Power transformer equipment defect intelligent diagnosis method and system based on knowledge graph

Abstract

The invention discloses an intelligent diagnosis method and system for defects of power transformer equipment based on a knowledge graph, and the method comprises the steps: carrying out the preprocessing of a plurality of obtained on-site defect description texts, and obtaining an initial word vector and an initial character vector; selecting a current field defect description text; adopting a defect key entity recognition model based on a CRF layer to recognize a key defect entity in the current field defect description text based on the initial character vector so as to perform entity alignment with the obtained power transformer defect knowledge graph to obtain a graph defect description text; based on the initial word vector, a twin neural network is adopted to match the graph defect description text with the current field defect description text, and standard defect description of the defective equipment or the defective part is obtained; and positioning a standard defect reason and a defect elimination method corresponding to the standard defect description in the power transformer defect knowledge graph. According to the invention, operation and inspection personnel can be more effectively helped to carry out rapid positioning and troubleshooting and defect elimination on equipment defects of the power transformer.

Description

Technical Field

[0001] The present invention belongs to the field of electric power operation and inspection, and in particular relates to a knowledge graph-based intelligent diagnosis method, system, electronic device and storage medium for power transformer equipment defects. Background Art

[0002] When defects occur in power transformer equipment, the operation and maintenance personnel are required to promptly analyze the weak links in the operation of the equipment, accurately and quickly determine the causes of the defects, and take corresponding measures to eliminate the defects. As the defect forms of power transformer equipment are becoming increasingly complex, and the same defect type may correspond to different defective components, this increases the difficulty of transformer defect diagnosis. The rapid diagnosis and effective resolution of power transformer equipment defects are crucial to the normal operation of the power system. Traditional transformer equipment defect diagnosis methods often rely on manual experience and cannot efficiently process a large amount of defect information.

[0003] In recent years, more and more researchers have begun to introduce artificial intelligence technology into the field of equipment defect diagnosis. For example, prior art 1 (CN116993043A) discloses a method and device for tracing the source of power equipment faults. The method includes the steps of extracting information from fault data to obtain triple data; importing triple data into a graph database to construct a knowledge graph of power equipment faults; designing and constructing a power equipment fault tracing model based on a fast graph convolutional network; inputting new power equipment fault data into the constructed power equipment fault tracing model to obtain historical fault cases and disposal measures that are the same or similar to the new power equipment fault data. However, this technology uses the GraphSAGE model to trace the source of power faults, and only samples some nodes in the graph. If the sampling is inaccurate, some key information will be lost, resulting in the failure of tracing. At the same time, this technology cannot judge the defect level, and thus cannot intelligently assist the operation and maintenance personnel to prioritize the defects of which equipment. Summary of the invention

[0004] In order to address the deficiencies in the prior art, the present invention provides a method, system, electronic device and storage medium for intelligent diagnosis of defects in power transformer equipment based on a knowledge graph, which can avoid the difficulty in diagnosing equipment defects due to inaccurate sampling, and can effectively judge the severity level of equipment defects, thereby more effectively helping operation and maintenance personnel to quickly locate and troubleshoot defects in power transformer equipment.

[0005] According to a first aspect of the present invention, a method for intelligent diagnosis of power transformer equipment defects based on a knowledge graph is provided. The method comprises:

[0006] Preprocessing the obtained multiple on-site defect description texts of the transformer to obtain the initial word vector and initial character vector of each on-site defect description text;

[0007] Selecting a current on-site defect description text from a plurality of on-site defect description texts of the transformer obtained;

[0008] The defect key entity recognition model based on the CRF layer is used to identify the key defect entities in the current on-site defect description text based on the initial character vector, and then align the entities with the acquired power transformer defect knowledge graph to obtain the graph defect description text;

[0009] Based on the initial word vector, the twin neural network is used to match the atlas defect description text with the current on-site defect description text to locate the standard defect description of the defective equipment or defective parts;

[0010] The standard defect causes and defect elimination methods corresponding to the standard defect descriptions are located in the power transformer defect knowledge graph.

[0011] Preferably, the multiple on-site defect description texts of the obtained transformer are preprocessed to obtain the initial word vector and initial character vector of each on-site defect description text, including:

[0012] According to the electric power field dictionary, the plurality of on-site defect description texts are respectively segmented;

[0013] The word vector model based on Word2vec is used to process the word segmentation results to obtain the initial word vector of each word in the word segmentation results.

[0014] Preferably, the multiple on-site defect description texts of the obtained transformer are preprocessed to obtain the initial word vector and initial character vector of each on-site defect description text, including:

[0015] Perform fine-grained character segmentation on each on-site defect description text to obtain the corresponding character set e = (e1, e2, ..., e m ), where e1, e2, ..., e m They are the 1st to mth characters after cutting respectively;

[0016] The BERT-based character vector model is used to obtain the dynamic vector of each character as the initial character vector.

[0017] Preferably, the current on-site defect description text is selected from a plurality of on-site defect description texts of the transformer obtained, including:

[0018] A BiGRU-based defect severity level recognition model is used to determine the severity level of each defect description text based on the initial word vector;

[0019] The current on-site defect description text is selected from the plurality of on-site defect description texts according to the severity level sorting.

[0020] Preferably, a BiGRU-based defect severity level model is used to determine the severity level of the defect corresponding to each on-site defect description text based on the initial word vector, including:

[0021] A transformer defect severity level recognition model based on BiGRU is adopted, and the hidden state of the corresponding word is obtained based on the initial word vector of each word; the transformer defect severity level recognition model based on BiGRU is expressed as:

[0022]

[0023] Among them, h t represents the hidden state of word t, x t Represents the initial word vector representation of word t, w t and v t Represents the forward hidden state of the BiGRU corresponding to word t and the reverse hidden state The corresponding weight, b t Represents the bias corresponding to the hidden state of word t;

[0024] According to the hidden states of all words in each on-site defect description text, the severity level of the defect corresponding to each on-site defect description text is obtained, and the formula is as follows:

[0025]

[0026] in, For the splicing operation, W o is the linear transformation matrix, softMax(·) is the normalized exponential function, and d is the final defect level probability corresponding to the on-site defect description text.

[0027] Preferably, a defect key entity recognition model based on a CRF layer is used to identify key defect entities in the current on-site defect description text based on the initial character vector, and to perform entity alignment with the acquired power transformer defect knowledge graph to obtain a graph defect description text, including:

[0028] The character sequence of the defect text is obtained by the linear layer dimensionality reduction processing of the initial character vector set of the current on-site defect description text X = (x1,...,x m );

[0029] According to X=(x1,...,x m ), using the defect key entity recognition model based on the CRF layer, determine the predicted label sequence corresponding to X as Y = (y1,....,y m ), and then find the predicted label sequence with the highest probability as the global optimal label sequence; where the predicted label sequence corresponding to X is Y = (y1,....,y m ) is expressed as:

[0030]

[0031] Where X=(x1,...,x m ) is the character sequence of the defective text, Y=(y1,....,y m ) is X=(x1,...,x m ) corresponding to the predicted label, W yi,yi+1 represents the label transfer score, Indicates that the character is Y i+1 The score of each label, s(X,Y) represents the emission score of the defect text, and P(Y|X) represents the probability that the predicted label sequence corresponding to X is Y after global normalization.

[0032] The identified key defect entities and the acquired power transformer defect knowledge graph are entity aligned to obtain the graph defect description text.

[0033] Preferably, based on the initial word vector, a twin neural network is used to match the atlas defect description text with the current on-site defect description text to locate the standard defect description of the defective equipment or defective parts, including:

[0034] Preprocess the graph defect description text to obtain the corresponding initial word vector;

[0035] The initial word vectors of the atlas defect description text and the initial word vectors of the current on-site defect description text are encoded by the twin neural network constructed by TextCNN to obtain the corresponding encoding representation:

[0036] c i =Max(ReLu(W s K i +bk i ))

[0037]

[0038] Among them, W s K represents the set matrix of the initial word vectors of the atlas defect description text or the current on-site defect description text, i represents the i-th convolution kernel, bki represents the bias corresponding to the convolution kernel, Max represents the maximum pooling layer, ReLu(·) is the activation function, c i represents the encoding representation corresponding to the i-th convolution kernel, and c represents the encoding representation corresponding to the atlas defect description text or the current on-site defect description text;

[0039] The coding representations corresponding to the atlas defect description text and the current on-site defect description text are aggregated using the Manhattan distance to obtain the standard defect description of the defective equipment or defective parts.

[0040] According to a second aspect of the present invention, there is provided a knowledge graph-based intelligent diagnosis system for power transformer equipment defects using the method described in the first aspect of the present invention. The system comprises:

[0041] A preprocessing unit, used to preprocess the obtained multiple on-site defect description texts of the transformer to obtain an initial word vector and an initial character vector of each on-site defect description text;

[0042] A sorting unit selects a current on-site defect description text from a plurality of on-site defect description texts of the transformer obtained;

[0043] An entity recognition and alignment unit is used to adopt a defect key entity recognition model based on a CRF layer to identify key defect entities in the current on-site defect description text based on the initial character vector, and to perform entity alignment with the acquired power transformer defect knowledge graph to obtain a graph defect description text;

[0044] The standard defect location unit is used to match the atlas defect description text with the current on-site defect description text based on the initial word vector using a twin neural network to locate the standard defect description of the defective equipment or defective parts;

[0045] A diagnostic unit is used to locate the standard defect cause and defect elimination method corresponding to the standard defect description in the power transformer defect knowledge graph.

[0046] According to a third aspect of the present invention, an electronic device is provided, which includes a processor and a storage medium, wherein the storage medium is used to store instructions, and the processor is used to operate according to the instructions to execute the steps of the method according to the first aspect of the present invention.

[0047] According to a fourth aspect of the present invention, a storage medium is provided, wherein the storage medium stores a computer program that can be loaded by a processor and execute the method according to the first aspect of the present invention.

[0048] The beneficial effects of the present invention are as follows:

[0049] 1. First, through the defect key entity recognition model based on the CRF layer, help the operation and maintenance personnel locate the key entities (parts of the transformer) in the defect text. The key entities are the entities in the corresponding graph; then, based on the defect description text matching algorithm of the twin network, match the defect text and the defect description associated with the key entity to locate the best defect description; then, locate the cause of the defect and the method of eliminating the defect through the defect description to assist the operation and maintenance personnel in on-site work. In this way, the identification of key entities is equivalent to a global positioning search in the knowledge graph, and the matching operation based on the twin neural network is equivalent to further positioning after global positioning, which can assist the operation and maintenance personnel to perform fast and efficient operations, and can effectively avoid the problem of difficulty in diagnosing equipment defects due to inaccurate sampling.

[0050] 2. By adopting the BiGRU-based defect severity level model, the defect severity level (e.g., critical, serious, general) of the on-site defect text can be identified, thereby assisting on-site operation and maintenance personnel to prioritize the defects occurring in the transformer and improve the efficiency of defect diagnosis and treatment.

[0051] In summary, the present invention combines the association and reasoning functions of the constructed power transformer equipment defect knowledge graph to automatically analyze and diagnose various transformer field defect description information. At the same time, compared with the traditional rule-based or expert system-based diagnosis method, the knowledge graph-based intelligent diagnosis method has stronger intelligence and adaptability, and can automatically infer the optimal diagnosis result according to different defect conditions, thus becoming an important means to improve the management and maintenance level of transformer equipment, and effectively improve the operation quality and safety of transformer equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is the overall framework of the intelligent diagnosis method of power transformer equipment defects based on knowledge graph of the present invention;

[0053] Figure 2 It is a flow chart of the intelligent diagnosis method of power transformer equipment defects based on knowledge graph of the present invention;

[0054] Figure 3 is a schematic diagram of a BiGRU-based defect severity level recognition model of the present invention;

[0055] Figure 4 It is a schematic diagram of a BERT-based character vector model and a CRF layer-based defect key entity recognition model of the present invention;

[0056] Figure 5 is a schematic diagram of a defect description text matching model based on a twin neural network of the present invention;

[0057] Figure 6 It is a schematic diagram of the interaction of the power transformer defect knowledge graph of the present invention;

[0058] Figure 7 It is a schematic diagram of the intelligent diagnosis system for power transformer equipment defects based on knowledge graph of the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0060] The present invention aims to provide a method, system, electronic device and storage medium for intelligent diagnosis of power transformer equipment defects based on knowledge graph, which can avoid the difficulty in diagnosing equipment defects due to inaccurate sampling, and can effectively judge the severity level of equipment defects, thereby more effectively helping operation and maintenance personnel to quickly locate and troubleshoot defects in power transformer equipment.

[0061] Embodiment 1

[0062] Embodiment 1 of the present application relates to an intelligent diagnosis method for power transformer equipment defects based on a knowledge graph. Figure 1 and Figure 2 , the method specifically comprises the following steps:

[0063] S1. Preprocess multiple on-site defect description texts of the obtained transformer to obtain the initial word vector and initial character vector of each on-site defect description text.

[0064] Among them, the on-site defect description text here includes the defect description text related to the transformer involved in the power grid resource business middle platform and the power grid production control system, as well as the defect description text generated during the operation of the transformer equipment.

[0065] In step S1, the process of obtaining the initial word vector through preprocessing specifically includes:

[0066] S101. According to the electric power field dictionary, the plurality of on-site defect description texts are respectively segmented.

[0067] For example, a word segmentation tool such as the Jieba word segmentation module can be used, and the electric power field dictionary can be used to segment the on-site defect description text. The specific word segmentation is shown in Table I:

[0068] Table I

[0069]

[0070] From the above word segmentation results, we can see that when the power field dictionary is not introduced, the ABC phases of the Furong main transformer are divided into four words. After the power field dictionary is introduced, the word is correctly divided.

[0071] S102, using a word vector model based on Word2vec to process the word segmentation result, and obtain an initial word vector for each word in the word segmentation result.

[0072] More specifically, based on the principle of Word2vec, a large amount of pre-processed transformer equipment defect record table data from the power grid resource business center and data collected by the power grid production management system can be used as a corpus to train the initial word vector for each word.

[0073] In step S1, the process of obtaining the initial character vector through preprocessing involves training using a BERT-based pre-trained language model, specifically including:

[0074] S111, perform fine-grained character segmentation on each on-site defect description text to obtain a corresponding character set e = (e1, e2, ..., e m ), where e1, e2, ..., e m They are the 1st to mth characters after cutting respectively.

[0075] S112. Use a BERT-based character vector model to obtain a dynamic vector for each character as an initial character vector.

[0076] More specifically, the BERT-based character vector model can be used to obtain each character e1, e2, ..., e m The corresponding dynamic vectors T1, T2, ..., T m , forming a dynamic vector set T = (T1, T2, ..., T m ), as shown in the following formula:

[0077] T i =T i_Token +T i_Segment +T i_Position

[0078] Among them, T i represents the dynamic vector corresponding to the i-th character, E i_Token Represents the static vector corresponding to the i-th character, E i_Position represents the position vector corresponding to the i-th character, E i_Segment Represents the sentence vector corresponding to the i-th character.

[0079] S2. Select a current field defect description text from a plurality of obtained field defect description texts of the transformer.

[0080] When there are multiple on-site defect description texts, it is usually necessary to prioritize all of the on-site defect description texts, so that the on-site defect description texts can be processed in sequence according to the sorting results. For example, each on-site defect description text can be analyzed to determine the severity level of the defect corresponding to each on-site defect description text, so that defects with high severity can be processed first according to the severity level. This step S2 is equivalent to the first positioning of intelligent diagnosis of transformer equipment defects.

[0081] Among them, the severity level of the defects corresponding to the multiple on-site defect description texts can be determined by experts or by operators searching and determining according to a pre-stored expert database. However, this manual search method is inefficient.

[0082] More preferably, a BiGRU-based defect severity level model may be used for determination. The corresponding step S2 specifically includes:

[0083] S201, using a BiGRU-based defect severity level model, and determining the severity level of each defect corresponding to each on-site defect description text based on the initial word vector. This step includes:

[0084] Firstly, a transformer defect severity level recognition model based on BiGRU is adopted to obtain the hidden state of the corresponding word based on the initial word vector of each word. The defect severity level model based on BiGRU is expressed as:

[0085]

[0086] Among them, x t Represents the initial word vector of word t, w t and v t Represents the forward hidden state of the BiGRU corresponding to word t and the reverse hidden state The corresponding weight, b t represents the bias corresponding to the hidden state of word t, h t represents the hidden state of word t.

[0087] Then, according to the hidden states of all words in each on-site defect description text, the severity level of the defect corresponding to each on-site defect description text is obtained, and the formula is as follows:

[0088]

[0089] in, For the splicing operation, W ois a linear transformation matrix, and the final defect level probability d corresponding to the on-site defect description text is obtained through softMax, where n is the total number of words.

[0090] S202: Select a current on-site defect description text from the plurality of on-site defect description texts according to severity level sorting.

[0091] As an example, Figure 3 As shown in the figure, a series of on-site defect description texts that have been preprocessed are used to identify the severity of on-site defects through the BiGRU defect level recognition module, and priority processing is performed based on the severity, for example:

[0092] Manual inspection found that the oil leakage rate of the ABC phase oil storage tank of Furong transformer # main transformer was faster than 5 seconds per drop and the oil level was below the lower limit (critical); manual inspection found that the expander indication of the ABC phase oil storage tank of Furong transformer # main transformer did not change when the ambient temperature changed significantly (serious); online monitoring found that the ABC phase oil storage tank of Furong transformer # main transformer did not form oil drops (normal). Therefore, the situation was handled in the order of critical - serious - normal.

[0093] S3. Adopt the defect key entity recognition model based on the CRF layer, identify the key defect entities in the current on-site defect description text based on the initial character vector, and align the entities with the acquired power transformer defect knowledge graph to obtain the graph defect description text.

[0094] Step S3 is equivalent to the second positioning of the intelligent diagnosis of transformer equipment defects. This step specifically includes:

[0095] Among them, the key defect entities in the defect description text are identified based on the CRF layer model, including:

[0096] S301, the character sequence X of the defective text is obtained by performing dimensionality reduction processing on the initial character vector set (i.e., the dynamic vector set T) through a linear layer. m ).

[0097] S302, according to X=(x1,...,x m ), using the defect key entity recognition model based on the CRF layer, determine the predicted label sequence corresponding to X as Y = (y1,....,y m ), and then find the predicted label sequence with the highest probability as the global optimal label sequence.

[0098] Specifically, suppose the predicted label sequence corresponding to X is Y = (y1, ...., y m ), then the probability that X is labeled as Y can be obtained by the following formula:

[0099]

[0100] Among them, s(X,Y) represents the emission score of defective text, represents the label transfer score, Indicates that the character is Y i+1 P(Y|X) represents the probability that the predicted label sequence corresponding to X is Y after global normalization.

[0101] The role of the CRF layer in the defect key entity recognition model based on the CRF layer is to consider the dependencies between labels (such as the constraints of adjacent labels) based on these probabilities, so as to obtain a global optimal label sequence. More specifically, the CRF decoding algorithm (such as the Viterbi algorithm) can be used to find the label sequence with the highest probability as the global optimal label sequence. This sequence is the label sequence of the defect key entity predicted by the model.

[0102] Next, the global optimal label sequence is decoded, and the decoded label sequence is mapped back to the current on-site defect description text, so that the key defect entities can be identified. This process also involves post-decoding processing, that is, post-processing the decoded label sequence to extract the specific information of the key entities, which may include merging adjacent identical labels (if they represent the same entity), removing unnecessary labels (such as padding labels), etc. Based on the post-processed label sequence, the key entities can be identified and extracted from the original text. The key entity extraction process usually also involves aligning each label in the global optimal label sequence with the corresponding character or character sequence of the current on-site defect description text, and extracting these characters or character sequences as key entities. In addition, the mapping process usually involves creating a mapping table from labels to actual key entities (such as defect type, location, etc.).

[0103] As an example, Figure 4 As shown, assuming that the current on-site defect description text "Online monitoring shows that no oil droplets are formed in the ABC phase oil storage cabinet of the main transformer of Furong transformer" is processed by fine-grained word segmentation to form "Online monitoring shows that no oil droplets are formed in the A, B, C phase oil storage cabinet of the main transformer of Furong transformer";

[0104] The final BIO label sequence “O,O,O,O,O,O,B-Trans,I-Trans,I-Trans,I-Trans,I-Trans,I-Trans,I-Trans,I-Trans,I-Trans,I-Trans,B-OilConser,I-OilConser,I-OilConser,O,O,O,O,O,O” can be obtained through the defect key entity recognition model based on the CRF layer, that is, the key defect entities “Furong main transformer ABC phases (transformer)” and “oil conservator (oil cabinet)” are identified.

[0105] S303, entity alignment is performed on the identified key defect entities and the acquired power transformer defect knowledge graph to obtain a graph defect description text.

[0106] The power transformer defect knowledge graph is a systematic knowledge representation that aims to organize and display various types of information about power transformer defects. Power transformer equipment mainly includes windings, leads, tap changers, outlet devices, cores, oil tanks, oil storage cabinets, desiccant, oil level gauges and other components. The power transformer defect knowledge graph can be constructed based on the above components. The power transformer defect knowledge graph can usually include multiple (main entity, relationship, guest entity) triples to describe the ontology relationship. Table 1 shows a specific example of the triples contained in the power transformer defect knowledge graph:

[0107] Table 1 Ontology relationship types

[0108]

[0109] As an example, according to the identified key defect entities "Furong transformer main transformer ABC phase" and "oil storage cabinet", the triple in the power defect knowledge graph is matched (oil storage cabinet, belongs to, Furong transformer main transformer ABC phase), and the graph defect description text corresponding to the oil storage cabinet is output (oil storage cabinet, defect occurs,?), where "?" can correspond to specific defect descriptions such as "slight leakage with signs of oil seepage", "oil leakage forms oil flow", "oil level gauge appearance is damaged", "oil level is higher than the upper limit of normal oil level" and "metal expander appearance is damaged".

[0110] S4. Based on the initial word vector, the twin neural network is used to match the atlas defect description text with the current on-site defect description text to locate the standard defect description of the defective equipment or defective parts.

[0111] Step S4 is equivalent to the second positioning of the intelligent diagnosis of transformer equipment defects. This step specifically includes:

[0112] S401, preprocessing the graph defect description text to obtain the corresponding initial word vector.

[0113] S402, encode the initial word vector of the atlas defect description text and the initial word vector of the current on-site defect description text respectively by the twin neural network constructed by TextCNN to obtain corresponding encoding representations; this process is expressed by the formula:

[0114] c i =Max(ReLu(W s K i +bk i ))

[0115]

[0116] Among them, W s K represents the set matrix of the initial word vectors of the atlas defect description text or the current on-site defect description text, i represents the i-th convolution kernel, bk i represents the bias corresponding to the convolution kernel, Max represents the maximum pooling layer, ReLu(·) is the activation function, c i represents the encoding representation corresponding to the i-th convolution kernel, and c represents c1~c k The concatenated encoded representation.

[0117] S403, using the Manhattan distance to aggregate information on the coding representations corresponding to the atlas defect description text and the current on-site defect description text, to obtain a standard defect description of the defective equipment or defective component.

[0118] As an example, Figure 5 As shown, the atlas defect description text of the oil storage tank mentioned as an example in step S3 and the initial on-site defect description text are matched through the twin neural network to obtain the standard defect description of the defective equipment or defective component corresponding to the key defect entity; for example, the on-site defect description text "online monitoring shows that no oil droplets are formed in the ABC phase oil storage tank of the Furong main transformer" is obtained through the twin neural network to obtain the standard defect description as "slight oil leakage with signs of oil leakage".

[0119] S5. Locate the standard defect causes and elimination methods corresponding to the standard defect descriptions in the power transformer defect knowledge graph.

[0120] Specifically, in the graph, there are triples such as (defect description, cause, defect cause), (defect elimination method, treatment, defect description), and the "defect description" in these triples is the standard defect description in the power transformer defect knowledge graph. Therefore, the defect cause and defect elimination description can be found through the standard defect description.

[0121] As an example, based on the standard defect description "minor oil leakage with signs of oil leakage", the triplet "minor oil leakage with signs of oil leakage, cause, oil pillow connection hose loose" and the triplet "standard installation of equipment, treatment, minor oil leakage with signs of oil leakage" can be located, thereby helping operation and maintenance personnel to perform intelligent diagnosis.

[0122] To further help users understand how the on-site defect description text interacts with the power transformer defect knowledge graph to achieve the effect of intelligent diagnosis, you can refer to Figure 6 .like Figure 6 As shown:

[0123] After achieving primary positioning, the current on-site defect description text "Online monitoring shows that no oil droplets are formed in the oil conservator of the ABC phase of the Furong transformer main transformer" can be used to identify the key defect entities (ABC phases of the Furong transformer main transformer, oil conservator) through the preprocessing of the BERT-based character vector model and the defect key entity recognition model based on the CRF layer. The entities are aligned with the entities in the transformer defect knowledge graph to obtain the graph defect description text corresponding to the oil conservator, and achieve secondary positioning;

[0124] Then the defect description text of the graph corresponding to the oil storage tank and the on-site defect description are matched by text similarity through the twin neural network to locate the standard defect description of the oil storage tank (slight leakage with signs of oil leakage);

[0125] Finally, based on the standard defect description, the cause and treatment relationship in the knowledge graph are used to locate the cause of the defect (loose oil pillow connection hose) and the method of eliminating the defect (standard installation of equipment), helping the operation and maintenance personnel to perform intelligent diagnosis of the transformer equipment.

[0126] Embodiment 2

[0127] The second embodiment of the present invention relates to an intelligent diagnosis system for power transformer equipment defects based on knowledge graph, such as Figure 7 As shown, the system includes:

[0128] A preprocessing unit, used to preprocess the obtained multiple on-site defect description texts of the transformer to obtain an initial word vector and an initial character vector of each on-site defect description text;

[0129] A sorting unit selects a current on-site defect description text from a plurality of on-site defect description texts of the transformer obtained;

[0130] An entity recognition and alignment unit is used to adopt a defect key entity recognition model based on a CRF layer to identify key defect entities in the current on-site defect description text based on the initial character vector, and to perform entity alignment with the acquired power transformer defect knowledge graph to obtain a graph defect description text;

[0131] The standard defect location unit is used to match the atlas defect description text with the current on-site defect description text based on the initial word vector using a twin neural network to locate the standard defect description of the defective equipment or defective parts;

[0132] A diagnostic unit is used to locate the standard defect cause and defect elimination method corresponding to the standard defect description in the power transformer defect knowledge graph.

[0133] Preferably, the preprocessing unit is further used to: perform word segmentation processing on the multiple field defect description texts respectively according to the power field dictionary; and process the word segmentation results using a word vector model based on Word2vec to obtain an initial word vector for each word in the word segmentation results.

[0134] Preferably, the preprocessing unit is further used to: perform fine-grained character segmentation on each on-site defect description text to obtain a corresponding character set e = (e1, e2, ..., e m ), where e1, e2, ..., e m They are the 1st to mth characters after cutting respectively; the BERT-based character vector model is used to obtain the dynamic vector of each character as the initial character vector.

[0135] Preferably, the sorting unit is also used to: adopt a BiGRU-based defect severity level recognition model to determine the severity level of the defect corresponding to each on-site defect description text based on the initial word vector; and select the current on-site defect description text from the multiple on-site defect description texts according to the severity level sorting.

[0136] Preferably, the sorting unit is further used to: adopt a BiGRU-based transformer defect severity level recognition model, based on the initial word vector of each word, obtain the hidden state of the corresponding word; the BiGRU-based transformer defect severity level recognition model is expressed as:

[0137]

[0138] Among them, h t represents the hidden state of word t, x t Represents the initial word vector of word t, w t and v t Represents the forward hidden state of the BiGRU corresponding to word t and the reverse hidden state The corresponding weight, b t Represents the bias corresponding to the hidden state of word t;

[0139] According to the hidden states of all words in each on-site defect description text, the severity level of the defect corresponding to each on-site defect description text is obtained, and the formula is as follows:

[0140]

[0141] in, For the splicing operation, W o is the linear transformation matrix, softMax(·) is the normalized exponential function, d is the final defect level probability of the corresponding on-site defect description text, and n is the total number of words.

[0142] Preferably, the entity recognition and alignment unit is further used for:

[0143] The character sequence of the defect text is obtained by the linear layer dimensionality reduction processing of the initial character vector set of the current on-site defect description text X = (x1,...,x m );

[0144] According to X=(x1,...,x m ), using the defect key entity recognition model based on the CRF layer, determine the predicted label sequence corresponding to X as Y = (y1,....,y m ), and then find the predicted label sequence with the highest probability as the global optimal label sequence; where the predicted label sequence corresponding to X is Y = (y1,....,y m ) is expressed as:

[0145]

[0146] Where X=(x1,...,x m ) is the character sequence of the defective text, Y=(y1,....,y m ) is X=(x1,...,x m ) corresponding to the predicted label, represents the label transfer score, Indicates that the character is Y i+1 The score of each label, s(X,Y) represents the emission score of the defect text, and P(Y|X) represents the probability that the predicted label sequence corresponding to X is Y after global normalization.

[0147] The identified key defect entities and the acquired power transformer defect knowledge graph are entity aligned to obtain the graph defect description text.

[0148] Preferably, the standard defect locating unit is also used for:

[0149] Preprocess the graph defect description text to obtain the corresponding initial word vector;

[0150] The initial word vectors of the atlas defect description text and the initial word vectors of the current on-site defect description text are encoded by the twin neural network constructed by TextCNN to obtain the corresponding encoding representation:

[0151] c i =Max(ReLu(W s K i +bk i ))

[0152]

[0153] Among them, W s K represents the set matrix of the initial word vectors of the atlas defect description text or the current on-site defect description text, i represents the i-th convolution kernel, bk i represents the bias corresponding to the convolution kernel, Max represents the maximum pooling layer, ReLu(·) is the activation function, c i represents the encoding representation corresponding to the i-th convolution kernel, and c represents c1~c k The concatenated coded representation;

[0154] The coding representations corresponding to the atlas defect description text and the current on-site defect description text are aggregated using the Manhattan distance to obtain the standard defect description of the defective equipment or defective parts.

[0155] Embodiment 3

[0156] This embodiment provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for intelligent diagnosis of defects of power transformer equipment based on knowledge graph as described in Embodiment 1 of the present invention are implemented.

[0157] The detailed steps are the same as the knowledge graph-based intelligent diagnosis method for power transformer equipment defects provided in Example 1, and will not be repeated here.

[0158] Embodiment 4

[0159] This embodiment relates to a storage medium on which a program is stored. When the program is executed by a processor, the steps of the intelligent diagnosis method for defects of power transformer equipment based on knowledge graph as described in Example 1 of the present invention are implemented.

[0160] The detailed steps are the same as the knowledge graph-based intelligent diagnosis method for power transformer equipment defects provided in Example 1, and will not be repeated here.

[0161] The beneficial effect of the present invention is that, compared with the prior art,

[0162] 1. First, through the defect key entity recognition model based on the CRF layer, help the operation and maintenance personnel locate the key entities (parts of the transformer) in the defect text. The key entities are the entities in the corresponding graph; then, based on the defect description text matching algorithm of the twin network, match the defect text and the defect description associated with the key entity to locate the best defect description; then, locate the cause of the defect and the method of eliminating the defect through the defect description to assist the operation and maintenance personnel in on-site work. In this way, the identification of key entities is equivalent to a global positioning search in the knowledge graph, and the matching operation based on the twin neural network is equivalent to further positioning after global positioning, which can assist the operation and maintenance personnel to perform fast and efficient operations, and can effectively avoid the problem of difficulty in diagnosing equipment defects due to inaccurate sampling.

[0163] 2. By adopting the BiGRU-based defect severity level model, the defect severity level (e.g., critical, serious, general) of the on-site defect text can be identified, thereby assisting on-site operation and maintenance personnel to prioritize the defects occurring in the transformer and improve the efficiency of defect diagnosis and treatment.

[0164] In summary, the present invention combines the association and reasoning functions of the constructed power transformer equipment defect knowledge graph to automatically analyze and diagnose various transformer field defect description information. At the same time, compared with the traditional rule-based or expert system-based diagnosis method, the knowledge graph-based intelligent diagnosis method has stronger intelligence and adaptability, and can automatically infer the optimal diagnosis result according to different defect conditions, thus becoming an important means to improve the management and maintenance level of transformer equipment, and effectively improve the operation quality and safety of transformer equipment. .

[0165] The present disclosure may be a system, method and / or computer program product. A computer program product may include a computer-readable storage medium on which are loaded computer-readable program instructions for causing a processor to implement various aspects of the present disclosure. A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, - but not limited to - an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a protruding structure in a groove on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., light pulses through a fiber optic cable), or an electrical signal transmitted through wires.

[0166] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0167] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A knowledge graph-based intelligent diagnosis method for power transformer equipment defects, characterized in that: include: Preprocessing the obtained multiple on-site defect description texts of the transformer to obtain the initial word vector and initial character vector of each on-site defect description text; Selecting a current on-site defect description text from a plurality of on-site defect description texts of the transformer obtained; The defect key entity recognition model based on the CRF layer is used to identify the key defect entities in the current on-site defect description text based on the initial character vector, and then align the entities with the acquired power transformer defect knowledge graph to obtain the graph defect description text; Based on the initial word vector, the twin neural network is used to match the atlas defect description text with the current on-site defect description text to locate the standard defect description of the defective equipment or defective parts; The standard defect causes and defect elimination methods corresponding to the standard defect descriptions are located in the power transformer defect knowledge graph.

2. The method for intelligent diagnosis of power transformer equipment defects based on knowledge graph according to claim 1 is characterized in that: Preprocess the multiple on-site defect description texts of the transformer to obtain the initial word vector and initial character vector of each on-site defect description text, including: According to the electric power field dictionary, the plurality of on-site defect description texts are respectively segmented; The word vector model based on Word2vec is used to process the word segmentation results to obtain the initial word vector of each word in the word segmentation results.

3. The intelligent diagnosis method for power transformer equipment defects based on knowledge graph according to claim 1 is characterized in that: Preprocess the multiple on-site defect description texts of the transformer to obtain the initial word vector and initial character vector of each on-site defect description text, including: Perform fine-grained character segmentation on each on-site defect description text to obtain the corresponding character set e = (e1, e2, ..., e m ), where e1, e2, ..., e m They are the 1st to mth characters after cutting respectively; The BERT-based character vector model is used to obtain the dynamic vector of each character as the initial character vector.

4. The power transformer equipment defect intelligent diagnosis system based on knowledge graph according to claim 1 is characterized in that: The current field defect description text is selected from the multiple field defect description texts of the transformer obtained, including: A BiGRU-based defect severity level recognition model is used to determine the severity level of each defect description text based on the initial word vector; The current on-site defect description text is selected from the plurality of on-site defect description texts according to the severity level sorting.

5. The power transformer equipment defect intelligent diagnosis system based on knowledge graph according to claim 4 is characterized in that: The BiGRU-based defect severity level model is used to determine the severity level of each defect description text based on the initial word vector, including: A transformer defect severity level recognition model based on BiGRU is adopted, and the hidden state of the corresponding word is obtained based on the initial word vector of each word; the transformer defect severity level recognition model based on BiGRU is expressed as: Among them, h t represents the hidden state of word t, x t Represents the initial word vector of word t, w t and v t Represents the forward hidden state of the BiGRU corresponding to word t and the reverse hidden state The corresponding weight, b t Represents the bias corresponding to the hidden state of word t; According to the hidden states of all words in each on-site defect description text, the severity level of the defect corresponding to each on-site defect description text is obtained, and the formula is as follows: in, For the splicing operation, W o is the linear transformation matrix, softMax(·) is the normalized exponential function, d is the final defect level probability of the corresponding on-site defect description text, and n is the total number of words.

6. The method for intelligent diagnosis of power transformer equipment defects based on knowledge graph according to claim 1 is characterized in that: The defect key entity recognition model based on the CRF layer is used to identify the key defect entities in the current on-site defect description text based on the initial character vector, and then align the entities with the acquired power transformer defect knowledge graph to obtain the graph defect description text, including: The character sequence of the defect text is obtained by the linear layer dimensionality reduction processing of the initial character vector set of the current on-site defect description text X = (x1,...,x m ); According to X=(x1,...,x m ), using the defect key entity recognition model based on the CRF layer, determine the predicted label sequence corresponding to X as Y = (y1,....,y m ), and then find the predicted label sequence with the highest probability as the global optimal label sequence; where the predicted label sequence corresponding to X is Y = (y1,....,y m ) is expressed as: Where X=(x1,...,x m ) is the character sequence of the defective text, Y=(y1,....,y m ) is X=(x1,...,x m ) corresponding to the predicted label, represents the label transfer score, Indicates that the character is Y i+1 The score of each label, s(X,Y) represents the emission score of the defect text, and P(Y|X) represents the probability that the predicted label sequence corresponding to X is Y after global normalization. The identified key defect entities and the acquired power transformer defect knowledge graph are entity aligned to obtain the graph defect description text.

7. The method for intelligent diagnosis of power transformer equipment defects based on knowledge graph according to claim 1 is characterized in that: Based on the initial word vector, the twin neural network is used to match the atlas defect description text with the current on-site defect description text to locate the standard defect description of the defective equipment or defective parts, including: Preprocess the graph defect description text to obtain the corresponding initial word vector; The initial word vectors of the atlas defect description text and the initial word vectors of the current on-site defect description text are encoded by the twin neural network constructed by TextCNN to obtain the corresponding encoding representation: c i =Max(ReLu(W s K i +bk i )) Among them, W s K represents the set matrix of the initial word vectors of the atlas defect description text or the current on-site defect description text, i represents the i-th convolution kernel, bk i represents the bias corresponding to the convolution kernel, Max represents the maximum pooling layer, ReLu(·) is the activation function, c i represents the encoding representation corresponding to the i-th convolution kernel, and c represents c1~c k The concatenated coded representation; The coding representations corresponding to the atlas defect description text and the current on-site defect description text are aggregated using the Manhattan distance to obtain the standard defect description of the defective equipment or defective parts.

8. An intelligent diagnosis system for power transformer equipment defects based on knowledge graph using the method as described in any one of claims 1 to 7, characterized in that: include: A preprocessing unit, used to preprocess the obtained multiple on-site defect description texts of the transformer to obtain an initial word vector and an initial character vector of each on-site defect description text; A sorting unit selects a current on-site defect description text from a plurality of on-site defect description texts of the transformer obtained; An entity recognition and alignment unit is used to adopt a defect key entity recognition model based on a CRF layer to identify key defect entities in the current on-site defect description text based on the initial character vector, and to perform entity alignment with the acquired power transformer defect knowledge graph to obtain a graph defect description text; The standard defect location unit is used to match the atlas defect description text with the current on-site defect description text based on the initial word vector using a twin neural network to locate the standard defect description of the defective equipment or defective parts; A diagnostic unit is used to locate the standard defect cause and defect elimination method corresponding to the standard defect description in the power transformer defect knowledge graph.

9. An electronic device, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-7.

10. A storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 7.

Images