CN113239143A - Power transmission and transformation equipment fault processing method and system fusing power grid fault case base - Google Patents

Power transmission and transformation equipment fault processing method and system fusing power grid fault case base Download PDF

Info

Publication number
CN113239143A
CN113239143A CN202110467599.XA CN202110467599A CN113239143A CN 113239143 A CN113239143 A CN 113239143A CN 202110467599 A CN202110467599 A CN 202110467599A CN 113239143 A CN113239143 A CN 113239143A
Authority
CN
China
Prior art keywords
fault
power grid
text
information
self
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110467599.XA
Other languages
Chinese (zh)
Other versions
CN113239143B (en
Inventor
秦佳峰
杨祎
白德盟
郑文杰
林颖
李程启
刘萌
辜超
吕学宾
周超
李龙龙
李�杰
王建
孙景文
贾然
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Corp of China SGCC
Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd
Original Assignee
State Grid Corp of China SGCC
Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by State Grid Corp of China SGCC, Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd filed Critical State Grid Corp of China SGCC
Priority to CN202110467599.XA priority Critical patent/CN113239143B/en
Publication of CN113239143A publication Critical patent/CN113239143A/en
Application granted granted Critical
Publication of CN113239143B publication Critical patent/CN113239143B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/335—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/334—Query execution
    • G06F16/3344—Query execution using natural language analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/951—Indexing; Web crawling techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Abstract

The invention belongs to the field of equipment fault processing, and provides a power transmission and transformation equipment fault processing method and system fusing a power grid fault case base. Inputting a fault phenomenon text into a multi-layer semantic and self-attention text matching model, and obtaining fault cases with similar set quantity based on a structured fault case library; calculating the keyword matching degree of the screened fault cases and a power grid standard library by using a keyword matching algorithm of self-adaptive weight, and screening power grid standards exceeding a set matching degree threshold value according to the matching degree; and inputting the fault phenomenon text and the screened power grid standard into a multi-layer semantic and self-attention text matching model, obtaining equipment fault information and a fault maintenance scheme based on a power grid domain database, and pushing the equipment fault information and the fault maintenance scheme to a client.

Description

Power transmission and transformation equipment fault processing method and system fusing power grid fault case base
Technical Field
The invention belongs to the field of equipment fault processing, and particularly relates to a power transmission and transformation equipment fault processing method and system fusing a power grid fault case base.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
In recent years, in fields with a large accumulation of historical text data such as medical care, telecommunication, finance and the like, natural language processing technology has shown a driving effect on the development and progress of the industry. With the continuous maturity of smart power grids and the promotion of ubiquitous power internet of things, the application requirements of natural language processing technology in the power industry are continuously expanded. With the continuous development of power grid services, the scale of text data related to the power grid services is continuously increased, the forms are continuously enriched, and the data sea quantization and diversification characteristics are presented. The power grid power transmission and transformation equipment fault text is accumulation of field overhaul experience in power grid fault overhaul work, contains detailed contents of equipment faults and overhaul, and has high professional value.
Some natural language processing technologies have been applied in the chinese grid text field, including: classifying the rank of the defective text based on text classification techniques, querying the text based on keyword matching, and the like. Although the applications are long, the inventor finds that equipment fault information has the characteristics of complexity and diversity, and the problems that a single technology is narrow in application range, text data is not deeply mined, a power grid equipment fault analysis scene is not provided and the like exist at present.
Disclosure of Invention
In order to solve the technical problems in the background art, the invention provides a power transmission and transformation equipment fault processing method and system fusing a power grid fault case base.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention provides a power transmission and transformation equipment fault processing method fusing a power grid fault case library.
A power transmission and transformation equipment fault processing method fusing a power grid fault case library comprises the following steps:
inputting fault phenomenon texts into a multi-layer semantic and self-attention text matching model, and obtaining fault cases with similar set quantity based on a structured fault case library;
calculating the keyword matching degree of the screened fault cases and a power grid standard library by using a keyword matching algorithm of self-adaptive weight, and screening power grid standards exceeding a set matching degree threshold value according to the matching degree;
and inputting the fault phenomenon text and the screened power grid standard into a multi-layer semantic and self-attention text matching model, obtaining equipment fault information and a fault maintenance scheme based on a power grid domain database, and pushing the equipment fault information and the fault maintenance scheme to a client.
As an embodiment, the structured fault case library is constructed by extracting equipment fault information from a power grid fault text.
In the process of constructing the structured fault case library, a text classification model embedded with the graph volume words is used for classifying the fault categories of the power grid fault texts, so that the case library attribute is enriched.
As an implementation mode, in the process of constructing the structured fault case library, the unified attribute template is adopted to extract the equipment fault information.
As an implementation mode, the multilayer semantic and self-attention text matching model is innovated in a coding layer and a sentence-level semantic interaction layer, the coding layer is composed of a Convolutional Neural Network (CNN) and a self-attention mechanism network structure, and the sentence-level semantic interaction layer can perform text semantic interaction from a sentence angle by using attention.
As an implementation manner, the process of calculating the keyword matching degree between the screened fault cases and the power grid standard library is as follows:
counting based on the fault cases with similar set number, recording the weights of different keywords, and obtaining a keyword list with the weights;
and calculating the matching degree of the current keyword list and the keywords of the power grid standard.
In one embodiment, the power grid domain database is constructed by acquiring power grid fault question and answer pair data based on a crawler technology.
The invention provides a power transmission and transformation equipment fault processing system fused with a power grid fault case library.
A power transmission and transformation equipment fault processing system fused with a power grid fault case library comprises:
the fault case screening module is used for inputting fault phenomenon texts into a multi-layer semantic and self-attention text matching model and obtaining fault cases with similar set quantity based on a structured fault case library;
the power grid standard screening module is used for calculating the matching degree of the screened fault cases and the keywords of the power grid standard library by using a keyword matching algorithm of self-adaptive weight, and screening the power grid standard exceeding a set matching degree threshold value according to the matching degree;
and the fault information and maintenance scheme acquisition module is used for inputting the fault phenomenon text and the screened power grid standard into a multi-layer semantic and self-attention text matching model, acquiring equipment fault information and a fault maintenance scheme based on the power grid domain database, and pushing the equipment fault information and the fault maintenance scheme to the client.
A third aspect of the invention provides a computer-readable storage medium.
A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps in the method for handling faults in a power transmission and transformation apparatus that merges grid fault cases as described above.
A fourth aspect of the invention provides a computer apparatus.
A computer device comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the program to realize the steps in the power transmission and transformation device fault processing method fusing the power grid fault case base.
Compared with the prior art, the invention has the beneficial effects that:
(1) the invention captures the global information of words through the graph convolution structure, captures the local information in the sample according to the currently input sample, and fuses the long-distance semantic information by using the attention mechanism, thereby realizing the new structure combining the global information and the local information of the words, breaking through the problem that the neural network is limited by the current sample and can not capture the global information, displaying the result on a plurality of data sets that the model achieves good effect, and having better effect than that of most models under the condition of not using extra word vectors.
(2) The invention provides a text matching model based on multilayer semantics and self-attention. A model structure combining a self-attention mechanism and a convolutional neural network is adopted as an encoder in the model, so that local features and context features are effectively unified, and the problem that the convolutional neural network only focuses on the local features is solved. Meanwhile, for better matching of text pairs, an interactive multilayer semantic structure is provided, and sentence-level interaction is better performed.
(3) The invention provides a keyword algorithm of self-adaptive weight, which screens out relevant power grid standards aiming at fault phenomena, can take the occurrence frequency of keywords into consideration range, and distributes weight according to the magnitude of word frequency.
(4) The invention uses the crawler technology to crawl fault analysis data from a network, processes the data by methods of noise elimination, screening and the like, and uses the processed fault analysis question-answer to construct a data set in the field of power grid faults, thereby helping the invention to provide fault analysis with easy comprehensibility.
(5) The method is based on the fault case library, combines the field database, the power grid standard library and the power grid field data set, can fully utilize the relevant text knowledge of the power grid, can provide possible fault reasons aiming at equipment fault phenomena, can timely make a targeted power grid power transmission and transformation equipment fault detection and maintenance scheme, provides relevant power grid standards, and provides fault analysis with easy comprehensibility.
Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the invention and together with the description serve to explain the invention and not to limit the invention.
Fig. 1 is a flowchart of a method for processing faults of power transmission and transformation equipment fusing a power grid fault case library according to an embodiment of the present invention;
FIG. 2 is a complete flow of text structuring according to an embodiment of the present invention;
FIG. 3 is a network structure of a text classification algorithm of an embodiment of the present invention;
FIG. 4 is a network structure of a text matching algorithm of an embodiment of the present invention;
FIG. 5 is the results of a standard topk experiment of an embodiment of the present invention;
FIG. 6 is a block diagram of an encoder of an embodiment of the present invention;
FIG. 7 is a system architecture diagram of an embodiment of the present invention;
fig. 8 is a fault classification flow diagram of an embodiment of the invention.
Detailed Description
The invention is further described with reference to the following figures and examples.
It is to be understood that the following detailed description is exemplary and is intended to provide further explanation of the invention as claimed. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
Example one
With reference to fig. 1 and fig. 7, the method for processing the fault of the power transmission and transformation equipment fusing the power grid fault case base in the embodiment specifically includes the following steps:
step S101: and inputting the fault phenomenon text into a multi-layer semantic and self-attention text matching model, and obtaining the fault cases with similar set quantity based on a structured fault case library.
In specific implementation, the structured fault case library is constructed according to equipment fault information extracted from a power grid fault text.
The structured fault case library is mainly constructed by a power grid equipment fault text data information extraction technology, and aims to extract information meaningful for power grid equipment faults and defect description by analyzing and processing unstructured text data to form structured data, so that subsequent operations such as fault analysis are facilitated.
Considering the diversity of the power grid fault text description, a unified attribute template needs to be manually constructed to extract attributes, the types of the attributes are mainly divided into numerical attributes, phrase type attributes and sentence description type attributes, and the specific flow is shown in fig. 2.
1) Numerical type Properties
2) Non-numeric attributes
The non-numerical attribute mainly refers to the attribute in text form, and there are two main expression forms: one is based on the attributes of the phrase form and one is based on the attributes of the sentence form.
The attributes of the phrase form include the season, the device origin, the device merchant and the like, and in addition, the keywords are also in the form of the phrase as expression. Aiming at the method mainly combining named entity recognition technology with word list matching, the named entity recognition technology adopts a Bert-LSTM-CRF model to recognize entities in a text and extracts attributes by combining common place names, seasons and other word lists. The key attribute is mainly extracted by a TF-IDF method.
And the sentence-form attributes comprise attributes of fault description, overhaul process and the like. Firstly, extracting corresponding description through a keyword matching technology, and then finding out the description of the fault phenomenon as a structural attribute by combining the keyword with a named entity identification method for fault description information, wherein the specific effect is shown in table 1. And then, for the maintenance process, classifying the maintenance method of each sentence by using a BERT classification model on the basis of sentence division, and taking the classified category as the maintenance flow attribute, which is specifically shown in the table 2.
Table 1 fault phenomenon extraction effect example
Figure BDA0003044741960000071
Figure BDA0003044741960000081
Table 2 examination method effect example
Figure BDA0003044741960000082
In the process of constructing the structured fault case library, the text classification model embedded with the graph volume words is used for classifying the fault categories of the power grid fault texts, and the case library attributes are enriched.
As shown in fig. 3, the fault category of the fault text is classified by using a text classification algorithm, and the classified result is used as the fault category attribute of the current fault case. Aiming at the fault category of the fault text, the invention provides a text classification algorithm embedded based on graph convolution words to perform classification tasks, and the specific flow is shown in fig. 8. The traditional classification algorithm can learn the syntactic information and the context information of the current text, and ignore the discontinuous global word co-occurrence information, long-distance semantic information and other global information in the corpus. The algorithm provided by the embodiment can fuse the syntactic information, the context information and the global information of the current text, breaks through the problem that the neural network is limited by the current sample and cannot capture the global information, and can have better performance on a less-labeled training set due to the structural characteristics of the graph.
The algorithm firstly constructs a simple network on all linguistic data, edges among word nodes are weights obtained based on a word co-occurrence method, edges of text nodes and word nodes are weights obtained based on TF-IDF, a word vector representation with global information and local information is obtained by capturing global co-occurrence information of words and combining current context by using a graph convolution neural network, then capturing time sequence characteristics of a text by using capturing time sequence characteristics of RNN, then capturing local characteristics among words by using CNN, and finally realizing text classification by using long-distance semantic information and text context information in an attention-driven efficient fusion network, and then mainly introducing a construction of a word embedding layer and a semantic fusion layer.
Constructing a word embedding layer:
the word embedding layer is mainly realized on a topological graph constructed based on a corpus. In the process of constructing the topological graph, weights of the document nodes and the word nodes are calculated by adopting a TF-IDF (term frequency-inverse document frequency) method, and compared with a method for calculating the weights based on word frequency, the TF-IDF method can well reflect the relationship between the document nodes and the word nodes. In order to make good use of global word co-occurrence information, a sliding fixed window method is adopted to obtain global word co-occurrence information in a corpus. Weights A for word node i and word node jijWe use the following formula to make the calculation.
Figure BDA0003044741960000091
Wherein f (i, j) represents the co-occurrence score of two words, and the specific calculation formula is as follows:
Figure BDA0003044741960000092
Figure BDA0003044741960000101
wherein C isW(i, j) represents slipNumber of simultaneous occurrences of word i and word j in the moving window, CWRepresenting the total number of sliding windows in the corpus. Using the above equations, an adjacency matrix A of the graph can be obtained, and a degree matrix D of the graph can be obtained from the adjacency matrix A, where D isii＝∑jAij。
In the topology graph constructed by the word and document nodes, the word node may transfer the node information to the adjacent word node and document node, and the document node may transfer the node information to the adjacent word node. By transmitting information from neighboring nodes, the byte-point information contains global word co-occurrence information, while the document nodes contain remote semantic information. Therefore, two layers of GCN networks are applied to a topological graph to extract node information in the graph, meanwhile, initial information of the nodes adopts a one-hot form, an initial node characteristic matrix X is set as an identity matrix, and a specific calculation formula is as follows:
Figure BDA0003044741960000102
wherein
Figure BDA0003044741960000103
Can be expressed as
Figure BDA0003044741960000104
Is a normalized symmetric adjacency matrix, W1And W0Representing a parameter matrix, and the ReLU is a ReLU activation function.
For the resulting node features, because the word node HWThe global co-occurrence information of the words is contained, and the word node characteristics are used as word vector representation of input data. And text node HDContains long-distance semantic information, and is fused into the semantic representation of the text through an attention mechanism in subsequent processing.
Thus, for an input sequence of n words
Figure BDA0003044741960000105
Is convolved by an embeddinAfter g, the text sequence state vector is
Figure BDA0003044741960000106
Wherein
Figure BDA0003044741960000107
k represents the node feature dimension of the graph convolution.
The word vectors obtained by the pre-training method in the large-scale corpus are further fused into the current word embedding layer to help, the word vectors are fused by adopting a connection method, and the specific formula is as follows
Figure BDA0003044741960000111
Wherein the content of the first and second substances,
Figure BDA0003044741960000112
representing the representation of the word i in the pre-training word vector,
Figure BDA0003044741960000113
a word vector representation representing the word i,
Figure BDA0003044741960000114
representing word vector representation after the word passes through the graph convolution word embedding layer, the text vector at this time is
Figure BDA0003044741960000115
And (3) semantic fusion layer:
the text utilizes an attention mechanism in a semantic fusion layer to help a model better fuse the context semantic information and the long-distance semantic information of the text. In the graph volume word embedding layer, the high-order vector representation z of the text can be obtained through the relation between the text and the wordgThe long-distance semantic information of the text is contained, and the high-order vector representation z of the current text can be obtained through a CNN coding layercWhich contains context information for the text. In order to better fuse the information in the two text vectorsTaken together, as much as possible provides the correct information needed for text classification, a sentence-level attention mechanism is introduced to aid in information fusion, and the detailed block diagram is shown in the figure.
In the process of semantic fusion, a vector containing text context information is used as a main body, and important information is extracted from a long-distance semantic vector by an attention mechanism to perform semantic fusion. The context semantic vector and the long-distance semantic vector are mapped into the same space through matrix transformation, then the correlation degree between the two vectors is calculated by using a Sigmoid function, and finally the final semantic vector of the text is obtained by using the correlation degree. The concrete implementation formula is as follows:
zs1＝W1zc+b1
zs2＝W2zg+b2
Figure BDA0003044741960000116
Figure BDA0003044741960000117
wherein, W1，W2，W3Is a parameter matrix, σ denotes a Sigmoid function, b1，b2，b3Which is indicative of a bias parameter that is,
Figure BDA0003044741960000121
representing a splicing operation, zsIs the last semantic representation vector of the current text.
The experimental results are as follows:
the model and other models are tested on a plurality of data sets, and the test result shows that the model provided by the embodiment obtains better recommendation effect than other models. We evaluated the effect of our model on 5 widely used text classification datasets, including: movie Review (MR), R52, Trec6, Trec50, AG. These data sets can be divided into two categories of text: sentiment classification and topic classification. MR is a data set about movie reviews, wherein each review only contains one text, and all data contain two emotion labels, and a divided training set and a divided testing set are adopted. Trec6 is a data set for questions containing data for the different question types in 6, 5452 samples in the training set, and 500 samples in the test set. Trec50 and Trec6 contain the same text, but possess more fine-grained labels. R52 is a subset of the Reuters-21578 dataset, containing different labels in 52, 6532 samples in the training set, and 2568 samples in the test set. The AG is a subset of the AG-News dataset and contains a total of 4 different News tags. The training set contained 7200 samples, each of 1800 tags, and the test set contained 4180 samples, each of 1045 tags. For the training set of these data sets, we divided them into a training set and a validation set, where the training set accounts for 80% and the validation set accounts for 20%, and the experiment used ACC as the criterion. The results of the experiment are shown in table 3:
table 3 effect of model on test set (%)
Figure BDA0003044741960000122
Figure BDA0003044741960000131
The model of the embodiment is called as GENET, and the experimental result shows that the effect of the model on a plurality of data sets is optimal.
In the embodiment, the multi-layer semantic and self-attention text matching model comprises a coding layer and a sentence-level semantic interaction layer, wherein the coding layer is composed of a convolutional neural network CNN and a self-attention mechanism network structure.
Most text classification models based on the neural network are mainly realized by a semantic interaction method of word granularity, and a semantic interaction method of sentence level is omitted. The multilayer semantic and self-attention text matching algorithm provided by the embodiment adopts a multilevel mode to perform semantic interaction, and provides a sentence-level semantic interaction method by combining an attention mechanism on the basis of semantic interaction with word granularity, so that a model can fully perform interaction tasks between texts. In addition, the algorithm adopts a network structure fusing the convolutional neural network CNN and the attention mechanism as an encoder, so that the local information of the text can be learned and the context information can be learned at the same time, and the defect that the CNN can only learn the local information and cannot learn the context information is effectively overcome. The following mainly introduces the coding layer and sentence-level semantic interaction layer of the model.
The coding layer mainly extracts semantic information of the text sentences through a coder. In the field of deep learning, a convolutional neural network is a common encoder, and can better learn local information of a text, but cannot learn context information of the text. The recurrent neural network is also commonly used in a text matching task and can better learn context information, but is limited by the fact that the training speed of a network structure is slow, the problem of long-distance dependence loss is easy to occur, and meanwhile, the extraction of local information of a text is lacked. On the basis of a convolutional neural network, a network structure which combines an attention-free mechanism through an enhanced residual error connection method is provided as an encoder, and the network structure mainly comprises three parts: the convolutional neural network layer, the self-attention layer and the enhanced residual are connected, and the specific structure is shown in fig. 6.
The convolutional neural network layer in the encoder performs convolution by adopting a padding method, and the convolution method can ensure that the input dimension of data does not change. When the input data is expressed as X ∈ Rl×kWhere 1 is expressed as the data length, k represents the dimension of the word vector, and the output data can be calculated by the following formula:
H＝CNNpadding(X)
wherein H ∈ Rl×mRepresents the convolved output, and m represents the number of convolution kernels.
Different from other attentions, attention of the user needs to introduce task vectors additionally to allocate attention weights, and the user himself is taken as data input and the sequence is inputEach element inside acts as a task vector itself to enable attention weight assignment to other vectors in the sequence. The self-attention mechanism can realize effective acquisition of context information, and an input text sequence is converted into H ═ H through coding of a CNN layer1，h2，h3，…，hn-1，hnIn which h ist∈RmThe output from the attention layer may be calculated by the following formula:
sij＝score(hi，hj)
aij＝softmax(sij)
Figure BDA0003044741960000141
wherein v isi∈RkRepresenting a context vector, s, derived based on the current element vectorijAnd aijRespectively representing the matching degree score and the attention weight of the current element vector and other element vectors.
The enhanced version residual connection is used for connecting the output of the CNN layer and the output of the attention layer, so that the context information of the text can be better learned while the local features of the text are acquired, and a specific calculation formula is as follows:
Figure BDA0003044741960000142
wherein u isiRepresenting the output sequence element of the encoder, hiAnd viThe output of the convolutional layer and the attention layer, respectively.
In the specific building process of the text matching model, the encoder adopts a multilayer structure to realize information extraction of input data, and N in the structure diagram represents an N-layer encoder.
Sentence-level semantic interaction layer:
firstly, before semantic interaction, the text proposes a concept of a basic semantic vector, which represents basic information of a current text, and uses the vector to perform subsequent semantic interaction operation. The basic semantic vector can be extracted from the text sequence through a pooling operation, and the specific formula is as follows:
Figure BDA0003044741960000151
Figure BDA0003044741960000152
wherein maxpool represents the maximum pooling function,
Figure BDA0003044741960000153
and
Figure BDA0003044741960000154
is a text sequence of two texts, abaseAnd bbaseA base semantic vector representing two texts.
Then, a multi-semantic extraction is carried out on the text sequence by using an attention mechanism and context vectors, each context vector represents a different semantic extraction mode, and the used context vector can be represented as [ c ]1，c2，...，cm]And m represents the number of context vectors, which can be obtained through random initialization. The weights of the context vector and the text sequence can be obtained through an attention mechanism, and the specific calculation formula is as follows:
Figure BDA0003044741960000155
Figure BDA0003044741960000156
wherein
Figure BDA0003044741960000157
Represents a pass context vector ciThe resulting attention weight of the user is obtained,softmax denotes the Softmax calculation function. Semantic vectors of two texts can be calculated by using the obtained attention weight
Figure BDA0003044741960000158
And
Figure BDA0003044741960000159
the specific calculation formula is as follows:
Figure BDA00030447419600001510
Figure BDA00030447419600001511
where j represents the number of vectors in the text sequence.
Then, semantic interaction between the two texts is performed using the aforementioned basic semantic vector. Calculating the weights of the current basic semantic vector and a plurality of semantic vectors of another text, and performing weighted summation operation by using the weights and the text sequence to obtain sentence-level interactive vectors, wherein the specific calculation formula is as follows:
Figure BDA0003044741960000161
Figure BDA0003044741960000162
Figure BDA0003044741960000163
Figure BDA0003044741960000164
wherein y isaAnd ybRepresenting through base semantic vectorsabaseAnd bbaseAnd obtaining the interaction vector.
And finally, fusing the basic semantic information and the sentence-level semantic interaction information to obtain the semantic information of the current text. The information fusion mode adopts a splicing method, hopefully, the existence of two kinds of information can be better ensured, the loss of the information is reduced, and the specific formula is as follows:
Figure BDA0003044741960000165
Figure BDA0003044741960000166
wherein
Figure BDA0003044741960000167
It is shown that the splicing operation is performed,
Figure BDA0003044741960000168
and
Figure BDA0003044741960000169
a fused semantic vector representing two texts.
The multilayer semantic and self-attention text matching model and other models of the embodiment are tested on a plurality of data sets, and the test result shows that the model provided by the invention obtains better effect than other models. We evaluated the effect of our model on the commonly used text matching data sets SNLI, Scitail and quadra. 400000 question pairs are included in the quadra dataset, and the task target needs to judge whether the two questions have the same meaning. SNLI is a reference data set for natural language reasoning, which contains 57 thousands of pairs of annotation statements, including inclusion, neutrality, contradiction 3 labels, Scitail is also a data set for natural language reasoning. For the training set of these data sets, we divided them into a training set and a validation set, where the training set accounts for 80% and the validation set accounts for 20%, and the experiment used ACC as the criterion. The results of the experiment are shown in Table 4.
Experimental results of model in Table 4 (%)
Model (model) SNLI Quora Scitail
BiMPM 86.9 88.2 70.6
SAN 88.6 89.4 77.5
CSRAN 88.7 89.2 86.7
MwAN 88.3 89.1 83.3
DIIN 88.0 89.1 84.4
MSNET(ours) 88.8 90.15 87.1
This example also compares the model effect on the data set actually applied in the project, and the specific results are shown in table 5. It can be seen that the effect of the algorithm presented herein is much higher than that of the other models.
Table 5 experimental results of model in application data set (%)
Model (model) Results of experiment (Acc)
BiMPM 87.2
SAN 88.5
CSRAN 88.9
MwAN 87.8
DIIN 88.1
MSNET(ours) 90.2
Step S102: and calculating the keyword matching degree of the screened fault cases and the power grid standard library by using a keyword matching algorithm of self-adaptive weight, and screening the power grid standard exceeding a set matching degree threshold according to the matching degree.
In specific implementation, the fault case library includes the keyword attributes of the cases, and similarly, each standard in the power grid standard library has its own keyword. And matching the fault case obtained in the fault analysis with the keywords in the standard by using a keyword matching technology of the self-adaptive weight, and taking the standard of the degree of correlation topk as a returned result.
As shown in fig. 4, the process of calculating the keyword matching degree between the screened fault cases and the grid standard library includes:
and counting based on the fault cases with similar set number, recording the weights of different keywords, and obtaining a keyword list with the weights.
The traditional keyword matching technology only considers whether the keywords occur or not, and ignores that the keywords with higher occurrence frequency have higher weights. The keyword matching algorithm provided by the text can better solve the problem. Assuming the k fault cases obtained in the foregoing, firstly, the k fault cases are counted, and the weights of different keywords are recorded, and the weights are calculated as follows:
wi＝ni/nall
wherein n isiIndicates the number of times of occurrence of the ith keyword, nallIndicating the number of times the keyword appears in its entirety. After the weight is calculated, a keyword list with the weight can be obtained
Calculating the matching degree of the current keyword list and the keywords of the power grid standard, wherein the calculation formula is as follows:
Figure BDA0003044741960000181
where m denotes the degree of matching, wiRepresenting the weight of the common key.
For the specific setting of k, we perform experiments on the test set, respectively consider the accuracy and coverage of the results, and finally select k to be 4, which is shown in fig. 5.
Step S103: and inputting the fault phenomenon text and the screened power grid standard into a multi-layer semantic and self-attention text matching model, obtaining equipment fault information and a fault maintenance scheme based on a power grid domain database, and pushing the equipment fault information and the fault maintenance scheme to a client.
The power grid field database is constructed by crawling relevant question-answer pair information related to faults on the internet by using a crawler technology, the sources of the question-answer pair data comprise community question-answer websites, encyclopedia websites and the like, and the crawled data are subjected to noise filtering and relevant screening, wherein the question-answer pair data comprise tens of thousands of pieces of question-answer pair data, the problem parts in the data are all relevant problems related to faults of power grid equipment, and the data are from the community question-answer websites and the like and are mainly responsible for puzzled answering, so that the question-answer pair data have higher comprehensibility than information in fault texts, and the fault information is analyzed from different angles.
Example two
The embodiment provides a power transmission and transformation equipment fault processing system fusing a power grid fault case library, which comprises:
the fault case screening module is used for inputting fault phenomenon texts into a multi-layer semantic and self-attention text matching model and obtaining fault cases with similar set quantity based on a structured fault case library;
the power grid standard screening module is used for calculating the matching degree of the screened fault cases and the keywords of the power grid standard library by using a keyword matching algorithm of self-adaptive weight, and screening the power grid standard exceeding a set matching degree threshold value according to the matching degree;
and the fault information and maintenance scheme acquisition module is used for inputting the fault phenomenon text and the screened power grid standard into a multi-layer semantic and self-attention text matching model, acquiring equipment fault information and a fault maintenance scheme based on the power grid domain database, and pushing the equipment fault information and the fault maintenance scheme to the client.
It should be noted that, each module in the power transmission and transformation equipment fault processing system integrated with the power grid fault case library in this embodiment is the same as the specific implementation process of each step in the power transmission and transformation equipment fault processing method integrated with the power grid fault case library in the first embodiment, and will not be described here again.
EXAMPLE III
The present embodiment provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor implements the steps in the power transmission and transformation equipment fault handling method fusing the grid fault case base as described above.
Example four
The embodiment provides a computer device, which includes a memory, a processor and a computer program stored in the memory and capable of running on the processor, and when the processor executes the program, the processor implements the steps in the power transmission and transformation device fault handling method fusing the power grid fault case base.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. A power transmission and transformation equipment fault processing method fused with a power grid fault case base is characterized by comprising the following steps:
inputting fault phenomenon texts into a multi-layer semantic and self-attention text matching model, and obtaining fault cases with similar set quantity based on a structured fault case library;
calculating the keyword matching degree of the screened fault cases and a power grid standard library by using a keyword matching algorithm of self-adaptive weight, and screening power grid standards exceeding a set matching degree threshold value according to the matching degree;
and inputting the fault phenomenon text and the screened power grid standard into a multi-layer semantic and self-attention text matching model, obtaining equipment fault information and a fault maintenance scheme based on a power grid domain database, and pushing the equipment fault information and the fault maintenance scheme to a client.
2. The power transmission and transformation equipment fault handling method fusing the power grid fault case library as claimed in claim 1, wherein the structured fault case library is constructed according to equipment fault information extracted from a power grid fault text.
3. The power transmission and transformation equipment fault processing method fusing the power grid fault case library as claimed in claim 2, wherein in the process of constructing the structured fault case library, the fault categories of the power grid fault texts are classified by using a text classification model embedded with graph convolution words, so that the case library attributes are enriched.
4. The power transmission and transformation equipment fault processing method fusing the power grid fault case library as claimed in claim 2, wherein in the process of constructing the structured fault case library, equipment fault information is extracted by adopting a uniform attribute template.
5. The power transmission and transformation equipment fault processing method fusing the power grid fault case base as claimed in claim 1, wherein the multilayer semantic and self-attention text matching model is innovated in a coding layer and a sentence-level semantic interaction layer, the coding layer is composed of a Convolutional Neural Network (CNN) and a network structure of a self-attention mechanism, and the sentence-level semantic interaction layer performs text semantic interaction from a sentence angle by using attention.
6. The method for processing the faults of the power transmission and transformation equipment fusing with the power grid fault case library as claimed in claim 1, wherein the process of calculating the keyword matching degree of the screened fault cases and the power grid standard library is as follows:
counting based on the fault cases with similar set number, recording the weights of different keywords, and obtaining a keyword list with the weights;
and calculating the matching degree of the current keyword list and the keywords of the power grid standard.
7. The power transmission and transformation equipment fault handling method fusing the power grid fault case library as claimed in claim 1, wherein the power grid domain database is constructed by acquiring power grid fault question and answer pair data based on a crawler technology.
8. The utility model provides a power transmission and transformation equipment fault processing system who fuses grid fault case storehouse which characterized in that includes:
the fault case screening module is used for inputting fault phenomenon texts into a multi-layer semantic and self-attention text matching model and obtaining fault cases with similar set quantity based on a structured fault case library;
the power grid standard screening module is used for calculating the matching degree of the screened fault cases and the keywords of the power grid standard library by using a keyword matching algorithm of self-adaptive weight, and screening the power grid standard exceeding a set matching degree threshold value according to the matching degree;
and the fault information and maintenance scheme acquisition module is used for inputting the fault phenomenon text and the screened power grid standard into a multi-layer semantic and self-attention text matching model, acquiring equipment fault information and a fault maintenance scheme based on the power grid domain database, and pushing the equipment fault information and the fault maintenance scheme to the client.
9. A computer-readable storage medium, on which a computer program is stored, which program, when executed by a processor, performs the steps in the method for handling faults in a power transmission and transformation device that merges grid fault cases according to any one of claims 1 to 7.
10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps in the method for handling faults in a power transmission and transformation device that merges grid fault cases according to any one of claims 1 to 7.
CN202110467599.XA 2021-04-28 2021-04-28 Power transmission and transformation equipment fault processing method and system fusing power grid fault case base Active CN113239143B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110467599.XA CN113239143B (en) 2021-04-28 2021-04-28 Power transmission and transformation equipment fault processing method and system fusing power grid fault case base

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110467599.XA CN113239143B (en) 2021-04-28 2021-04-28 Power transmission and transformation equipment fault processing method and system fusing power grid fault case base

Publications (2)

Publication Number Publication Date
CN113239143A true CN113239143A (en) 2021-08-10
CN113239143B CN113239143B (en) 2022-09-30

Family

ID=77129770

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110467599.XA Active CN113239143B (en) 2021-04-28 2021-04-28 Power transmission and transformation equipment fault processing method and system fusing power grid fault case base

Country Status (1)

Country Link
CN (1) CN113239143B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113779996A (en) * 2021-08-31 2021-12-10 中国中医科学院中医药信息研究所 Standard entity text determination method and device based on BilSTM model and storage medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107578106A (en) * 2017-09-18 2018-01-12 中国科学技术大学 A kind of neutral net natural language inference method for merging semanteme of word knowledge
CN109902153A (en) * 2019-04-02 2019-06-18 杭州安脉盛智能技术有限公司 Equipment fault diagnosis method and system based on natural language processing and reasoning by cases
CN110765240A (en) * 2019-10-31 2020-02-07 中国科学技术大学 Semantic matching evaluation method for multiple related sentence pairs
CN112035661A (en) * 2020-08-24 2020-12-04 北京大学深圳研究生院 Text emotion analysis method and system based on graph convolution network and electronic device
US20210089594A1 (en) * 2019-09-25 2021-03-25 Microsoft Technology Licensing, Llc Inter-document attention mechanism

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107578106A (en) * 2017-09-18 2018-01-12 中国科学技术大学 A kind of neutral net natural language inference method for merging semanteme of word knowledge
CN109902153A (en) * 2019-04-02 2019-06-18 杭州安脉盛智能技术有限公司 Equipment fault diagnosis method and system based on natural language processing and reasoning by cases
US20210089594A1 (en) * 2019-09-25 2021-03-25 Microsoft Technology Licensing, Llc Inter-document attention mechanism
CN110765240A (en) * 2019-10-31 2020-02-07 中国科学技术大学 Semantic matching evaluation method for multiple related sentence pairs
CN112035661A (en) * 2020-08-24 2020-12-04 北京大学深圳研究生院 Text emotion analysis method and system based on graph convolution network and electronic device

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
杜修明等: "电力设备典型故障案例的文本挖掘", 《高电压技术》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113779996A (en) * 2021-08-31 2021-12-10 中国中医科学院中医药信息研究所 Standard entity text determination method and device based on BilSTM model and storage medium
CN113779996B (en) * 2021-08-31 2023-10-10 中国中医科学院中医药信息研究所 Standard entity text determining method and device based on BiLSTM model and storage medium

Also Published As

Publication number Publication date
CN113239143B (en) 2022-09-30

Similar Documents

Publication Publication Date Title
Snyder et al. Interactive learning for identifying relevant tweets to support real-time situational awareness
CN107066446B (en) Logic rule embedded cyclic neural network text emotion analysis method
US20190073420A1 (en) System for creating a reasoning graph and for ranking of its nodes
CN111444709A (en) Text classification method, device, storage medium and equipment
CN111143576A (en) Event-oriented dynamic knowledge graph construction method and device
CN114417004A (en) Method, device and system for fusing knowledge graph and case graph
Zhao et al. ZYJ123@ DravidianLangTech-EACL2021: Offensive language identification based on XLM-RoBERTa with DPCNN
CN113312480A (en) Scientific and technological thesis level multi-label classification method and device based on graph convolution network
CN114880307A (en) Structured modeling method for knowledge in open education field
CN114742071A (en) Chinese cross-language viewpoint object recognition and analysis method based on graph neural network
Hasanati et al. Implementation of support vector machine with lexicon based for sentimenT ANALYSIS ON TWITter
Sendhilkumar Developing a conceptual framework for short text categorization using hybrid CNN-LSTM based Caledonian crow optimization
Manik et al. Out-of-Scope Intent Detection on A Knowledge-Based Chatbot.
CN113239143B (en) Power transmission and transformation equipment fault processing method and system fusing power grid fault case base
Sheeba et al. A fuzzy logic based on sentiment classification
Clarizia et al. Sentiment analysis in social networks: A methodology based on the latent dirichlet allocation approach
CN112749556B (en) Multi-language model training method and device, storage medium and electronic equipment
CN115129807A (en) Fine-grained classification method and system for social media topic comments based on self-attention
Angdresey et al. Classification and Sentiment Analysis on Tweets of the Ministry of Health Republic of Indonesia
CN113516198A (en) Cultural resource text classification method based on memory network and graph neural network
Liu et al. Aspect sentiment mining of short bullet screen comments from online TV series
Maladkar Content based hierarchical URL classification with Convolutional Neural Networks
CN111538898A (en) Web service package recommendation method and system based on combined feature extraction
Fu et al. A study on recursive neural network based sentiment classification of Sina Weibo
Javeed An lstm model for extracting hierarchical relations between words for better topic modeling

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant