Fault Joint Diagnosis Method Based on Multi-Source Data Fusion of Transformers

Abstract

A fault joint diagnosis method based on multi-source data fusion of transformers. In the method, fault diagnosis data of transformers are collected and data cleaning is completed. The fault diagnosis data of transformers include structured data and unstructured data. Knowledge extraction is performed on the cleaned unstructured data, and multi-source data fusion is completed by combining data fusion technology to realize the association between multi-source data. Based on the extracted knowledge triples and structured data, a transformer fault knowledge graph is constructed. Fault diagnosis is carried out based on the constructed transformer fault knowledge graph, and fault diagnosis of the transformer is realized with a single parameter to obtain a diagnosis result. Based on the diagnosis result, joint diagnosis of transformer faults is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of on-line monitoring of power transformers in electrical engineering, and particularly relates to a fault joint diagnosis method based on multi-source data fusion of transformers. Background Art

[0002] The power system is one of the important infrastructure for the operation of modern society. Among them, the transformer, as an important power transmission and distribution device of electric energy, undertakes core tasks such as voltage lifting and current distribution in the power system. With the continuous development of the power system, the normal operation of transformers is crucial for the stability and reliability of the power system. However, due to the complex and changeable operating environment and potential internal and external factors, transformers face various potential fault risks, such as winding short circuits, insulation aging, partial discharges, etc. These faults may lead to a decline in equipment performance or even damage, seriously affecting the normal operation of the power system.

[0003] Traditional fault diagnosis methods often rely on manual experience and regular inspections. This method has disadvantages such as being unable to capture hidden dangers in a timely manner, low efficiency, and relying on manual experience. Therefore, it is necessary to introduce advanced technical means to improve the accuracy and timeliness of transformer fault diagnosis.

[0004] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present invention. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] The present invention proposes a fault joint diagnosis method based on multi-source data fusion of transformers. Based on the fault diagnosis technology of knowledge graphs, with the help of advanced technical means such as artificial intelligence and big data analysis, it realizes intelligent perception and rapid judgment of complex and changeable fault modes, improves the accuracy and timeliness of fault diagnosis, and conducts dynamic joint diagnosis on transformer equipment faults.

[0006] The fault joint diagnosis method based on multi-source data fusion of transformers includes the following steps:

[0007] S100: Collect transformer fault diagnosis data and complete data cleaning. The transformer fault diagnosis data includes structured data and unstructured data. The structured data is data stored in the form of a table with a data model and relationships;

[0008] S200: Extract knowledge from the cleaned unstructured data and complete the fusion of multi-source data by combining data fusion technology to realize the association between multi-source data;

[0009] S300: Build a transformer fault knowledge graph based on the extracted knowledge triples and structured data;

[0010] S400: Conduct fault diagnosis based on the constructed transformer fault knowledge graph, and achieve fault diagnosis of the transformer with a single parameter to obtain a diagnosis result.

[0011] S500: Achieve joint diagnosis of transformer faults based on the diagnosis result.

[0012] In addition, the present invention also discloses a device for joint fault diagnosis based on multi-source data fusion of transformers, which is characterized in that it includes:

[0013] A collection unit, which is used for: collecting transformer fault diagnosis data and completing data cleaning. The transformer fault diagnosis data includes structured data and unstructured data, and the structured data is data stored in a tabular form with a data model and relationships.

[0014] A fusion unit, which is used for: extracting knowledge from the cleaned unstructured data and completing the fusion of multi-source data by combining data fusion technology to achieve the association between multi-source data.

[0015] A construction unit, which is used for: constructing a transformer fault knowledge graph based on the extracted knowledge triples and structured data.

[0016] A single diagnosis unit, which is used for: conducting fault diagnosis based on the constructed transformer fault knowledge graph, and achieving fault diagnosis of the transformer with a single parameter to obtain a diagnosis result.

[0017] A joint diagnosis unit, which is used for: achieving joint diagnosis of transformer faults based on the diagnosis result.

[0018] Beneficial effects:

[0019] With the help of the knowledge graph, the present invention can achieve intelligent perception and rapid judgment of complex and changeable fault modes, improve the accuracy and timeliness of fault diagnosis. It not only helps to improve the reliability of the power system, reduce the impact of faults on the system, but also can effectively extend the service life of equipment, improve the economy and safety of power grid operation. It can obtain very high-accuracy equipment status analysis and fault diagnosis as well as other applications based on the transformer fault knowledge graph, discover potential fault signs in advance, take timely maintenance measures, and reduce the maintenance cost and operation and maintenance risk of equipment. This has a positive promoting effect on the safe and stable operation of the power system. Description of the drawings

[0020] Figure 1 It is a schematic flow diagram of knowledge extraction for unstructured data;

[0021] Figure 2 It is a schematic flow diagram of constructing a transformer fault knowledge graph;

[0022] Figure 3It is a flow chart of transformer fault diagnosis based on a transformer fault knowledge graph;

[0023] Figure 4 A partial example schematic diagram of the transformer fault knowledge graph;

[0024] Figure 5 A fault joint diagnosis interface diagram based on the transformer fault knowledge graph;

[0025] Figure 6 It is a schematic diagram of the process of the present invention.

[0026] By reading the detailed descriptions in the preferred specific embodiments below, various other advantages and benefits of the present invention will become clear to those of ordinary skill in the art. The accompanying drawings in the specification are only for the purpose of showing the preferred embodiments, and are not considered as limiting the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts. Moreover, throughout the drawings, the same reference numerals are used to denote the same components. Specific Embodiments

[0027] The specific embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the specific embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0028] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art should understand that technicians may use different terms to refer to the same component. The specification and claims do not use the difference in terms as a way to distinguish components, but use the difference in function of components as the criterion for distinction. As mentioned throughout the specification and claims, "comprising" or "including" is an open-ended term and should be interpreted as "including but not limited to". The subsequent description in the specification is for the purpose of implementing the preferred embodiments of the present invention, but the description is based on the general principles of the specification and is not intended to limit the scope of the present invention. The protection scope of the present invention shall be defined by the appended claims.

[0029] For ease of understanding of the embodiments of the present invention, the following will further explain with specific embodiments as examples in conjunction with the accompanying drawings, and the accompanying drawings do not constitute a limitation to the embodiments of the present invention.

[0030] As Figures 1 to 6 shown, the fault joint diagnosis method based on transformer multi-source data fusion includes the following steps:

[0031] S100: Collect transformer fault diagnosis data and complete data cleaning. The transformer fault diagnosis data includes structured data and unstructured data. The structured data is stored in a tabular form with a data model and relationships.

[0032] S200: Extract knowledge from the cleaned unstructured data and complete the fusion of multi-source data by combining data fusion technology to realize the association between multi-source data.

[0033] S300: Build a transformer fault knowledge graph based on the extracted knowledge triples and structured data.

[0034] S400: Conduct fault diagnosis based on the built transformer fault knowledge graph, and realize the fault diagnosis of the transformer with a single parameter to obtain a diagnosis result.

[0035] S500: Realize the joint diagnosis of transformer faults based on the diagnosis result.

[0036] In the preferred implementation of the method, in step S100, the data collection and data cleaning include the following steps:

[0037] S1011: Collect data including transformer ledger information, fault reports, specification standards, on-line monitoring data, and meteorological data.

[0038] S1012: Complete data cleaning for the collected data, and clean the structured data and unstructured data respectively.

[0039] S1013: Cleaning unstructured data includes removing special characters from the text, unifying the format, and removing stop words. Among them, cleaning picture data includes removing noise, adjusting the size, and image enhancement.

[0040] S1014: Structured data cleaning includes data type conversion and cleaning of duplicate data, missing values, and outliers.

[0041] In the preferred implementation of the method, in step S200, the knowledge extraction and combination of data fusion technology to complete the fusion of multi-source data include the following steps:

[0042] S2011: Extract knowledge from unstructured data, extract entities, relationships, events, and attributes, and combine them into knowledge triples according to the relationships between entities. The knowledge triples include "entity - relationship - entity" or "entity - relationship - attribute"; see Figure 1 , Figure 1It is a schematic diagram of the process for unstructured data knowledge extraction. First, preprocess the unstructured text data to complete data cleaning. Then, gradually extract entities, relationships, and attributes from the cleaned text data. During the extraction process, due to different technologies used, this process transforms the text into information. Finally, integrate the extracted information into triples and store them in a graph database, specifically represented as entity-relationship-entity, entity-attribute-attribute value, relationship-attribute-attribute value, etc.

[0043] S2012: Store the knowledge triples extracted from unstructured data in the graph database. For the entities of the cleaned structured data, store them in the graph database, construct the schema of the transformer fault knowledge graph, establish relationships between the structured data entities based on the same fields in the data, and establish relationships with the previously stored unstructured knowledge triples to complete multi-source data fusion.

[0044] In the preferred embodiment of the described method, in step S300, the construction of the transformer fault knowledge graph includes the following steps:

[0045] S3011: Based on historical data, analyze and extract information from the transformer fault knowledge graph schema, automatically cluster the information that cannot be accurately identified, supplement and improve the transformer fault knowledge graph through semi-supervised learning, and improve the knowledge base of the transformer fault knowledge graph by analyzing point table information, historical alarm information, defect information, and monitoring data.

[0046] S3012: Improve the transformer fault knowledge graph through knowledge reasoning and graph evaluation.

[0047] In the preferred embodiment of the described method, in step S400, the fault diagnosis of the transformer based on the constructed transformer fault knowledge graph using a single parameter includes:

[0048] S4011: Based on online monitoring data, diagnose the operating conditions of the transformer using dissolved gas in oil, partial discharge, core grounding current, and oil temperature data respectively.

[0049] S4012: Based on the transformer fault knowledge graph, achieve fault diagnosis of the transformer according to each monitored quantity.

[0050] In the preferred embodiment of the described method, in step S500, the joint diagnosis of transformer faults based on the diagnosis results includes:

[0051] S5011: Based on the diagnosis results of a single parameter, determine the joint diagnosis result based on the weight of each parameter result.

[0052] S5012: Joint diagnosis of transformer faults by combining monitoring data of multiple parameters.

[0053] In one embodiment, the on-line monitoring data at least includes dissolved gas in oil data, partial discharge data, core grounding current data, and oil temperature. Further, knowledge extraction includes entity recognition (NER), relation extraction, event extraction, text classification, etc., and fault diagnosis uses similarity, statistical analysis, machine learning, as well as classification algorithms and the analytic hierarchy process; knowledge extraction models include dictionary matching, regular expressions, hidden Markov models (HMMs), conditional random fields (CRFs), pre-trained model BERT, classification model SVM, etc.

[0054] In one embodiment, the data comes from different systems of the substation, manufacturers or suppliers, industry standards and specifications, etc.

[0055] Perform data cleaning on the collected data. Unstructured data and structured data have different cleaning methods. Unstructured data mainly removes special characters in the text, unifies case, and eliminates stop words, etc. Structured data mainly solves data type conversion, duplicate data, missing values, and outliers, etc. After cleaning the data, entities, relations, events, and attributes can be extracted from unstructured data, and entities and attributes can be extracted from structured data. Different entities can be linked through the same fields.

[0056] Extract entities, relations, events, and attributes from unstructured data, and construct a transformer fault knowledge graph schema with equipment as the core based on expert experience and structured data, which facilitates the conversion of data from the relational data format of the table to the graph database format. The designed transformer fault knowledge graph schema contains entity types: substation, transformer, fault, maintenance personnel, meteorological conditions, standards, dissolved gas in oil, partial discharge, core grounding current, and oil temperature. Each type of entity contains multiple attributes, and the attributes reflect the information of the entity. Entities are linked through relations, and some relations also contain some attributes. Store entities, relations, and attributes based on Neo4j, complete entity alignment and coreference resolution, and initially construct a transformer fault knowledge graph.

[0057] Figure 3 It is a flow chart of transformer fault diagnosis based on the transformer fault knowledge graph. Based on the transformer fault knowledge graph, first, perform fault diagnosis on the transformer according to a single monitoring quantity. Then, obtain the final diagnosis result by weighted calculation based on the weights of the influence degrees of each parameter determined by the analytic hierarchy process on the diagnosis result.

[0058] Based on knowledge reasoning and graph evaluation, the initially constructed transformer fault knowledge graph is improved. Through knowledge reasoning, the existing information in the graph is used for inference to fill potential knowledge gaps and improve the knowledge coverage of the graph. At the same time, through graph evaluation, the consistency, accuracy, and effectiveness of the graph can be tested, potential errors or inconsistencies can be identified, and corrections can be made. This process is an iterative one. Through the injection of professional domain knowledge and systematic evaluation, the transformer fault knowledge graph can be continuously optimized and improved. Figure 4 It is a part of the constructed complete transformer fault knowledge graph. It is a part of the transformer fault knowledge graph constructed by the method described in this patent. This part involves entities such as transformers, faults, monitoring data, and staff, as well as the relationships existing between them. Clicking on an entity can display the attributes of that entity.

[0059] Based on the constructed complete transformer fault knowledge graph, the data of one sensor among dissolved gases in oil, partial discharge, core grounding current, and oil temperature is used as a diagnostic parameter for fault diagnosis. Taking the dissolved gas in oil data as an example, when the online monitoring data (concentrations of multiple gases) of a certain dissolved gas in oil is input, the system will judge the operating condition of the transformer. The judgment basis is the standards stored in the transformer fault knowledge graph and the operating state of the transformer with the data most similar to the input gas concentration data found based on graph algorithms. By synthesizing the results of the standards and graph algorithms, the operating state of the transformer under the condition of the dissolved gas in oil data can be known.

[0060] The data of multiple or all sensors among dissolved gases in oil, partial discharge, core grounding current, and oil temperature are used as parameters for combined transformer fault diagnosis. When the data of multiple sensors are input, such as the data of dissolved gases in oil, partial discharge, and oil temperature, the system will give the operating state of the transformer based on the input data. The diagnosis basis is to diagnose separately with the data of a single sensor and synthesize the diagnosis results using the analytic hierarchy process. Figure 5 It is an example of the interface graph for combined fault diagnosis of the data of a certain transformer at a certain moment. Through the interface, when the ledger information of the transformer and the monitoring data (dissolved gases in oil, partial discharge, oil temperature, core grounding current) at a certain time are input, the operating condition of the transformer at that time can be output.

[0061] In one embodiment, structured data is stored in tabular form and has a clear data model and relationships, while unstructured data refers to data without a clear structure or format and usually requires special processing methods for analysis and utilization.

[0062] The data involved in this patent includes transformer ledger information, fault reports, specifications and standards, on-line monitoring data, meteorological data, etc. Transformer ledger information, meteorological data, and on-line monitoring data are usually structured data, while fault reports and specifications and standards are usually unstructured data. The acquisition methods of these data are different and need to be selected and acquired according to specific situations.

[0063] Transformer ledger information includes the basic information of the transformer, such as the manufacturer, model, rated capacity, installation location, commissioning date, etc. It is usually stored in a database or spreadsheet in the form of a table, managed and maintained by the power company or equipment operation unit, and can be obtained through regular inspections, maintenance records, and equipment procurement information. Fault reports include information such as fault descriptions, causes, handling processes, and maintenance records of the transformer. These information are usually recorded in text form in reports or logs, and may include attachments in formats such as pictures and charts. They are compiled by maintenance personnel or engineers and can be obtained through fault reports and maintenance logs. Specifications and standards include documents such as technical specifications, industry standards, safety standards, and operating procedures related to transformers. These information are usually recorded in text form and may include attachments in the form of charts and drawings. They can be obtained from industry associations, standardization organizations, government departments, or equipment manufacturers, or can be obtained by purchasing or downloading electronic documents. On-line monitoring data includes the real-time monitoring data of the transformer, such as the measured values of parameters such as partial discharge, temperature, current, and voltage. In addition to numerical data, it also includes information such as timestamps and event reports. Generally, it is collected in real time by sensors and monitoring equipment installed on the transformer and then transmitted to the monitoring system for storage and analysis. Meteorological data includes various parameters related to meteorology, such as temperature, humidity, wind speed, rainfall, etc. These data are usually stored in tabular form and include information such as timestamps. They are obtained through meteorological stations, satellite remote sensing, meteorological forecasting agencies, etc., and can be collected in real time by meteorological sensors.

[0064] In one embodiment, for the knowledge extraction of unstructured data (documents, reports) and the fusion of multi-source data, the following steps and methods are taken:

[0065] 1) Preprocess the unstructured data, including text cleaning, word segmentation, part-of-speech tagging, entity recognition, etc., for subsequent knowledge extraction and fusion.

[0066] 2) Use natural language processing (NLP) technology to extract key information from unstructured texts such as fault reports, such as fault descriptions, causes, handling processes, etc. Convert the extracted information into a structured knowledge representation form, such as triples (entity, relationship, entity), entity relationship diagrams, etc.

[0067] 3) Integrate the structured data and the extracted knowledge representation to construct a comprehensive data model for multi-source data. Establish the associations between different data sources through shared entities or attributes. For example, match the device IDs mentioned in the fault reports with those in the monitoring data, or associate the keywords in the fault descriptions with the relevant clauses in the specification standards.

[0068] Through the above process, knowledge extraction from unstructured data can be achieved, and combined with data fusion technology to complete the fusion with structured data, realizing the association between data. The fault knowledge graph of transformers can be enriched to provide more comprehensive and accurate support for business decision-making and applications.

[0069] Knowledge extraction extracts useful information from unstructured text, including entity recognition, relation extraction, event extraction, and attribute extraction, etc. Entity recognition identifies entities with specific meanings from text, such as transformers, substations, faults, rules, etc., which is achieved through named entity recognition (NER) technology. NER technology uses methods based on rules, statistical models, or deep learning to identify entities in text. Relation extraction is to extract the relationships or connections between entities from text, which is achieved through methods based on rules, pattern matching, statistical models, or deep learning. Event extraction is to extract events or actions with specific semantics from text, which is achieved through the method based on semantic role labeling (SRL). SRL matches the verbs in the text with their corresponding semantic roles to extract event information. Attribute extraction is to extract the features or attributes of entities from text, which is achieved through methods based on pattern matching, rules, statistical models, or deep learning.

[0070] Combine these extracted entities, relations, events, and attributes into knowledge triples. First, assign a unique identifier to each entity. Then, represent each relation as a triple, which includes two entities and the type of relation between them. Next, represent each event as a triple, which includes the trigger word of the event, the participating entities, and the event type. Finally, represent each attribute as a triple, which includes the entity, the attribute name, and the attribute value.

[0071] In one embodiment, in the transformer fault-related data, the association between multi-source data is the association between data from different sources or types. The association can be one-to-one or one-to-many, and all associations are directional, depending on the relationship between the data and the business requirements. Through the association analysis of multi-source data, potential associations and rules between data can be discovered, providing a deeper understanding and insight for business decision-making and problem-solving.

[0072] The acquired data is divided into structured data and unstructured data. For structured data such as transformer ledger information, it is stored in the form of a table, including information such as transformer ID, name, model, and manufacturer. In the graph database, it is stored in the form of entity-attribute-attribute value, that is, the transformer is the entity, and transformer ID, name, model, manufacturer, etc. are attributes. For unstructured data such as fault reports, they are stored in text form. Through the aforementioned knowledge extraction techniques, information such as transformer ID, transformer name, fault name, fault type, fault description, and fault cause can be extracted from them. In the graph database, it is stored in the form of entity-attribute-attribute value, that is, the transformer fault is the entity, and transformer ID, transformer name, fault name, fault type, fault description, fault cause, etc. are attributes. In the graph database, the two entities are related through the common attribute transformer ID, that is, there is a "fault occurred" relationship between the transformer entity and the fault entity, that is, (transformer - fault occurred - fault), thus connecting the structured data and unstructured data. For unstructured data, entity-relationship-entity and entity-attribute-attribute value triples can be extracted from it according to the aforementioned knowledge extraction techniques and stored in the graph database. For structured data, the two data are regarded as two entities. If the two data have common attributes, they are directly related using a relationship.

[0073] In a one-to-one association, there is a one-to-one relationship between one type of data and another type of data. For example, the data collected by each sensor may be associated with a specific transformer. In a one-to-many association, there is a one-to-many relationship between one type of data and other types of data. For example, the ledger information can be associated with multiple fault information because a transformer may have multiple faults. In a many-to-one association, multiple types of data are all associated with another type of data. For example, the data collected by multiple sensors may all be associated with the operating status of the same transformer because these sensors are all used to monitor different parameters of the transformer.

[0074] In one embodiment, refer to Figure 2 , Figure 2 is a schematic diagram of the process for constructing a transformer fault knowledge graph. This diagram includes all the processes for constructing a transformer fault knowledge graph, including Figure 1 extracting information from unstructured data. The transformer fault knowledge graph established through this process can achieve the fusion of multi-source data and the fault diagnosis of transformers. The process for constructing a transformer fault knowledge graph is as follows:

[0075] 1) Collect various data related to transformers, including ledger information, fault reports, standards and specifications, monitoring data, meteorological data, etc. This data can come from multiple sources such as monitoring devices, maintenance records, and operation and maintenance systems.

[0076] 2) Perform preprocessing operations such as cleaning, deduplication, and formatting on the collected data to ensure data quality and consistency. This step can include operations such as missing value handling, outlier detection, and data transformation.

[0077] 3) Extract useful information from the preprocessed data. Use methods such as natural language processing technology, text mining technology, and entity recognition technology to extract key information from text data, such as fault types, fault causes, and handling measures.

[0078] 4) For information that cannot be accurately identified, use machine learning or deep learning techniques for automatic classification or clustering. Achieve this through methods such as text classification, clustering algorithms, and topic modeling.

[0079] 5) Use semi-supervised learning techniques to supplement and improve the knowledge graph. By training with existing labeled data and a large amount of unlabeled data, the system's understanding and processing ability of unknown data can be improved, thereby enhancing the coverage and accuracy of the knowledge graph.

[0080] 6) Construct the extracted information and processed data into the structure of a knowledge graph, and combine expert experience and pattern matching methods to solve entity disambiguation and co-reference resolution problems. Use technologies such as graph databases to store and manage the nodes and edges of the knowledge graph for querying and reasoning. And continuously update and improve the knowledge base, including adding new knowledge, correcting errors, and updating data. Achieve this through regular data updates, manual reviews, and automated algorithms, etc.

[0081] Through statistical analysis of the data, some features and information can be extracted. Current, voltage, temperature, etc. can be extracted from the point table information of the transformer. From the historical alarm information, the fault types, frequencies, time periods, etc. that the transformer has experienced in the past can be extracted. Information such as the faults that the transformer has experienced and the maintenance records can be analyzed from the defect information. Information such as the operating status and performance parameters of the transformer can be extracted from the monitoring data.

[0082] Use natural language processing (NLP) techniques to process text data and extract key information such as fault type, description, time, etc. For information that cannot be accurately identified, use clustering algorithms (such as K-means, hierarchical clustering, etc.) to automatically classify similar data for further processing later. Input known labeled data (such as known fault types, fault causes, etc.) and unlabeled data into the model for training. Use semi-supervised learning algorithms (such as autoencoders, generative adversarial networks, etc.) to predict and classify unlabeled data, thereby supplementing and improving the information in the knowledge graph. And construct the structure of the knowledge graph with the extracted information and processed data, including entities, attributes, relationships, etc. Continuously update and improve the knowledge base through methods such as manual review and automated algorithms, including adding new knowledge, correcting errors, updating data, etc.

[0083] In one embodiment, for dissolved gases in oil, analyze the types and concentrations of dissolved gases in oil. According to the generation mechanisms and concentration change trends of different gases, judge the state of the transformer insulation system, such as whether there are problems like partial discharge, overheating, insulation aging, etc. For partial discharge, monitor indicators such as the amplitude, number of times, and duration of partial discharge in the transformer, and analyze the spectral characteristics and time-domain characteristics of the partial discharge signal to judge the location, degree, and type of discharge, thereby evaluating the insulation state and health condition of the transformer. For the core grounding current, monitor the magnitude and waveform of the core grounding current of the transformer, as well as the change trend of the grounding current. According to the change of the grounding current, judge the insulation state and grounding situation of the transformer, whether there are insulation faults or grounding faults. For the oil temperature, monitor the real-time value, change trend, over-temperature situation, etc. of the transformer oil temperature. According to the change of the oil temperature, evaluate the load condition of the transformer, whether the operating temperature is normal, and whether there is an overheating risk.

[0084] Based on the constructed transformer fault knowledge graph for fault diagnosis, in the transformer fault knowledge graph, the concentration of dissolved gases in oil has been associated with fault modes such as insulation aging and discharge, the amplitude of partial discharge has been associated with fault modes such as partial discharge and insulation damage, the core grounding current has been associated with fault modes such as grounding fault and insulation damage, and the oil temperature has been associated with fault modes such as overload and insulation aging. For a single parameter, the process is as follows:

[0085] 1) Collect and clean the transformer monitoring data to ensure the accuracy and integrity of the data. Extract the single parameter required for diagnosis.

[0086] 2) Extract features from the monitoring data of the single parameter, and extract features related to faults, such as peak value, mean value, variance, frequency-domain features, etc.

[0087] 3) Use the constructed transformer fault knowledge graph to match the extracted features with the fault patterns in the knowledge graph. Based on the matching results, find the fault mode or fault entity that matches the features.

[0088] 4) Perform fault diagnosis based on the matching results. If the feature highly matches a certain fault mode, the corresponding fault type can be diagnosed. Analyze and interpret the diagnosis results to determine the severity and impact of the fault.

[0089] 5) Based on feedback and experience from actual applications, continuously optimize and update the transformer fault knowledge graph, including adding new fault modes, correcting errors, updating data, etc., to improve the accuracy and efficiency of diagnosis.

[0090] In one embodiment, the process of implementing joint diagnosis of transformer faults based on single parameter diagnosis results is as follows:

[0091] 1) Determine the weight based on expert experience, historical data analysis, actual case studies, etc. The weight indicates the importance of each parameter for fault diagnosis, so that the parameter results can be comprehensively considered in the subsequent joint diagnosis process.

[0092] 2) After determining the weights of multiple monitoring parameters such as dissolved gas concentration in oil, partial discharge amplitude, core grounding current, oil temperature, etc., the selected parameters are jointly diagnosed using a weighted summation method. According to the weight of each parameter, the diagnostic results of each parameter are weighted to obtain a joint diagnosis result.

[0093] 3) In the joint diagnosis process, there may be contradictions between the diagnostic results of different parameters. At this time, the contradictory results are adjusted or further analyzed according to the weights. Expert judgment, historical case comparison and other methods are added to determine the final diagnostic result and obtain the joint diagnosis result.

[0094] In one embodiment, for cleaning of unstructured data text data, first remove special characters, such as punctuation marks, HTML tags, etc., using regular expressions or string processing methods. Then, unify the format, convert the text into a unified format, such as unifying lowercase or uppercase, removing extra spaces, etc. Finally, remove stop words, such as common words such as "的", "是", "在", etc., in the text based on domain knowledge or a predefined stop word list.

[0095] For the cleaning of unstructured data images, first, remove the noise by using image processing techniques such as median filtering and Gaussian filtering. Then, resize the image to a uniform size for subsequent processing and analysis. Finally, enhance the image by using techniques such as histogram equalization and contrast enhancement to enhance the quality and readability of the image.

[0096] For the cleaning of structured data, first, convert the data into the correct data types, such as converting strings to numbers, converting date and time formats, etc., for subsequent analysis and modeling. Then, remove duplicate data rows and retain only unique data rows to avoid affecting analysis and modeling. Secondly, handle missing values by choosing to fill in missing values, delete data rows or columns containing missing values, or use interpolation and other methods. Finally, use statistical methods or visualization methods to detect outliers, and then process them according to business rules or domain knowledge, which can be deletion, replacement, or correction.

[0097] In one embodiment, the schema of the transformer fault knowledge graph refers to the structured description of entities and relations in the graph, including the attributes of entities and the connections between relations. To construct the schema of the transformer fault knowledge graph and associate various entities, first, identify the entities and attributes in the transformer fault knowledge graph based on the information in the data. Then, based on the identified entities and attributes, define the relations between the entities. The relations describe the connections and interactions between entities. For example, there can be relations such as "fault occurred" and "maintenance" between a transformer and a fault report. Finally, according to the identified entities, attributes, and relations, construct the schema of the graph, and use the representation method of the schema language of the graph database to describe the structure of the graph. For example, the entity-attribute-relation (EAV) model can be used to represent the schema of the graph.

[0098] During the process of constructing the graph, add diagnostic criteria as part of the graph. The diagnostic criteria are represented as entities and establish relations with other entities to guide fault diagnosis and prediction. Define the diagnostic criteria entities and establish relations with entities such as fault reports and online monitoring data to describe the diagnostic criteria for specific fault states.

[0099] When establishing entity relations, first, establish the relations between structured data entities. Based on the same fields or attributes in the data, establish the relations between structured data entities. Then, establish the relations between structured data and unstructured knowledge triples, and establish relations between the stored unstructured knowledge triples and structured data entities. For example, establish a relation between the transformer entity through the description field in the fault report to associate the fault situation described in the fault report with the actual information of the transformer.

[0100] Based on the collected data, a comprehensive schema of the transformer fault knowledge graph is constructed as follows:

[0101] Entities include:

[0102] Transformer ledger, including attributes such as transformer ID (unique identifier), model, manufacturer, rated capacity, etc.;

[0103] Fault report, including attributes such as report ID (unique identifier), transformer ID (foreign key associated with transformer ledger information), report time, report content, fault category, fault description, etc.;

[0104] Diagnostic criteria, including attributes such as standard ID (unique identifier), standard name, standard description, reference documents, etc.;

[0105] Online monitoring data, which are dissolved gases in oil, partial discharge, oil temperature, and core grounding current information respectively, with attributes such as monitoring ID (unique identifier), transformer ID (foreign key associated with transformer ledger information), monitoring time, monitoring parameters, etc.;

[0106] Meteorological data, including attributes such as data ID (unique identifier), monitoring time, temperature, humidity, wind speed, etc

[0107] Maintenance data, including attributes such as maintenance ID (unique identifier), transformer ID (foreign key associated with transformer ledger information), maintenance time, maintenance content, maintenance personnel, etc.;

[0108] Staff members, including attributes such as personnel ID, name, department, etc.;

[0109] Relationships include:

[0110] Transformer and fault report: One transformer can have multiple fault reports.

[0111] Transformer and online monitoring data: One transformer can have multiple online monitoring data.

[0112] Transformer and maintenance data: One transformer can have multiple maintenance records.

[0113] Fault report and diagnostic criteria: One fault report may involve one or more diagnostic criteria.

[0114] Monitoring data and diagnostic criteria; Monitoring data has corresponding diagnostic criteria.

[0115] Relationship between staff members and fault reports, maintenance data: The relationship between staff members and faults and maintenance.

[0116] By constructing the above schema, a comprehensive transformer fault knowledge graph can be established, covering various relevant information and relationships, facilitating comprehensive analysis and management of transformer fault diagnosis, monitoring, and maintenance. Usually, according to the usage scenario and specific business, a transformer fault knowledge graph with more entities, relationships, and attributes can be constructed.

[0117] In one embodiment, the point table information refers to the data acquisition points and monitoring parameters defined in the transformer monitoring system, usually provided by equipment manufacturers or the monitoring system, including information such as parameter names, data types, units, etc. The historical alarm information records the faults, abnormalities, and alarm information that occurred during the operation of the transformer, from channels such as the transformer monitoring system, SCADA system, or maintenance records. The defect information refers to the defects, abnormalities, and fault phenomena found during the overhaul and maintenance of the transformer, usually recorded by maintenance personnel in maintenance logs, overhaul reports, or repair records. The monitoring data is the real-time operating data of the transformer collected through the transformer monitoring system, including monitoring parameters such as oil temperature, dissolved gases in oil, partial discharge, etc., collected by sensors and stored in the database.

[0118] First, clean and preprocess the collected transformer point table information, historical alarm information, defect information, and monitoring data, including removing duplicate data, handling missing values, handling outliers, etc., to ensure the quality and accuracy of the data. Then, use data analysis and mining techniques to analyze the cleaned data, discover the patterns, trends, and abnormal situations therein, using methods such as statistical analysis, machine learning, and deep learning. Next, according to the results of the data analysis, extract and update the relevant information in the transformer fault knowledge graph, including fault types, causes, characteristics, diagnostic methods, etc., using techniques such as natural language processing and text mining for knowledge extraction. Finally, according to the correlation between different data, establish a relationship model between the data, update the structure of the transformer fault knowledge graph, and verify and evaluate the updated transformer fault knowledge graph, checking its accuracy and integrity, using methods such as expert review and cross-validation for evaluation.

[0119] In another embodiment, the present invention also discloses a fault joint diagnosis device based on transformer multi-source data fusion, which includes:

[0120] A collection unit, which is used for: collecting transformer fault diagnosis data and completing data cleaning, where the transformer fault diagnosis data includes structured data and unstructured data, and the structured data is data stored in a tabular form with a data model and relationships;

[0121] A fusion unit, which is used for: extracting knowledge from the cleaned unstructured data and completing the fusion of multi-source data by combining data fusion techniques to achieve the association between multi-source data;

[0122] A construction unit, which is used for: constructing a transformer fault knowledge graph based on the extracted knowledge triples and structured data;

[0123] A single diagnosis unit, which is used for: performing fault diagnosis based on the constructed transformer fault knowledge graph, and achieving fault diagnosis of the transformer with a single parameter to obtain a diagnosis result;

[0124] A combined diagnosis unit, which is used to: achieve a combined diagnosis of transformer faults based on the diagnosis results.

[0125] Preferably, the collection unit is further used to:

[0126] Collect data including transformer ledger information, fault reports, specification standards, on-line monitoring data, and meteorological data;

[0127] Complete data cleaning for the collected data, and clean structured data and unstructured data respectively;

[0128] Among them,

[0129] Cleaning unstructured data includes removing special characters of the text, unifying the format, and removing stop words. Among them, cleaning picture data includes removing noise, adjusting the size, and image enhancement;

[0130] Structured data cleaning includes data type conversion and cleaning duplicate data, missing values, and outliers.

[0131] Preferably, the fusion unit is further used to:

[0132] Extract knowledge from unstructured data, extract entities, relationships, events, attributes, and combine them into knowledge triples according to the relationships between entities. The knowledge triples include "entity-relationship-entity" or "entity-relationship-attribute";

[0133] Store the knowledge triples extracted from unstructured data into a graph database. For the entities of the structured data after cleaning, store them in the graph database, construct a schema for the transformer fault knowledge graph, establish relationships between structured data entities based on the same fields in the data, and establish relationships with the already stored unstructured knowledge triples to complete multi-source data fusion.

[0134] Preferably, the construction unit is further used to:

[0135] According to historical data, analyze and extract information from the data based on the transformer fault knowledge graph schema, automatically cluster the information that cannot be accurately identified, supplement and improve the transformer fault knowledge graph through semi-supervised learning, and improve the knowledge base of the transformer fault knowledge graph by analyzing point table information, historical alarm information, defect information, and monitoring data.

[0136] Improve the transformer fault knowledge graph through knowledge reasoning and graph evaluation.

[0137] Preferably, the single diagnosis unit is further used to:

[0138] Based on on-line monitoring data, the operating conditions of the transformer are diagnosed respectively by dissolved gases in oil, partial discharge, core grounding current, and oil temperature data.

[0139] Based on the transformer fault knowledge graph, fault diagnosis of the transformer is realized according to each monitored quantity.

[0140] Preferably, the joint diagnosis unit is further configured to:

[0141] Based on the diagnosis results of single parameters, determine the joint diagnosis result based on the weights of the results of each parameter.

[0142] Joint diagnosis of transformer faults by combining the monitoring data of multiple parameters.

[0143] Although the embodiments of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the above specific embodiments and application fields. The above specific embodiments are merely illustrative and guiding, rather than restrictive. Those of ordinary skill in the art can also make many forms under the inspiration of this specification and without departing from the scope protected by the claims of the present invention, and these all belong to the scope of protection of the present invention.

Claims

1. A fault joint diagnosis method based on transformer multi-source data fusion, characterized in that It includes the following steps: S100: Collect relevant data for transformer fault diagnosis and complete data cleaning. The transformer fault diagnosis data includes structured data and unstructured data. The structured data is stored in the form of a table with a data model and relationships. In step S100, data collection and data cleaning include the following steps: S1011: Collect data including transformer inventory information, fault reports, specification standards, online monitoring data, and meteorological data; S1012: Complete data cleaning for the collected data, and clean structured data and unstructured data respectively; Among them, Cleaning unstructured data includes removing special characters from the text, unifying the format, and removing stop words. Among them, cleaning image data includes removing noise, adjusting the size, and image enhancement; Structured data cleaning includes data type conversion and handling duplicate data, missing values, and outliers; S200: Extract knowledge from the cleaned unstructured data and complete the fusion of multi-source data by combining data fusion technology to achieve the association between multi-source data. S2011: Extract knowledge from unstructured data, extract entities, relationships, events, and attributes, and combine them into knowledge triples according to the relationships between entities. Knowledge triples include "entity - relationship - entity" or "entity - relationship - attribute"; S2012: Store the knowledge triples extracted from unstructured data in a graph database. For the entities of the cleaned structured data, store them in the graph database, construct the schema of the transformer fault knowledge graph, establish relationships between structured data entities based on the same fields in the data, and establish relationships with the previously stored unstructured knowledge triples to complete the fusion of multi-source data; S300: Based on the extracted knowledge triples and structured data, construct a transformer fault knowledge graph. The construction of the transformer fault knowledge graph includes the following steps: S3011: According to historical data, analyze and extract information based on the schema of the transformer fault knowledge graph, and automatically cluster the information that cannot be accurately identified. Supplement and improve the transformer fault knowledge graph through semi-supervised learning. Improve the knowledge base of the transformer fault knowledge graph by analyzing point table information, historical alarm information, defect information, and monitoring data. Among them, the schema of the transformer fault knowledge graph refers to the structured description of entities and relationships in the graph, including the attributes of entities and the connections between relationships. Construct the schema of the transformer fault knowledge graph and associate various entities. First, identify the entities and attributes in the transformer fault knowledge graph according to the information in the data. Then, on the basis of identifying entities and attributes, define the relationships between entities. Relationships describe the connections and interactions between entities. Finally, according to the identified entities, attributes, and relationships, construct the schema of the transformer fault knowledge graph, and use the representation method of the schema language of the graph database to describe the structure of the graph. Entities include: Transformer inventory, including transformer ID, model, manufacturer, and rated capacity; Fault report, including report ID, transformer ID, report time, report content, fault category, and fault description; Diagnostic criteria, including standard ID, standard name, standard description, and reference documents; Online monitoring data, which are dissolved gases in oil, partial discharge, oil temperature, and core grounding current information respectively, each with monitoring ID, transformer ID, monitoring time, and monitoring parameters; Meteorological data, including data ID, monitoring time, temperature, humidity, and wind speed, Maintenance data, including maintenance ID, transformer ID, maintenance time, maintenance content, and maintenance personnel; Staff members, including personnel ID, name, and department; Relationships include: Transformer and fault report: One transformer has multiple fault reports, Transformer and online monitoring data: One transformer has multiple online monitoring data, Transformer and maintenance data: One transformer has multiple maintenance records, Fault report and diagnostic criteria: One fault report involves one or more diagnostic criteria, Monitoring data and diagnostic criteria; Monitoring data has corresponding diagnostic criteria, Relationship between staff members and fault reports, maintenance data: The relationship between staff members and faults and maintenance, S3012: Improve the transformer fault knowledge graph through knowledge reasoning and graph evaluation, and use graph database technology to store and manage the nodes and edges of the knowledge graph for querying and reasoning, and continuously update and improve the knowledge base through regular data updates, manual reviews, and automated algorithms; S400: Conduct fault diagnosis based on the constructed transformer fault knowledge graph, and achieve fault diagnosis of the transformer with a single parameter to obtain a diagnostic result. The fault diagnosis of the transformer with a single parameter based on the constructed transformer fault knowledge graph includes, S4011: Based on online monitoring data, diagnose the operating conditions of the transformer with dissolved gases in oil, partial discharge, core grounding current, and oil temperature data respectively, S4012: Based on the transformer fault knowledge graph, achieve fault diagnosis of the transformer according to each monitored quantity; S500: Achieve joint diagnosis of transformer faults based on the diagnostic result, S5011: Based on the diagnostic result of a single parameter, determine the joint diagnostic result based on the weight of each parameter result, S5012: Conduct joint diagnosis of transformer faults by combining the monitoring data of multiple parameters.

2. A fault joint diagnosis device based on transformer multi-source data fusion for implementing the method according to claim 1, characterized in that, It includes: Collection unit, which is used to: Collect transformer fault diagnosis data and complete data cleaning. The transformer fault diagnosis data includes structured data and unstructured data. The structured data is data stored in tabular form with a data model and relationships; Fusion unit, which is used to: Extract knowledge from the cleaned unstructured data and complete the fusion of multi-source data by combining data fusion technology to achieve the association between multi-source data; Construction unit, which is used to: Construct a transformer fault knowledge graph based on the extracted knowledge triples and structured data; Single diagnosis unit, which is used to: Conduct fault diagnosis based on the constructed transformer fault knowledge graph, and achieve fault diagnosis of the transformer with a single parameter to obtain a diagnostic result; Joint diagnosis unit, which is used to: Achieve joint diagnosis of transformer faults based on the diagnostic result.

3. The device according to claim 2, characterized in that The collection unit is also used for: collecting data including transformer ledger information, fault reports, specifications and standards, on-line monitoring data, and meteorological data; performing data cleaning on the collected data, and cleaning structured data and unstructured data respectively; wherein, cleaning unstructured data includes removing special characters of the text, unifying formats, and removing stop words. Among them, cleaning picture data includes removing noise, adjusting sizes, and image enhancement; structured data cleaning includes data type conversion and processing of duplicate data, missing values, and outliers.

4. The device according to claim 2, characterized in that, The construction unit is also used for: analyzing and extracting information from data based on the transformer fault knowledge graph schema according to historical data, automatically clustering information that cannot be accurately identified, supplementing and improving the transformer fault knowledge graph through semi-supervised learning, and improving the knowledge base of the transformer fault knowledge graph by analyzing point table information, historical alarm information, defect information, and monitoring data; perfecting the transformer fault knowledge graph through knowledge reasoning and graph evaluation.

5. The device according to claim 2, characterized in that, The single diagnosis unit is also used for: diagnosing the operation status of the transformer based on on-line monitoring data, using dissolved gas in oil, partial discharge, core grounding current, and oil temperature data respectively; realizing fault diagnosis of the transformer based on the transformer fault knowledge graph according to each monitored quantity.

6. The device according to claim 2, characterized in that, The joint diagnosis unit is also used for: determining the joint diagnosis result based on the weights of the results of each parameter according to the diagnosis results of single parameters; performing joint diagnosis of transformer faults by combining the monitoring data of multiple parameters.

Images