Fault text processing method and device for power equipment

Abstract

The invention provides a fault text processing method and device for power equipment, and the method comprises the steps: carrying out the text preprocessing of a fault description text, and generating a high-dimensional word vector and a low-dimensional word vector; performing fault grading judgment on the high-dimensional word vector through a pre-constructed hybrid deep neural network model to generate a fault grade; semantic definition matching is conducted on the low-dimensional word vectors through a pre-constructed classification model, a semantic structure conversion table is generated, and the semantic structure conversion table comprises semantic element missing values; constructing a structure association tree according to a preset association rule table and the semantic structure conversion table; through a text distance estimation method combining semantics and structures, similarity calculation is carried out on a structure association tree and a preset standard text library, semantic element missing values are filled up, fault semantic structure information is generated, accurate analysis and grading can be carried out on unstructured fault texts, information missing values are complemented, information integrity is improved, and the fault semantic structure information is obtained. And the text analysis efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, in particular to the field of artificial intelligence technology, and more particularly to a method and device for processing fault text of electric power equipment. Background Art

[0002] Technological innovation is driving the power grid industry's transformation towards digitalization and intelligence, particularly achieving significant progress in monitoring grid equipment status and predicting potential failures. With the continued rise in electricity consumption and the increasing complexity of grid architecture, coupled with the rapid development of renewable energy generation technologies, increasingly stringent requirements are being placed on grid reliability and stability. Currently, data-driven grid operations, powered by big data and artificial intelligence, have become a core element of smart grid development. The operation of power grid equipment generates vast amounts of data. The unique properties of Chinese text data and the intermixing of numbers and letters within text create unique challenges for text analysis in the power industry. Text records of power equipment failures detail the fault conditions encountered during operation. Manually recorded fault texts often suffer from colloquialisms, inconsistent length, and missing key information. These issues complicate information extraction and reduce completeness. Fault severity and information often rely on manual judgment, which is susceptible to subjective factors. This not only impacts the efficiency of text analysis but also makes it difficult to ensure accuracy. Summary of the Invention

[0003] One object of the present invention is to provide a method for processing fault texts for power equipment, capable of accurately analyzing and grading unstructured fault texts, completing missing information, improving information completeness, and enhancing text analysis efficiency and accuracy. Another object of the present invention is to provide a device for processing fault texts for power equipment. Another object of the present invention is to provide a computer-readable medium. Yet another object of the present invention is to provide a computer device.

[0004] In order to achieve the above objectives, the present invention discloses a method for processing fault text of electric power equipment, comprising:

[0005] Get the fault description text of the power equipment;

[0006] Perform text preprocessing on the fault description text to generate high-dimensional word vectors and low-dimensional word vectors;

[0007] A pre-built hybrid deep neural network model is used to perform fault classification on high-dimensional word vectors and generate fault levels.

[0008] Through the pre-built classification model, the low-dimensional word vectors are matched with semantic definitions to generate a semantic structure conversion table, which includes missing values ​​of semantic elements;

[0009] According to the preset association rule table and the semantic structure conversion table, a structural association tree is constructed;

[0010] Through the text distance estimation method that combines semantics and structure, the similarity between the structural association tree and the preset standard text library is calculated, and the missing values ​​of semantic elements are filled to generate fault semantic structure information.

[0011] Preferably, text preprocessing is performed on the fault description text to generate high-dimensional word vectors and low-dimensional word vectors, including:

[0012] The pre-built word metamodel is used to segment and clean the fault description text to generate a fault word sequence.

[0013] Use vector embedding technology to vectorize the fault word sequence and generate high-dimensional word vectors;

[0014] Through the principal component analysis method, the high-dimensional word vector is reduced to generate a low-dimensional word vector.

[0015] Preferably, the hybrid deep neural network model includes a convolutional neural network and a bidirectional long short-term memory network;

[0016] A pre-built hybrid deep neural network model is used to perform fault classification on high-dimensional word vectors and generate fault levels, including:

[0017] Through convolutional neural networks, feature extraction is performed on high-dimensional word vectors to generate local features;

[0018] Through the bidirectional long short-term memory network, feature extraction is performed on high-dimensional word vectors to generate context features;

[0019] Combine local features with context features to generate features to be classified;

[0020] Through the attention mechanism, the classification features are calculated to generate feature classification vectors;

[0021] The feature classification vector is calculated through the normalized activation function to generate multiple risk levels and the classification probability corresponding to each risk level;

[0022] The risk level corresponding to the maximum classification probability is determined as the fault level of the fault description text.

[0023] Preferably, the association rule table includes semantic definitions and corresponding association relationships;

[0024] According to the preset association rule table and the semantic structure conversion table, a structural association tree is constructed, including:

[0025] According to the association relationship, the core semantic elements and associated semantic elements in the semantic structure conversion table are determined;

[0026] The core semantic element is determined as the root node, and the associated semantic elements are determined as child nodes to obtain a structural association tree.

[0027] Preferably, a text distance estimation method combining semantics and structure is used to calculate the similarity between the structural association tree and the preset standard text library, and missing values ​​of semantic elements are filled to generate fault semantic structure information, including:

[0028] The semantic weight of the structure association tree is generated according to the height of each node in the structure association tree through the exponential decay method;

[0029] Through the hierarchical analysis method, the structure weight of the structure association tree is generated by calculating the low-dimensional word vector corresponding to each node of the preset judgment matrix and the structure association tree;

[0030] Calculate semantic similarity between the structural association tree and the standard text library to generate semantic similarity;

[0031] Calculate the structural similarity between the structural association tree and the standard text library to generate structural similarity;

[0032] Generate similarity results based on the semantic weight and semantic similarity of the structural association tree and the structural weight and structural similarity of the structural association tree;

[0033] The missing values ​​of semantic elements are filled in according to the semantic elements corresponding to the similarity results to generate the fault semantic structure information.

[0034] The present invention also discloses a fault text processing device for electric power equipment, comprising:

[0035] An acquisition unit, used for acquiring a fault description text of the power equipment;

[0036] A text preprocessing unit is used to perform text preprocessing on the fault description text and generate high-dimensional word vectors and low-dimensional word vectors;

[0037] The fault classification unit is used to perform fault classification on high-dimensional word vectors using a pre-built hybrid deep neural network model to generate fault levels;

[0038] The semantic structure conversion unit is used to match the semantic definitions of low-dimensional word vectors through a pre-built classification model and generate a semantic structure conversion table, which includes missing values ​​of semantic elements;

[0039] A structure association tree construction unit is used to construct a structure association tree according to a preset association rule table and a semantic structure conversion table;

[0040] The fault semantic structure information generation unit is used to calculate the similarity between the structural association tree and the preset standard text library through a text distance estimation method that combines semantics and structure, and fill in the missing values ​​of semantic elements to generate fault semantic structure information.

[0041] Preferably, the text preprocessing unit is specifically used to perform word segmentation and cleaning processing on the fault description text through a pre-built word meta model to generate a fault vocabulary sequence; vectorize the fault vocabulary sequence through vector embedding technology to generate a high-dimensional word vector; and perform dimensionality reduction processing on the high-dimensional word vector through the principal component analysis method to generate a low-dimensional word vector.

[0042] Preferably, the hybrid deep neural network model includes a convolutional neural network and a bidirectional long short-term memory network;

[0043] The fault grading unit is specifically used to extract features from high-dimensional word vectors through convolutional neural networks to generate local features; extract features from high-dimensional word vectors through bidirectional long short-term memory networks to generate context features; concatenate local features with context features to generate features to be classified; calculate features to be classified through an attention mechanism to generate feature classification vectors; calculate feature classification vectors through a normalized activation function to generate multiple risk levels and the classification probability corresponding to each risk level; and determine the risk level corresponding to the maximum classification probability as the fault level of the fault description text.

[0044] Preferably, the association rule table includes semantic definitions and corresponding association relationships;

[0045] The structure association tree construction unit is specifically used to determine the core semantic elements and associated semantic elements in the semantic structure conversion table according to the association relationship; determine the core semantic elements as root nodes and the associated semantic elements as child nodes to obtain the structure association tree.

[0046] Preferably, the fault semantic structure information generating unit is specifically used to generate the semantic weight of the structure association tree according to the height of each node in the structure association tree through the exponential decay method; generate the structural weight of the structure association tree through the hierarchical analysis method according to the preset judgment matrix and the low-dimensional word vector corresponding to each node of the structure association tree; perform semantic similarity calculation between the structure association tree and the standard text library to generate semantic similarity; perform structural similarity calculation between the structure association tree and the standard text library to generate structural similarity; generate similarity results based on the semantic weight, semantic similarity, structural weight and structural similarity of the structure association tree; fill in the missing values ​​of the semantic elements according to the semantic elements corresponding to the similarity results to generate fault semantic structure information.

[0047] The present invention also discloses a computer-readable medium on which a computer program is stored. When the program is executed by a processor, the method described above is implemented.

[0048] The present invention also discloses a computer device, including a memory and a processor, wherein the memory is used to store information including program instructions, the processor is used to control the execution of program instructions, and the processor implements the above method when executing the program.

[0049] The present invention also discloses a computer program product, comprising a computer program / instruction, which implements the above method when the computer program / instruction is executed by a processor.

[0050] The present invention obtains a fault description text of an electric power device; performs text preprocessing on the fault description text to generate a high-dimensional word vector and a low-dimensional word vector; performs fault grading judgment on the high-dimensional word vector through a pre-built hybrid deep neural network model to generate a fault level; performs semantic definition matching on the low-dimensional word vector through a pre-built classification model to generate a semantic structure conversion table, the semantic structure conversion table including missing values ​​of semantic elements; constructs a structural association tree according to a preset association rule table and the semantic structure conversion table; calculates similarity between the structural association tree and a preset standard text library through a text distance estimation method combining semantics and structure, fills in missing values ​​of semantic elements, and generates fault semantic structure information, which can accurately analyze and grade unstructured fault texts, fill in missing values ​​of information, improve information completeness, and enhance text analysis efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 A flowchart of a method for processing fault text of an electric power device provided by an embodiment of the present invention;

[0053] Figure 2 A flowchart of another method for processing fault text of an electric power device provided by an embodiment of the present invention;

[0054] Figure 3 A schematic structural diagram of a fault text processing device for electric power equipment provided by an embodiment of the present invention;

[0055] Figure 4 A schematic structural diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0057] It should be noted that the fault text processing method and device of an electric power equipment disclosed in the present application can be used in the field of artificial intelligence technology, and can also be used in any field other than the field of artificial intelligence technology. The application field of the fault text processing method and device of an electric power equipment disclosed in the present application is not limited.

[0058] In order to facilitate the understanding of the technical solution provided by this application, the relevant contents of the technical solution of this application are first explained below. The massive data generated during the operation of power grid equipment includes structured data that is easy to process, such as real-time monitoring data and power load information, and unstructured data that is more difficult to process, such as audio, video or text. Due to its complex structure and unclear characteristics, preprocessing of unstructured data becomes a major challenge. Therefore, the effective collection and processing of unstructured data has become the key to ensuring the stable operation of smart grids.

[0059] The specific properties of Chinese text data, such as the lack of spaces between words, the complex and variable meanings of sentences and words, and the intermingling of numbers and letters in text, pose particular challenges for text analysis in the power industry. Text records of power equipment failures detail the fault conditions encountered during operation, including information such as voltage level, substation name, equipment name, fault location, fault discovery and repair time, fault description, and severity level. Fault description and severity are key elements in assessing equipment condition. Operations and maintenance personnel use this information to determine the specific location and severity of the fault and develop appropriate repair plans.

[0060] Natural Language Processing (NLP) technology significantly improves the efficiency and accuracy of text processing by converting difficult-to-parse text data into numerical data, reducing reliance on human experience. This technology helps integrate multi-source information on power equipment, improving the scientific nature and efficiency of operation and maintenance, and promoting the power grid's advancement toward a higher level of intelligence and automation.

[0061] The following uses a fault text processing device for power equipment as an example to illustrate the implementation process of the fault text processing method for power equipment provided by an embodiment of the present invention. It is understood that the execution subject of the fault text processing method for power equipment provided by an embodiment of the present invention includes but is not limited to the fault text processing device for power equipment.

[0062] Figure 1 A flowchart of a method for processing fault text of an electric power device provided by an embodiment of the present invention is shown as follows: Figure 1 As shown, the method includes:

[0063] Step 101: Obtain a fault description text of the power equipment.

[0064] Step 102: Perform text preprocessing on the fault description text to generate high-dimensional word vectors and low-dimensional word vectors.

[0065] In this embodiment of the present invention, the textual characteristics of the fault description text were deeply analyzed, and a general vocabulary library, a specialized vocabulary library, and a list of excluded words were compiled to facilitate preliminary word segmentation and word removal of the data. Vectorization was used to convert the text data into a machine-processable numerical form.

[0066] Step 103: Use a pre-built Attention-based Hybrid DeepNatural Netword (AB-HDNN) model to perform fault classification on the high-dimensional word vector and generate a fault level.

[0067] In an embodiment of the present invention, the AB-HDNN model is used to determine the fault level and risk level, where the fault levels include general, important, and urgent.

[0068] In an embodiment of the present invention, the AB-HDNN model includes a convolutional neural network (CNN) and a bidirectional long short-term memory network (Bidirectional Long Short Term Memory, Bi-LSTM).

[0069] Step 104: Using a pre-built classification model, semantic definition matching is performed on the low-dimensional word vectors to generate a semantic structure conversion table, which includes missing values ​​of semantic elements.

[0070] In the embodiment of the present invention, the common problems of unclear description and difficulty in information extraction in fault texts are addressed by mapping texts into semantic structure transformations, thereby converting unstructured data into a structured form and improving data organization.

[0071] Step 105: Construct a structural association tree according to the preset association rule table and the semantic structure conversion table.

[0072] In the embodiment of the present invention, the association rule table includes semantic definitions and corresponding association relationships.

[0073] Step 106 : Using a text distance estimation method that combines semantics and structure, similarity is calculated between the structural association tree and a preset standard text library, and missing values ​​of semantic elements are filled to generate fault semantic structure information.

[0074] In an embodiment of the present invention, a calculation strategy of differentiated text distance estimation is used to fill in missing elements of the structural association tree and accurately extract fault information, thereby improving the level of data normalization, solving the problems of ambiguity and incomplete information in text descriptions, and thus more accurately extracting key fault information.

[0075] In the technical solution provided by the embodiment of the present invention, a fault description text of the power equipment is obtained; the fault description text is preprocessed to generate high-dimensional word vectors and low-dimensional word vectors; a fault classification judgment is performed on the high-dimensional word vectors through a pre-built hybrid deep neural network model to generate a fault level; a semantic definition matching is performed on the low-dimensional word vectors through a pre-built classification model to generate a semantic structure conversion table, which includes missing values ​​of semantic elements; a structural association tree is constructed according to a preset association rule table and the semantic structure conversion table; a text distance estimation method combining semantics and structure is used to calculate the similarity between the structural association tree and a preset standard text library, and the missing values ​​of semantic elements are filled to generate fault semantic structure information, which can accurately analyze and grade unstructured fault texts, fill in missing information values, improve information completeness, and enhance text analysis efficiency and accuracy.

[0076] Figure 2 A flowchart of another method for processing fault text of an electric power device provided by an embodiment of the present invention is shown as follows: Figure 2 As shown, the method includes:

[0077] Step 201: Obtain a fault description text of the power equipment.

[0078] In the embodiment of the present invention, each step is executed by a fault text processing device of the electric power equipment.

[0079] In an embodiment of the present invention, the text record of the power equipment fault records in detail the fault conditions that occurred during the operation of the power equipment, including but not limited to voltage level, substation name, equipment name, fault location, fault discovery event, repair time, fault description and other information.

[0080] It is worth noting that if the unstructured data is audio or video, the audio or video can be converted into fault description text through a text conversion tool.

[0081] Step 202: Use the pre-built word meta model to perform word segmentation and cleaning on the fault description text to generate a fault vocabulary sequence.

[0082] In this embodiment of the present invention, a lexical metamodel is a pre-trained text processing model based on a general vocabulary library, a specialized vocabulary library, and an exclusion list, used for text segmentation and cleaning. The general vocabulary library can be obtained from a publicly available large-scale data collection platform; the specialized vocabulary library is derived from information such as equipment names and faults provided by frontline operations and maintenance teams; and the exclusion list includes, but is not limited to, names of people, places, punctuation marks, and prepositions.

[0083] Specifically, the fault description text is input into the word meta-model for word segmentation and cleaning, and continuous sentences are decomposed into independent word sequences. Based on the exclusion word list, noise information that affects the model performance, such as names of people, places, punctuation marks and prepositions, is removed from the sentences, and the fault word sequence is output.

[0084] In the embodiment of the present invention, by constructing a general vocabulary library, a professional vocabulary library and an exclusion word list, the data content coverage is made wider; redundant information is processed and non-fault description information is removed to improve analysis efficiency.

[0085] Step 203: Vectorize the fault word sequence using vector embedding technology to generate a high-dimensional word vector.

[0086] In this embodiment of the present invention, vector embedding technology is used to capture the semantic attributes of each word in the fault word sequence based on the surrounding words. This preprocessed word sequence is converted into a machine-understandable numerical representation, and the high-dimensional and sparse word vectors are compressed into a high-dimensional word vector with a fixed dimension. As an optional solution, the high-dimensional word vector with a fixed dimension is a 64-word dimension vector.

[0087] In the embodiment of the present invention, a word meta-model is introduced to realize text segmentation and de-wording, and a vector embedding method is combined with PCA processing to make the data easy to handle and observable.

[0088] Step 204: Perform dimensionality reduction processing on the high-dimensional word vectors through the principal component analysis (PCA) method to generate low-dimensional word vectors.

[0089] In this embodiment of the present invention, since 64-dimensional high-dimensional data is difficult to display intuitively, the high-dimensional word vectors are reduced in dimension using the PCA method to generate low-dimensional word vectors. As an optional solution, the low-dimensional word vectors are 2-dimensional vectors.

[0090] At the same time, referencing fault terminology standards, we standardized semantically similar terms. For example, we unified "knife switch" and "disconnector" into "disconnector," and created a unified word list. We also unified synonyms and similar expressions to prevent the model from misclassifying different terms as different features.

[0091] Step 205: Use the pre-built hybrid deep neural network model to perform fault classification judgment on the high-dimensional word vector to generate a fault level.

[0092] In an embodiment of the present invention, the AB-HDNN model includes a convolutional neural network (CNN) and a bidirectional long short-term memory network (Bidirectional Long Short Term Memory, Bi-LSTM).

[0093] In the embodiment of the present invention, step 205 specifically includes:

[0094] Step 2051: Use a convolutional neural network (CNN) to extract features from the high-dimensional word vectors and generate local features.

[0095] In the embodiment of the present invention, CNN relies on sliding windows to perform convolution operations and uses maximum pooling to perform feature dimensionality reduction, which is suitable for extracting local features in text. Specifically, high-dimensional word vectors are input into CNN for local feature extraction and output.

[0096] Step 2052: Use a bidirectional long short-term memory network (Bi-LSTM) to extract features from the high-dimensional word vectors and generate context features.

[0097] In the embodiment of the present invention, the advantage of Bi-LSTM is that it captures the contextual information of the text, thereby supplementing the text information. Specifically, the high-dimensional word vector is input into Bi-LSTM to extract contextual features and output contextual features.

[0098] Step 2053: Concatenate the local features with the context features to generate features to be classified.

[0099] In the embodiment of the present invention, CNN and Bi-LSTM are combined to perform feature extraction using their respective advantages. The combination of the two enhances the integrity of feature extraction, which can clarify the fault feature description without affecting the feature extraction efficiency.

[0100] Specifically, the local features and context features are concatenated through the fully connected layer to generate features to be classified.

[0101] Step 2054: Calculate the features to be classified through the attention mechanism to generate a feature classification vector.

[0102] Specifically, in order to strengthen the influence of key features on the classification results, the features to be classified are input into Attention and feature weights are output; matrix multiplication operation is performed on the feature weights and the features to be classified to generate a feature classification vector.

[0103] Step 2055: Calculate the feature classification vector using a normalized activation function to generate multiple risk levels and the classification probability corresponding to each risk level.

[0104] In the embodiment of the present invention, the feature classification vector is subjected to calculation of a normalized activation function to generate the classification probability of the fault description text belonging to each risk level.

[0105] Step 2056: Determine the risk level corresponding to the maximum classification probability as the fault level of the fault description text.

[0106] In the embodiment of the present invention, a greater classification probability indicates a greater probability that the fault description text belongs to the risk level; a smaller classification probability indicates a smaller probability that the fault description text belongs to the risk level.

[0107] Specifically, the classification probabilities corresponding to each risk level are compared, and the maximum classification probability is selected; the risk level corresponding to the maximum classification probability is determined as the fault level of the fault description text.

[0108] As an optional solution, in order to facilitate the classification of fault levels in the model text output, mapping output can be performed according to a preset mapping table of risk levels and fault levels. For example, the mapping table of risk levels and fault levels is shown in Table 1:

[0109] Table 1

[0110] Fault level Risk Level generally 1 important 2 urgent 3

[0111] For example, if the risk level is determined to be 1, according to the output method of the model, the fault level can be output as general or as 1.

[0112] Step 206: Using the pre-built classification model, semantic definition matching is performed on the low-dimensional word vectors to generate a semantic structure conversion table, which includes missing values ​​of semantic elements.

[0113] In an embodiment of the present invention, the classification model can be obtained by training a neural network using a pre-set semantic definition set and a low-dimensional word vector training set. The semantic definition set includes but is not limited to the overall device name, specific location, attributes, fault problem, and severity.

[0114] Specifically, the low-dimensional word vector is input into the classification model for semantic definition matching, and the semantic definition that matches each word in the low-dimensional word vector is extracted from the semantic definition set, and the semantic definition is used as the part-of-speech information of the word for subsequent processing.

[0115] Specifically, the position information of each semantic definition and each word in the text is obtained. For multiple words corresponding to the same semantic definition, the position information of the semantic definition and the position information of the word in the text are calculated using the distance formula dis(X,Y) = |xy| to generate the lexeme distance. The smallest lexeme distance is selected, and the word corresponding to the smallest lexeme distance is determined as the semantic element corresponding to the semantic definition. Where dis(X,Y) is the lexeme distance, x is the position information of the semantic definition in the text, and y is the position information of the word in the text.

[0116] In the embodiment of the present invention, the smaller the word position distance is, the higher the semantic association between the two words is, and thus the stronger the semantic association between the two words is numerically reflected, and they should be listed in the same semantic structure conversion table. For example, the semantic structure conversion table is shown in Table 2:

[0117] Table 2

[0118] Semantic Definition Semantic elements Total device name Cooler system Specific location Oil pillow property Oil temperature Failure issues rust degree Too high

[0119] In the embodiment of the present invention, unstructured text data is converted into a structured form, which can improve the efficiency of subsequent processing.

[0120] It is worth noting that the semantic structure conversion table shown in Table 2 is a conversion table without missing values. However, in actual production scenarios, since unstructured information is basically artificially formed, it is difficult to accurately analyze and obtain semantic elements. The problems of colloquialism and missing elements in actual texts lead to unclear descriptions, and there will be problems of missing values ​​of semantic elements in the semantic structure conversion table. In response to missing values ​​of semantic elements, the present invention constructs a structural association tree to include the semantic and structural information of the text in the tree, and proposes a text distance difference estimation strategy based on semantic and structural association to improve the quality of fault text and accurately identify it.

[0121] Step 207: Determine the core semantic elements and associated semantic elements in the semantic structure conversion table according to the association relationship.

[0122] In the embodiment of the present invention, the association relationship can be set according to actual needs, and the embodiment of the present invention does not limit this. As an optional solution, the association relationship is shown in Table 3:

[0123] Table 3

[0124] Semantic Definition Element Association Total device name subject It has a modifying relationship with the attributive and a supplementing relationship with the adverbial. Specific location adverbial Complementary relationship with the subject property attribute Modifying relationship with adverbials Failure issues object Core relationship / modifying relationship with attributive degree adverbial Core relationship / complementary relationship with the subject

[0125] In an embodiment of the present invention, according to the association relationship, the fault problem is determined as the core semantic element, and the total equipment name, specific location, attributes and degree are determined as associated semantic elements; or, according to the association relationship, the degree is determined as the core semantic element, and the total equipment name, specific location, attributes and fault problem are determined as associated semantic elements.

[0126] Step 208: Determine the core semantic element as the root node and the associated semantic elements as child nodes to obtain a structural association tree.

[0127] Specifically, the core semantic element is determined as the root node of the tree, and the associated semantic elements are determined as the child nodes of the tree to construct a structural association tree, wherein the arrows in the tree structure point to dependency relationships, which include supplementary relationships and modified relationships.

[0128] Step 209 : Using a text distance estimation method that combines semantics and structure, similarity is calculated between the structural association tree and a preset standard text library, and missing values ​​of semantic elements are filled to generate fault semantic structure information.

[0129] In this embodiment of the present invention, after constructing a structural association tree, a text distance estimation method combining semantics and structure is used to align the fault information text to be extracted with the element set of a standard text library, and to complete any missing key element information for more accurate problem extraction. This invention introduces different weighting parameters when estimating semantic and structural text distances, which determines the use of differentiated assignment strategies.

[0130] In the embodiment of the present invention, step 209 specifically includes:

[0131] Step 2091: Generate the semantic weight of the structure association tree according to the height of each node in the structure association tree through the exponential decay method.

[0132] In this embodiment of the present invention, when evaluating semantic similarity, an exponential decay method is used to assign weights to nodes. The core semantic element is located at the top of the tree as the root node, while other nodes are gradually ranked lower in the tree based on their degree of association with the root node. Different weights are assigned to nodes based on their height in the tree. The specific calculation formula is as follows:

[0133]

[0134] Among them, W t is the weight of node t, H t is the height of node t in the tree.

[0135] Combine the weights of each node according to the structural association tree to obtain the semantic weight W of the structural association tree i .

[0136] Step 2092: Calculate the structure weight of the structure association tree by using the hierarchical analysis method based on the preset judgment matrix and the low-dimensional word vector corresponding to each node of the structure association tree.

[0137] In the embodiment of the present invention, when assigning structural association weights, the hierarchical analysis method is used to determine the weights of different elements through hierarchical processing. Complex problems are regarded as a collection of multiple elements, and weights are assigned based on the relative importance of each element. The specific implementation steps are as follows:

[0138] Step a: Determine the judgment matrix A.

[0139] Determine the key elements that can determine fault information based on the problem. i , a j (i,j=1,2,3,...,n), a ij for a i to a j The strength of association.

[0140]

[0141] Step b: determine the relative weight of each association relationship, that is, the structural weight.

[0142] Determine the maximum eigenvalue a by the judgment matrix A max , and the low-dimensional word vectors corresponding to each node of the structure association tree are normalized to obtain the structural weight of the structure association tree.

[0143] Step 2093: Calculate semantic similarity between the structural association tree and the standard text library to generate semantic similarity.

[0144] In the embodiment of the present invention, the standard text library is a database of pre-stored standard semantics and association relationships.

[0145] Specifically, semantic similarity is calculated for each node in the structure association tree T1 and the standard text library T2 to generate semantic similarity. The calculation formula is as follows:

[0146]

[0147]

[0148]

[0149]

[0150] Among them, S x (T1, T2) is the semantic similarity between the structure association tree T1 and the standard text library T2; Hx12 The information content of the common semantic elements of the structure association tree and the standard text library; H xt1 is the total amount of semantic information of the structure association tree; H xt2 is the total amount of semantic information in the standard text library; p i1 is the frequency of occurrence of nodes (core semantic elements and associated semantic elements) in the structure association tree in the fault description text; i2 is the frequency of occurrence of semantic elements in the standard text library in the fault description text; i12 The frequency of occurrence of common semantic elements of the structure association tree and standard text library in the fault description text; h i1 h is the height of the node (core semantic element and associated semantic element) in the structure association tree; i2 is the height of the semantic element in the standard text library in the tree; h i12 The height of the common semantic elements of the structure association tree and the standard text library in the tree.

[0151] Step 2094: Calculate the structural similarity between the structural association tree and the standard text library to generate structural similarity.

[0152] In the embodiment of the present invention, the standard text library is a database of pre-stored standard semantics and association relationships.

[0153] Specifically, semantic similarity is calculated for each node in the structure association tree T1 and the standard text library T2 to generate semantic similarity. The calculation formula is as follows:

[0154]

[0155]

[0156]

[0157]

[0158] Among them, S y (T1, T2) is the structural similarity between the structural association tree T1 and the standard text library T2; H y12 H is the information volume of the common structural elements (association relationships) of the structure association tree and the standard text library; yt1 is the total amount of structure association tree structure information; H yt2 is the total amount of standard text library structure information; p j1 is the frequency of occurrence of the structural elements (association relationships) in the structural association tree in the fault description text; j2 is the frequency of occurrence of structural elements (association relationships) in the fault description text in the standard text library; j12is the frequency of occurrence of common structural elements (association relationships) of the structure association tree and the standard text library in the fault description text; W j is the structural weight of the structural association tree.

[0159] Step 2095: Generate a similarity result based on the semantic weight and semantic similarity of the structure association tree, and the structural weight and structural similarity of the structure association tree.

[0160] Specifically, by S(T1, T2)=W i S x (T1,T2)+W j S y (T1, T2), calculate the semantic weight, semantic similarity, structural weight and structural similarity of the structural association tree to generate the similarity result. Among them, S(T1, T2) is the similarity result, W i is the semantic weight of the structure association tree, W j is the structural weight of the structural association tree, S y (T1, T2) is the structural similarity between the structural association tree T1 and the standard text library T2, S x (T1, T2) is the semantic similarity between the structure association tree T1 and the standard text base T2.

[0161] Step 2096: Fill in missing values ​​of semantic elements according to the semantic elements corresponding to the similarity results to generate fault semantic structure information.

[0162] In an embodiment of the present invention, based on the similarity results, the comparison differences between the fault text to be processed and the standard vocabulary are obtained, and the semantic element values ​​of the standard vocabulary corresponding to the highest similarity result are filled to the missing values ​​of the semantic elements to generate fault semantic structure information.

[0163] In the embodiment of the present invention, missing elements are completed through difference alignment, and a series of processes such as text analysis, processing and optimization are completed, so that the data in the power professional text library is more complete and accurate, and the fault information points can be located more clearly.

[0164] It is worth noting that the acquisition, storage, use, and processing of data in the technical solutions of this application are in compliance with the relevant provisions of laws and regulations. The user information in the embodiments of this application is obtained through legal and compliant channels, and the acquisition, storage, use, and processing of user information are authorized and agreed by the customer.

[0165] It is worth noting that the information collected in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of relevant countries and regions, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0166] It is worth noting that the technical solution provided in this application provides users with corresponding operation entrances for them to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered.

[0167] In the technical solution of the fault text processing method of the power equipment provided by the embodiment of the present invention, the fault description text of the power equipment is obtained; the text preprocessing is performed on the fault description text to generate high-dimensional word vectors and low-dimensional word vectors; the high-dimensional word vectors are subjected to fault classification judgment through a pre-built hybrid deep neural network model to generate a fault level; the low-dimensional word vectors are subjected to semantic definition matching through a pre-built classification model to generate a semantic structure conversion table, and the semantic structure conversion table includes missing values ​​of semantic elements; according to a preset association rule table and the semantic structure conversion table, a structural association tree is constructed; through a text distance estimation method combining semantics and structure, the structural association tree and a preset standard text library are calculated for similarity, and the missing values ​​of semantic elements are filled to generate fault semantic structure information, which can accurately analyze and grade unstructured fault texts, fill in missing information values, improve information completeness, and enhance text analysis efficiency and accuracy.

[0168] Figure 3 A schematic diagram of a fault text processing device for an electric power device according to an embodiment of the present invention is provided. The device is used to execute the fault text processing method for the electric power device. Figure 3 As shown, the device includes: an acquisition unit 11, a text preprocessing unit 12, a fault grading unit 13, a semantic structure conversion unit 14, a structure association tree construction unit 15 and a fault semantic structure information generation unit 16.

[0169] The acquiring unit 11 is used to acquire a fault description text of the electric power equipment.

[0170] The text preprocessing unit 12 is used to perform text preprocessing on the fault description text to generate high-dimensional word vectors and low-dimensional word vectors.

[0171] The fault grading unit 13 is used to perform fault grading judgment on the high-dimensional word vector through a pre-built hybrid deep neural network model to generate a fault level.

[0172] The semantic structure conversion unit 14 is used to perform semantic definition matching on low-dimensional word vectors through a pre-built classification model to generate a semantic structure conversion table, which includes missing values ​​of semantic elements.

[0173] The structure association tree construction unit 15 is used to construct a structure association tree according to a preset association rule table and a semantic structure conversion table.

[0174] The fault semantic structure information generating unit 16 is used to calculate the similarity between the structure association tree and the preset standard text library through a text distance estimation method combining semantics and structure, and fill in the missing values ​​of semantic elements to generate fault semantic structure information.

[0175] In an embodiment of the present invention, the text preprocessing unit 12 is specifically used to perform word segmentation and cleaning processing on the fault description text through a pre-constructed word meta model to generate a fault vocabulary sequence; vectorize the fault vocabulary sequence through vector embedding technology to generate a high-dimensional word vector; and perform dimensionality reduction processing on the high-dimensional word vector through the principal component analysis method to generate a low-dimensional word vector.

[0176] In an embodiment of the present invention, the hybrid deep neural network model includes a convolutional neural network and a bidirectional long short-term memory network; the fault grading unit 13 is specifically used to perform feature extraction on high-dimensional word vectors through the convolutional neural network to generate local features; perform feature extraction on high-dimensional word vectors through the bidirectional long short-term memory network to generate context features; splice the local features with the context features to generate features to be classified; calculate the features to be classified through the attention mechanism to generate feature classification vectors; calculate the feature classification vectors through the normalized activation function to generate multiple risk levels and the classification probability corresponding to each risk level; determine the risk level corresponding to the maximum classification probability as the fault level of the fault description text.

[0177] In an embodiment of the present invention, the association rule table includes semantic definitions and corresponding association relationships; the structural association tree construction unit 15 is specifically used to determine the core semantic elements and associated semantic elements in the semantic structure conversion table according to the association relationships; the core semantic elements are determined as root nodes, and the associated semantic elements are determined as child nodes to obtain a structural association tree.

[0178] In an embodiment of the present invention, the fault semantic structure information generation unit 16 is specifically used to generate the semantic weight of the structure association tree according to the height of each node in the structure association tree through the exponential decay method; generate the structural weight of the structure association tree through the hierarchical analysis method according to the preset judgment matrix and the low-dimensional word vector corresponding to each node of the structure association tree; perform semantic similarity calculation between the structure association tree and the standard text library to generate semantic similarity; perform structural similarity calculation between the structure association tree and the standard text library to generate structural similarity; generate similarity results based on the semantic weight, semantic similarity, structural weight and structural similarity of the structure association tree; fill in the missing values ​​of the semantic elements according to the semantic elements corresponding to the similarity results to generate fault semantic structure information.

[0179] In the scheme of the embodiment of the present invention, a fault description text of the power equipment is obtained; the fault description text is preprocessed to generate high-dimensional word vectors and low-dimensional word vectors; a fault classification judgment is performed on the high-dimensional word vectors through a pre-built hybrid deep neural network model to generate a fault level; a semantic definition matching is performed on the low-dimensional word vectors through a pre-built classification model to generate a semantic structure conversion table, which includes missing values ​​of semantic elements; a structural association tree is constructed according to a preset association rule table and the semantic structure conversion table; a text distance estimation method combining semantics and structure is used to calculate the similarity between the structural association tree and a preset standard text library, and the missing values ​​of semantic elements are filled to generate fault semantic structure information, which can accurately analyze and grade unstructured fault texts, fill in missing information values, improve information completeness, and enhance text analysis efficiency and accuracy.

[0180] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer device. Specifically, the computer device may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0181] An embodiment of the present invention provides a computer device including a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions. When the program instructions are loaded and executed by the processor, the steps of the embodiment of the fault text processing method of the above-mentioned power equipment are implemented. For a specific description, please refer to the embodiment of the fault text processing method of the above-mentioned power equipment.

[0182] Reference below Figure 4 , which shows a structural diagram of a computer device 600 suitable for implementing an embodiment of the present application.

[0183] like Figure 4 As shown, the computer device 600 includes a central processing unit (CPU) 601, which can perform various appropriate tasks and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. Various programs and data required for the operation of the computer device 600 are also stored in the RAM 603. The CPU 601, ROM 602, and 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.

[0184] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, and the like; an output section 607 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 608 including devices such as a hard disk; and a communication section 609 including a network interface card such as a LAN card or a modem. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. Removable media 611, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 610 as needed, so that computer programs read therefrom can be installed in the storage section 608 as needed.

[0185] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program including program code for executing the methods illustrated in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication portion 609 and / or installed from removable media 611.

[0186] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0187] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0188] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0189] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0190] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0191] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0192] The acquisition, storage, use, and processing of data in this application's technical solution comply with relevant national laws and regulations.

[0193] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned. They should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0194] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0195] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0196] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0197] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for processing fault text of electric power equipment, characterized in that: The method comprises: Get the fault description text of the power equipment; Performing text preprocessing on the fault description text to generate high-dimensional word vectors and low-dimensional word vectors; Performing fault classification judgment on the high-dimensional word vector through a pre-built hybrid deep neural network model to generate a fault level; By using a pre-built classification model, the low-dimensional word vector is matched with a semantic definition to generate a semantic structure conversion table, wherein the semantic structure conversion table includes missing values ​​of semantic elements; According to the preset association rule table and the semantic structure conversion table, a structure association tree is constructed; By using a text distance estimation method that combines semantics and structure, similarity calculation is performed between the structural association tree and a preset standard text library, and missing values ​​of the semantic elements are filled to generate fault semantic structure information.

2. The method for processing fault text of electric power equipment according to claim 1, characterized in that: The performing text preprocessing on the fault description text to generate a high-dimensional word vector and a low-dimensional word vector includes: The fault description text is segmented and cleaned by using a pre-built word meta model to generate a fault vocabulary sequence; Vectorizing the fault vocabulary sequence through vector embedding technology to generate the high-dimensional word vector; The high-dimensional word vector is subjected to dimensionality reduction processing by a principal component analysis method to generate the low-dimensional word vector.

3. The method for processing fault text of electric power equipment according to claim 1, characterized in that: The hybrid deep neural network model includes a convolutional neural network and a bidirectional long short-term memory network; The pre-built hybrid deep neural network model is used to perform fault classification judgment on the high-dimensional word vector to generate a fault level, including: By using the convolutional neural network, feature extraction is performed on the high-dimensional word vector to generate local features; Through the bidirectional long short-term memory network, feature extraction is performed on the high-dimensional word vector to generate context features; The local features are combined with the context features to generate features to be classified; The features to be classified are calculated through an attention mechanism to generate a feature classification vector; The feature classification vector is calculated by a normalized activation function to generate a plurality of risk levels and a classification probability corresponding to each risk level; The risk level corresponding to the maximum classification probability is determined as the fault level of the fault description text.

4. The method for processing fault text of electric power equipment according to claim 1, characterized in that: The association rule table includes semantic definitions and corresponding association relationships; The step of constructing a structure association tree according to the preset association rule table and the semantic structure conversion table includes: According to the association relationship, determining the core semantic elements and the associated semantic elements in the semantic structure conversion table; The core semantic element is determined as a root node, and the associated semantic element is determined as a child node to obtain the structural association tree.

5. The method for processing fault text of electric power equipment according to claim 1, characterized in that: The text distance estimation method combining semantics and structure calculates the similarity between the structural association tree and a preset standard text library, fills in the missing values ​​of the semantic elements, and generates fault semantic structure information, including: Generate a semantic weight of the structure association tree according to the height of each node in the structure association tree by using an exponential decay method; By using the hierarchical analysis method, the structure weight of the structure association tree is generated by calculating according to the preset judgment matrix and the low-dimensional word vector corresponding to each node of the structure association tree; Calculate semantic similarity between the structural association tree and a standard text library to generate semantic similarity; Calculate the structural similarity between the structural association tree and the standard text library to generate structural similarity; Generate a similarity result according to the semantic weight and semantic similarity of the structure association tree, the structure weight and structure similarity of the structure association tree; The missing values ​​of the semantic elements are filled in according to the semantic elements corresponding to the similarity results to generate fault semantic structure information.

6. A fault text processing device for electric power equipment, characterized in that: The device comprises: An acquisition unit, used for acquiring a fault description text of the electric power equipment; A text preprocessing unit, used to perform text preprocessing on the fault description text to generate a high-dimensional word vector and a low-dimensional word vector; A fault grading unit, used to perform fault grading judgment on the high-dimensional word vector through a pre-built hybrid deep neural network model to generate a fault level; A semantic structure conversion unit, used to perform semantic definition matching on the low-dimensional word vector through a pre-built classification model to generate a semantic structure conversion table, wherein the semantic structure conversion table includes missing values ​​of semantic elements; A structure association tree construction unit, used to construct a structure association tree according to a preset association rule table and the semantic structure conversion table; The fault semantic structure information generating unit is used to calculate the similarity between the structural association tree and a preset standard text library through a text distance estimation method combining semantics and structure, and fill in the missing values ​​of the semantic elements to generate fault semantic structure information.

7. The fault text processing device for electric power equipment according to claim 6, characterized in that: The text preprocessing unit is specifically used to perform word segmentation and cleaning processing on the fault description text through a pre-constructed word meta model to generate a fault vocabulary sequence; to vectorize the fault vocabulary sequence through vector embedding technology to generate the high-dimensional word vector; and to perform dimensionality reduction processing on the high-dimensional word vector through a principal component analysis method to generate the low-dimensional word vector.

8. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the fault text processing method of the electric power equipment described in any one of claims 1 to 5 is implemented.

9. A computer device comprising a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions, characterized in that: When the program instructions are loaded and executed by the processor, the fault text processing method of the electric power equipment according to any one of claims 1 to 5 is implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the fault text processing method of the electric power equipment according to any one of claims 1 to 5 is implemented.

Images