Server alarm information classification method and system and related equipment

Abstract

The invention discloses a server alarm information classification method and system and related equipment, and relates to the field of server operation and maintenance, and the method comprises the steps: obtaining real-time server alarm information, inputting the real-time server alarm information into a preset encoder, obtaining a semantic coding vector outputted by the preset encoder, and outputting the semantic coding vector to a server; the preset encoder is obtained based on operation and maintenance manual data training after random full-word masking; inputting the semantic coding vector into a preset classification model to obtain a category label output by the preset classification model, the preset classification model being obtained by training based on a category label of historical server alarm information and a historical semantic coding vector output by a preset encoder; and taking the category corresponding to the category label as the fault category corresponding to the real-time server alarm information. According to the method, the real-time server alarm information is understood based on the encoder adapting to the statement characteristics in the server operation and maintenance field, the semantic coding vectors are output, then the semantic coding vectors are classified through the preset classification model, and the operation and maintenance time of the server is shortened.

Description

Technical Field

[0001] The present application relates to the field of server operation and maintenance technology, and in particular to a classification method, system and related equipment for server alarm information. Background Art

[0002] With the development of Internet technology, the digital transformation of enterprises has gradually expanded, resulting in an increasing number of types and numbers of servers used by enterprises to process business. Enterprises need to manage and maintain the computer rooms where the servers are located to ensure the stable operation of the servers.

[0003] Traditional operation and maintenance work is mainly done by operation and maintenance personnel. They determine the operation problems of the server through the server alarm information and maintain the server based on their experience. Server alarm information refers to the notification or warning information automatically sent by the server when an abnormal situation occurs during the operation of the server or exceeds the set threshold. Server alarm information is an important basis for maintaining and restoring the safe and stable operation of the equipment in the computer room.

[0004] Traditional operation and maintenance work is mainly based on manual work and manual experience. The operation and maintenance personnel determine the corresponding server fault category based on the server alarm information based on their experience, and then locate the fault for maintenance. Therefore, the operation and maintenance personnel need to spend time to accurately locate the server fault, making the traditional operation and maintenance method time-consuming. Summary of the invention

[0005] In view of the above problems, this application provides a server alarm information classification method, system and related equipment to achieve the purpose of reducing the time consumption of traditional operation and maintenance methods. The specific solution is as follows:

[0006] A first aspect of the present application provides a method for classifying server alarm information, the method comprising:

[0007] Acquire real-time server alarm information, and input the real-time server alarm information into a preset encoder to obtain a semantic encoding vector output by the preset encoder, wherein the preset encoder is obtained after training based on operation and maintenance manual data after random full-word masking;

[0008] Input the semantic coding vector into a preset classification model to obtain a category label output by the preset classification model, wherein the preset classification model is trained based on the category labeling of the historical server alarm information and the historical semantic coding vector output by the preset encoder, and the historical semantic coding vector is a semantic coding vector obtained after the historical server alarm information with the category label is input into the preset encoder;

[0009] The category corresponding to the category label is used as the fault category corresponding to the real-time server alarm information.

[0010] In a possible implementation, the training process of the preset encoder includes:

[0011] Obtain the server's operation and maintenance manual data and organize the data;

[0012] Performing random full-word masking on the operation and maintenance manual data after data sorting to obtain operation and maintenance training data;

[0013] The encoder is trained using the operation and maintenance training data to obtain the preset encoder, the input of the preset encoder is: the real-time server alarm information, and the output of the preset encoder is: a semantic coding vector representing the real-time server alarm information.

[0014] In a possible implementation, the obtaining and arranging the operation and maintenance manual data of the server includes:

[0015] Obtain the operation and maintenance manual data from the operation and maintenance manuals of various manufacturers;

[0016] Removing all structured data in the operation and maintenance manual data that is not related to the server, removing character strings in the operation and maintenance manual data that affect the training of the encoder using regular matching, and dividing the operation and maintenance manual data into multiple sentences according to periods in the operation and maintenance manual data;

[0017] A word segmentation tool is used to segment the multiple sentences obtained by segmentation, and the word segmentation of the word segmentation tool is based on word data in the server operation and maintenance field dictionary.

[0018] In a possible implementation, the training process of the preset classification model includes:

[0019] Obtain historical server alarm information and mark its categories;

[0020] Obtaining classification training data, wherein the classification training data is historical server alarm information with category annotations;

[0021] Inputting the historical server alarm information with category annotation into the preset encoder to obtain the historical semantic encoding vector output by the preset encoder;

[0022] According to the category labeling of the historical server alarm information and the historical semantic coding vector, the constructed classification model is trained to obtain the preset classification model, the input of the preset classification model is: the semantic coding vector of the real-time server alarm information, and the output of the preset classification model is: the category label corresponding to the real-time server alarm information.

[0023] In a possible implementation, obtaining the historical server alarm information and marking the category includes:

[0024] The historical server alarm information is obtained, and the historical server alarm information is categorized using a multi-round iterative labeling technique.

[0025] A second aspect of the present application provides a classification system for server alarm information, the classification system for server alarm information comprising:

[0026] An acquisition unit, used for acquiring real-time server alarm information, and inputting the real-time server alarm information into a preset encoder to obtain a semantic encoding vector output by the preset encoder, wherein the preset encoder is obtained after training based on the operation and maintenance manual data after random full-word masking;

[0027] A classification unit, used for inputting the semantic coding vector into a preset classification model to obtain a category label output by the preset classification model, wherein the preset classification model is trained based on the category labeling of the historical server alarm information and the historical semantic coding vector output by the preset encoder, and the historical semantic coding vector is a semantic coding vector obtained after the historical server alarm information with the category labeling is input into the preset encoder;

[0028] The output unit is used to use the category corresponding to the category label as the fault category corresponding to the real-time server alarm information.

[0029] In a possible implementation, the first training module of the preset encoder is further included, and the first training module is configured to execute:

[0030] Acquire the operation and maintenance manual data of the server and organize the data, perform random full-word masking on the organized operation and maintenance manual data to obtain operation and maintenance training data, use the operation and maintenance training data to train an encoder to obtain the preset encoder, the input of the preset encoder is: the real-time server alarm information, and the output of the preset encoder is: a semantic encoding vector representing the real-time server alarm information.

[0031] In a possible implementation, the first training module obtains the operation and maintenance manual data of the server and organizes the data, and the specific configuration is as follows:

[0032] The operation and maintenance manual data is obtained from the operation and maintenance manuals of various manufacturers, and structured data in the operation and maintenance manual data that is not related to the server is removed as a whole. Regular matching is used to remove character strings in the operation and maintenance manual data that affect the training of the encoder, and the operation and maintenance manual data is segmented into multiple sentences according to the periods in the operation and maintenance manual data. A word segmentation tool is used to segment the multiple sentences obtained by segmentation, and the word segmentation basis of the word segmentation tool is the word data in the server operation and maintenance field dictionary.

[0033] A third aspect of the present application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0034] The memory is used to store computer programs;

[0035] The processor is used to execute the computer program so that the electronic device can implement the method for classifying server alarm information of the above-mentioned first aspect or any implementation manner of the first aspect.

[0036] The fourth aspect of the present application provides a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements the method for classifying server alarm information of the above-mentioned first aspect or any implementation method of the first aspect.

[0037] By means of the above technical scheme, the present application provides a classification method, system and related equipment for server alarm information. The method uses a preset encoder and a preset classification model to classify real-time server alarm information. The preset encoder is trained with the operation and maintenance manual data of the server, so that the preset encoder can adapt to the sentence characteristics in the field of server operation and maintenance, and output a semantic coding vector representing the sentence meaning of the real-time server alarm information. The preset classification model is trained according to the category labeling of the server alarm information and the preset encoder for the historical semantic coding vector of the historical server alarm information, so that the preset classification model can adapt to the semantic coding vector of the preset encoder for classification, realize the effective classification of service alarm information, help operation and maintenance personnel to perform operation and maintenance work, and effectively reduce the operation and maintenance time of traditional operation and maintenance methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0039] Figure 1 A flow chart of a method for classifying server alarm information provided in an embodiment of the present application;

[0040] Figure 2 A schematic diagram of the structure of an OP-BERT model provided in an embodiment of the present application;

[0041] Figure 3 A schematic diagram of a process for processing operation and maintenance manual data provided in an embodiment of the present application;

[0042] Figure 4 A schematic diagram of the structure of a DPCNN provided in an embodiment of the present application;

[0043] Figure 5 A schematic diagram of the structure of a server alarm information classification system provided in an embodiment of the present application;

[0044] Figure 6 A hardware structure block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] The following describes the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation method section of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0046] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0047] The terms "first", "second" etc. in the specification of the application and the above-mentioned drawings are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable in appropriate circumstances, and this is only to describe the distinction mode adopted by the objects of the same attributes when describing in the embodiments of the application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0048] With the rapid development of Internet technology and the gradual entry of enterprise digital transformation, the types and number of servers used by enterprises to process business are increasing year by year, which requires more server operation and maintenance personnel. Since the 21st century, we have actively promoted the development of digital infrastructure and promoted research in the field of basic IT operation and maintenance.

[0049] Server alarm information refers to the notification or warning information that the server automatically sends to the monitoring platform when an abnormal situation occurs during server operation or exceeds the set threshold. It can include alarm information at the hardware, operating system, and application levels. Server alarm information is an important basis for maintaining and restoring the safe and stable operation of equipment in the computer room. Operation and maintenance personnel analyze service alarm information based on experience, determine the corresponding server fault category based on server alarm information, locate faults, or deal with potential safety hazards.

[0050] Traditional operation and maintenance work usually relies mainly on manual labor and manual experience, and there is a phenomenon of "treating the symptoms rather than the root cause". There is a lack of statistics and analysis of different types of historical service alarm information. Operation and maintenance personnel need to spend time to accurately locate server failures, making it difficult for traditional operation and maintenance models to keep up with the pace of digital development. Especially since the information technology application innovation industry has been put into practice, the development momentum of servers has been rapid, and servers of different types and models have been gradually put into production and use.

[0051] Furthermore, in recent years, natural language processing (NLP) technology has been widely used in various fields (such as machine translation, speech recognition, question-answering systems, and text classification) and has achieved good results. Its application method is mainly to use pre-trained models to fine-tune specific tasks. However, the application of natural language processing technology to the automatic classification of server alarm information has the following difficulties: the corpus in the field of server operation and maintenance has the characteristics of fuzzy entity boundaries, strong terminology, and long-tail distribution of data; the model trained with corpus in general fields is difficult to generalize; the classification system of server alarm information is complex and needs to be customized according to actual conditions; there is little research specifically on server alarm information. There is a lack of large-scale, high-quality annotated data.

[0052] In order to solve the above problems, the embodiment of the present application provides a classification method for server alarm information, which realizes the effective classification of service alarm information through a preset encoder and a preset classification model, helps operation and maintenance personnel to perform operation and maintenance work, and effectively reduces the operation and maintenance time of traditional operation and maintenance methods. In addition, the preset encoder and the preset classification model can train a model suitable for automatic classification of server alarm information by using large-scale domain text and historical server alarm information as training corpus. The classification method for server alarm information of the embodiment of the present application is described in detail below in conjunction with the accompanying drawings.

[0053] Reference Figure 1 , Figure 1 A flow chart of a method for classifying server alarm information provided in an embodiment of the present application is shown in FIG. Figure 1As shown, a method for classifying server alarm information provided by an embodiment of the present application may include steps S10 to S12, and these steps are described in detail below.

[0054] S10, obtaining real-time server alarm information, and inputting the real-time server alarm information into a preset encoder to obtain a semantic encoding vector output by the preset encoder, wherein the preset encoder is obtained after training based on operation and maintenance manual data after random full-word masking.

[0055] Among them, the encoder is an important device, which is mainly used to compile and convert signals (such as bit streams) or data into signal forms that can be used for communication, transmission and storage. The preset encoder used in this embodiment is the trained OP-BERT model. Figure 2 As shown in the figure, the main structure of the OP-BERT model is the encoding layer of the Transformer model. The Transformer model is a neural network model based on the self-attention mechanism. The Transformer model learns long-range dependencies in data based on the multi-head self-attention mechanism. After adding the results of the multi-head self-attention layer and the input before the multi-head self-attention layer, layer normalization is used to normalize the features to obtain faster and more stable gradient back propagation. After feature normalization, better feature expression capabilities are obtained through nonlinear changes in the feedforward neural network. The algorithm formula of the multi-head self-attention mechanism can be:

[0056]

[0057]

[0058]

[0059] in; Can represent the output of multi-head attention mechanism; Can represent vector concatenation operations; that is, multiple Connect in the column dimension; It is The attention mechanism results of each head; Can represent the Query matrix; Can represent Key matrix; Can represent the Value matrix; Can be expressed with The corresponding coefficient matrix; Can be expressed with The corresponding coefficient matrix; Can be expressed with The corresponding coefficient matrix; Can indicate the size of the dimension; Can represent the attention mechanism calculation process; The normalized vector probability distribution can be represented so that the sum of the vectors remains 1 and its value range is [0, 1].

[0060] In the attention mechanism calculation process, the similarity value between the current Query matrix and the Key matrix is ​​calculated and the similarity value is passed through Get the value weight, multiply the value weight with the corresponding Value matrix to get After the result of the multi-head self-attention layer is added to the input before the multi-head self-attention layer, the layer normalization is used to normalize the features. The specific calculation formula can be shown as follows:

[0061]

[0062]

[0063] in, Can represent layer normalization results; It can represent the layer normalization calculation process; It can represent the input before the multi-head self-attention layer; compared with the statistical normal distribution standardization calculation formula, the layer normalization calculation formula adds , and Three parameters, Can represent a very small constant; and Can represent trainable parameters.

[0064] After feature normalization, better feature expression ability is obtained through the nonlinear change of the feedforward neural network. The specific calculation formula can be:

[0065]

[0066] in, Can represent the output of a feedforward neural network; Can represent the input of a feedforward neural network; and can all represent coefficient matrices; and can represent bias terms.

[0067] A semantic coding vector refers to a vector representation that maps the meaning or content of a text to a high-dimensional space. The semantic coding vector enables the text data to be subsequently processed and analyzed by a computer while retaining the semantic information of the text.

[0068] The training process of the preset encoder in this embodiment can be specifically shown as steps 1 to 3:

[0069] Step 1: Obtain the server's operation and maintenance manual data and organize the data;

[0070] Step 2: Perform random full-word masking on the sorted operation and maintenance manual data to obtain operation and maintenance training data;

[0071] Step 3: Use the operation and maintenance training data to train the encoder to obtain a preset encoder. The input of the preset encoder is: real-time server alarm information, and the output of the preset encoder is: a semantic encoding vector representing the real-time server alarm information.

[0072] Among them, Figure 3 As shown, this embodiment can obtain operation and maintenance manual data from the operation and maintenance manuals of various manufacturers. In order to improve the field adaptability of the preset encoder, the operation and maintenance manual data may include data on a series of devices associated with the server (such as storage, load balancing, minicomputers, tape libraries, network switches, etc.). After obtaining the operation and maintenance manual data, this embodiment can clean the operation and maintenance manual data, remove the structured data in the operation and maintenance manual data that is not related to the server as a whole, use regular matching to remove the strings (such as special characters, garbled characters, and commands, etc.) in the operation and maintenance manual data that affect the encoder training, and divide the operation and maintenance manual data into multiple sentences according to the periods in the operation and maintenance manual data; use a word segmentation tool to segment the multiple sentences obtained by segmentation, and the word segmentation tool is based on the word data in the server operation and maintenance field dictionary. This embodiment completes the data sorting of the operation and maintenance manual data through the processing in the aforementioned content.

[0073] Among them, the structured, server-independent data in the operation and maintenance manual data may be: copyright statements, promotional advertisements, and other data that are independent of the server itself. In the present embodiment, the word segmentation tool used may be FOOLNLTK. FOOLNLTK is an open source word segmentation tool trained based on the BiLSTM model, which can accurately cut Chinese text into words or phrases and annotate each word in the text with its part of speech, such as noun, verb, etc. The BiLSTM model can capture contextual information in the text, thereby improving the accuracy of word segmentation, part-of-speech tagging, and entity recognition. Among them, the algorithm formula of the BiLSTM model can be:

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080] in, It can represent the input gate; It can represent the forget gate; Can represent output gate; Can represent candidate memory cells; Can represent memory units; can represent hidden vectors; , , , , , , and can all represent coefficient matrices; , , and can represent bias terms.

[0081] The server operation and maintenance domain dictionary can be constructed manually and may include special terms and machine models in various server operation and maintenance, such as "active-active", "hot standby" and "NETAPP8200". The server operation and maintenance domain dictionary can help the word segmentation tool to better segment words and improve the accuracy of the word segmentation tool, thereby providing better data for subsequent random full-word masking and encoder pre-training.

[0082] Random Whole Word Masking refers to a technology used in the field of natural language processing, especially in pre-trained language models (such as BERT), which aims to train the model's ability to predict these masked words based on the context by masking some or all of the words in the text. Random whole word masking can effectively increase the difficulty of training, allowing the encoder to learn more in-depth semantic information during the training process, so that the encoder can distinguish the same noun in different contexts. Of course, this embodiment can also use traditional character masking to obtain operation and maintenance training data.

[0083] In this embodiment, random full-word masking is performed with a probability of p, and the words selected to be covered in the operation and maintenance manual data after the data is sorted can be replaced with special characters [MASK], or replaced with random words, or remain unchanged. Specifically, in this embodiment, the operation and maintenance manual data after the data is sorted is randomly masked with a probability of 15%, and the masked words have an 80% probability of being replaced with special characters [MASK], a 10% probability of being replaced with random words, and a 10% probability of remaining unchanged. Of course, the probability p of random full-word masking is optional. Experimental data show that training the encoder with a probability of 15% to 40% makes the encoder training faster and has better training effects.

[0084] For example, in HEN_5600V3, the disk status is abnormal and the hard disk needs to be replaced. The word data after data sorting is: "HEN_5600V3", "occurrence", "disk", "status abnormality", "need to be replaced" and "hard disk", and the word selected by random full-word masking is "disk". When the probability of replacing with special characters [MASK] is 80%, the operation and maintenance manual data can be expressed as: HEN_5600V3, [MASK] status abnormality occurs, and the hard disk needs to be replaced; when the probability of replacing with random words is 10%, the operation and maintenance manual data can be expressed as: HEN_5600V3, memory status abnormality occurs, and the hard disk needs to be replaced; when the probability of 10% remains unchanged, the operation and maintenance manual data can be expressed as: HEN_5600V3, disk status abnormality occurs, and the hard disk needs to be replaced.

[0085] After obtaining the operation and maintenance training data through random full-word masking, this embodiment adds identifiers before and after each text in the operation and maintenance training data to perform text identification. Specifically, this embodiment adds "[CLS]" (start identifier) ​​before each text and "[SEP]" (end identifier) ​​after each text. After text identification, this embodiment constructs the operation and maintenance training data into a word embedding vector , fragment embedding vector and position embedding vector , the three embedding vectors are added together and input into the constructed OP-BERT model, and the OP-BERT model is pre-trained for unsupervised domain adaptation. The pre-training tasks are the masked language task and the next sentence prediction task, and the OP-BERT model that completes the domain training is obtained.

[0086] Among them, the masked language task refers to the OP-BERT model predicting masked words in a self-supervised manner through bidirectional context information. For example, in HEN_5600V3, the disk status is abnormal and the hard disk needs to be replaced. The probability of "disk" being replaced by special characters [MASK] is 80%, the probability of being replaced by random words is 10%, and the probability of remaining unchanged is 10%. The OP-BERT model needs to predict the true value "disk" through the previous context (HEN_5600V3, occurred) and the following context (status is abnormal, hard disk needs to be replaced).

[0087] The OP-BERT model is trained through the masked language task, and the OP-BERT model can encode bidirectional context representation words. In order for the OP-BERT model to further understand the logical relationship between texts, this embodiment also needs to train the OP-BERT model through the next sentence prediction task. The next sentence prediction task can train the OP-BERT model to understand the relationship between two text sequences. Specifically, this embodiment constructs sentence pairs with continuous contexts based on the operation and maintenance training data, and uses them to train the OP-BERT model. The sentence pairs have multiple sentences with contextual relationships. When this embodiment generates sentence pairs, the sentence pairs with a probability of 50% are continuous sentences labeled "true"; the sentence pairs with a probability of 50% are continuous sentences labeled "false", and the second sentence in the sentence pairs labeled "false" can be randomly extracted from the operation and maintenance training data.

[0088] For example, "HEN_5600V3, the disk status is abnormal, and the hard disk needs to be replaced" is a text with correct context, so it can be marked with a "true" label; a sentence ("Please check the fan status") is randomly extracted from the operation and maintenance training data for replacement, and "HEN_5600V3, the disk status is abnormal, please check the fan status" is obtained. The context of this generated sentence is incorrect, so it can be marked with a "false" label.

[0089] The loss function of the OP-BERT model for the masked language task is a negative log-likelihood function, which can be expressed as follows:

[0090] ,

[0091] in, and Can represent model parameters; Can represent a set of masked words; Can represent specific masking words; It can represent the size of the dictionary in the server operation and maintenance field; It can represent the predicted probability value.

[0092] The loss function of the OP-BERT model for the next sentence prediction task can be expressed as follows:

[0093] ,

[0094] in, and Can represent model parameters; Can represent a set of sentence pairs; Can represent specific sentence pairs; It can represent the predicted probability value.

[0095] Since the OP-BERT model jointly learns the masked language task and the next sentence prediction task, the final loss function of the OP-BERT model can be a linear combination of the loss function of the masked language task and the loss function of the next sentence prediction task. The final loss function can be expressed as follows:

[0096]

[0097] S11, inputting the semantic coding vector into a preset classification model to obtain a category label output by the preset classification model, wherein the preset classification model is trained based on the category labeling of the historical server alarm information and the historical semantic coding vector output by the preset encoder, and the historical semantic coding vector is a semantic coding vector obtained by inputting the historical server alarm information with the category labeling into the preset encoder;

[0098] S12. Use the category corresponding to the category label as the fault category corresponding to the real-time server alarm information.

[0099] The classification model is an important model in machine learning, which is used to classify instances in a data set into different categories. The preset classification model in this embodiment is as follows: Figure 4 The DPCNN model shown in the figure. Unsupervised embedding refers to constructing the input text into a vector that can be input into the model. In this embodiment, unsupervised embedding refers to the semantic encoding vector output by OP-BERT. Regional embedding refers to converting the text vector into a vector by multi-scale convolution. Features, the convolution layer is an equal-length convolution layer (the operation input dimension is n, the convolution kernel size is m, the compensation s=1, the p at both ends of the input is (m-1) / 2, and the output dimension is the same as the input dimension, both are n). After passing through two layers of equal-length convolution layers, the downsampling pooling layer is used for pooling, where the operation kernel size of the pooling layer is size=(3, 1) and the stride is stride=2. After the semantic encoding vector output by OP-BERT passes through the downsampling pooling layer, the length of the semantic encoding vector output by OP-BERT is compressed to half of the original, but its perception field of view is doubled. Then the linear layer and the softmax layer are used to map the compressed semantic encoding vector to the label domain, and the classification result of the semantic encoding vector is determined according to the category corresponding to the label.

[0100] When training the classification model in this embodiment, it is necessary to combine the trained preset encoder with the classification model to be trained, and train the classification model through the training data set and the output result of the preset encoder. Figure 2 As shown, this embodiment constructs an OP-BERT-DPCNN classification model, and OP-BERT is used as an encoder to output a semantic coding vector containing semantic information to DPCNN. The training model of the OP-BERT-DPCNN classification model can be shown in steps 4 to 7:

[0101] Step 4: Obtain historical server alarm information and mark the category;

[0102] Step 5: Obtain classification training data, which is historical server alarm information with category labels;

[0103] Step 6: Input the historical server alarm information with category annotations into a preset encoder to obtain a historical semantic encoding vector output by the preset encoder;

[0104] Step 7: According to the category labeling and historical semantic coding vector of the historical server alarm information, the constructed classification model is trained to obtain the preset classification model. The input of the preset classification model is: the semantic coding vector of the real-time server alarm information, and the output of the preset classification model is: the category label corresponding to the real-time server alarm information.

[0105] Among them, this embodiment constructs a server alarm information classification system, and by analyzing historical server alarm information, divides the server alarm information into multiple categories: hard disk abnormality, memory abnormality, power supply abnormality, fan abnormality, network connection abnormality, CPU occupancy rate is too high, server temperature is too high, memory occupancy rate is too high, hard disk occupancy rate is too high and application abnormality, and constructs category labels corresponding to the above categories.

[0106] This embodiment obtains historical server alarm information, and uses multiple rounds of iterative labeling technology to classify the historical server alarm information, so that each piece of historical server alarm information has a category label. Specifically, each piece of historical server alarm information is labeled by two labelers at the same time. First, labeler A preliminarily labels the historical server alarm information, and then labeler B verifies and labels the historical server alarm information. If the preliminary labeling result of labeler A is inconsistent with the verification labeling result of labeler B, labeler A and labeler B discuss and determine the final labeling result. Thus, the classification training data in this embodiment is obtained, and the classification training data is divided into a training set, a verification set, and a test set in a ratio of 8:1:1. The training set is used for training fine-tuning of OP-BERT-DPCNN, the verification set is used to verify the training effect of OP-BERT-DPCNN during the training process, and the test set is used to evaluate the training performance of OP-BERT-DPCNN after the training is completed.

[0107] This embodiment uses the training set to train OP-BERT-DPCNN, and uses the validation set to verify the training effect of OP-BERT-DPCNN. When the overall loss of the validation set decreases, it can be determined that the effect of OP-BERT-DPCNN has improved. When there is no improvement in the model training effect after more than 5000 batches, the training of OP-BERT-DPCNN can be stopped in advance. After completing the training of OP-BERT-DPCNN, the classification effect of OP-BERT-DPCNN is tested using the test set, and the precision rate p, recall rate R and F1 value are used for evaluation. The specific formula can be shown as follows:

[0108]

[0109]

[0110]

[0111] in, It can represent a true positive sample, that is, the model predicts that the sample is a positive sample, and the actual sample is also a positive sample; It can represent a false positive, that is, a sample that the model predicts is a positive sample, but is actually a negative sample; It can represent a false negative example, that is, the model predicts that the result is a negative sample, but it is actually a positive sample. It will also affect the precision and recall rate. It can represent a true negative example, that is, the model predicts that the result is a negative sample, and the actual sample is also a negative sample.

[0112] During the training process, this embodiment inputs the historical server alarm information with the labeled category labels in the test set into OP-BERT-DPCNN, and the historical server alarm information enters the trained OP-BERT, and OP-BERT outputs a semantic coding vector with a vector dimension of (batch_size, seq_len, hidden_size), and inputs the semantic coding vector into DPCNN. In DPCNN, first, a classification feature vector with a vector dimension of (batch_size, num_filters) is obtained based on the semantic coding vector, and the category vector corresponding to the classification feature vector is obtained through the linear layer and the softmax layer. Among them, batch_size represents the batch size, seq_len represents the sequence length, hidden_size represents the number of hidden units of the encoder, and num_filters represents the number of filters.

[0113] This embodiment completes the fine-tuning training of DPCNN through continuous forward propagation, back propagation and iterative training. Among them, forward propagation refers to the entire intermediate process of transferring text from the input model to the output model, while back propagation traverses the gradient of the entire intermediate process in the opposite order according to the chain rule in calculus, and uses the gradient given by back propagation to update the parameters in the model to obtain better model training results. Iterative training means: inputting data into the model multiple times and updating the model parameters to achieve local optimality.

[0114] At this point, this embodiment completes the training of OP-BERT-DPCNN. By inputting the real-time server alarm information into the trained OP-BERT-DPCNN, the category label of the real-time server alarm information can be obtained, thereby determining the fault category corresponding to the real-time server alarm information.

[0115] The embodiment of the present application provides a method for classifying server alarm information, which uses a preset encoder and a preset classification model to classify real-time server alarm information. The preset encoder is trained with the operation and maintenance manual data of the server, so that the preset encoder can adapt to the sentence characteristics in the server operation and maintenance field, and output a semantic coding vector representing the sentence meaning of the real-time server alarm information. The preset classification model is trained based on the category labeling of the server alarm information and the historical semantic coding vector of the historical server alarm information by the preset encoder, so that the preset classification model can adapt to the semantic coding vector of the preset encoder for classification, realize the effective classification of service alarm information, help operation and maintenance personnel to perform operation and maintenance work, and effectively reduce the operation and maintenance time of traditional operation and maintenance methods.

[0116] Furthermore, in this embodiment, a large amount of corpus that conforms to the server operation and maintenance field is obtained from the operation and maintenance manuals of various manufacturers, so that the preset encoder can adapt to the sentence characteristics of the server operation and maintenance field, and the data used to train the preset encoder and the preset classification model in this embodiment has also undergone multiple data processing to improve the quality of the training data. In addition, this embodiment also classifies and organizes historical service alarm information, and counts multiple categories of server alarm information, which is convenient for use as a classification basis for the preset classification model.

[0117] A method for classifying server alarm information provided by an embodiment of the present application is introduced above. A system applying the method for classifying server alarm information is introduced below.

[0118] See also Figure 5 , Figure 5 A schematic diagram of a server alarm information classification system provided in an embodiment of the present application. Figure 5 As shown, the classification system of the server alarm information includes:

[0119] The acquisition unit 100 is used to acquire real-time server alarm information and input the real-time server alarm information into a preset encoder to obtain a semantic encoding vector output by the preset encoder, wherein the preset encoder is obtained after training based on the operation and maintenance manual data after random full-word masking;

[0120] A classification unit 110 is used to input the semantic coding vector into a preset classification model to obtain a category label output by the preset classification model, wherein the preset classification model is trained based on the category labeling of the historical server alarm information and the historical semantic coding vector output by a preset encoder, and the historical semantic coding vector is a semantic coding vector obtained by inputting the historical server alarm information with the category label into the preset encoder;

[0121] The output unit 120 is used to use the category corresponding to the category label as the fault category corresponding to the real-time server alarm information.

[0122] In a possible implementation, the server alarm information classification system further includes a first training module of a preset encoder, and the first training module may be configured to execute:

[0123] The operation and maintenance manual data of the server is obtained and the data is sorted. The operation and maintenance manual data after the data sorting is randomly full-word masked to obtain the operation and maintenance training data. The operation and maintenance training data is used to train the encoder to obtain the preset encoder. The input of the preset encoder is: real-time server alarm information, and the output of the preset encoder is: a semantic encoding vector representing the real-time server alarm information.

[0124] In a possible implementation, the first training module obtains the operation and maintenance manual data of the server and organizes the data, which can be specifically configured as follows:

[0125] Obtain operation and maintenance manual data from the operation and maintenance manuals of various manufacturers, remove all structured data that is not related to the server, use regular matching to remove the character strings in the operation and maintenance manual data that affect the encoder training, and segment the operation and maintenance manual data into multiple sentences according to the periods in the operation and maintenance manual data. Use a word segmentation tool to segment the multiple sentences obtained by segmentation. The word segmentation tool is based on the word data in the server operation and maintenance field dictionary.

[0126] In a possible implementation, the server alarm information classification system further includes a second training module of a preset classification model, and the second training module may be configured to execute:

[0127] Obtain historical server alarm information and perform category labeling; obtain classification training data, which is historical server alarm information with category labeling; input the historical server alarm information with category labeling into a preset encoder to obtain the historical semantic coding vector output by the preset encoder; train the constructed classification model based on the category labeling and historical semantic coding vector of the historical server alarm information to obtain the preset classification model, the input of the preset classification model is: the semantic coding vector of the real-time server alarm information, and the output of the preset classification model is: the category label corresponding to the real-time server alarm information.

[0128] In a possible implementation, the second training module obtains historical server alarm information and performs category labeling, which can be specifically configured as follows:

[0129] Obtain historical server alarm information and use multi-round iterative labeling technology to classify the historical server alarm information.

[0130] The present application also provides an electronic device in an embodiment. Figure 6 As shown, it shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiment of the present application. The electronic device in the embodiment of the present application may include but is not limited to fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 6 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0131] like Figure 6As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 to a random access memory (RAM) 603. When the electronic device is powered on, various programs and data required for the operation of the electronic device are also stored in the RAM 603. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0132] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a memory card, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 An electronic device having various devices is shown, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0133] An embodiment of the present application also provides a computer program product including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any method for classifying server alarm information provided in the embodiment of the present application.

[0134] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When one or more computer programs are executed by an electronic device, the electronic device can implement any server alarm information classification method provided in an embodiment of the present application.

[0135] It should also be noted that the system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the system embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.

[0136] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0137] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0138] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a training device, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, training device, or data center. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0139] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0140] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0141] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A classification method for server alarm information, characterized in that: The classification method of the server alarm information includes: Acquire real-time server alarm information, and input the real-time server alarm information into a preset encoder to obtain a semantic encoding vector output by the preset encoder, wherein the preset encoder is obtained after training based on operation and maintenance manual data after random full-word masking; Input the semantic coding vector into a preset classification model to obtain a category label output by the preset classification model, wherein the preset classification model is trained based on the category labeling of the historical server alarm information and the historical semantic coding vector output by the preset encoder, and the historical semantic coding vector is a semantic coding vector obtained after the historical server alarm information with the category label is input into the preset encoder; The category corresponding to the category label is used as the fault category corresponding to the real-time server alarm information.

2. The classification method of server alarm information according to claim 1, characterized in that: The training process of the preset encoder includes: Obtain the server's operation and maintenance manual data and organize the data; Performing random full-word masking on the operation and maintenance manual data after data sorting to obtain operation and maintenance training data; The encoder is trained using the operation and maintenance training data to obtain the preset encoder, the input of the preset encoder is: the real-time server alarm information, and the output of the preset encoder is: a semantic coding vector representing the real-time server alarm information.

3. The classification method of server alarm information according to claim 2, characterized in that: The obtaining of the operation and maintenance manual data of the server and collating the data includes: Obtain the operation and maintenance manual data from the operation and maintenance manuals of various manufacturers; Removing all structured data in the operation and maintenance manual data that is not related to the server, removing character strings in the operation and maintenance manual data that affect the training of the encoder using regular matching, and dividing the operation and maintenance manual data into multiple sentences according to periods in the operation and maintenance manual data; A word segmentation tool is used to segment the multiple sentences obtained by segmentation, and the word segmentation of the word segmentation tool is based on word data in the server operation and maintenance field dictionary.

4. The classification method of server alarm information according to claim 1, characterized in that: The training process of the preset classification model includes: Obtain historical server alarm information and mark its categories; Obtaining classification training data, wherein the classification training data is historical server alarm information with category annotations; Inputting the historical server alarm information with category annotation into the preset encoder to obtain the historical semantic encoding vector output by the preset encoder; According to the category labeling of the historical server alarm information and the historical semantic coding vector, the constructed classification model is trained to obtain the preset classification model, the input of the preset classification model is: the semantic coding vector of the real-time server alarm information, and the output of the preset classification model is: the category label corresponding to the real-time server alarm information.

5. The classification method of server alarm information according to claim 1, characterized in that: The obtaining of historical server alarm information and marking of categories includes: The historical server alarm information is obtained, and the historical server alarm information is categorized using a multi-round iterative labeling technique.

6. A classification system for server alarm information, characterized in that: The classification system of the server alarm information includes: An acquisition unit, used for acquiring real-time server alarm information, and inputting the real-time server alarm information into a preset encoder to obtain a semantic encoding vector output by the preset encoder, wherein the preset encoder is obtained after training based on the operation and maintenance manual data after random full-word masking; A classification unit, used for inputting the semantic coding vector into a preset classification model to obtain a category label output by the preset classification model, wherein the preset classification model is trained based on the category labeling of the historical server alarm information and the historical semantic coding vector output by the preset encoder, and the historical semantic coding vector is a semantic coding vector obtained after the historical server alarm information with the category labeling is input into the preset encoder; The output unit is used to use the category corresponding to the category label as the fault category corresponding to the real-time server alarm information.

7. The classification system of server alarm information according to claim 6, characterized in that: The system also includes a first training module of a preset encoder, wherein the first training module is configured to execute: Acquire the operation and maintenance manual data of the server and organize the data, perform random full-word masking on the organized operation and maintenance manual data to obtain operation and maintenance training data, use the operation and maintenance training data to train an encoder to obtain the preset encoder, the input of the preset encoder is: the real-time server alarm information, and the output of the preset encoder is: a semantic encoding vector representing the real-time server alarm information.

8. The classification system of server alarm information according to claim 6, characterized in that: The first training module obtains the server operation and maintenance manual data and organizes the data. The specific configuration is as follows: The operation and maintenance manual data is obtained from the operation and maintenance manuals of various manufacturers, and structured data in the operation and maintenance manual data that is not related to the server is removed as a whole. Regular matching is used to remove character strings in the operation and maintenance manual data that affect the training of the encoder, and the operation and maintenance manual data is segmented into multiple sentences according to the periods in the operation and maintenance manual data. A word segmentation tool is used to segment the multiple sentences obtained by segmentation, and the word segmentation basis of the word segmentation tool is the word data in the server operation and maintenance field dictionary.

9. An electronic device, characterized in that: The method comprises at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electronic device can implement the server alarm information classification method as described in any one of claims 1 to 5.

10. A computer program product, characterized in that It comprises computer-readable instructions, and when the computer-readable instructions are executed on an electronic device, the electronic device implements the method for classifying server alarm information as claimed in any one of claims 1 to 5.

Images