A method and device for power equipment operation and maintenance based on knowledge graph
By building an abnormal knowledge graph of power equipment, the problems of low operation and maintenance efficiency and decision-making errors in the existing technology are solved, and automated operation and maintenance decisions are realized, ensuring the stable operation of the power communication network.
Patent Information
- Application Number
- CN202410805256.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-21
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-06-21
AI Technical Summary
The existing power equipment operation and maintenance methods rely on manual inspection and text content analysis, making it difficult to adapt to complex engineering sites, resulting in low operation and maintenance efficiency and decision-making errors, affecting the safety of the power system.
Using a knowledge graph-based method, by obtaining abnormal operating status data of power equipment, pre-processing, entity identification and relationship extraction, constructing an abnormal knowledge graph, identifying severity levels and formulating operation and maintenance disposal sequences, and achieving automated operation and maintenance decisions.
It improves the automation level of power equipment operation and maintenance, reduces manual errors, ensures the stable operation of the power communication network, and improves operation and maintenance efficiency and accuracy.
Smart Images

Figure CN118378194B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power equipment operation and maintenance, and specifically relates to a power equipment operation and maintenance method and device based on knowledge graph. Background Art
[0002] The operation of power grid is developing in the direction of digitalization and intelligence, and data-driven power grid fault diagnosis has become a practical need for smart grid dispatching. At present, the operation and maintenance of power equipment in the power grid is mainly based on text content analysis and abnormal state information feature extraction. For example, the patent application text CN117251789A discloses a method and system for evaluating the health status of power equipment, which obtains the historical operation data of the power equipment to be evaluated and the operation monitoring data of the power equipment to be evaluated through multiple sensors, wherein the historical operation data and the operation monitoring data both include: current value, voltage value, temperature value, humidity value, pressure value, power supply frequency and power, and then performs feature extraction and comparative analysis on them respectively through a convolutional neural network model to obtain a classification result for indicating whether the power equipment to be evaluated is in a healthy state. Due to the lack of models and methods for quantitative representation of abnormal state information, the existing models cannot adapt to the complex and changeable engineering site practices. In addition, when there is a demand for operation and maintenance of power equipment, manual inspection is the main way to monitor the status of power equipment. The operation and maintenance personnel rely on their own experience and knowledge reserves to understand and judge the defect level and category of the power equipment from the abnormal state on site, and subjective decision-making errors are inevitable. Due to the large area of the maintenance site and the large number of components, the process of manually troubleshooting equipment abnormalities on site is very complicated and time-consuming, which will affect the operation and maintenance efficiency of power equipment and pose a threat to the safety of the power system. This has become a problem that needs to be urgently solved by technical personnel in this field. Summary of the invention
[0003] In view of the defects existing in the prior art, the present invention provides a method and device for power equipment operation and maintenance based on knowledge graph. According to the maintenance personnel information display module, relevant maintenance personnel can be contacted to handle abnormalities in a timely manner, thereby effectively ensuring the stable operation of the power communication network.
[0004] The present invention adopts the following technical solution:
[0005] In a first aspect, the present invention provides a method for operation and maintenance of electric power equipment based on a knowledge graph, the method comprising the following steps:
[0006] Step S1, obtaining abnormal operation status data of power equipment and preprocessing it to generate a data set of abnormal operation status of power equipment;
[0007] Step S2: performing entity recognition on the abnormal operation status data set of the power equipment, and extracting entity relationships based on the characteristics of the entity recognition results;
[0008] Step S3, obtaining the final annotation results of entity recognition and combining them with the extracted entity relationships to construct a knowledge graph of power equipment anomalies;
[0009] Step S4: Obtain abnormal alarm information of power equipment, identify the severity level according to the constructed power equipment abnormality knowledge graph, and sort out the order of operation and maintenance treatment for equipment abnormality;
[0010] Step S5: According to the order of operation and maintenance treatment of equipment abnormalities, the results exceeding the abnormal diagnosis credibility threshold are obtained through similarity calculation, and abnormal operation and maintenance treatment is performed on the power equipment corresponding to the abnormal information exceeding the severity level threshold.
[0011] Furthermore, preprocessing of abnormal operation status data of power equipment includes a first preprocessing and a second preprocessing, the first preprocessing includes text segmentation, text filtering and / or text replacement, and the second preprocessing includes text quality improvement.
[0012] Furthermore, in the above steps, the text segmentation specifically includes: calculating the degree of aggregation of the constituent phrases in the professional corpus in the electric power field, and screening out the professional candidate phrases in the electric power field through the aggregation threshold; calculating the autonomy of the professional candidate phrases in the electric power field, and performing secondary screening on the screened professional candidate phrases in the electric power field according to the autonomy threshold; determining the professional phrases in the electric power field from the secondary screened professional candidate phrases in the electric power field according to the preset word grouping rules in the electric power field, and creating a professional dictionary in the electric power field; adding the professional dictionary in the electric power field to the Jieba word segmentation library, and performing auxiliary word segmentation on each abnormal operation status data of electric power equipment based on the Jieba word segmentation library.
[0013] Furthermore, the text filtering specifically includes: creating a basic filtering vocabulary based on a professional filtering vocabulary in the power field; comparing and filtering the word segmentation result of each abnormal operation status data with the data in the basic filtering vocabulary, if the word segmentation result is not in the stop word vocabulary, retaining the word segmentation result; otherwise, deleting the word segmentation result; outputting the abnormal operation status data after filtering.
[0014] Furthermore, the text replacement specifically includes: creating a synonym thesaurus based on a professional synonym table in the electric power field; comparing and judging the word segmentation result of each abnormal operation status data with the data in the synonym thesaurus, and if the word segmentation result is in the synonym thesaurus, replacing the word segmentation result with a phrase in the synonym thesaurus; until the entire word segmentation result traversal is completed; and outputting the abnormal operation status data after replacement.
[0015] Furthermore, the text quality improvement preprocessing includes three parts: text feature extraction, reconstruction activation function gradient compensation and text classification weight balancing.
[0016] Furthermore, the text feature extraction specifically includes: after a first preprocessing of the abnormal operating status data, a feature extraction convolutional neural network model uses weight vectors with different weight values but the same size to perform convolution calculations and slide on the word vector data of the abnormal operating status data to obtain feature word vectors that represent the meaning of words in the abnormal operating status data, thereby realizing the extraction of text feature values; the feature word vectors are input into a bidirectional text classification model for sentence semantic understanding, and then text classification is performed.
[0017] Furthermore, in the above step S2, the entity recognition includes dynamic word vector conversion, bidirectional encoding of text features, and obtaining the global optimal sequence.
[0018] Furthermore, in the above step S3, an ontology construction tool is used to establish an ontology model of the power equipment anomaly knowledge graph, which includes four-tuple information of equipment anomaly level, equipment anomaly cause, equipment anomaly situation description, and equipment anomaly operation and maintenance measures. The graph database is used to store and visualize the power equipment anomaly knowledge graph ontology model to complete the construction of the power equipment anomaly knowledge graph.
[0019] Furthermore, according to the order of operation and maintenance treatment of equipment abnormalities, the results exceeding the abnormal diagnosis credibility threshold are obtained through similarity calculation, including:
[0020] Based on the order of operation and maintenance treatment of equipment abnormalities, the operation characteristic words and alarm characteristic words of abnormal operation status data of power equipment and abnormal alarm information of power equipment are respectively obtained;
[0021] Use TFIDF model to weight the operation feature words and alarm feature words;
[0022] Based on the weighted results, similarity calculation is performed to obtain the abnormal diagnosis credibility of the abnormal alarm information of the power equipment;
[0023] The abnormal diagnosis credibility of the obtained abnormal alarm information of the electric power equipment is compared with the preset abnormal diagnosis credibility threshold, and the abnormal alarm information of the electric power equipment exceeding the abnormal diagnosis credibility threshold is given.
[0024] Furthermore, similarity calculation is performed based on the weighted results, satisfying the following relationship:
[0025] ;
[0026] In the formula, Indicates The abnormal operation status data of power equipment k The number of occurrences of sports characteristic words; Indicates kThe number of times the operation characteristic words appear in the abnormal operation status data of all power equipment; Indicates j The abnormal alarm information of power equipment The number of times the alarm feature word appears; Indicates The number of times the alarm feature word appears in all abnormal alarm information of power equipment, N is the number of abnormal operation status data of power equipment, N' is the number of abnormal alarm information of power equipment, M is the number of operation feature words, M' is the number of alarm feature words, D represents all abnormal operation status data of power equipment, and D' represents all abnormal alarm information of power equipment.
[0027] In a second aspect, the present invention further provides a power equipment operation and maintenance device based on a knowledge graph, comprising:
[0028] A preprocessing module, which is used to obtain and preprocess the abnormal operation status data of the power equipment to generate a data set of abnormal operation status of the power equipment;
[0029] The recognition and extraction module is used to perform entity recognition on the abnormal operation status data set of power equipment and extract entity relationships based on the characteristics of the entity recognition results;
[0030] The model building module is used to obtain the final annotation results of entity recognition and combine them with the extracted entity relationships to build a knowledge graph of power equipment anomalies;
[0031] The level identification module is used to obtain abnormal alarm information of power equipment, identify the severity level according to the constructed power equipment abnormality knowledge graph, and sort out the order of operation and maintenance treatment for equipment abnormalities;
[0032] The operation and maintenance processing module is used to perform operation and maintenance processing according to the order of equipment abnormalities, obtain results that exceed the abnormal diagnosis credibility threshold through similarity calculation, and perform abnormal operation and maintenance processing on the power equipment corresponding to the abnormal alarm information that exceeds the severity level threshold.
[0033] The preprocessing modules include:
[0034] Calculate the aggregation degree of the constituent phrases in the professional corpus in the electric power field, and screen out the professional candidate phrases in the electric power field through the aggregation degree threshold; calculate the autonomy of the professional candidate phrases in the electric power field, and perform secondary screening on the screened professional candidate phrases in the electric power field according to the autonomy threshold; determine the professional phrases in the electric power field from the secondary screened professional candidate phrases in the electric power field according to the preset word grouping rules in the electric power field, and create a professional dictionary in the electric power field; add the professional dictionary in the electric power field to the Jieba word segmentation library, and perform auxiliary word segmentation on each abnormal operation status data of the electric power equipment based on the Jieba word segmentation library.
[0035] The preprocessing modules include:
[0036] Based on the professional filtering vocabulary in the power field, a basic filtering vocabulary is created; the word segmentation result of each abnormal operation status data is compared and filtered with the data in the basic filtering vocabulary. If the word segmentation result is not in the stop word vocabulary, the word segmentation result is retained; otherwise, the word segmentation result is deleted; and the abnormal operation status data after filtering is output.
[0037] The preprocessing module also includes:
[0038] Based on the professional synonym table in the electric power field, a synonym word library is created; the word segmentation result of each abnormal operation status data is compared and judged with the data in the synonym word library. If the word segmentation result is in the synonym word library, the word segmentation result is replaced with the phrase in the synonym word library; until the entire word segmentation result traversal is completed; and the abnormal operation status data after replacement is output.
[0039] Model building modules include:
[0040] An ontology construction tool is used to establish an ontology model of the power equipment anomaly knowledge graph, which includes four-tuple information of equipment anomaly level, equipment anomaly cause, equipment anomaly situation description, and equipment anomaly operation and maintenance measures. The graph database is used to store and visualize the power equipment anomaly knowledge graph ontology model to complete the construction of the power equipment anomaly knowledge graph.
[0041] In a third aspect, the present invention also provides an electric power equipment operation and maintenance system based on a knowledge graph, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the above-mentioned method steps are implemented.
[0042] The knowledge graph-based power equipment operation and maintenance method and device provided by the present invention include at least the following beneficial effects: the present invention can support power grid staff to view the knowledge graph of power communication equipment anomalies and import power communication equipment anomaly data. It can also help staff to identify the severity level of the anomaly and the detailed information of the anomaly based on the abnormal alarm information, plan the abnormal operation and maintenance handling sequence according to the severity level of the anomaly and give abnormal handling measures according to the abnormal cause in the abnormal detailed information. It can also contact relevant maintenance personnel according to the maintenance personnel information display module to handle the anomaly in a timely manner, thereby effectively ensuring the stable operation of the power communication network. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 A schematic diagram of the flow of a knowledge graph-based power equipment operation and maintenance method according to an embodiment of the present invention.
[0044] Figure 2This is a schematic diagram of a knowledge graph-based power equipment operation and maintenance device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0045] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. 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.
[0046] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.
[0047] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the commodity or device including the elements.
[0048] The present invention provides a power equipment operation and maintenance method based on knowledge graph, the method comprising the following steps:
[0049] Step S1, obtaining abnormal operation status data of power equipment and preprocessing it to generate a data set of abnormal operation status of power equipment;
[0050] In the above steps, the preprocessing of the abnormal operation status data of the power equipment includes a first preprocessing and a second preprocessing. The first preprocessing includes text segmentation, and text filtering and / or text replacement, that is, the first preprocessing can be performed after the text segmentation, and the text filtering and / or text replacement processing is performed. The second preprocessing includes text quality improvement. The first preprocessing step can be regarded as a physical operation or external processing of the abnormal operation status data of the power equipment. The second preprocessing step is an internal improvement from the perspective of data quality, which lays the foundation for subsequent data processing. They are introduced below:
[0051] (1) Text segmentation preprocessing
[0052] Jieba word segmentation has the advantage of being easy to use. You only need to call the word segmentation function to perform text segmentation. Jieba word segmentation first uses a dictionary-based method for word segmentation, and then uses a statistical-based method for auxiliary word segmentation. This method is easy to implement, has a fast word segmentation speed, is simple and efficient, and the word segmentation results are relatively accurate. Considering that the abnormal operating status data of power equipment is highly professional, only using the Jieba word segmentation tool for word segmentation cannot achieve effective word segmentation of professional vocabulary in the power field. In the present invention, a professional dictionary in the power field is introduced, and the dictionary is used as a custom dictionary for Jieba word segmentation to perform auxiliary word segmentation. In the above steps, the text segmentation preprocessing specifically includes:
[0053] 1) Calculate the aggregation degree of the phrases in the professional corpus in the power field, and select the professional alternative phrases in the power field through the aggregation degree threshold;
[0054] The calculation formula of the degree of polymerization is as follows:
[0055] ;
[0056] in, Expressive words , Expressive words ; Expressive words and words frequency of co-occurrence in the corpus; Respectively represent words and words frequency of occurrence in the corpus;
[0057] The degree of polymerization calculated by the above formula It is a value between -1 and 1. For two given words, the greater the possibility that they form a phrase, the greater their aggregation degree.
[0058] 2) Calculating the autonomy of the professional candidate phrases in the power field, and performing secondary screening on the selected professional candidate phrases in the power field according to the autonomy threshold;
[0059] The calculation formula of the autonomy is as follows:
[0060] ;
[0061] in, Indicates a phrase; Indicates a word in a phrase; express The set of all adjacent words on the left; express The set of all adjacent words on the right; Representation word Adjacent phrase The probability on the left; Representation word Adjacent phrase The probability on the right;
[0062] Pick As a phrase The autonomy of the phrase is compared with the preset autonomy threshold. If it is greater than the preset autonomy threshold, it means that the phrase The characters immediately adjacent to the left and right of As a professional alternative phrase in the power field; otherwise, the phrase It cannot be used as a professional alternative phrase in the power field. The degree of autonomy is proportional to the probability of the phrase being recognized.
[0063] 3) According to the preset word grouping rules in the power field, determine the professional phrases in the power field from the secondary screened professional alternative phrases in the power field, and create a professional dictionary in the power field;
[0064] 4) Add the professional dictionary in the electric power field to the Jieba word segmentation database for auxiliary word segmentation.
[0065] In the above text segmentation process, the fixed collocation of professional phrases with greater probability is determined by calculating the degree of aggregation and autonomy. The word combination rules further filter out phrases that do not conform to the standards and do not meet the professional expressions in the power field, and then obtain standardized professional phrases in the power field, and form a professional dictionary in the power field.
[0066] (2) Text filtering preprocessing
[0067] When filtering the text of abnormal operation status data of power equipment, a professional filtering vocabulary in the power field is selected as the basic filtering vocabulary. By combining the characteristics of the power field text being proprietary, some meaningless words such as prepositions, quantifiers, auxiliary words, punctuation marks, etc. are selected for filtering. In the above steps, the text filtering preprocessing specifically includes:
[0068] 1) Create a basic filtering vocabulary based on the professional filtering vocabulary in the power field;
[0069] 2) Perform Chinese word segmentation on each abnormal operation status data, compare and filter the word segmentation result with the data in the basic filtering word library, and if the word segmentation result is not in the stop word word library, retain the word segmentation result; otherwise, delete the word segmentation result;
[0070] 3) Output the abnormal operation status data after filtering.
[0071] (3) Text replacement preprocessing
[0072] Text replacement methods can include similarity calculation and synonym dictionary. The similarity calculation method mainly calculates the similarity between two words to determine whether the two words are the same word. The disadvantage of this calculation method is that the text replacement accuracy and efficiency are not high. Another method based on synonym dictionary is to summarize the synonyms that may appear in the power field into a dictionary. When judging whether two words are synonyms, it is only necessary to traverse the synonym dictionary to complete the synonym replacement of abnormal operation status data.
[0073] In the above steps, the text replacement preprocessing specifically includes:
[0074] 1) Create a synonym database based on the professional synonym table in the power field;
[0075] 2) Perform Chinese word segmentation on each abnormal operation status data, compare and judge the word segmentation result with the data in the synonym word library, and if the word segmentation result is in the synonym word library, replace the word segmentation result with the phrase in the synonym word library; until the entire word segmentation result traversal is completed;
[0076] 3) Output the abnormal operation status data after replacement.
[0077] (4) Text quality improvement preprocessing
[0078] The text quality improvement preprocessing includes three parts: text feature extraction, reconstruction activation function gradient compensation and text classification weight balancing.
[0079] 1) Text feature extraction
[0080] Taking into account the differences between abnormal operation status data texts, in order to accurately parse the data, a feature extraction convolutional neural network model is used to extract features from the text words after word segmentation.
[0081] After the abnormal operation status data is preprocessed for the first time, the feature extraction convolutional neural network model uses weight vectors with different weight values and the same size to perform convolution calculation and slide on the abnormal operation status data to obtain feature word vectors representing the meaning of words in the abnormal operation status data, thereby realizing the extraction of text feature values; the feature word vectors are input into the bidirectional text classification model for sentence semantic understanding, and then text classification is performed;
[0082] The calculation formula of the feature extraction convolutional neural network model is as follows:
[0083] ;
[0084] in, Represents the extracted text feature value; Represents the weight vector middle The weight of represents the original word vector, is the size of the original word vector; Represents the position in the original word vector Abnormal operation status data of the location; Represents dimensional data;
[0085] Feature word vector size The calculation formula is as follows:
[0086] ;
[0087] in Indicates the sliding stride; Indicates the number of fill layers.
[0088] The bidirectional text classification model is composed of a forward text classification sub-model and a backward text classification sub-model to respectively realize bidirectional learning for the previous information and the following information. The output of the bidirectional text classification model is expressed as:
[0089] ;
[0090] in, Represents the output merge result of the forward text classification sub-model and the backward text classification sub-model; represents the time step; They represent the outputs of the forward text classification sub-model and the backward text classification sub-model respectively, and the calculation formula is as follows:
[0091] ;
[0092] in, Represent the input and output of the model respectively; represents the weight matrix; represents the bias vector; represents the logistic regression activation function; tanh represents the hyperbolic tangent activation function; Represent the input gate, forget gate and output gate functions respectively; Display alternative values and updated status values;
[0093] 2) Reconstruct activation function gradient compensation
[0094] The output of the bidirectional text classification model is transferred to the reconstruction activation function to ensure that the gradient exists at non-zero, so as to achieve rapid convergence of the bidirectional text classification model;
[0095] The calculation formula of the reconstruction activation function is as follows:
[0096] ;
[0097] in, Represents the input of the reconstruction activation function; represents the output of the reconstruction activation function, is the reverse gradient of the activation function.
[0098] The above activation function can prevent the gradient from disappearing while accelerating the convergence of the model solution, thereby improving the text classification accuracy of abnormal operating status data.
[0099] 3) Text classification weight balance
[0100] There is an obvious imbalance problem in the abnormal operation status data of power equipment. There is a large difference in the amount of data of various categories, which also makes the feature data extracted based on the abnormal operation status data of power equipment also unbalanced and the data amount gap is large. This will also lead to more abnormal operation status data with larger data volume being paid more attention and processed, while fewer status data with smaller data volume are ignored, which ultimately affects the accuracy of text classification. In the present invention, a balance function is used to weight the samples of each category of text classification output by the bidirectional text classification model to ensure the accuracy of text classification;
[0101] The calculation formula of the equilibrium function is as follows:
[0102] ;
[0103] In the formula, represents the equilibrium function; represents the output probability and , is the sample label; Represents the balancing parameter, which is used to balance the number of samples of different categories. The size of is inversely proportional to the sample size; Represents the sensitivity coefficient, which is used to adjust the sensitivity of the equalization function to the classification category.
[0104] Step S2: performing entity recognition on the abnormal operation status data set of the power equipment, and extracting entity relationships based on the characteristics of the entity recognition results;
[0105] In the above steps, the entity recognition includes three processing steps: dynamic word vector conversion, bidirectional encoding of text features, and obtaining the global optimal sequence. Specifically, they include:
[0106] Step S21: Encode a single character in the abnormal operation status data set of power equipment through a word vector pre-training language model, train to obtain a word vector and a position vector corresponding to the single character, and realize dynamic word vector conversion of the text;
[0107] The word vector formed by the transformation will be used as the initialization input information of the bidirectional gated recurrent network and will be used for subsequent related processing;
[0108] The core component of the word vector pre-training language model is the bidirectional encoding structure, which consists of multiple encoding units. Each encoding unit contains an attention mechanism for finding relevant information in the input sequence and weighting it. The attention mechanism can help the model capture long-distance dependencies in the input sequence, thereby improving the model's language understanding ability. In the model, each encoding unit simultaneously considers different semantic features through a multi-head attention mechanism, thereby further improving the performance of the model. In addition to using encoding units, the model also uses pre-training and fine-tuning methods to optimize the performance of the model.
[0109] The attention mechanism calculates the correlation between each word and all other words, and then uses these correlations to adjust the importance of each word to generate a new semantic feature expression. The calculation formula of the attention mechanism is as follows:
[0110] ;
[0111] in, represents the input vector; Represents the dimension of the input vector;
[0112] In order to expand the expressive power of the word vector pre-trained language model, a multi-head attention mode is used in the model to expand the representation subspace of the attention mechanism. The calculation formula of the multi-head attention mode is as follows:
[0113] ;
[0114] In the process of processing fault text, the meanings of words in different positions in the sentence will be different. The position information of words in the sentence is introduced into the model to form a position vector. The calculation formula of the position vector is as follows:
[0115] ;
[0116] in, Indicates the position of a word in a text; Represents the dimension of the word vector.
[0117] Step S22, bidirectionally encoding the input dynamic word vector through a bidirectional gate controlled recurrent network, calculating and obtaining bidirectional semantic feature information related to the text context, so as to obtain a text feature encoding containing the bidirectional semantic feature information;
[0118] Among them, the bidirectional gate-controlled recurrent network includes forward and reverse gate-controlled recurrent networks to simultaneously process forward and backward text information, thereby extracting bidirectional semantic features. The output result of the bidirectional gate-controlled recurrent network is obtained by the joint action of the forward and reverse gate-controlled recurrent networks. The calculation formula is as follows:
[0119] ;
[0120] in, express Input of time; Represents the logistic regression activation function for vector addition or multiplication; Represents the dot product of two vectors; Respectively represent weight matrices; * represents the product of matrices; represents the update gate; Reset gate. Represent the outputs of the forward and reverse gate-controlled recurrent networks respectively;
[0121] In a bidirectional gate-controlled recurrent network, the update gate and reset gate are used to adjust the contribution range of the state information at the previous moment at the current time. The larger the value of the update gate, the greater the contribution range of the state information at the previous moment to the current state; and the smaller the value of the reset gate, the smaller the contribution of the state information at the previous moment to the current state.
[0122] Step S23, inputting the text feature encoding containing the bidirectional semantic feature information into the global optimal decoding model for decoding calculation to obtain the global optimal result and output the label sequence with the highest probability, and obtaining the entity category of each character in the abnormal operation status data;
[0123] Specifically, given For the input sequence, the probability score of the predicted output sequence result y is calculated using the following formula:
[0124] ;
[0125] in, represents the transfer matrix element; Represents the i-th character in the text;
[0126] The total score of the correct annotation is summed with the scores of all possible annotations and normalized to obtain the probability of generating a label sequence under the condition of a given input sequence. , the calculation formula is as follows:
[0127] ;
[0128] in, represents the input sequence; represents a tag sequence;
[0129] During the training process, the loss function is calculated by the logarithmic maximum likelihood estimation method to obtain the logarithmic probability of the model output sequence that is most consistent with the correct label sequence, and the highest score result is output as the final labeling result of entity recognition. The calculation formula is as follows:
[0130] ;
[0131] in, Represents the output sequence.
[0132] After entity recognition, step S24 is also included, specifically: based on the entity recognition results, a process for extracting entity relationships of abnormal operation status data is designed to extract entity relationships of unstructured text. When extracting entity relationships of abnormal operation status data of power equipment, the entity type is directly determined by the entity label, and then entity relationship extraction rules are established according to the defined entity relationships. The extraction rules are as follows:
[0133] (1) Determine the entity tag of the abnormally operating device based on the identifier of the abnormally operating device and associate it with the identifier of the abnormally operating device;
[0134] (2) Determine the relevant abnormal part identification according to the entity tag; and determine the relevant equipment abnormality level, equipment abnormality cause, equipment abnormality description, and equipment abnormality category according to the determined abnormal part identification;
[0135] (3) Based on the description of the equipment abnormality, extract the equipment abnormality operation and maintenance measures.
[0136] Step S3, obtaining the final annotation results of entity recognition and combining them with the extracted entity relationships to construct a knowledge graph of power equipment anomalies;
[0137] In the above steps, an ontology construction tool is used to establish an electric power equipment abnormality knowledge graph ontology model including four-tuple information of equipment abnormality level, equipment abnormality cause, equipment abnormality description, and equipment abnormality operation and maintenance measures, and a graph database is used to store and visualize the electric power equipment abnormality knowledge graph ontology model to complete the construction of the electric power equipment abnormality knowledge graph;
[0138] In the above steps, the Protégé ontology construction tool is used to complete the construction of the relationship between the entities of the power communication equipment fault, so as to establish the knowledge graph ontology model of the power communication equipment fault, and Neo4j is used to store and visualize the acquired quadruple knowledge. In the knowledge graph, the storage method of the knowledge base depends on the form of knowledge representation. Neo4j is used to optimize the knowledge representation method, and then the storage knowledge base is optimized according to the knowledge representation method, laying the foundation for the knowledge management of the subsequent knowledge graph. In view of the problem that the data sources are wide and difficult to merge, Neo4j is used to realize the knowledge fusion of multi-source heterogeneous abnormal operation status data; in view of the problem that the data volume is large and difficult to store and query, Neo4j is used to integrate and store the acquired knowledge in a suitable form; in view of the problem that the real-time data is updated slowly, Neo4j is used to implement the knowledge update strategy so that the knowledge can be updated in time.
[0139] Step S4: Obtain abnormal alarm information of power equipment, identify the severity level according to the constructed power equipment abnormality knowledge graph, and sort out the order of operation and maintenance treatment for equipment abnormality;
[0140] In the above steps, severity level identification is performed, including:
[0141] 1) The abnormal alarm information of power equipment is represented by word embedding vectors and used as the input of the bidirectional text classification model for bidirectional encoding training;
[0142] The acquired abnormal alarm information of the power equipment is expressed in the following form:
[0143] ;
[0144] in, Represents text The A vocabulary.
[0145] For each word During the training process, the abnormal warning information needs to be preprocessed before entering the bidirectional text classification model, and the word vector matrix is used to represent the abnormal warning information:
[0146] ;
[0147] Abnormal warning information Convert it into a real number matrix, add the abnormal level information into it, and further process the obtained real number matrix.
[0148] (2) Enhance the relevant features in the abnormal alarm information of power equipment through the attention mechanism;
[0149] The bidirectional text classification model is composed of a forward long-term short-term network and a backward long-term short-term network. The long-term short-term network mainly uses the forget gate, input gate and output gate to protect and control the input information. In the forward network, the input analysis feature vector is The state at the previous moment Calculate together to calculate the current state , the calculation formula is as follows:
[0150] ;
[0151] in, t Indicates the moment; i represents the input gate; f represents the forget gate; o represents the output gate; b represents the bias term; w (•) represents the parameter matrix;
[0152] The forward long-term short-term network processes the abnormal warning information in the previous text to input the word segmentation vector x 1 to xt , combined with xt The previous information is based on xt Perform semantic encoding and output information as Similarly, the reverse long-term short-term network processes the abnormal warning information in the following text to input the word segmentation feature vector nx arrive xt , combined with xt The following information is for xt Perform semantic encoding and output information as Finally, the obtained and To represent the information after the t-th word segmentation encoding, the calculation formula is as follows:
[0153] ;
[0154] ;
[0155] in, represents vector connection; represents the dimension of a unidirectional long-term and short-term network; H represents the set of output vectors.
[0156] (3) Use the Softmax activation function to identify the severity level of anomalies.
[0157] In the attention mechanism Attention, the attention probability distribution of the abnormal level feature vector in the abnormal warning information text output by the bidirectional long-term short-term network at time n to the final state The calculation process is as follows:
[0158] ;
[0159] in, Represents the complete hidden state sequence of abnormal alarm information at time n; represents the jth hidden state vector in the abnormal alarm information at time n; N represents the number of input abnormal alarm information text sequences; Represents the hidden state at time n for distribution of attention.
[0160] The calculation formula of the final feature of abnormal alarm information of power communication equipment at time n based on the attention mechanism is as follows:
[0161] ;
[0162] Among them, when The higher the value of , the higher the representation ability of the corresponding verb embedding is, and the more abnormal and important discriminative information of power communication equipment can be obtained from the attention mechanism.
[0163] Finally, through the Softmax activation function calculation of the output layer, the dimension of the abnormal level type label is 1×4, and its probability distribution calculation formula is as follows:
[0164] ;
[0165] Where T represents the number of abnormal level labels, T=4, including Minor, Warning, Major, and Critical; V represents the weight of abnormal warning information generated during the training process.
[0166] Each abnormal alarm information input into the bidirectional long-term and short-term network has a corresponding label in the training set data, which is the abnormal severity level corresponding to the abnormal alarm information; the probability of each label is calculated , where the label with the highest probability is the severity level of the anomaly caused by the abnormal alarm information.
[0167] Step S5: According to the order of operation and maintenance treatment of equipment abnormalities, the results exceeding the abnormal diagnosis credibility threshold are obtained through similarity calculation, and abnormal operation and maintenance treatment is performed on the power equipment corresponding to the abnormal alarm information exceeding the severity level threshold.
[0168] On the basis of planning the abnormal operation and maintenance disposal sequence, the operation feature words obtained after preprocessing the abnormal operation status data of the power equipment and the alarm feature words obtained after the aforementioned preprocessing of the abnormal alarm information of the power equipment are respectively vectorized, and the operation feature words and the alarm feature words are weighted using the TFIDF model, which can measure the importance of the operation feature words and the alarm feature words in the abnormal operation status data of the power equipment and the abnormal alarm information of the power equipment respectively. By calculating the word frequency and inverse document frequency of the feature words, a weight can be assigned to each feature word to indicate its importance in the abnormal text. The cosine similarity function is used to perform similarity calculation to obtain a result with a high reliability of abnormal diagnosis, and timely handle the abnormal information with a higher severity level, so as to realize the intelligent diagnosis of abnormalities of power communication equipment.
[0169] The vector space model is used to vectorize the feature words of the abnormal text (power equipment abnormal operation status data and power equipment abnormal alarm information) after preprocessing, and the TFIDF factor is used to weight the feature words. Then, the cosine similarity function is used to calculate the similarity of the abnormal text, so as to obtain a more reliable abnormal diagnosis result. d j And each power equipment abnormal alarm information d' j Mapped into a feature vector, the calculation formula is as follows:
[0170] ;
[0171] in, represents the abnormal operation status data of the ith power equipment among all the abnormal operation status data of the power equipment, Indicates abnormal operation status data of power equipment The corresponding k-th running feature word; express exist The corresponding weight in . represents the jth abnormal alarm information of the power equipment among all abnormal alarm information of the power equipment, Indicates abnormal alarm information of power equipment The corresponding Alarm feature words; express exist The corresponding weight in . and Defined as TFIDF factor, the calculation formula is as follows:
[0172] ;
[0173] in, represents the number of occurrences of the kth motion feature word in the abnormal operation status data of the i-th power equipment; represents the number of times the kth operation feature word appears in all abnormal operation status data of power equipment; Indicates the jth abnormal alarm information of the power equipment The number of times the alarm feature word appears; Indicates The number of times the alarm feature word appears in all abnormal alarm information of power equipment, N is the number of abnormal operation status data of power equipment, N' is the number of abnormal alarm information of power equipment, D represents all abnormal operation status data of power equipment, and D' represents all abnormal alarm information of power equipment; abnormal operation status data of power equipment d i And power equipment abnormal alarm information d j The similarity calculation is calculated using the cosine similarity formula, which is as follows:
[0174] ;
[0175] Among them, M is the number of running feature words, M' is the number of alarm feature words, sim ( d i ,d j )∈[0,1], the closer the cosine value is to 1, the higher the credibility of the abnormal diagnosis result is, and the closer the cosine value is to 0, the lower the credibility of the abnormal diagnosis is.
[0176] like Figure 2 As shown, the present invention also provides a power equipment operation and maintenance device based on knowledge graph, comprising:
[0177] A preprocessing module, which is used to obtain and preprocess the abnormal operation status data of the power equipment to generate a data set of abnormal operation status of the power equipment;
[0178] The recognition and extraction module is used to perform entity recognition on the abnormal operation status data set of power equipment and extract entity relationships based on the characteristics of the entity recognition results;
[0179] The model building module is used to obtain the final annotation results of entity recognition and combine them with the extracted entity relationships to build a knowledge graph of power equipment anomalies;
[0180] The level identification module is used to obtain abnormal alarm information of power equipment, identify the severity level according to the constructed power equipment abnormality knowledge graph, and sort out the order of operation and maintenance treatment for equipment abnormalities;
[0181] The operation and maintenance processing module is used to perform operation and maintenance processing according to the order of equipment abnormalities, obtain results that exceed the abnormal diagnosis credibility threshold through similarity calculation, and perform abnormal operation and maintenance processing on the power equipment corresponding to the abnormal alarm information that exceeds the severity level threshold.
[0182] The preprocessing modules include:
[0183] Preprocessing of abnormal operation status data of power equipment includes text segmentation, text filtering and / or text replacement, and text quality improvement.
[0184] The preprocessing modules include:
[0185] Calculate the aggregation degree of the constituent phrases in the professional corpus in the electric power field, and screen out the professional candidate phrases in the electric power field through the aggregation degree threshold; calculate the autonomy of the professional candidate phrases in the electric power field, and perform secondary screening on the screened professional candidate phrases in the electric power field according to the autonomy threshold; determine the professional phrases in the electric power field from the secondary screened professional candidate phrases in the electric power field according to the preset word grouping rules in the electric power field, and create a professional dictionary in the electric power field; add the professional dictionary in the electric power field to the Jieba word segmentation library, and perform auxiliary word segmentation on each abnormal operation status data of the electric power equipment based on the Jieba word segmentation library.
[0186] The preprocessing modules include:
[0187] Based on the professional filtering vocabulary in the power field, a basic filtering vocabulary is created; Chinese word segmentation is performed on each abnormal operation status data, and the word segmentation result is compared and filtered with the data in the basic filtering vocabulary. If the word segmentation result is not in the stop word vocabulary, the word segmentation result is retained; otherwise, the word segmentation result is deleted; and the abnormal operation status data after filtering is output.
[0188] The preprocessing module also includes:
[0189] Based on the professional synonym table in the power field, a synonym word library is created; Chinese word segmentation is performed on each abnormal operation status data, and the word segmentation result is compared and judged with the data in the synonym word library. If the word segmentation result is in the synonym word library, the word segmentation result is replaced with the phrase in the synonym word library; until the entire word segmentation result traversal is completed; and the abnormal operation status data after replacement is output.
[0190] Model building modules include:
[0191] An ontology construction tool is used to establish an ontology model of the power equipment anomaly knowledge graph, which includes four-tuple information of equipment anomaly level, equipment anomaly cause, equipment anomaly situation description, and equipment anomaly operation and maintenance measures. The graph database is used to store and visualize the power equipment anomaly knowledge graph ontology model to complete the construction of the power equipment anomaly knowledge graph.
[0192] The present invention provides a power equipment operation and maintenance system based on knowledge graph. In view of the complex and changeable characteristics of power equipment anomaly detection data, a power equipment defect knowledge graph is established in a targeted manner. According to the constructed power equipment defect knowledge graph and power equipment anomaly detection model, power equipment anomaly detection is realized, and the power equipment related anomaly knowledge is associated, integrated and saved in a graphical manner. Managers can not only obtain the anomaly details through abnormal alarm information, but also judge the level or grade of the power equipment anomaly, and can propose corresponding treatment measures in a linked manner, so as to ensure the stable operation of the power equipment and the safety of the power grid system.
[0193] Various implementations of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
[0194] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.
[0195] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0196] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).
[0197] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.
Claims
1. A power equipment operation and maintenance method based on knowledge graph, characterized in that: The method comprises the following steps: Step S1, obtaining abnormal operation status data of power equipment and preprocessing it to generate a data set of abnormal operation status of power equipment; Step S2: performing entity recognition on the abnormal operation status data set of the power equipment, and extracting entity relationships based on the characteristics of the entity recognition results, including: Step S21: Encode a single character in the abnormal operation status data set of power equipment through a word vector pre-training language model, train to obtain a word vector and a position vector corresponding to the single character, and realize dynamic word vector conversion of the text; Step S22, bidirectionally encoding the input dynamic word vector through a bidirectional gate controlled recurrent network, calculating and obtaining bidirectional semantic feature information related to the text context, so as to obtain a text feature encoding containing the bidirectional semantic feature information; Step S23, inputting the text feature encoding containing the bidirectional semantic feature information into the global optimal decoding model for decoding calculation to obtain the global optimal result and output the label sequence with the highest probability, and obtaining the entity category of each character in the abnormal operation status data; Step S24: according to the identification of the abnormally operating device, determine the entity tag of the abnormally operating device and associate it with the identification of the abnormally operating device; determine the entity category according to the entity tag and determine the related abnormal part identification; according to the determined abnormal part identification, determine the related equipment abnormality level, equipment abnormality cause, equipment abnormality description, and equipment abnormality category in sequence; according to the equipment abnormality description, extract equipment abnormal operation and maintenance measures; Step S3, obtaining the final annotation results of entity recognition and combining them with the extracted entity relationships to construct a knowledge graph of power equipment anomalies; Step S4: Obtain abnormal alarm information of power equipment, identify the severity level according to the constructed power equipment abnormality knowledge graph, and sort out the order of operation and maintenance treatment for equipment abnormality; Step S5: According to the order of operation and maintenance treatment of equipment abnormalities, the results exceeding the abnormal diagnosis credibility threshold are obtained through similarity calculation, and abnormal operation and maintenance treatment is performed on the power equipment corresponding to the abnormal alarm information exceeding the severity level threshold.
2. The method according to claim 1, characterized in that Preprocessing the abnormal operation status data of the power equipment includes a first preprocessing and a second preprocessing, the first preprocessing includes text segmentation, text filtering and / or text replacement, and the second preprocessing includes text quality improvement.
3. The method according to claim 2, characterized in that The text segmentation specifically includes: Calculate the aggregation degree of the phrases in the professional corpus in the power field, and select the professional alternative phrases in the power field through the aggregation degree threshold; Calculating the autonomy of the professional candidate phrases in the electric power field, and performing secondary screening on the screened professional candidate phrases in the electric power field according to the autonomy threshold; According to the preset word grouping rules in the power field, professional phrases in the power field are determined from the secondary screened professional alternative phrases in the power field, and a professional dictionary in the power field is created; The professional dictionary in the electric power field is added to the Jieba word segmentation library, and auxiliary word segmentation is performed on each abnormal operation status data of the electric power equipment based on the Jieba word segmentation library.
4. The method according to claim 2 or 3, characterized in that The text filtering specifically includes: Create a basic filtering vocabulary based on the professional filtering vocabulary in the power field; Compare and filter the word segmentation result of each abnormal operation status data with the data in the basic filtering word library. If the word segmentation result is not in the stop word word library, the word segmentation result is retained; otherwise, the word segmentation result is deleted; Outputs the abnormal operation status data after filtering.
5. The method according to claim 2 or 3, characterized in that: The text replacement specifically includes: Create a synonym database based on the professional synonym table in the power field; Compare and judge the word segmentation result of each abnormal operation status data with the data in the synonym word library. If the word segmentation result is in the synonym word library, replace the word segmentation result with the phrase in the synonym word library; until the entire word segmentation result traversal is completed; Outputs the abnormal operation status data after replacement.
6. The method according to claim 2, characterized in that The text quality improvement includes text feature extraction, reconstruction activation function gradient compensation and text classification weight balancing; The text feature extraction specifically includes: after a first preprocessing of the abnormal operation status data, a feature extraction convolutional neural network model uses weight vectors with different weight values but the same size to perform convolution calculations and slide on the abnormal operation status data to obtain feature word vectors that represent the meaning of words in the abnormal operation status data, thereby extracting text feature values; and inputting the feature word vectors into a bidirectional text classification model to perform sentence semantic understanding, thereby performing text classification.
7. The method according to claim 1, characterized in that In the above step S3, an ontology construction tool is used to establish an electric power equipment anomaly knowledge graph ontology model including four-tuple information of equipment anomaly level, equipment anomaly cause, equipment anomaly situation description, and equipment anomaly operation and maintenance measures, and a graph database is used to store and visualize the electric power equipment anomaly knowledge graph ontology model to complete the construction of the electric power equipment anomaly knowledge graph.
8. The method according to claim 1, characterized in that According to the order of operation and maintenance treatment of equipment anomalies, similarity calculation is used to obtain results that exceed the anomaly diagnosis credibility threshold, including: Based on the order of operation and maintenance treatment of equipment abnormalities, the operation characteristic words and alarm characteristic words of abnormal operation status data of power equipment and abnormal alarm information of power equipment are respectively obtained; Use TFIDF model to weight the operation feature words and alarm feature words; Based on the weighted results, similarity calculation is performed to obtain the abnormal diagnosis credibility of the abnormal alarm information of the power equipment; The abnormal diagnosis credibility of the obtained abnormal alarm information of the electric power equipment is compared with the preset abnormal diagnosis credibility threshold, and the abnormal alarm information of the electric power equipment exceeding the abnormal diagnosis credibility threshold is given.
9. The method according to claim 8, characterized in that Similarity calculation is performed based on the weighted results, satisfying the following relationship: ; In the formula, Indicates i The abnormal operation status data of power equipment k The number of occurrences of the sports characteristic words; Indicates k The number of times the operation characteristic words appear in the abnormal operation status data of all power equipment; Indicates j The abnormal alarm information of power equipment l The number of times the alarm feature word appears; Indicates l The number of times the alarm feature word appears in all abnormal alarm information of power equipment, N is the number of abnormal operation status data of power equipment, N' is the number of abnormal alarm information of power equipment, M is the number of operation feature words, M' is the number of alarm feature words, D represents all abnormal operation status data of power equipment, and D' represents all abnormal alarm information of power equipment.
10. A power equipment operation and maintenance device based on knowledge graph, characterized in that: include: A preprocessing module, which is used to obtain and preprocess the abnormal operation status data of the power equipment to generate a data set of abnormal operation status of the power equipment; The identification and extraction module is used to perform entity recognition on the abnormal operation status data set of power equipment and extract entity relationships based on the characteristics of the entity recognition results, including: Step S21: Encode a single character in the abnormal operation status data set of power equipment through a word vector pre-training language model, train to obtain a word vector and a position vector corresponding to the single character, and realize dynamic word vector conversion of the text; Step S22, bidirectionally encoding the input dynamic word vector through a bidirectional gate controlled recurrent network, calculating and obtaining bidirectional semantic feature information related to the text context, so as to obtain a text feature encoding containing the bidirectional semantic feature information; Step S23, inputting the text feature encoding containing the bidirectional semantic feature information into the global optimal decoding model for decoding calculation to obtain the global optimal result and output the label sequence with the highest probability, and obtaining the entity category of each character in the abnormal operation status data; Step S24: according to the identification of the abnormally operating device, determine the entity tag of the abnormally operating device and associate it with the identification of the abnormally operating device; determine the entity category according to the entity tag and determine the related abnormal part identification; according to the determined abnormal part identification, determine the related equipment abnormality level, equipment abnormality cause, equipment abnormality description, and equipment abnormality category in sequence; according to the equipment abnormality description, extract equipment abnormal operation and maintenance measures; The model building module is used to obtain the final annotation results of entity recognition and combine them with the extracted entity relationships to build a knowledge graph of power equipment anomalies; The level identification module is used to obtain abnormal alarm information of power equipment, identify the severity level according to the constructed power equipment abnormality knowledge graph, and sort out the order of operation and maintenance treatment for equipment abnormalities; The operation and maintenance processing module is used to perform operation and maintenance processing according to the order of equipment abnormalities, obtain results that exceed the abnormal diagnosis credibility threshold through similarity calculation, and perform abnormal operation and maintenance processing on the power equipment corresponding to the abnormal alarm information that exceeds the severity level threshold.
