Alarm convergence method and system for IT operation and maintenance

Through the neural network model, the alarm information of the railway intelligent operation and maintenance system is embedded vectorized and clustered, which solves the problems of complexity and redundancy of alarm information in the system, and realizes efficient convergence and accurate correlation of alarm information.

CN119026090APending Publication Date: 2024-11-26SHAANXI JINGSHEN RAILWAY CO LTD +1
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202410984566.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The alarm information generated in the railway intelligent operation and maintenance system is large in scale, many types, and complex in correlation. The existing technology is difficult to effectively converge, resulting in too much redundant alarm information and difficult to identify key alarms.

Method used

The Skip-Gram model based on neural network is used to preprocess and embed the alarm information vectorization. The clustering algorithm divides the alarm information with high correlation into different types of clusters to achieve the convergence of the alarm information.

Benefits of technology

Effectively reduce redundant alarm information, improve the compression rate and accuracy of alarm data association, and is suitable for railway intelligent operation and maintenance systems with complex relationships.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119026090A_ABST
    Figure CN119026090A_ABST
Patent Text Reader

Abstract

The invention discloses an alarm convergence method and system for IT operation and maintenance, and the method comprises the steps: storing collected alarm information into an established alarm database, and carrying out the preprocessing of the alarm information, and obtaining alarm information, namely an alarm sequence, combined according to a standard; obtaining a training sample according to the alarm sequence, training a Skip-Gram model constructed based on a neural network by using the training sample to obtain a weight matrix of a hidden layer, and further obtaining an embedded vector of the alarm information in a multi-dimensional space; the alarm information embedding vectors are divided into different clusters through clustering, the alarm information with high correlation is associated, and convergence of the alarm information is achieved; convergence of the alarm information is achieved, and gathering of related alarm information is facilitated.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of railway operation and maintenance, and in particular to an alarm convergence method and system for IT operation and maintenance. Background Art

[0002] With the advent of the digital age, the cloud-network integrated railway infrastructure and the intelligent operation and maintenance systems it carries have brought in a massive amount of alarm event information. Alarm processing consumes a lot of valuable operation and maintenance manpower. At the same time, it is accompanied by problems such as it is difficult to handle all the alarms, the massive alarms easily overwhelm the actual problems, and it is difficult to identify key alarms.

[0003] In the existing technology, related research on using association analysis algorithms such as Apriori to mine association rules between alarms and clustering algorithms to cluster alarms with high similarity to achieve the purpose of alarm convergence has been applied to operation and maintenance systems in different industries. However, there is still relatively little research on alarm convergence in railway intelligent operation and maintenance systems. The mining of alarm data is mostly focused on the prediction of equipment life, safety status prediction, etc., and is mostly targeted at a certain professional equipment, such as the overhead line.

[0004] The railway intelligent operation and maintenance system involves multi-professional equipment, and the alarm information generated is large in scale, diverse in types, and complex in correlation. There is an urgent need for an alarm information convergence method that can reduce redundant alarm information and facilitate the aggregation of related alarm information. Summary of the invention

[0005] The present invention aims to provide an alarm convergence method and system for IT operation and maintenance, so as to realize the convergence of alarm information and facilitate the aggregation of related alarm information.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention discloses an alarm convergence method for IT operation and maintenance, including:

[0008] The collected alarm information is stored in the established alarm database, and the alarm information is pre-processed to obtain the alarm information combined according to the standard, i.e., the alarm sequence;

[0009] Obtain training samples according to the alarm sequence, use the training samples to train the Skip-Gram model built based on the neural network to obtain the weight matrix of the hidden layer, and further obtain the embedding vector of the alarm information in the multi-dimensional space;

[0010] The warning information embedding vector is divided into different clusters through clustering, and the warning information with high correlation is associated to achieve the convergence of warning information;

[0011] Among them, the distance between the warning information embedding vectors indicates the strength of the correlation between the warning information.

[0012] Furthermore, the collected alarm information is stored in the established alarm database, including:

[0013] Collect railway-related warning information data and retain the timestamp, description information and generation object in the warning information;

[0014] According to the object generating the alarm information, the device-level tag of the alarm information is associated, the associated alarm information is stored in the established alarm database, and a unique ID is set for each alarm information;

[0015] If the description information and device-level labels of multiple alarm messages are the same, the type IDs of the multiple alarm messages are set to the same value;

[0016] The attributes of the alarm information include id, timestamp, description information, device-level tag and type id;

[0017] Equipment-level labels include railway bureau information, power supply section information, line information, station area information, uplink and downlink information, professional information and equipment number.

[0018] Furthermore, training samples are obtained according to the alarm sequence, and the Skip-Gram model constructed based on the neural network is trained using the training samples to obtain the weight matrix of the hidden layer, and further the embedded vector of the alarm information in the multi-dimensional space is obtained, including:

[0019] The training sample includes a sample label and input sample data, and the window size is specified to sample and obtain alarm information from the alarm sequence as the sample label, and any alarm information is extracted from an alarm sequence as the input sample data;

[0020] The Skip-Gram model includes an input layer, a hidden layer, an output layer and an activation layer. The training objective of the Skip-Gram model is to maximize the conditional probability of predicting a sample label using input sample data;

[0021] The training sample is input into the Skip-Gram model, the input layer X performs one-hot encoding on the input sample data, and the warning information is represented as an n-dimensional vector X; wherein n represents the total number of non-repetitive warning information;

[0022] The input layer maps the input sample data to the hidden layer H through the input matrix; each row of the input matrix represents an embedding of an input sample data, and the dimension of the input matrix W is (n×d); each hidden layer unit in the hidden layer H corresponds to the value of the input sample data under the set dimension in the embedding space, that is, H=WT X; where d represents the set embedding dimension;

[0023] The hidden layer H maps the input sample data represented by the hidden layer to the output layer Y through the output matrix W', each column of the output matrix represents a weight vector of a hidden layer unit, and the dimension of the output matrix W' is (d×n);

[0024] During the training process, the Skip-Gram model uses the back-propagation algorithm and the gradient descent algorithm, and through the joint work of the input matrix and the input matrix, the two weight matrices W and W' are continuously updated to achieve the purpose of minimizing the loss function, so that the Skip-Gram model is optimized to obtain the weight information of the two weight matrices W and W'; wherein, the weight information of the input matrix W at this time is the required alarm information embedding vector.

[0025] Furthermore, the hidden layer H outputs the value of each hidden layer unit corresponding to the input sample data under the set dimension in the embedding space to the output layer Y through the output matrix W', then each node in the output layer Y represents the score of the corresponding input sample data as the sample label, that is, Y = W' T H and dimension is n;

[0026] The output layer Y normalizes the scores through the sigmoid function and converts them into score probability values, that is, the input sample data is the score probability value of the sample label.

[0027] Furthermore, the objective function of the Skip-Gram model is to maximize the conditional probability of obtaining the sample label by using the input sample prediction. The objective function of the Skip-Gram model is:

[0028]

[0029] In the formula, N represents the total number of training samples, n′ represents the number of training samples, that is, the n′th training sample; A l represents the label sample, C(A c ) represents the input sample data A c The corresponding sample label set, A c represents the input sample, represents the input sample data in the n′th training sample, represents the sample label in the n′th training sample, Represents the Skip-Gram model based on the input sample data of the n′th training sample Predicted sample labels The conditional probability of .

[0030] Furthermore, the Skip-Gram model uses a negative sampling method to set a positive example and multiple negative examples for each training sample; wherein the sample label is selected as the positive example, and the randomly sampled non-sample label is used as the negative example;

[0031] The Skip-Gram model is trained using each training sample and the corresponding positive and negative examples, wherein the training objective of the Skip-Gram model is to maximize the probability of positive examples and minimize the probability of negative examples, so that the Skip-Gram model can accurately distinguish between positive examples and negative examples;

[0032] The training objective of the Skip-Gram model is transformed from a maximization problem to a minimization problem, that is, the training objective of the Skip-Gram model is to minimize the loss function:

[0033]

[0034] In the formula, Loss represents the minimization loss function, N represents the total number of training samples, and n′ represents the number of training samples, that is, the n′th training sample; Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Predicted sample labels The probability value of is the probability of a positive example; Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Predicted sample labels The score,

[0035] Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Non-sample labels are not predicted The sum of the probability values ​​of Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Predicting non-sample labels The score, Indicates the input sample data in the n′th training sample A collection of randomly sampled negative examples.

[0036] Furthermore, the warning information embedding vector is divided into different clusters through clustering, and the warning information with high correlation is associated to achieve the convergence of the warning information, including:

[0037] All warning information embedding vectors are regarded as points in space W = {W1…W n}, the connecting edges of any two points in W are grouped into a set E, that is, an undirected graph G(W,E) is constructed;

[0038] Based on the undirected graph G, an adjacency matrix V is established, and the adjacency matrix V is used to store the weights of each connecting edge in the undirected graph G;

[0039] Based on the undirected graph G, a degree matrix D is established, and the degree matrix D is used to store the degree of each point in the undirected graph G;

[0040] Calculate a normalized Laplace matrix L according to the adjacency matrix V and the degree matrix D, perform eigenvalue decomposition on the Laplace matrix L, and specify the dimension k after dimensionality reduction to obtain an eigenvector matrix corresponding to the first k smallest eigenvalues ​​of the Laplace matrix L; wherein the dimension of the eigenvector matrix is ​​(n×k);

[0041] The K-means clustering algorithm is used to cluster the feature vectors, and the clustering results are obtained by dividing them according to the set number of clusters;

[0042] The clustering result indicates the connection relationship between the alarm information. The alarm information in the same cluster has a strong correlation. Different alarm information sets are divided into different clusters and stored in the database to achieve convergence of the alarm information.

[0043] Furthermore, in spectral clustering, the closer the distance between any two points in the undirected graph, the greater the weight value of the connecting edge between the two points, and the weight of the connecting edge is:

[0044] V k,j =σ(-||W k -W j ||2),

[0045] Where V k,j Represents the warning information embedding vector W k and the warning information embedding vector W j The weight of the edge connecting them, ||W k -W j ||2 represents two warning information embedding vectors W k With W j The function of the Euclidean distance between k With W j They represent two warning information embedding vectors respectively, σ(x) represents the sigmoid function, which is a monotonically increasing function, so the input is set to the negative of the Euclidean distance between the vectors;

[0046] The degree of each point is the sum of the weights of the edges connecting the point, that is, the degree of each warning information embedding vector is:

[0047]

[0048] Where D kRepresents the warning information embedding vector W k degree, n represents the total number of non-repetitive alarm information, j represents the jth alarm information embedding vector; V k,j Represents the warning information embedding vector W k and the warning information embedding vector W j The weight of the connecting edge;

[0049] The Laplace matrix L is:

[0050] L=ID -1 / 2 VD -1 / 2 ,

[0051] Where I represents the identity matrix, D represents the degree matrix, and V represents the adjacency matrix.

[0052] In a second aspect, an embodiment of the present invention discloses an alarm convergence system for IT operation and maintenance, including:

[0053] A collection and processing unit, used to store the collected alarm information into an established alarm database, and pre-process the alarm information to obtain alarm information combined according to standards, i.e., an alarm sequence;

[0054] An embedding unit is used to obtain training samples according to the alarm sequence, use the training samples to train the Skip-Gram model built based on the neural network to obtain the weight matrix of the hidden layer, and further obtain the embedding vector of the alarm information in the multi-dimensional space;

[0055] A clustering convergence unit, used to divide the alarm information embedding vector into different clusters by clustering, and associate the alarm information with higher correlation to achieve convergence of the alarm information;

[0056] Among them, the distance between the warning information embedding vectors indicates the strength of the correlation between the warning information.

[0057] On the third aspect, based on the same inventive concept, an embodiment of the present invention also provides an electronic device, including a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program or instructions to implement the steps of the aforementioned alarm convergence method for IT operation and maintenance.

[0058] In a fourth aspect, based on the same inventive concept, an embodiment of the present invention further provides a computer storage medium, in which a computer program or instruction is stored. When the computer program or instruction is executed by a processor, the steps of the aforementioned alarm convergence method for IT operation and maintenance are implemented.

[0059] In a fifth aspect, based on the same inventive concept, an embodiment of the present invention further provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of the aforementioned alarm convergence method for IT operation and maintenance.

[0060] The technical effects and advantages of the present invention are as follows: compared with the prior art using association analysis algorithms such as Apriori, the present invention has more advantages in processing complex and large-scale data through a neural network model. By learning advanced features and patterns, it is suitable for scenarios where complex relationships exist in alarm information, and is more suitable for alarm information processing in a railway intelligent operation and maintenance system with complex association forms. Compared with mining frequent items, the present invention embeds alarm information as a vector in a high-dimensional space, which can more accurately analyze the correlation between alarm information and effectively improve the alarm data association compression rate and accuracy.

[0061] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0063] Figure 1 The process of an alarm convergence method for IT operation and maintenance according to an embodiment of the present invention is as follows Figure 1 ;

[0064] Figure 2 The process of an alarm convergence method for IT operation and maintenance according to an embodiment of the present invention is as follows Figure 2 ;

[0065] Figure 3 It is a schematic diagram of an embedded process of an alarm convergence method for IT operation and maintenance according to an embodiment of the present invention;

[0066] Figure 4 A schematic diagram of a clustering process of an alarm convergence method for IT operation and maintenance according to an embodiment of the present invention;

[0067] Figure 5 This is a structural diagram of an alarm convergence system for IT operation and maintenance according to an embodiment of the present invention;

[0068] Figure 6The figure is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] In order to solve the shortcomings of the prior art, the present invention discloses an alarm convergence method for IT operation and maintenance, such as Figure 1 and Figure 2 As shown, the following steps are included:

[0071] Step S1: storing the collected alarm information into the established alarm database, and preprocessing the alarm information to obtain alarm information combined according to the standard, i.e., an alarm sequence;

[0072] Step S2: obtaining training samples according to the alarm sequence, using the training samples to train the Skip-Gram model constructed based on the neural network to obtain the weight matrix of the hidden layer, and further obtaining the embedding vector of the alarm information in the multi-dimensional space;

[0073] Step S3: dividing the warning information embedding vector into different clusters through clustering, and associating the warning information with higher correlation to achieve convergence of the warning information;

[0074] Among them, the distance between the warning information embedding vectors indicates the strength of the correlation between the warning information.

[0075] In some specific embodiments, step S1: storing the collected alarm information into an established alarm database, and preprocessing the alarm information to obtain alarm information combined according to a standard, i.e., an alarm sequence, includes:

[0076] Step S11: storing the collected alarm information into the established alarm database. The specific operations are as follows:

[0077] Collect the alarm information data transmitted from the railway related information management system, and retain the timestamp, description information and generated object in the alarm information;

[0078] According to the object generating the alarm information, the device-level tag of the alarm information is associated, the associated alarm information is stored in the established alarm database, and a unique id is set for each alarm information as an identifier in the alarm database;

[0079] In order to converge the alarm information later, the data in the alarm database needs to be further processed, that is, if the description information and device-level labels of multiple alarm information are the same, the type IDs of multiple alarm information are set to the same value;

[0080] The attributes of the alarm information include id, timestamp, description information, device-level tag and type id, as shown in Table 1:

[0081] Table 1 Alarm information attribute table

[0082] Alarm Information Attributes illustrate alert_id Alarm information id alert_time Alarm information timestamp alert_message Alarm information description alert_tag Alarm information associated device tag alert_type_id Alarm information type id ;

[0083] Equipment-level labels include railway bureau information, power supply section information, line information, station area information, uplink and downlink information, professional information and equipment number.

[0084] Step S12: pre-process the alarm information to obtain alarm information combined according to the standard, that is, an alarm sequence. The specific operation is as follows:

[0085] Preprocessing includes alarm data cleaning and alarm sequence segmentation;

[0086] First, the alarm information is cleaned, mainly to fill in the missing data, remove blank data, and delete redundant data. Redundant data refers to excessively repeated alarm information in a short period of time. Excessively repeated alarm information will affect the subsequent embedding results.

[0087] Then the alarm information is segmented into alarm sequences, which is mainly to combine the alarm information into alarm sequences according to certain standards, so as to generate training samples later;

[0088] When the time interval between two alarm messages is too long, it is meaningless to assume that there is a relationship between the two alarm messages. Similarly, if the devices associated with the two alarm messages are too far apart, it is meaningless to assume that there is a relationship between the two alarm messages.

[0089] Therefore, it is necessary to set a time threshold and a subordinate relationship threshold for the alarm information, collect the alarm information under the same time window and the same subordinate relationship, and combine them to generate an alarm sequence;

[0090] For example, if the time threshold is set to 60s, then the time windows are such as [0,60], [60,120]..., and the alarm information in the same time window meets the time threshold constraint of being combined into the same alarm sequence; similarly, if the subordinate relationship threshold is set to "professional", then as long as the alarm information of the device-level label "professional" and the subordinate "railway bureau", "power supply section", "line", "station area", "uplink and downlink" are the same, they all meet the subordinate threshold constraint of being combined into the same alarm sequence.

[0091] For the alarm information in the segmented alarm sequence, only the type id attribute of the alarm information is concerned, and the alarm information with the same attribute is considered to be the same entity.

[0092] In some specific embodiments, step S2: obtaining training samples according to the alarm sequence, using the training samples to train the Skip-Gram model constructed based on the neural network to obtain the weight matrix of the hidden layer, and further obtaining the embedded vector of the alarm information in the multidimensional space; Figure 3 As shown, the specific operations are as follows:

[0093] The present invention adopts the modeling idea and training method of the Word2Vec semantic model to describe the correlation between warning information. By establishing an embedding space, the warning information is mapped to a new multidimensional space, so that the warning information with strong correlation has a smaller distance in the space.

[0094] The modeling idea and training method of the Word2Vec semantic model is to map each word in the vocabulary into a word vector in a multi-dimensional real vector space, and use the central word as input data to predict the surrounding context words;

[0095] In the present invention, the alarm sequence corresponds to a "sentence", the alarm information corresponds to a "word", and different alarm information constitutes a "lexicon". In order to make the vectors of "words" that frequently appear in the same "sentence" at the same time have a higher similarity in the embedding space, the present invention adopts the Skip-Gram model constructed by a neural network trained with negative sampling.

[0096] Sampling the alarm sequence to obtain training samples, where the training samples include sample labels and input sample data;

[0097] Among them, the alarm information is sampled from the alarm sequence by specifying the window size as the sample label; and any alarm information is extracted from an alarm sequence as the input sample data.

[0098] The training goal of the Skip-Gram model is to maximize the conditional probability of predicting sample labels using input sample data; the training is completed by inputting paired training samples into the neural network, and after the Skip-Gram model training is completed, the parameters learned by the training samples, that is, the weight matrix of the hidden layer, are obtained, and the embedded vector of the alarm information in the multidimensional space is further obtained; wherein, the distance between the embedded vectors of the alarm information indicates the strength of the correlation between the alarm information.

[0099] like Figure 3As shown, the Skip-Gram model includes an input layer, a hidden layer, an output layer and an activation layer. The training sample is input into the Skip-Gram model. The input layer X performs one-hot encoding on the input sample data to facilitate machine recognition. At the same time, the alarm information can be represented as an n-dimensional vector X, and n is the total number of non-repeated alarm information. For example, the alarm information A k Corresponding to the input layer node x k =1, and the other input layer nodes are vectors of 0; that is, X represents the alarm information vector, and x represents each flag bit of the alarm information vector, for example, the first alarm information X1={1,0,0,...0}.

[0100] The input layer maps the input sample data to the hidden layer H through the input matrix; each row of the input matrix represents the embedding of an input sample data, and the embedding dimension is set to d, then the dimension of the input matrix W is (n×d).

[0101] Each hidden layer unit in the hidden layer H corresponds to the value of the input sample data under the set dimension in the embedding space, that is, H = W T ·X; where the dimension is d.

[0102] The hidden layer H maps the input sample data represented by the hidden layer to the output layer Y through the output matrix W', each column of the output matrix represents a weight vector of a hidden layer unit, and the dimension of the output matrix W' is (d×n);

[0103] The hidden layer H outputs the value of each hidden layer unit corresponding to the input sample data under the set dimension in the embedding space to the output layer Y through the output matrix W', then each node in the output layer Y represents the score of the corresponding input sample data as the sample label, that is, Y = W' T H and dimension is n;

[0104] The output layer Y normalizes the scores through the sigmoid function and converts them into score probability values, that is, the input sample data is the score probability value of the sample label.

[0105] The objective function of the Skip-Gram model is to maximize the conditional probability of obtaining the sample label by predicting the input sample. The objective function of the Skip-Gram model is:

[0106]

[0107] In the formula, N represents the total number of training samples, n′ represents the number of training samples, that is, the n′th training sample; Al represents the label sample, C(A c ) represents the input sample data A cThe corresponding sample label set, A c represents the input sample, represents the input sample data in the n′th training sample, represents the sample label in the n′th training sample, Represents the Skip-Gram model based on the input sample data of the n′th training sample Predicted sample labels The conditional probability of .

[0108] The Skip-Gram model uses a negative sampling method to set a positive example and multiple negative examples for each training sample; wherein the sample label is selected as the positive example, and the randomly sampled non-sample label is used as the negative example;

[0109] The Skip-Gram model is trained using each training sample and the corresponding positive and negative examples. The training objective of the Skip-Gram model is to maximize the probability of positive examples (i.e., maximize the conditional probability of obtaining sample labels by using the input samples to predict) and minimize the probability of negative examples (i.e., minimize the conditional probability of obtaining non-sample labels by using the input samples to predict), so that the Skip-Gram model can accurately distinguish between positive and negative examples.

[0110] The training objective of the Skip-Gram model is transformed from a maximization problem to a minimization problem, that is, the training objective of the Skip-Gram model is to minimize the loss function:

[0111]

[0112] In the formula, Loss represents the minimization loss function, N represents the total number of training samples, and n′ represents the number of training samples, that is, the n′th training sample; Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Predicted sample labels The probability value of is the probability of a positive example; Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Predicted sample labels The score, Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Non-sample labels are not predicted The sum of the probability values ​​of Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Predicting non-sample labels The score, Indicates the input sample data in the n′th training sample A collection of randomly sampled negative examples.

[0113] During the training process, the Skip-Gram model uses the back-propagation algorithm and the gradient descent algorithm, and through the joint work of the input matrix and the input matrix, the two weight matrices W and W' are continuously updated to achieve the purpose of minimizing the loss function, so that the Skip-Gram model is optimized to obtain the weight information of the two weight matrices W and W'; wherein, the weight information of the input matrix W at this time is the required alarm information embedding vector.

[0114] In some specific embodiments, step S3: divide the warning information embedding vector into different clusters by clustering, and associate the warning information with higher correlation to achieve convergence of the warning information; Figure 4 As shown, the specific operations are as follows:

[0115] All warning information embedding vectors are regarded as points in space W = {W1…W n}; Connect any two points in W through edges to form a set E, that is, construct an undirected graph G(W,E);

[0116] An adjacency matrix V is established based on the undirected graph G, and the adjacency matrix V is used to store the weights of each connecting edge in the undirected graph G; wherein, in spectral clustering, the closer the distance between any two points in the undirected graph, the greater the weight value of the connecting edge between the two points, and the weight of the connecting edge is:

[0117] V k,j =σ(-||W k -W j ||2),

[0118] Where V k,j Represents the warning information embedding vector W k and the warning information embedding vector W j The weight of the edge connecting them, ||W k -W j ||2 represents two warning information embedding vectors W k With W j The function of the Euclidean distance between k With W j They represent two warning information embedding vectors respectively, and σ(x) represents the sigmoid function, which is a monotonically increasing function. Therefore, the input is set to the negative of the Euclidean distance between the vectors.

[0119] A degree matrix D is established based on the undirected graph G, and the degree matrix D is used to store the degree of each point in the undirected graph G; wherein the degree of each point is the sum of the weights of the edges connecting the point, that is, the degree of each warning information embedding vector is:

[0120]

[0121] Where D k Represents the warning information embedding vector W k degree, n represents the total number of non-repetitive alarm information, j represents the jth alarm information embedding vector; V k,j Represents the warning information embedding vector W k and the warning information embedding vector W j The weight of the connecting edge.

[0122] The normalized Laplace matrix L is calculated based on the adjacency matrix V and the degree matrix D:

[0123] L=ID -1 / 2 WD -1 / 2 ,

[0124] Where I represents the identity matrix, D represents the degree matrix, and V represents the adjacency matrix.

[0125] Perform eigenvalue decomposition on the Laplace matrix L, and specify the dimension k after dimensionality reduction to obtain the eigenvector matrix corresponding to the first k smallest eigenvalues ​​of the Laplace matrix L; wherein the dimension of the eigenvector matrix is ​​(n×k).

[0126] The K-means clustering algorithm is used to cluster the feature vectors, and the clustering results are obtained by dividing them according to the set number of clusters;

[0127] The clustering result indicates the connection relationship between the alarm information. The alarm information in the same cluster has a strong correlation. Different alarm information sets are divided into different clusters and stored in the database to achieve convergence of the alarm information.

[0128] As shown in Table 2, at this time, the alarm information set includes two attributes: the alarm information set id and the alarm information type id set, wherein the alarm information set id is the unique identifier of the alarm information set in the database, and the alarm information type id set is the set of alarm information type id attributes of the clustered alarm information;

[0129] Table 2 Alarm information set attribute table

[0130] Alarm information set illustrate alert_relationship_id Alarm information set id alert_type_ids Alarm information type id set

[0131] There are a large number of alarm information production formulas under the same event window and the same subordinate relationship. The alarm information with the attribute alarm information type id in the same alarm information set is summarized and combined into a work order and sent to the operation and maintenance personnel.

[0132] The complexity of multidimensional data clustering processed by spectral clustering in the embodiment of the present invention is better than that of traditional clustering algorithms. Spectral clustering divides a graph into different subgraphs, so that the connections within the subgraphs are tighter and the connections between subgraphs are sparser, so as to achieve effective clustering.

[0133] Based on the same inventive concept, the embodiment of the present invention also provides an alarm convergence system for IT operation and maintenance, such as Figure 5 As shown, including:

[0134] A collection and processing unit, used to store the collected alarm information into an established alarm database, and pre-process the alarm information to obtain alarm information combined according to standards, i.e., an alarm sequence;

[0135] An embedding unit is used to obtain training samples according to the alarm sequence, use the training samples to train the Skip-Gram model built based on the neural network to obtain the weight matrix of the hidden layer, and further obtain the embedding vector of the alarm information in the multi-dimensional space;

[0136] A clustering convergence unit, used to divide the alarm information embedding vector into different clusters by clustering, and associate the alarm information with higher correlation to achieve convergence of the alarm information;

[0137] Among them, the distance between the warning information embedding vectors indicates the strength of the correlation between the warning information.

[0138] Regarding the system in the above embodiment, the specific manner in which each unit module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0139] Based on the same inventive concept, an embodiment of the present invention further provides an electronic device, whose structure is as follows: Figure 6 As shown, it includes a memory, a processor and a computer program stored in the memory, and the processor executes the computer program or instructions to implement the aforementioned alarm convergence method for IT operation and maintenance.

[0140] Based on the same inventive concept, an embodiment of the present invention further provides a computer storage medium, in which a computer program or instruction is stored. When the computer program or instruction is executed by a processor, the steps of the aforementioned alarm convergence method for IT operation and maintenance are implemented.

[0141] Based on the same inventive concept, an embodiment of the present invention further provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements the steps of the aforementioned alarm convergence method for IT operation and maintenance.

[0142] The embodiment of the present invention provides an alarm convergence method for analyzing, processing and summarizing alarm information, thereby reducing redundant alarm information and summarizing relevant alarm information, thereby helping the operation and maintenance team to identify and solve problems more quickly.

[0143] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An alarm convergence method for IT operation and maintenance, characterized in that: include: The collected alarm information is stored in the established alarm database, and the alarm information is pre-processed to obtain the alarm information combined according to the standard, i.e., the alarm sequence; Obtain training samples according to the alarm sequence, use the training samples to train the Skip-Gram model built based on the neural network to obtain the weight matrix of the hidden layer, and further obtain the embedding vector of the alarm information in the multi-dimensional space; The warning information embedding vector is divided into different clusters through clustering, and the warning information with high correlation is associated to achieve the convergence of warning information; Among them, the distance between the warning information embedding vectors indicates the strength of the correlation between the warning information.

2. According to the IT operation and maintenance-oriented alarm convergence method of claim 1, it is characterized in that: The collected alarm information is stored in the established alarm database, including: Collect railway-related warning information data and retain the timestamp, description information and generation object in the warning information; According to the object generating the alarm information, the device-level tag of the alarm information is associated, the associated alarm information is stored in the established alarm database, and a unique ID is set for each alarm information; If the description information and device-level labels of multiple alarm messages are the same, the type IDs of the multiple alarm messages are set to the same value; The attributes of the alarm information include id, timestamp, description information, device-level tag and type id; Equipment-level labels include railway bureau information, power supply section information, line information, station area information, uplink and downlink information, professional information and equipment number.

3. The alarm convergence method for IT operation and maintenance according to claim 1 is characterized in that: According to the alarm sequence, training samples are obtained, and the Skip-Gram model constructed based on the neural network is trained using the training samples to obtain the weight matrix of the hidden layer, and further the embedded vector of the alarm information in the multi-dimensional space is obtained, including: The training sample includes a sample label and input sample data, and the window size is specified to sample and obtain alarm information from the alarm sequence as the sample label, and any alarm information is extracted from an alarm sequence as the input sample data; The Skip-Gram model includes an input layer, a hidden layer, an output layer and an activation layer. The training objective of the Skip-Gram model is to maximize the conditional probability of predicting a sample label using input sample data; The training sample is input into the Skip-Gram model, the input layer X performs one-hot encoding on the input sample data, and the warning information is represented as an n-dimensional vector X; wherein n represents the total number of non-repetitive warning information; The input layer maps the input sample data to the hidden layer H through the input matrix; each row of the input matrix represents an embedding of an input sample data, and the dimension of the input matrix W is (n×d); each hidden layer unit in the hidden layer H corresponds to the value of the input sample data under the set dimension in the embedding space, that is, H=W T X; where d represents the set embedding dimension; The hidden layer H maps the input sample data represented by the hidden layer to the output layer Y through the output matrix W', each column of the output matrix represents a weight vector of a hidden layer unit, and the dimension of the output matrix W' is (d×n); During the training process, the Skip-Gram model uses the back-propagation algorithm and the gradient descent algorithm, and through the joint work of the input matrix and the input matrix, the two weight matrices W and W' are continuously updated to achieve the purpose of minimizing the loss function, so that the Skip-Gram model is optimized to obtain the weight information of the two weight matrices W and W'; wherein, the weight information of the input matrix W at this time is the required alarm information embedding vector.

4. The alarm convergence method for IT operation and maintenance according to claim 3 is characterized in that: The hidden layer H outputs the value of each hidden layer unit corresponding to the input sample data under the set dimension in the embedding space to the output layer Y through the output matrix W', then each node in the output layer Y represents the score of the corresponding input sample data as the sample label, that is, Y = W' T H and dimension is n; The output layer Y normalizes the scores through the sigmoid function and converts them into score probability values, that is, the input sample data is the score probability value of the sample label.

5. The alarm convergence method for IT operation and maintenance according to claim 1, characterized in that: The objective function of the Skip-Gram model is to maximize the conditional probability of obtaining the sample label by predicting the input sample. The objective function of the Skip-Gram model is: In the formula, N represents the total number of training samples, n′ represents the number of training samples, that is, the n′th training sample; A l represents the label sample, C(A c ) represents the input sample data A c The corresponding sample label set, A c represents the input sample, represents the input sample data in the n′th training sample, represents the sample label in the n′th training sample, Represents the Skip-Gram model based on the input sample data of the n′th training sample Predicted sample labels The conditional probability of .

6. The alarm convergence method for IT operation and maintenance according to claim 4 or 5, characterized in that: The Skip-Gram model uses a negative sampling method to set a positive example and multiple negative examples for each training sample; wherein the sample label is selected as the positive example, and the randomly sampled non-sample label is used as the negative example; The Skip-Gram model is trained using each training sample and the corresponding positive and negative examples, wherein the training objective of the Skip-Gram model is to maximize the probability of positive examples and minimize the probability of negative examples, so that the Skip-Gram model can accurately distinguish between positive examples and negative examples; The training objective of the Skip-Gram model is transformed from a maximization problem to a minimization problem, that is, the training objective of the Skip-Gram model is to minimize the loss function: In the formula, Loss represents the minimized loss function, N represents the total number of training samples, and n′ represents the number of training samples, that is, the n′th training sample; Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Predicted sample labels The probability value of is the probability of a positive example; Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Predicted sample labels The score, Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Non-sample labels are not predicted The sum of the probability values ​​of Indicates that the Skip-Gram model is based on the input sample data in the n′th training sample Predicting non-sample labels The score, Indicates the input sample data in the n′th training sample A collection of randomly sampled negative examples.

7. The alarm convergence method for IT operation and maintenance according to claim 1, characterized in that: The alarm information embedding vector is divided into different clusters through clustering, and the alarm information with high correlation is associated to achieve the convergence of alarm information, including: All warning information embedding vectors are regarded as points in space W = {W1…W n }, the connecting edges of any two points in W are grouped into a set E, that is, an undirected graph G(W,E) is constructed; Based on the undirected graph G, an adjacency matrix V is established, and the adjacency matrix V is used to store the weights of each connecting edge in the undirected graph G; Based on the undirected graph G, a degree matrix D is established, and the degree matrix D is used to store the degree of each point in the undirected graph G; Calculate a normalized Laplace matrix L according to the adjacency matrix V and the degree matrix D, perform eigenvalue decomposition on the Laplace matrix L, and specify the dimension k after dimensionality reduction to obtain an eigenvector matrix corresponding to the first k smallest eigenvalues ​​of the Laplace matrix L; wherein the dimension of the eigenvector matrix is ​​(n×k); The K-means clustering algorithm is used to cluster the feature vectors, and the clustering results are obtained by dividing them according to the set number of clusters; The clustering result indicates the connection relationship between the alarm information. The alarm information in the same cluster has a strong correlation. Different alarm information sets are divided into different clusters and stored in the database to achieve convergence of the alarm information.

8. The alarm convergence method for IT operation and maintenance according to claim 7, characterized in that: In spectral clustering, the closer the distance between any two points in the undirected graph, the greater the weight value of the connecting edge between the two points, and the weight of the connecting edge is: Where V k,j Represents the warning information embedding vector W k and the warning information embedding vector W j The weight of the edge connecting them, ||W k -W j ||2 represents two warning information embedding vectors W k With W j The function of the Euclidean distance between k With W j They represent two warning information embedding vectors respectively, σ(x) represents the sigmoid function, which is a monotonically increasing function, so the input is set to the negative of the Euclidean distance between the vectors; The degree of each point is the sum of the weights of the edges connecting the point, that is, the degree of each warning information embedding vector is: Where D k Represents the warning information embedding vector W k degree, n represents the total number of non-repetitive alarm information, j represents the jth alarm information embedding vector; V k,j Represents the warning information embedding vector W k and the warning information embedding vector W j The weight of the connecting edge; The Laplace matrix L is: L=I-D -1 / 2 VD -1 / 2 , Where I represents the identity matrix, D represents the degree matrix, and V represents the adjacency matrix.

9. An alarm convergence system for IT operation and maintenance, characterized in that: include: A collection and processing unit, used to store the collected alarm information into an established alarm database, and pre-process the alarm information to obtain alarm information combined according to standards, i.e., an alarm sequence; An embedding unit is used to obtain training samples according to the alarm sequence, use the training samples to train the Skip-Gram model built based on the neural network to obtain the weight matrix of the hidden layer, and further obtain the embedding vector of the alarm information in the multi-dimensional space; A clustering convergence unit, used to divide the alarm information embedding vector into different clusters by clustering, and associate the alarm information with higher correlation to achieve convergence of the alarm information; Among them, the distance between the warning information embedding vectors indicates the strength of the correlation between the warning information.

10. An electronic device, characterized in that: It includes a memory, a processor and a computer program stored in the memory, and the processor executes the computer program or instructions to implement the steps of an alarm convergence method for IT operation and maintenance as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program or instruction, and when the computer program or instruction is executed by the processor, the steps of the alarm convergence method for IT operation and maintenance described in any one of claims 1 to 8 are implemented.

12. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by the processor, the steps of an alarm convergence method for IT operation and maintenance as described in any one of claims 1-8 are implemented.

Citation Information

Cited By

  • Service work order classification method, electronic equipment, storage medium and program product

    CN119377412A

  • Service work order classification method, electronic device, storage medium and program product

    CN119377412B

  • Process alarm optimization and disposal method and device based on word vector alarm clustering

    CN119848259A