Alarm incentive studying and judging method for automatic operation process of power dispatching

Through the BERT+BiLSTM+Resnet+CRF model and the power scheduling automation alarm knowledge graph, the actual characteristics of alarm information in power scheduling automation operations are extracted and matched, and the problem of low efficiency of traditional manual analysis is solved, and the rapid and accurate alarm positioning and solution are achieved. Measures provided.

CN118917657AActive Publication Date: 2024-11-08STATE GRID FUJIAN ELECTRIC POWER CO LTD +1
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202410938470.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-13
Publication Date
2024-11-08
Estimated Expiration
2044-07-13

AI Technical Summary

Technical Problem

The analysis and judgment of alarm information in traditional power scheduling automation operations relies on manual intervention, which is inefficient and susceptible to human factors, making it difficult to quickly and accurately locate the causes and solutions of alarms.

Method used

The BERT+BiLSTM+Resnet+CRF entity extraction model is used to extract text entity features, combine the power scheduling automation alarm knowledge graph, and search related work orders through breadth-first search and time descending sorting, and calculate matching using TF-IDF, cosine similarity and Siamese network. The most relevant work order returns the work items that result in alarms and solutions.

Benefits of technology

Effectively identify the entities in the alarm information, quickly locate the work items that lead to alarms, corresponding solutions and safety measures, reduce human interference, and improve the efficiency of alarm analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917657A_ABST
    Figure CN118917657A_ABST
Patent Text Reader

Abstract

The invention provides a power dispatching automation operation process-oriented alarm inducement study and judgment method, which comprises a text entity extraction module, a knowledge graph-based similar work order retrieval module, a similarity calculation-based work order matching module and a graph perfection module, and is characterized in that different entity features of a text are extracted by constructing a BERT + BiLSTM + Resnet + CRF entity extraction model; the method comprises the following steps of: selecting a power dispatching automation alarm knowledge graph, retrieving top-n related alarm work orders by using breadth-first search and time descending sort on the basis of the power dispatching automation alarm knowledge graph, matching the most related work orders by comprehensively considering TF-IDF, cosine similarity and Siamese network, and mapping misoperation causing alarm and research and judgment information of corresponding solutions and safety measures on the basis of the work orders; according to the method, the related entities of the alarm summary can be obtained by mining the potential entity characteristics of the alarm summary, the alarm entities are positioned by retrieving the historical power dispatching automatic alarm knowledge graph, the potential work items causing the occurrence of the alarm are mined, and the alarm reasons, the solutions and the solutions corresponding to the work items are returned.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power grid operation and maintenance, and in particular to an alarm cause analysis and judgment method for an electric power dispatching automation operation process. Background Art

[0002] The power system is one of the infrastructures that supports industrial production and daily life. Various alarm information may appear in the actual automation operation process. The traditional power automation alarm cause judgment relies on manual intervention, which is inefficient and easily affected by human factors. Based on the alarm knowledge graph in the field of power dispatching automation, various entities in the power system and their associations are modeled and stored. By extracting alarm entities based on the alarm phenomenon overview, searching the alarm entities in the knowledge graph, tracing the alarm work items from the alarm entities, and mapping the alarm causes, solutions and safety measures based on the work items, it can locate and judge the alarm causes of power dispatching automation operations, and provide auxiliary support for fault handling. Summary of the invention

[0003] The present invention proposes a method for analyzing and judging the cause of alarms for the process of automated power dispatching operations. The method can obtain entities related to the alarm overview by mining the potential entity features of the alarm overview, locate the alarm entity by retrieving the historical knowledge graph of automated power dispatching alarms, discover the potential work items that cause the alarm, return the cause of the alarm, the solution and the safety measures, and provide auxiliary support for the analysis and judgment of automated power dispatching operations alarms.

[0004] The present invention adopts the following technical solutions.

[0005] A method for analyzing and judging alarm causes in an automated power dispatching operation process, the method comprising a text entity extraction module, a similar work order retrieval module based on a knowledge graph, a work order matching module based on similarity calculation, and a graph improvement module. The method extracts different entity features of text by constructing a BERT+BiLSTM+Resnet+CRF entity extraction model, and retrieves top-n related alarm work orders based on a knowledge graph of automated power dispatching alarms by using breadth-first search and time descending sorting, and comprehensively considers TF-IDF, cosine similarity, and Siamese network matching to match the most relevant work orders, and maps out the analysis and judgment information of the erroneous operation that causes the alarm and the corresponding solution and safety measures based on the work order.

[0006] The method comprises the following steps:

[0007] Step S1, extracting and identifying text entity features of electric power dispatching automation operation alarm information to obtain text entity objects related to electric power dispatching automation operation;

[0008] Step S2: perform graph keyword search based on text entity objects to find a set of work orders with high matching degree;

[0009] Step S3: Calculate the text similarity between the alarm information and the alarm summary of the matching work order to obtain the most similar historical work order record;

[0010] Step S4: retrieve the work item that caused the alarm in the most similar work order record, return the work item and the corresponding solution and safety measures, and after the actual cause of the current alarm is found out, form an alarm work order record and import it into the knowledge graph.

[0011] In step S1, text entity features are extracted and identified and classified for the electric power dispatching automation operation alarm information to obtain text entity objects related to the electric power dispatching automation operation, which specifically includes the following steps:

[0012] Step S1A1, using the work order records, historical data, ledger information, and operation manual generated during the power dispatching automation operation as source data, and constructing a power dispatching automation operation data set through data cleaning;

[0013] Step S1A2: Build a BERT+BiLSTM+Resnet+CRF entity extraction model and train it based on the power dispatching automation operation dataset.

[0014] Step S1A3: Use the model trained in S1A2 to perform entity recognition on the electric power dispatching automation operation alarm information, and return various entities involved in the alarm information.

[0015] In step S1A1, during the data cleaning process, for text data, the python-docx library is used to extract text; for pdf file data, the OCR technology is used to extract text; for excel file table data, pandas is used to extract text, and all extracted text data are classified according to the text structure information, and the string is cut and deduplicated according to the text length to obtain the initial power dispatch automation operation data set;

[0016] In step S1A2, the BERT model construction formula is as follows:

[0017]

[0018] V=BERT([CLS],T1,...,T n ,[SEP])=(e1,...,e n )

[0019] Where Q = {Q1, Q2, ..., Q n} represents the initial text sequence, T={T1,T2,...,Tn} represents the initial text vector obtained by encoding. (e1,...,e n ) represents the vector after adding separators before and after the initial text vector;

[0020] In step S1A2, the BiLSTM model consists of a vector representation layer, a bidirectional LSTM layer, and a linear layer. The LSTM model consists of the input word X at time t. t , cell state C t , hidden layer state h t , forget gate f t , Memory Gate t , output gate o t The calculation process of LSTM includes: forgetting the information in the cell state and remembering the information at the new moment, retaining the information that is useful for the subsequent time, discarding the useless information, and outputting the hidden state h at each time step. t The output information of the forget gate, memory gate, and output gate is composed of the hidden state h at the previous moment t-1 and the current input X t control;

[0021] In step S1A2, the specific formula of the Resnet model is as follows:

[0022] y l =h(x l )+F(x l ,W l )

[0023] x l+1 =f(y l )

[0024] In the formula, x l , x l+1 They represent the input and output of the Lth residual unit, f represents the ReLU function, and h(x l ) represents x l The identity mapping of , F represents the residual function;

[0025] In step S1A2, the specific formula of the CRF model is as follows:

[0026]

[0027] Where Z(X) represents the normalization factor, f k (y i-1 ,y i ,X,i) is the feature function, which means the feature y of the i-th position given the input sequence x and the output sequence y i The value of λ k Represents the weight parameter.

[0028] In step S2, a graph keyword search is performed based on the text entity to find a set of work orders with a high matching degree. The specific steps include:

[0029] Step S2A1: construct a to-be-matched work order set Set, and store the work order set that is relatively matched with the alarm information by matching the alarm phenomenon descriptions that are relatively similar to the alarm information and the historical work order records;

[0030] Step S2A2, setting the hop count h of breadth-first search, based on the constructed electric power dispatching automation alarm knowledge graph, using the various entities extracted in step S1A3 to retrieve the corresponding alarm object entities in the graph, starting from each alarm object entity to perform breadth-first search, and adding all the work orders involved within the maximum hop count to the to-be-matched work order set Set;

[0031] Step S2A3: Sort the work orders in the to-be-matched work order set in descending order of time, and retain the top-n work orders with the most recent timeline.

[0032] The constructed electric power dispatching automation alarm knowledge graph described in step S2 includes but is not limited to the following entity types and relationship types:

[0033] Entities mainly include alarm work order record name, alarm phenomenon description, alarm cause, solution, and safety measures; relationships mainly include triggering, solution, and sequence. The information corresponding to the entity is stored in the form of attributes, where the alarm work order record name includes two attributes: entity id and entity name. The entity id is the unique identifier of the alarm work order record.

[0034] The work orders in the to-be-matched work order set described in step S2 are sorted in descending order of time, and the top-n work orders with the latest timeline are retained, so that the records in the to-be-matched work orders are highly relevant to the alarm causes mentioned in the alarm overview, and the top-n work orders with the latest timeline are sorted according to the severity of the problem, the scope of impact, and the urgency of solving the problem, so that the retained work order records reflect the latest system status and problem trends;

[0035] In step S3, text similarity calculation is performed between the alarm information and the alarm summary of the matching work order to obtain the most similar historical work order record, which specifically includes the following steps:

[0036] Step S3A1, setting a weight set λ=(λ1,λ2,λ3), which respectively represents the weights of the scores of the text after being calculated by TF-IDF, cosine similarity, and Siamese neural network;

[0037] Step S3A2: Calculate the TF-IDF value of the alarm information and the alarm overview of the matching work order to obtain the similarity S1;

[0038] Step S3A3, calculating the cosine similarity between the alarm information and the alarm overview of the matching work order to obtain similarity S2;

[0039] Step S3A4: Use the Siamese network to calculate the similarity between the alarm information and the alarm overview in the matching work order to obtain similarity S3;

[0040] Step S3A5, obtaining a similarity score S of the text by weighted calculation of the similarity;

[0041] Step S3A6: Take the work order with the highest similarity score as the best matching work order.

[0042] In step S3A2, the TF-IDF similarity calculation formula is as follows:

[0043]

[0044] TF-IDF=TF·IDF

[0045] Where n i,j Indicates the term t i In the document d j The number of occurrences in TF i,j Indicates the term t i In the document d j The frequency of occurrence in. |D| represents the number of all documents, |j:t i ∈d j | indicates that the term is included i The number of documents;

[0046] In step S3A3, the specific formula of cosine similarity is as follows:

[0047]

[0048] In the formula, A i , B i Represents the values ​​of each dimension of vector A and B, It means the inner product of vectors A and B. represents the product of A and B modulo;

[0049] In step S3A4, the specific structure of the Siamese network is as follows:

[0050] The Siamese network consists of two convolutional neural networks with the same structure. Its network structure consists of a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer consists of a 3x3 convolution kernel, and the pooling layer uses average pooling.

[0051] In step S3A5, the specific formula for similarity weighted calculation is as follows:

[0052] S=λ1S1+λ2S2+λ3S3

[0053] Where λ1, λ2, λ3 are weight parameters set in S3A1, and S1, S2, S3 are similarities calculated in steps S3A2, S3A3, and S3A4.

[0054] In step S4, the work item that causes the alarm phenomenon in the most similar work order record is retrieved, and the work item and the corresponding solution and safety measure are returned. After the cause of the current alarm is actually found out, the work order is imported to improve the knowledge graph, which specifically includes the following steps:

[0055] Step S4A1: retrieve the most matching work order item obtained in step S3, find the alarm part in the alarm phenomenon content described in the work order item that matches the current alarm phenomenon description, query the corresponding cause, and obtain the alarm cause reason;

[0056] Step S4A2: Retrieve the solution and corresponding safety measures for the alarm reason in the most matching work item, and return the solution and measures;

[0057] Step S4A3: Return the alarm reason, solution and safety measures to the user in the form of a work ticket, to provide auxiliary support for alarm information processing in the process of power dispatch automation operation;

[0058] Step S4A4: The alarm causes and corresponding solutions, safety measures, etc. that are finally clarified are formed into work orders, imported into the system, and the knowledge graph is improved.

[0059] The method obtains entities related to the alarm summary by mining the potential entity features summarized in the alarm information, locates the alarm entity by searching the historical electric power dispatching automation alarm knowledge graph, traces and discovers the potential work items that cause the alarm, returns the alarm causes, solutions and safety measures corresponding to the potential work items, and imports the alarm records formed by actual operation and maintenance operations into the knowledge graph;

[0060] The text entity extraction module uses an entity extraction model to extract entities from the alarm text to reduce human interference when extracting and analyzing the alarm overview generated during the power dispatch automation operation;

[0061] The similar work order retrieval module based on the knowledge graph performs related entity retrieval based on the power dispatch automation alarm knowledge graph to quickly locate the most relevant top-n work orders that cause the alarm phenomenon;

[0062] The work order matching module based on similarity calculation is used to find the work order that best matches the alarm information, and map the work items that are most likely to cause the alarm and the corresponding solutions and safety measures from the work order, providing auxiliary support for the positioning and analysis of alarm information in power dispatching automation operations.

[0063] Compared with the prior art, the present invention has the following beneficial effects: better identification of different entities in the alarm overview, retrieval of the work order record that best matches the alarm overview based on the electric power dispatching automation alarm knowledge graph, and mapping of the work items that cause the alarm phenomenon and the corresponding solutions and safety measures from the work order. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0065] Attached Figure 1 It is a schematic diagram of the principle framework of the present invention. DETAILED DESCRIPTION

[0066] As shown in the figure, a method for analyzing and judging the cause of alarms in the process of automated power dispatching operations is provided. The method includes a text entity extraction module, a similar work order retrieval module based on a knowledge graph, a work order matching module based on similarity calculation, and a graph improvement module. The method extracts different entity features of the text by constructing a BERT+BiLSTM+Resnet+CRF entity extraction model, and retrieves the top-n related alarm work orders based on the knowledge graph of automated power dispatching alarms using breadth-first search and time descending sorting. The method also considers TF-IDF, cosine similarity, and Siamese network matching to be the most relevant work orders, and maps out the analysis information of the erroneous operation that causes the alarm and the corresponding solutions and safety measures based on the work orders.

[0067] The method comprises the following steps:

[0068] Step S1, extracting and identifying text entity features of electric power dispatching automation operation alarm information to obtain text entity objects related to electric power dispatching automation operation;

[0069] Step S2: perform graph keyword search based on text entity objects to find a set of work orders with high matching degree;

[0070] Step S3: Calculate the text similarity between the alarm information and the alarm summary of the matching work order to obtain the most similar historical work order record;

[0071] Step S4: retrieve the work item that caused the alarm in the most similar work order record, return the work item and the corresponding solution and safety measures, and after the actual cause of the current alarm is found out, form an alarm work order record and import it into the knowledge graph.

[0072] In step S1, text entity features are extracted and identified and classified for the electric power dispatching automation operation alarm information to obtain text entity objects related to the electric power dispatching automation operation, which specifically includes the following steps:

[0073] Step S1A1, using the work order records, historical data, ledger information, and operation manual generated during the power dispatching automation operation as source data, and constructing a power dispatching automation operation data set through data cleaning;

[0074] Step S1A2: Build a BERT+BiLSTM+Resnet+CRF entity extraction model and train it based on the power dispatching automation operation dataset.

[0075] Step S1A3: Use the model trained in S1A2 to perform entity recognition on the electric power dispatching automation operation alarm information, and return various entities involved in the alarm information.

[0076] In step S1A1, during the data cleaning process, for text data, the python-docx library is used to extract text; for pdf file data, the OCR technology is used to extract text; for excel file table data, pandas is used to extract text, and all extracted text data are classified according to the text structure information, and the string is cut and deduplicated according to the text length to obtain the initial power dispatch automation operation data set;

[0077] In step S1A2, the BERT model construction formula is as follows:

[0078]

[0079] V=BERT([CLS],T1,...,T n ,[SEP])=(e1,...,e n )

[0080] Where Q = {Q1, Q2, ..., Q n} represents the initial text sequence, T={T1,T2,...,T n} represents the initial text vector obtained by encoding. (e1,...,e n ) represents the vector after adding separators before and after the initial text vector;

[0081] In step S1A2, the BiLSTM model consists of a vector representation layer, a bidirectional LSTM layer, and a linear layer. The LSTM model consists of the input word X at time t. t , cell state C t , hidden layer state h t , forget gate f t , Memory Gatet , output gate o t The calculation process of LSTM includes: forgetting the information in the cell state and remembering the information at the new moment, retaining the information that is useful for the subsequent time, discarding the useless information, and outputting the hidden state h at each time step. t The output information of the forget gate, memory gate, and output gate is composed of the hidden state h at the previous moment t-1 and the current input X t control;

[0082] In step S1A2, the specific formula of the Resnet model is as follows:

[0083] y l =h(x l )+F(x l ,W l )

[0084] x l+1 =f(y l )

[0085] In the formula, x l , x l+1 They represent the input and output of the Lth residual unit, f represents the ReLU function, and h(x l ) represents x l The identity mapping of , F represents the residual function;

[0086] In step S1A2, the specific formula of the CRF model is as follows:

[0087]

[0088] Where Z(X) represents the normalization factor, f k (y i-1 ,y i ,X,i) is the feature function, which means the feature y of the i-th position given the input sequence x and the output sequence y i The value of λ k Represents the weight parameter.

[0089] In step S2, a graph keyword search is performed based on the text entity to find a set of work orders with a high matching degree. The specific steps include:

[0090] Step S2A1: construct a to-be-matched work order set Set, and store the work order set that is relatively matched with the alarm information by matching the alarm phenomenon descriptions that are relatively similar to the alarm information and the historical work order records;

[0091] Step S2A2, setting the hop count h of breadth-first search, based on the constructed electric power dispatching automation alarm knowledge graph, using the various entities extracted in step S1A3 to retrieve the corresponding alarm object entities in the graph, starting from each alarm object entity to perform breadth-first search, and adding all the work orders involved within the maximum hop count to the to-be-matched work order set Set;

[0092] Step S2A3: Sort the work orders in the to-be-matched work order set in descending order of time, and retain the top-n work orders with the most recent timeline.

[0093] The constructed electric power dispatching automation alarm knowledge graph described in step S2 includes but is not limited to the following entity types and relationship types:

[0094] Entities mainly include alarm work order record name, alarm phenomenon description, alarm cause, solution, and safety measures; relationships mainly include triggering, solution, and sequence. The information corresponding to the entity is stored in the form of attributes, where the alarm work order record name includes two attributes: entity id and entity name. The entity id is the unique identifier of the alarm work order record.

[0095] The work orders in the to-be-matched work order set described in step S2 are sorted in descending order of time, and the top-n work orders with the latest timeline are retained, so that the records in the to-be-matched work orders are highly relevant to the alarm causes mentioned in the alarm overview, and the top-n work orders with the latest timeline are sorted according to the severity of the problem, the scope of impact, and the urgency of solving the problem, so that the retained work order records reflect the latest system status and problem trends;

[0096] In step S3, text similarity calculation is performed between the alarm information and the alarm summary of the matching work order to obtain the most similar historical work order record, which specifically includes the following steps:

[0097] Step S3A1, setting a weight set λ=(λ1,λ2,λ3), which respectively represents the weights of the scores of the text after being calculated by TF-IDF, cosine similarity, and Siamese neural network;

[0098] Step S3A2: Calculate the TF-IDF value of the alarm information and the alarm overview of the matching work order to obtain the similarity S1;

[0099] Step S3A3, calculating the cosine similarity between the alarm information and the alarm overview of the matching work order to obtain similarity S2;

[0100] Step S3A4: Use the Siamese network to calculate the similarity between the alarm information and the alarm overview in the matching work order to obtain similarity S3;

[0101] Step S3A5, obtaining a similarity score S of the text by weighted calculation of the similarity;

[0102] Step S3A6: Take the work order with the highest similarity score as the best matching work order.

[0103] In step S3A2, the TF-IDF similarity calculation formula is as follows:

[0104]

[0105] TF-IDF=TF·IDF

[0106] Where n i,j Indicates the term t i In the document d j The number of occurrences in TF i,j Indicates the term t i In the document d j The frequency of occurrence in. |D| represents the number of all documents, |j:t i ∈d j | indicates that the term is included i The number of documents; in step S3A3, the specific formula of cosine similarity is as follows:

[0107]

[0108] In the formula, A i , B i Represents the values ​​of each dimension of vector A and B, It means the inner product of vectors A and B. represents the product of A and B modulo;

[0109] In step S3A4, the specific structure of the Siamese network is as follows:

[0110] The Siamese network consists of two convolutional neural networks with the same structure. Its network structure consists of a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer consists of a 3x3 convolution kernel, and the pooling layer uses average pooling.

[0111] In step S3A5, the specific formula for similarity weighted calculation is as follows:

[0112] S=λ1S1+λ2S2+λ3S3

[0113] Where λ1, λ2, λ3 are weight parameters set in S3A1, and S1, S2, S3 are similarities calculated in steps S3A2, S3A3, and S3A4.

[0114] In step S4, the work item that causes the alarm phenomenon in the most similar work order record is retrieved, and the work item and the corresponding solution and safety measure are returned. After the cause of the current alarm is actually found out, the work order is imported to improve the knowledge graph, which specifically includes the following steps:

[0115] Step S4A1: retrieve the most matching work order item obtained in step S3, find the alarm part in the alarm phenomenon content described in the work order item that matches the current alarm phenomenon description, query the corresponding cause, and obtain the alarm cause reason;

[0116] Step S4A2: Retrieve the solution and corresponding safety measures for the alarm reason in the most matching work item, and return the solution and measures;

[0117] Step S4A3: Return the alarm reason, solution and safety measures to the user in the form of a work ticket, to provide auxiliary support for alarm information processing in the process of power dispatch automation operation;

[0118] Step S4A4: The alarm causes and corresponding solutions, safety measures, etc. that are finally clarified are formed into work orders, imported into the system, and the knowledge graph is improved.

[0119] The method obtains entities related to the alarm summary by mining the potential entity features summarized in the alarm information, locates the alarm entity by searching the historical electric power dispatch automation alarm knowledge graph, discovers the potential work items that cause the alarm, returns the alarm causes, solutions and safety measures corresponding to the potential work items, and imports the alarm records formed by actual operation and maintenance operations into the knowledge graph;

[0120] The text entity extraction module uses an entity extraction model to extract entities from the alarm text to reduce human interference when extracting and analyzing the alarm overview generated during the power dispatch automation operation;

[0121] The similar work order retrieval module based on the knowledge graph performs related entity retrieval based on the power dispatch automation alarm knowledge graph to quickly locate the most relevant top-n work orders that cause the alarm phenomenon;

[0122] The work order matching module based on similarity calculation is used to find the work order that best matches the alarm information, and map the work items that are most likely to cause the alarm and the corresponding solutions and safety measures from the work order, providing auxiliary support for the positioning and analysis of alarm information in power dispatching automation operations.

[0123] This embodiment includes the following functional modules running on a server cluster or server: a text entity extraction module, a similar work order retrieval module based on a knowledge graph, a work order matching module based on similarity calculation, and a graph improvement module;

[0124] Among them, for the alarm overview generated during the power dispatching automation operation, the traditional manual extraction and analysis method is easily affected by human factors. The entity extraction model is used to extract the entity of the alarm text, which can reduce the interference of human factors; the relevant entity retrieval based on the power dispatching automation alarm knowledge graph can quickly locate the most relevant top-n work orders that cause the alarm phenomenon; the alarm text matching based on similarity calculation can find the work order that best matches the alarm, and map the work items that are most likely to cause the alarm and the corresponding solutions and safety measures from the work order, providing auxiliary support for the positioning and analysis of alarm information in power dispatching automation operations.

[0125] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for analyzing and judging alarm causes in an electric power dispatching automation operation process, characterized in that: The method includes a text entity extraction module, a similar work order retrieval module based on a knowledge graph, a work order matching module based on similarity calculation, and a graph improvement module. The method extracts different entity features of the text by constructing a BERT+BiLSTM+Resnet+CRF entity extraction model, and retrieves the top-n related alarm work orders based on the power dispatching automation alarm knowledge graph by using breadth-first search and time descending sorting, and comprehensively considers TF-IDF, cosine similarity and Siamese network matching of the most relevant work orders, and maps out the misoperation that causes the alarm and the corresponding solution and safety measures based on the work order.

2. According to claim 1, a method for analyzing and determining the cause of an alarm in an automated power dispatching operation process, characterized in that: The steps include: Step S1, extracting and identifying text entity features of electric power dispatching automation operation alarm information to obtain text entity objects related to electric power dispatching automation operation; Step S2: perform graph keyword search based on text entity objects to find a set of work orders with high matching degree; Step S3: Calculate the text similarity between the alarm information and the alarm summary of the matching work order to obtain the most similar historical work order record; Step S4: retrieve the work item that caused the alarm in the most similar work order record, return the work item and the corresponding solution and safety measures, and after the actual cause of the current alarm is found out, form an alarm work order record and import it into the knowledge graph.

3. The method for analyzing and determining the alarm cause of an electric power dispatching automation operation process according to claim 2 is characterized in that: In step S1, text entity features are extracted and identified and classified for the electric power dispatching automation operation alarm information to obtain text entity objects related to the electric power dispatching automation operation, which specifically includes the following steps: Step S1A1, using the work order records, historical data, ledger information, and operation manual generated during the power dispatching automation operation as source data, and constructing a power dispatching automation operation data set through data cleaning; Step S1A2: Build a BERT+BiLSTM+Resnet+CRF entity extraction model and train it based on the power dispatching automation operation dataset. Step S1A3: Use the model trained in S1A2 to perform entity recognition on the electric power dispatching automation operation alarm information, and return various entities involved in the alarm information.

4. The method for analyzing and determining the alarm cause of an electric power dispatching automation operation process according to claim 3 is characterized in that: In step S1A1, during the data cleaning process, for text data, the python-docx library is used to extract text; for pdf file data, the OCR technology is used to extract text; for excel file table data, pandas is used to extract text, and all extracted text data are classified according to the text structure information, and the string is cut and deduplicated according to the text length to obtain the initial power dispatch automation operation data set; In step S1A2, the BERT model construction formula is as follows: V=BERT([CLS],T1,...,T n ,[SEP])=(e1,...,e n ) Where Q = {Q1, Q2, ..., Q n } represents the initial text sequence, T={T1,T2,...,T n } represents the initial text vector obtained by encoding. (e1,...,e n ) represents the vector after adding separators before and after the initial text vector; In step S1A2, the BiLSTM model consists of a vector representation layer, a bidirectional LSTM layer, and a linear layer. The LSTM model consists of the input word X at time t. t , cell state C t , hidden layer state h t , forget gate f t , Memory Gate t , output gate o t The calculation process of LSTM includes: forgetting the information in the cell state and remembering the information at the new moment, retaining the information that is useful for the subsequent time, discarding the useless information, and outputting the hidden state h at each time step. t The output information of the forget gate, memory gate, and output gate is composed of the hidden state h at the previous moment t-1 and the current input X t control; In step S1A2, the specific formula of the Resnet model is as follows: y l =h(x l )+F(x l ,W l ) x l+1 =f(y l ) In the formula, x l , x l+1 They represent the input and output of the Lth residual unit, f represents the ReLU function, and h(x l ) represents x l The identity mapping of , F represents the residual function; In step S1A2, the specific formula of the CRF model is as follows: Where Z(X) represents the normalization factor, f k (y i-1 ,y i ,X,i) is the feature function, which means the feature y of the i-th position given the input sequence x and the output sequence y i The value of λ k Represents the weight parameter.

5. The method for analyzing and determining the alarm cause of an electric power dispatching automation operation process according to claim 2 is characterized in that: In step S2, a graph keyword search is performed based on the text entity to find a set of work orders with a high matching degree. The specific steps include: Step S2A1: construct a to-be-matched work order set Set, and store the work order set that is relatively matched with the alarm information by matching the alarm phenomenon descriptions that are relatively similar to the alarm information and the historical work order records; Step S2A2, setting the hop count h of breadth-first search, based on the constructed electric power dispatching automation alarm knowledge graph, using the various entities extracted in step S1A3 to retrieve the corresponding alarm object entities in the graph, starting from each alarm object entity to perform breadth-first search, and adding all the work orders involved within the maximum hop count to the to-be-matched work order set Set; Step S2A3: Sort the work orders in the to-be-matched work order set in descending order of time, and retain the top-n work orders with the most recent timeline.

6. The method for analyzing and determining the alarm cause of an electric power dispatching automation operation process according to claim 2 is characterized in that: The constructed electric power dispatching automation alarm knowledge graph described in step S2 includes but is not limited to the following entity types and relationship types: Entities mainly include alarm work order record name, alarm phenomenon description, alarm cause, solution, and safety measures; relationships mainly include triggering, solution, and sequence. The information corresponding to the entity is stored in the form of attributes, where the alarm work order record name includes two attributes: entity id and entity name. The entity id is the unique identifier of the alarm work order record.

7. The method for analyzing and determining the alarm cause of an electric power dispatching automation operation process according to claim 2 is characterized in that: The work orders in the to-be-matched work order set described in step S2 are sorted in descending order of time, and the top-n work orders with the latest timeline are retained, so that the records in the to-be-matched work orders are highly relevant to the alarm causes mentioned in the alarm overview, and the top-n work orders with the latest timeline are sorted according to the severity of the problem, the scope of impact, and the urgency of solving the problem, so that the retained work order records reflect the latest system status and problem trends; In step S3, text similarity calculation is performed between the alarm information and the alarm summary of the matching work order to obtain the most similar historical work order record, which specifically includes the following steps: Step S3A1, setting a weight set λ=(λ1,λ2,λ3), which respectively represents the weights of the scores of the text after being calculated by TF-IDF, cosine similarity, and Siamese neural network; Step S3A2, calculating the TF-IDF value of the alarm information and the alarm summary of the matching work order, and obtaining the similarity S1; Step S3A3, calculating the cosine similarity of the alarm information and the alarm summary of the matching work order, and obtaining the similarity S2; Step S3A4, using the Siamese network to calculate the similarity of the alarm information and the alarm summary in the matching work order, and obtaining the similarity S3; Step S3A5, obtaining a similarity score S of the text by weighted calculation of the similarity; Step S3A6: Take the work order with the highest similarity score as the best matching work order.

8. The method for analyzing and determining the alarm cause of an electric power dispatching automation operation process according to claim 2 is characterized in that: In step S3A2, the TF-IDF similarity calculation formula is as follows: TF-IDF=TF·IDF Where n i,j Indicates the term t i In the document d j The number of occurrences in TF i,j Indicates the term t i In the document d j The frequency of occurrence in. |D| represents the number of all documents, |j:t i ∈d j | indicates that the term is included i The number of documents; In step S3A3, the specific formula of cosine similarity is as follows: In the formula, A i , B i Represents the values ​​of each dimension of vector A and B, It means the inner product of vectors A and B. represents the product of A and B modulo; In step S3A4, the specific structure of the Siamese network is as follows: The Siamese network consists of two convolutional neural networks with the same structure. Its network structure consists of a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer consists of a 3x3 convolution kernel, and the pooling layer uses average pooling. In step S3A5, the specific formula for similarity weighted calculation is as follows: S=λ1S1+λ2S2+λ3S3 Where λ1, λ2, λ3 are weight parameters set in S3A1, and S1, S2, S3 are similarities calculated in steps S3A2, S3A3, and S3A4.

9. The method for analyzing and determining the alarm cause of an electric power dispatching automation operation process according to claim 2 is characterized in that: In step S4, the work item that causes the alarm phenomenon in the most similar work order record is retrieved, and the work item and the corresponding solution and safety measure are returned. After the cause of the current alarm is actually found out, the work order is imported to improve the knowledge graph, which specifically includes the following steps: Step S4A1: retrieve the most matching work order item obtained in step S3, find the alarm part in the alarm phenomenon content described in the work order item that matches the current alarm phenomenon description, query the corresponding cause, and obtain the alarm cause reason; Step S4A2: Retrieve the solution and corresponding safety measures for the alarm reason in the most matching work item, and return the solution and measures; Step S4A3: Return the alarm reason, solution and safety measures to the user in the form of a work ticket, to provide auxiliary support for alarm information processing in the process of power dispatch automation operation; Step S4A4: The alarm causes and corresponding solutions, safety measures, etc. that are finally clarified are formed into work orders, imported into the system, and the knowledge graph is improved.

10. The method for analyzing and determining the alarm cause of an electric power dispatching automation operation process according to claim 2, characterized in that: The method obtains entities related to the alarm summary by mining the potential entity features summarized in the alarm information, locates the alarm entity by searching the historical electric power dispatch automation alarm knowledge graph, discovers the potential work items that cause the alarm, returns the alarm causes, solutions and safety measures corresponding to the potential work items, and imports the alarm records formed by actual operation and maintenance operations into the knowledge graph; The text entity extraction module uses an entity extraction model to extract entities from the alarm text to reduce human interference when extracting and analyzing the alarm overview generated during the power dispatch automation operation; The similar work order retrieval module based on the knowledge graph performs related entity retrieval based on the power dispatch automation alarm knowledge graph to quickly locate the most relevant top-n work orders that cause the alarm phenomenon; The work order matching module based on similarity calculation is used to find the work order that best matches the alarm information, and map the work items that are most likely to cause the alarm and the corresponding solutions and safety measures from the work order, providing auxiliary support for the positioning and analysis of alarm information in power dispatching automation operations.

Citation Information

Patent Citations

  • Method and system for automatically constructing knowledge maps for mass unstructured texts

    CN108875051A

  • Power grid infrastructure construction risk management and control method based on image recognition and knowledge graph

    CN114694098A

  • Power grid alarm information auxiliary decision-making method based on knowledge graph

    CN116957236A

  • Tax field-oriented knowledge map construction method and system

    WO2021196520A1

  • Affair information query method and apparatus, and computer device and storage medium

    WO2023134057A1