encoding vector, and the third group of network layers includes at least one fully connected layer.

[0183] The splicing module 405 is configured to splice the third sample feature data and the fourth sample feature data to obtain the fifth sample feature data.

[0184] The determination module 406 is configured to determine the prototype of the Gaussian prototype network model corresponding to each fault category based on the sixth sample feature data of each sample service data in the first group of sample service data.

[0185] In some embodiments, the prototype acquisition device further includes other modules to perform the prototype acquisition methods of the other embodiments described above.

[0186] In some embodiments, the fault diagnosis apparatus further includes a prototype acquisition device of any of the above embodiments.

[0187] Figure 5 This is a schematic diagram of the structure of an electronic device according to some embodiments of the present disclosure.

[0188] like Figure 5 As shown, the electronic device 500 includes a memory 501 and a processor 502 coupled to the memory 501. The processor 502 is configured to execute the method of any of the foregoing embodiments based on instructions stored in the memory 501.

[0189] The memory 501 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory may store, for example, an operating system, application programs, a boot loader, and other programs.

[0190] Electronic device 500 may also include input / output interface 503, network interface 504, storage interface 505, etc. These interfaces 503, 504, and 505, as well as the memory 501 and processor 502, can be connected via, for example, a bus 506. Input / output interface 503 provides a connection interface for input / output devices such as monitors, mice, keyboards, and touchscreens. Network interface 504 provides a connection interface for various networked devices. Storage interface 505 provides a connection interface for external storage devices such as SD cards and USB flash drives.

[0191] This disclosure also provides a computer-readable storage medium including computer program instructions that, when executed by a processor, implement the method of any of the above embodiments.

[0192] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the method of any of the above embodiments.

[0193] The embodiments of this disclosure have now been described in detail. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein based on the above description.

[0194] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0195] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that the functions specified in one or more flowchart illustrations and / or one or more blocks in a block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate functions for implementing the functions in the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0196] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0197] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0198] While specific embodiments of this disclosure have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments or equivalent substitutions can be made to some technical features without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A fault diagnosis method, comprising: Acquire business data, which includes data of multiple first indicators within each time window; The business data is first processed to obtain first feature data and second feature data corresponding to each first indicator and the plurality of time windows; The first feature data is input into the first set of network layers of the Gaussian prototype network model to obtain the third feature data; The second feature data is input into the second group of network layers of the Gaussian prototype network model to obtain the fourth feature data. At least one type of network layer in the first group of network layers does not belong to the second group of network layers. The number of network layers in the first group of network layers is greater than the number of network layers in the second group of network layers. The third feature data and the fourth feature data are concatenated to obtain the fifth feature data; The fifth feature data is input into the third group of network layers of the Gaussian prototype network model to obtain the sixth feature data. The third group of network layers includes at least one fully connected layer. Based on the sixth feature data and the prototype of the Gaussian prototype network model, fault diagnosis is performed. The first process includes: The business data is decomposed into multiple time windows for each first indicator by seasonal trend decomposition to obtain the first trend item, the first period item, and the first residual item corresponding to each time window. The first feature data is determined based on the first trend term and the first period term; The second feature data is determined based on the first residual term. The business data mentioned above refers to IPTV-related business data. The data of the multiple first indicators include IPTV user perception indicators, which include one or more of the following: IPTV network jitter count, number of stutters, duration exceeding the preset time required for alarm, media packet loss rate, latency factor, and number of handover failures.

2. The method as described in claim 1, wherein, The sixth feature data includes the encoding vector and the covariance matrix of the encoding vector; Fault diagnosis based on the sixth feature data and the prototype of the Gaussian prototype network model includes: performing fault diagnosis based on the encoding vector and the prototype of the Gaussian prototype network model.

3. The method as described in claim 1, wherein, The second feature data includes one or more of the mean, standard deviation, kurtosis, and skewness of the first residual term.

4. The method of claim 1, wherein, The prototype of the Gaussian prototype network model is obtained in the following way: Obtain multiple sets of first sample business data labeled with at least one fault category, each set of first sample business data including data of multiple second indicators within each first time window; At least one set of first sample business data is amplified to obtain at least one set of second sample business data. A second processing is performed on each sample business data in each of the at least one set of sample business data to obtain the first sample feature data and the second sample feature data corresponding to each second indicator and the plurality of first time windows. The at least one set of sample business data includes a first set of sample business data, which belongs to a sample business dataset. The sample business dataset is composed of the plurality of first sample business data and the at least one second sample business data. The first sample feature data is input into the first group of network layers to obtain the third sample feature data; The second sample feature data is input into the second group of network layers to obtain the fourth sample feature data; The third sample feature data and the fourth sample feature data are concatenated to obtain the fifth sample feature data; The fifth sample feature data is input into the third group of network layers to obtain the sixth sample feature data, which includes the sample encoding vector and the sample covariance matrix of the sample encoding vector. Based on the sixth sample feature data of each sample business data in the first group of sample business data, the prototype of the Gaussian prototype network model corresponding to each fault category is determined. The second process includes: By using seasonal trend decomposition, the data of each second indicator of each sample business data in each group of sample business data is decomposed into multiple first time windows to obtain the second trend term, second period term and second residual term corresponding to each first time window. The first sample feature data is determined based on the second trend term and the second period term; The second sample feature data is determined based on the second residual term.

5. A method for obtaining prototypes in a Gaussian prototype network model, comprising: Obtain multiple sets of first sample business data labeled with at least one fault category, each set of first sample business data including data of multiple second indicators within each first time window; At least one set of first sample business data is amplified to obtain at least one set of second sample business data. A second processing is performed on each sample business data in each of the at least one set of sample business data to obtain the first sample feature data and the second sample feature data corresponding to each second indicator and the plurality of first time windows. The at least one set of sample business data includes a first set of sample business data, which belongs to a sample business dataset. The sample business dataset is composed of the plurality of first sample business data and the at least one second sample business data. The first sample feature data is input into the first group of network layers of the Gaussian prototype network model to obtain the third sample feature data. The second sample feature data is input into the second group of network layers of the Gaussian prototype network model to obtain the fourth sample feature data. At least one type of network layer in the first group of network layers does not belong to the second group of network layers. The number of network layers in the first group of network layers is greater than the number of network layers in the second group of network layers. The third sample feature data and the fourth sample feature data are concatenated to obtain the fifth sample feature data; The fifth sample feature data is input into the third group of network layers of the Gaussian prototype network model to obtain the sixth sample feature data. The sixth sample feature data includes the sample encoding vector and the sample covariance matrix of the sample encoding vector. The third group of network layers includes at least one fully connected layer. Based on the sixth sample feature data of each sample business data in the first group of sample business data, the prototype of the Gaussian prototype network model corresponding to each fault category is determined. The second process includes: By using seasonal trend decomposition, the data of each second indicator of each sample business data in each group of sample business data is decomposed into multiple first time windows to obtain the second trend term, second period term and second residual term corresponding to each first time window. The first sample feature data is determined based on the second trend term and the second period term; The second sample feature data is determined based on the second residual term. The multiple sets of first sample business data are IPTV-related business data, and the multiple sets of second indicator data include IPTV user perception indicators, which include one or more of the following: IPTV network jitter count, number of stutters, duration exceeding the preset time required for alarm, media packet loss rate, latency factor, and number of handover failures.

6. The method of claim 5, wherein, The second sample feature data includes one or more of the mean, standard deviation, kurtosis, and skewness of the second residual term.

7. The method as described in any one of claims 1-5, wherein, The first group of network layers includes the second group of network layers.

8. The method of claim 7, wherein: The first group of network layers includes one-dimensional convolutional layers, max pooling layers, deferred layers, fully connected layers, and linear rectified layers; and / or The second group of network layers includes fully connected layers and linear rectifier layers.

9. The method according to any one of claims 1-5, wherein, The third group of network layers includes three fully connected layers.

10. The method of claim 4 or 5, wherein, Determining the Gaussian prototype network model and the prototype corresponding to each fault category includes: The covariance inverse matrix of the sample encoding vector of the i-th sample service data in the first group of sample service data for the c-th fault category is calculated using formula (1). , （1） in, Let M be the sample covariance matrix of the i-th sample business data in the first group of sample business data for the c-th fault category, where 1≤c≤M, M is the number of fault categories, 1≤i≤N, and N is the number of sample business data in the first group of sample business data for the c-th fault category. Formula (2) is used to determine the Gaussian prototype network model corresponding to the c-th fault category. , （2） in, express A vector consisting of the diagonal elements. The sample encoding vector represents the i-th sample service data in the first group of sample service data for the c-th fault category; Based on the Gaussian prototype network model and the c-th fault category The prototype of the Gaussian prototype network model corresponding to the c-th fault category is determined.

11. The method of claim 10, wherein: The at least one set of sample service data also includes a second set of sample service data, and the second set of sample service data and the first set of sample service data constitute the sample service dataset. Based on the Gaussian prototype network model and the c-th fault category The prototypes corresponding to the Gaussian prototype network model and the c-th fault category include: Using formula (3), calculate the sample encoding vector of the i-th sample service data in the second group of sample service data for the c-th fault category. distance , ；（3） Using formula (4), calculate the i-th sample service data in the second group of sample service data belonging to the c-th fault category. probability : （4） Calculate the value of the c-th fault category using formula (5). Corresponding losses : ；（5） The Gaussian prototype network model is determined based on the loss to correspond to the prototype of the c-th fault category.

12. The method of claim 4 or 5, wherein, Augmentation of at least one set of first-sample business data to obtain at least one set of second-sample business data includes: For any set of first sample business data, replace the data of multiple second indicators within at least one first time window in the set of first sample business data with the corresponding data from another set.

13. The method of claim 12, wherein, The first sample of business data is data within a certain unit of time, and the other data is data of multiple second indicators within the at least one first time window within another unit of time before the first unit of time.

14. The method of claim 13, wherein, The other unit of time is the preceding unit of time adjacent to the first unit of time.

15. The method of claim 12, wherein, The plurality of first time windows include the at least one first time window and other first time windows preceding the at least one first time window, and the other data is a prediction based on data from a plurality of second indicators within the other first time windows.

16. The method of claim 15, wherein, The other set of data was predicted using a differentially integrated moving average autoregressive model.

17. The method of claim 12, wherein, The other set of data corresponding to each of the at least one first time window is the data of multiple second indicators within the corresponding other first time window.

18. The method of claim 17, wherein, The other first time window is an adjacent first time window.

19. The method of claim 18, wherein, The at least one first time window includes at least one pair of first time windows, and each pair of first time windows includes two adjacent first time windows.

20. The method of claim 19, wherein, The data of multiple second indicators are replaced within two adjacent first time windows.

21. A fault diagnosis device, comprising: The acquisition module is configured to acquire business data, which includes data of multiple first indicators within each time window of multiple time windows; The processing module is configured to perform a first processing on the business data to obtain first feature data and second feature data corresponding to each first indicator and the plurality of time windows; The input module is configured to input the first feature data into the first group of network layers of the Gaussian prototype network model to obtain the third feature data; input the second feature data into the second group of network layers of the Gaussian prototype network model to obtain the fourth feature data, wherein at least one type of network layer in the first group of network layers does not belong to the second group of network layers; and input the fifth feature data into the third group of network layers of the Gaussian prototype network model to obtain the sixth feature data, wherein the third group of network layers includes at least one fully connected layer, and wherein the number of network layers in the first group of network layers is greater than the number of network layers in the second group of network layers. The splicing module is configured to splice the third feature data and the fourth feature data to obtain the fifth feature data; The diagnostic module is configured to perform fault diagnosis based on the sixth feature data and the prototype of the Gaussian prototype network model. The first process includes: The business data is decomposed into multiple time windows for each first indicator by seasonal trend decomposition to obtain the first trend item, the first period item, and the first residual item corresponding to each time window. The first feature data is determined based on the first trend term and the first period term. The second feature data is determined based on the first residual term. The business data mentioned above refers to IPTV-related business data. The data of the multiple first indicators include IPTV user perception indicators, which include one or more of the following: IPTV network jitter count, number of stutters, duration exceeding the preset time required for alarm, media packet loss rate, latency factor, and number of handover failures.

22. A device for obtaining prototypes in a Gaussian prototype network model, comprising: The acquisition module is configured to acquire multiple sets of first sample business data labeled with at least one fault category, each set of first sample business data including data of multiple second indicators within each first time window; The amplification module is configured to amplify at least one first sample of business data to obtain at least one second sample of business data. The processing module is configured to perform a second processing on each sample business data in each of at least one set of sample business data to obtain first sample feature data and second sample feature data corresponding to each second indicator and the plurality of first time windows. The at least one set of sample business data includes a first set of sample business data, which belongs to a sample business dataset. The sample business dataset is composed of the plurality of first sample business data and the at least one second sample business data. The input module is configured to input the first sample feature data into the first set of network layers of the Gaussian prototype network model to obtain the third sample feature data. The second sample feature data is input into the second group of network layers of the Gaussian prototype network model to obtain the fourth sample feature data, wherein at least one type of network layer in the first group of network layers does not belong to the second group of network layers; the fifth sample feature data is input into the third group of network layers of the Gaussian prototype network model to obtain the sixth sample feature data, wherein the sixth sample feature data includes a sample encoding vector and the sample covariance matrix of the sample encoding vector, wherein the third group of network layers includes at least one fully connected layer, wherein the number of network layers in the first group of network layers is greater than the number of network layers in the second group of network layers; The splicing module is configured to splice the third sample feature data and the fourth sample feature data to obtain the fifth sample feature data; The determination module is configured to determine the prototype of the Gaussian prototype network model corresponding to each fault category based on the sixth sample feature data of each sample service data in the first group of sample service data. The second process includes: By using seasonal trend decomposition, the data of each second indicator in each sample of business data in each group of sample business data is decomposed into multiple first time windows to obtain the second trend term, second period term and second residual term corresponding to each first time window. The first sample feature data is determined based on the second trend term and the second period term. The second sample feature data is determined based on the second residual term. The multiple sets of first sample business data are IPTV-related business data, and the multiple sets of second indicator data include IPTV user perception indicators, which include one or more of the following: IPTV network jitter count, number of stutters, duration exceeding the preset time required for alarm, media packet loss rate, latency factor, and number of handover failures.

23. An electronic device, comprising: Memory; as well as A processor coupled to the memory is configured to execute the method of any one of claims 1-20 based on instructions stored in the memory.

24. A computer-readable storage medium comprising computer program instructions, wherein, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1-20.

25. A computer program product comprising a computer program, wherein, When the computer program is executed by a processor, it implements the method described in any one of claims 1-20.

Images