A method for constructing knowledge graph for transformer fault diagnosis

Abstract

The present invention discloses a method for constructing a transformer fault diagnosis knowledge graph, which belongs to the technical field of transformer fault diagnosis, and solves the problem that the transformer fault diagnosis knowledge graph in the prior art cannot adapt to the dynamic changes of relevant fault feature data over time, and does not consider the differences in applicable conditions of different fault diagnosis algorithms, resulting in inaccurate diagnosis results. The method includes: separately constructing the ontology layer and data layer of the transformer fault diagnosis knowledge graph; wherein the ontology layer is a class of the knowledge graph, which is composed of four subject domains, including: transformer, fault, maintenance and algorithm, and each subject domain includes entities, attributes and relationships; the data layer is an instance of the knowledge graph, which is used to store instance data of the ontology layer; collecting data related to transformer fault diagnosis, and according to the type of the collected data, selecting a matching processing method, and mapping the collected data to the corresponding subject domain of the data layer; so as to construct a transformer fault diagnosis knowledge graph.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer fault diagnosis, and in particular to a method for constructing a transformer fault diagnosis knowledge graph. Background Art

[0002] As a key equipment in the power system, the safe and stable operation of the transformer is crucial to the reliability of the entire power grid. Dissolved Gas Analysis (DGA) is one of the most widely used transformer fault diagnosis technologies. By monitoring the dissolved gas content in transformer oil, faults inside the transformer, such as partial discharge, overheating, and insulation aging, can be effectively detected. However, with the continuous expansion of the scale of power systems and the advancement of data acquisition technology, traditional transformer fault diagnosis methods have shown many limitations in processing large amounts of historical data and complex fault modes, as described below.

[0003] 1. Static diagnostic model: Most existing methods use static diagnostic models that fail to fully consider the dynamic changes and historical trends of equipment status. This makes it difficult for the system to identify complex patterns of progressive failures or multiple failures.

[0004] 2. Insufficient knowledge representation: Traditional methods often store diagnostic knowledge in the form of rules or simple models, lacking the ability to represent complex knowledge structures. This limits the system's description and analysis of multi-dimensional and multi-level fault characteristics.

[0005] 3. Algorithm applicability issues: Existing dissolved gas analysis algorithms in oil (such as Rogers ratio method, Duval triangle diagram method, etc.) are often designed for specific types of faults or specific operating conditions, lacking a comprehensive analysis and systematic representation of algorithm application scenarios. This makes it difficult to select the most suitable diagnostic algorithm in practical applications.

[0006] In response to these limitations, in recent years, the application of knowledge graph technology in the industrial field has provided new ideas for solving the above problems. As a semantic network, knowledge graph can represent the complex relationships between entities in a graph structure, providing a powerful tool for the representation and reasoning of complex knowledge. In the industrial field, the application of knowledge graph has shown great potential, as described below.

[0007] 1. Equipment health management: Researchers have begun to apply knowledge graphs to the health status assessment and predictive maintenance of industrial equipment. A knowledge graph-based rotating machinery fault diagnosis method is proposed. By constructing a knowledge graph containing equipment structure, failure mode and diagnosis rules, more accurate fault location and diagnosis are achieved.

[0008] 2. Knowledge reasoning and decision support: Reasoning technology based on knowledge graphs provides new possibilities for intelligent decision-making in industrial systems. A power system fault diagnosis method combining knowledge graphs and deep learning has been developed, which achieves accurate identification and interpretation of complex fault modes through graph neural networks.

[0009] 3. Cross-domain knowledge transfer: The semantic representation capability of knowledge graphs makes it possible to transfer knowledge between different fields or different types of equipment. The knowledge graph is used to realize the transfer of fault diagnosis knowledge from large transformers to distribution transformers, which significantly improves the diagnosis accuracy in small sample scenarios.

[0010] Although knowledge graphs have shown great potential in the industrial field, their application in transformer fault diagnosis is still in its early stages. How to effectively construct a transformer fault diagnosis knowledge graph and organically combine it with the traditional dissolved gas in oil analysis method is still an urgent problem to be solved. In particular, there is still a lack of systematic solutions for generating knowledge graphs that include full-time, full-volume, and dynamic equipment data, as well as diversified dissolved gas in oil analysis algorithms. Summary of the invention

[0011] In view of the above analysis, an embodiment of the present invention aims to provide a method for constructing a transformer fault diagnosis knowledge graph, so as to solve the problem in the prior art that the transformer fault diagnosis knowledge graph cannot adapt to the dynamic changes of relevant fault feature data over time and does not consider the differences in applicable conditions of different fault diagnosis algorithms, resulting in inaccurate diagnosis results.

[0012] The present invention discloses a method for constructing a transformer fault diagnosis knowledge graph, the method comprising:

[0013] The ontology layer and data layer of the transformer fault diagnosis knowledge graph are constructed separately; the ontology layer is the class of the knowledge graph, which consists of four subject domains, including: transformer, fault, maintenance and algorithm. Each subject domain includes entities, attributes and relationships; the data layer is an instance of the knowledge graph, which is used to store the instance data of the ontology layer;

[0014] Collect data related to transformer fault diagnosis, and select a matching processing method according to the type of the collected data to map the collected data to the corresponding subject domain of the data layer; so as to construct a transformer fault diagnosis knowledge graph; the data related to transformer fault diagnosis includes dynamic operation data, fault and maintenance data;

[0015] For dynamic operation data, fault and maintenance data, data extraction and time aggregation are performed, and the extraction results of dynamic operation data, fault and maintenance data are stored in the form of quadruple under the fault and maintenance subject domain in the data layer, and associated with the time node through relationship connection.

[0016] On the basis of the above scheme, the present invention also makes the following improvements:

[0017] Furthermore, for dynamic running data, execute:

[0018] The dynamic operation data of the transformer is collected periodically and preprocessed, and entities, attributes and relationships are extracted from the preprocessed dynamic operation data to obtain corresponding extraction results; after adding time nodes to each group of entities, attributes and relationships extracted from the dynamic operation data, they are stored in the form of quadruple under the fault and maintenance subject domain in the data layer; multiple quadruple groups corresponding to the dynamic operation data of the same time node are associated with the only time node through relationship lines.

[0019] Further, for fault and repair data, execute:

[0020] The fault and maintenance data of the fault and maintenance nodes are collected and preprocessed, and the entities, attributes and relationships of the preprocessed fault and maintenance data are extracted to obtain the corresponding extraction results; after adding a unique time node to each group of entities, attributes and relationships extracted from the fault and maintenance data, they are stored in the form of quadruple under the fault and maintenance subject domain in the data layer; multiple quadruple groups corresponding to the dynamic operation data, fault and maintenance data of the same time node are associated with the unique time node through relationship lines.

[0021] Furthermore, the four-tuple form is expressed as: entity-attribute-attribute value-time;

[0022] For the same time, the year, month, day, hour, minute and second are used to represent the time hierarchically, and the quarter and week are connected as separate nodes through relationships.

[0023] Furthermore, the data related to transformer fault diagnosis also includes a fault diagnosis algorithm;

[0024] For the fault diagnosis algorithm, entities, relationships and attributes are extracted based on expert experience knowledge, and the extraction results are stored under the algorithm subject domain in the data layer.

[0025] Further, for the fault diagnosis algorithm, execute:

[0026] The fault diagnosis algorithm is sequentially subjected to algorithm entity, algorithm feature description, algorithm integration and algorithm relationship modeling to realize entity, relationship and attribute extraction based on expert experience knowledge, and the algorithm extraction results are stored in the algorithm's subject domain in the data layer.

[0027] Further, the fault diagnosis algorithm is sequentially subjected to algorithm instantiation, algorithm feature description, algorithm integration and algorithm relationship modeling, and executed:

[0028] Create independent algorithm nodes for each fault diagnosis algorithm, identify the entities of the corresponding fault diagnosis algorithm by analyzing the data set and algorithm logic of the corresponding fault diagnosis algorithm, and add basic attributes to each algorithm node to realize the algorithm entity;

[0029] Describe the characteristics and applicable conditions of the fault diagnosis algorithm to achieve algorithm feature description; encapsulate, register, and map parameters of the fault diagnosis algorithm to achieve algorithm integration; identify the attributes of the corresponding fault diagnosis algorithm based on the algorithm feature description and algorithm integration, and describe the relationship between entities and between entities and attributes in the fault diagnosis algorithm;

[0030] Describe the relationship between different fault diagnosis algorithms, implement algorithm relationship modeling, and identify the relationship between corresponding fault diagnosis algorithms.

[0031] Furthermore, the data related to transformer fault diagnosis also includes static data;

[0032] For static data, preprocessing, entity, attribute and relationship extraction are performed in sequence, and the data extraction results are stored in the form of triples in the transformer subject domain in the data layer;

[0033] The triple form is expressed as: entity-attribute-attribute value.

[0034] Furthermore, the subject domain of transformer is used to describe the basic characteristics and operating status of the transformer;

[0035] The subject domain of faults is used to describe various fault types and characteristics that may occur in transformers;

[0036] The subject domain of maintenance is used to describe the maintenance and preventive measures for transformer failures;

[0037] The subject domain of algorithms is used to describe various analysis algorithms for transformer fault diagnosis.

[0038] Furthermore, each subject domain includes entities, attributes and relationships of the current subject domain; among them, under the transformer subject domain,

[0039] Entities include: transformers, components, attributes and states;

[0040] Attributes include: static attributes and dynamic attributes;

[0041] Relationships include: containment relationship, attribute relationship and status relationship;

[0042] Under the Fault subject area,

[0043] Entities include: fault type, fault symptoms and fault cause;

[0044] Attributes include: fault severity, frequency of occurrence, typical characteristics, and time attributes;

[0045] Relationships include: causing relationships, expressing relationships, and associating relationships;

[0046] Under the Overhaul subject area,

[0047] The entities include: maintenance activities, maintenance types, maintenance plans and maintenance tools;

[0048] Attributes include: maintenance cycle, maintenance time, maintenance cost and time attributes;

[0049] Relationships include: targeting relationships, using relationships, and preventing relationships;

[0050] Under the Algorithms subject area,

[0051] Entities include: diagnostic algorithms, algorithm parameters, and diagnostic results;

[0052] Attributes include: algorithm rules, applicable conditions and accuracy;

[0053] Relationships include: input relationships, output relationships, optimization relationships and combination relationships.

[0054] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0055] The method for constructing a transformer fault diagnosis knowledge graph provided by the present invention constructs the body layer and data layer of the transformer fault diagnosis knowledge graph, and introduces four subject domains of transformer, fault, maintenance and algorithm, thereby forming a complete, dynamic and closed-loop transformer fault diagnosis knowledge graph, realizing the unified and full life cycle management of transformer ledger, operation and maintenance data, providing full-time, full-volume and dynamic data for the fault diagnosis algorithm, and providing a data basis for accurate prediction. And by improving the specific process of the transformer fault diagnosis knowledge graph, it is convenient for relevant technical personnel to efficiently and accurately construct the transformer fault diagnosis knowledge graph, which can adapt to the dynamic changes of fault diagnosis data and integrate fault diagnosis algorithms, effectively improving the accuracy and response speed of transformer fault diagnosis.

[0056] In addition, in the present invention, a time dimension is introduced for dynamic operation data, and the dynamic operation data extraction results are associated with time nodes through relationship lines to form a four-tuple representation containing time, thereby realizing dynamic representation of dynamic operation data, thereby being able to fully describe the state and changes of the transformer throughout its life cycle, and effectively solving the problem in the prior art of lacking a transformer fault diagnosis knowledge graph that can adapt to the dynamic changes of fault diagnosis data.

[0057] In addition, by dividing the fault diagnosis algorithm into separate subject domains, various transformer fault diagnosis algorithms are comprehensively sorted out and knowledge assets are formed, and the logical relationship, applicable conditions, advantages and disadvantages of each algorithm are given, which facilitates operation and maintenance personnel to quickly and accurately apply the algorithm, and facilitates algorithm personnel to study and improve the algorithm. The algorithm connects historical data, operating data, faults, and maintenance in series, forming a business closed loop of historical data accumulation, fault diagnosis, and operation and maintenance.

[0058] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. In the entire drawings, the same reference symbols represent the same components;

[0060] Figure 1 A flowchart of a method for constructing a transformer fault diagnosis knowledge graph provided by an embodiment of the present invention;

[0061] Figure 2 A schematic diagram of a transformer fault diagnosis knowledge graph provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0063] A specific embodiment of the present invention discloses a method for constructing a transformer fault diagnosis knowledge graph, the flow chart of which is as follows: Figure 1 As shown, the following steps are included:

[0064] Step S1: Build the ontology layer and data layer of the transformer fault diagnosis knowledge graph respectively; the ontology layer is the class of the knowledge graph, which consists of four subject domains, including: transformer, fault, maintenance and algorithm; the data layer is an instance of the knowledge graph, which is used to store the instance data of the ontology layer.

[0065] Step S2: Collect data related to transformer fault diagnosis, and select a matching processing method according to the type of the collected data, map the collected data to the corresponding subject domain of the data layer, so as to construct a transformer fault diagnosis knowledge graph; the data related to transformer fault diagnosis includes dynamic operation data, fault and maintenance data, static data and fault diagnosis algorithm; wherein,

[0066] For dynamic operation data, fault and maintenance data, data extraction and time aggregation are performed, and the extraction results of dynamic operation data, fault and maintenance data are stored in the form of four-tuples under the fault and maintenance subject domain in the data layer, and are associated with time nodes through relationship lines;

[0067] For static data, preprocessing, entity, attribute and relationship extraction are performed in sequence, and the data extraction results are stored in the form of triples in the transformer subject domain in the data layer;

[0068] For the fault diagnosis algorithm, entities, relationships and attributes are extracted based on expert experience knowledge, and the extraction results are stored under the algorithm subject domain in the data layer.

[0069] With respect to step S1, in the transformer fault diagnosis knowledge graph constructed in this embodiment, the ontology layer defines multiple subject domains involved in transformer fault diagnosis, as well as entities, attributes and relationships under each subject domain. The data layer, according to the definition of the ontology layer, maps the preprocessed static data and dynamic operation data to the data layer, thereby realizing the construction of the transformer fault diagnosis knowledge graph.

[0070] The following is a detailed description of the entity layer and the data layer in this embodiment.

[0071] (1) Ontology layer design

[0072] The ontology layer is the skeleton of the transformer fault diagnosis knowledge graph, which defines the subject domains involved in transformer fault diagnosis, as well as the entities, attributes and relationships under each subject domain. Preferably, in this embodiment, starting from the four dimensions in the transformer fault diagnosis process, the ontology layer is divided into four subject domains: transformer, fault, maintenance and algorithm.

[0073] The advantages of dividing the subject domains in this way are as follows: a) Comprehensive coverage: The four subject domains cover all the core elements involved in transformer fault diagnosis. The transformer is the object of fault diagnosis, the fault is the diagnostic target, the maintenance is the application of the diagnostic results, and the algorithm is the method of fault diagnosis; b) Clear boundaries: This division makes each subject domain have relatively independent concepts and attributes, which is convenient for the organization and management of knowledge; c) Practicality: This division method directly corresponds to the actual process of fault diagnosis, from equipment to problem identification, to solutions and diagnostic methods; d) Scalability: Each subject domain can be expanded independently. For example, new diagnostic algorithms can be added to the algorithm domain without affecting other domains. Below, the four subject domains are explained in detail.

[0074] 1) Transformer subject domain

[0075] The transformer subject domain is used to describe the basic characteristics and operating status of the transformer.

[0076] a) Entity:

[0077] Transformer: As the central entity, it represents the specific transformer equipment;

[0078] Components: such as windings, core, oil tank, etc., describing the main components of the transformer;

[0079] Attributes: including static parameters such as rated voltage, rated current, capacity, etc.

[0080] Status: describes the operating status of the transformer, such as normal, overload, fault, etc.

[0081] b) Key attributes:

[0082] Static attributes: model, production date, rated capacity, rated voltage, etc.;

[0083] Dynamic properties: oil temperature, load current, dissolved gas content in oil, etc.

[0084] c) Important relationships:

[0085] Containment relationship: The transformer contains multiple components;

[0086] Attribute relations: Transformers have multiple attributes;

[0087] State relationship: The transformer is in a certain operating state.

[0088] 2) Fault subject domain

[0089] The fault subject domain is used to describe the various fault types and characteristics that may occur in transformers.

[0090] a) Entity:

[0091] Fault type: such as partial discharge, overheating, insulation aging, etc.;

[0092] Fault symptoms: describe the external manifestations of the fault;

[0093] Fault cause: The root cause of the fault.

[0094] b) Key attributes:

[0095] Fault severity: such as minor, moderate, or severe;

[0096] Frequency of occurrence: describes how common a certain type of failure is;

[0097] Typical characteristics: such as specific gas ratio, temperature anomaly, etc.;

[0098] Time attribute: the time when the fault occurred.

[0099] c) Important relationships:

[0100] Cause relationship: a certain cause leads to a specific type of failure;

[0101] Manifestation relationship: the fault is manifested through certain symptoms;

[0102] Correlation: The correlation or mutual exclusivity between different fault types.

[0103] 3) Overhaul the subject domain

[0104] The maintenance subject domain is used to describe the maintenance and preventive measures for transformer failures.

[0105] a) Entity:

[0106] Maintenance activities: specific maintenance operations;

[0107] Maintenance type: maintenance targets specific fault types;

[0108] Maintenance plan: a systematic solution for specific fault types;

[0109] Service Tools: Equipment and tools needed to perform service.

[0110] b) Key attributes:

[0111] Maintenance cycle: the time interval for regular maintenance;

[0112] Maintenance time: the time required to complete the maintenance;

[0113] Maintenance cost: including labor, material and other costs;

[0114] Time attribute: maintenance time.

[0115] c) Important relationships:

[0116] Targeted relationship: the specific fault type targeted by the repair plan;

[0117] Usage relationship: Maintenance activities use specific maintenance tools;

[0118] Prevention relationship: preventable failure types that maintenance activities can prevent.

[0119] 4) Algorithm subject domain

[0120] The algorithm subject domain is used to describe various analysis algorithms for transformer fault diagnosis.

[0121] a) Entity:

[0122] Diagnostic algorithms: such as Rogers ratio method, Duval triangle diagram method, etc.

[0123] Algorithm parameters: input parameters required for the algorithm to run;

[0124] Diagnostic result: fault diagnosis conclusion output by the algorithm.

[0125] b) Key attributes:

[0126] Algorithm type: such as rule-based, machine learning, deep learning, etc.;

[0127] Applicable conditions: the data types and scenarios to which the algorithm is applicable;

[0128] Accuracy: The diagnostic accuracy of the algorithm under specific conditions.

[0129] c) Important relationships:

[0130] Input relation: The algorithm accepts a specific type of data as input;

[0131] Output relations: The algorithm produces a specific type of diagnostic result;

[0132] Optimization relationship: describes the improvement or optimization relationship between algorithms;

[0133] Combination relationship: An integrated algorithm is composed of multiple basic algorithms.

[0134] (2) Data layer

[0135] The data layer is an instance of the transformer fault diagnosis knowledge graph, which contains instances of actual data defined based on the ontology layer. The ontology layer and the data layer are in the relationship of class and instance. The construction process of the data layer is the process of extracting entities and attributes and data from the pre-processed static data and dynamic data, and mapping them to the data layer of transformer fault diagnosis. In addition, the relationship between the ontology layer and the data layer can also be understood as the relationship between "template" and "instance". The ontology layer defines the structure and organization of knowledge, while the data layer fills in specific data based on this structure. In the specific implementation process, the corresponding data can be stored in the data layer based on the collected data related to transformer fault diagnosis, so as to construct and update the transformer fault diagnosis knowledge graph. The specific process is described as follows:

[0136] (1) Static data

[0137] Since static data is relatively fixed, in the early stage of constructing the transformer fault diagnosis knowledge graph, the static data of the transformer is collected, and preprocessing, entity, attribute and relationship extraction are performed in sequence, and the data extraction results are stored in the form of triples in the transformer subject domain in the data layer. Preferably, in this embodiment, the static data of the transformer mainly includes the equipment hardware parameter data and equipment usage record data of the transformer. Among them, the equipment hardware parameter data generally does not change over time, while the frequency of change of the equipment usage record parameters is relatively low. In this embodiment, they are all regarded as static data, which is very important for transformer fault diagnosis and equipment management.

[0138] The device hardware parameter data is mainly collected through the following methods:

[0139] 1) Equipment ledger import: Import the following equipment hardware parameter data of the transformer from the power company's asset management system: the transformer's unique identification code, model, manufacturer, production date, rated capacity, rated voltage (high voltage side and low voltage side), rated current and cooling method, etc.;

[0140] 2) Technical document scanning: Scan and perform optical character recognition (OCR) on the transformer's technical manuals, design drawings and other documents to extract the following equipment hardware parameter data of the transformer: insulation level, impedance voltage and connection group, etc.

[0141] The collection of equipment usage record data is mainly carried out in the following ways:

[0142] 3) Manual data entry: For some equipment usage record data that is difficult to obtain automatically, such as the physical location, installation date, expected service life, etc. of the transformer, a special data entry interface can be set up for manual input by professionals;

[0143] 4) Historical record compilation: Collect and organize equipment usage record data including historical maintenance records, upgrade and renovation information, previous fault conditions, etc. of the transformer.

[0144] It should be noted that although the device usage record information is not strictly static data, it can be used as important background information considering that its frequency of change is low. Therefore, in this embodiment, the device usage record information is used as static data.

[0145] Based on the above description, it can be seen that in this embodiment, the static data acquisition unit comprehensively collects the equipment hardware parameter data and equipment usage record data of the transformer through a combination of automation and manual methods, providing solid data support for the operation management of the transformer.

[0146] In this embodiment, the static data of the transformer is preprocessed, including: data cleaning, data standardization, data structuring, data verification, metadata management and data integration. Data cleaning includes duplicate data processing, missing value processing and correction of erroneous values. Data standardization includes unit unification, naming standardization, date format unification, etc. Data structuring includes attribute decomposition and hierarchical relationship establishment. Data verification includes parameter range check, logical consistency check, time consistency check, etc. Metadata management includes data source record, data quality mark, update history tracking, etc. Data integration includes multi-source data integration and conflict resolution, etc.

[0147] It should be noted that by completing the data collection and preprocessing of static data, high-quality, standardized preprocessed static data can be obtained. These preprocessed static data are easier to identify and extract, laying a solid foundation for the subsequent knowledge graph construction.

[0148] (2) Dynamic operation data, fault and maintenance data

[0149] Since the dynamic operation data changes with time, the dynamic operation data of the transformer can be periodically collected and preprocessed, and the entities, attributes and relationships of the preprocessed dynamic operation data can be extracted to obtain the corresponding extraction results; after adding a time node to each group of entities, attributes and relationships extracted from the dynamic operation data, they are stored in the form of a quadruple under the fault and maintenance subject domain in the data layer; multiple quadruple groups corresponding to the dynamic operation data at the same time node are associated with the unique time node through relationship lines. In addition, for fault and maintenance data, the fault and maintenance data of the fault and maintenance node are collected and preprocessed, and the entities, attributes and relationships of the preprocessed fault and maintenance data are extracted to obtain the corresponding extraction results; after adding a unique time node to each group of entities, attributes and relationships extracted from the fault and maintenance data, they are stored in the form of a quadruple under the fault and maintenance subject domain in the data layer; multiple quadruple groups corresponding to the dynamic operation data, fault and maintenance data at the same time node are associated with the unique time node through relationship lines.

[0150] In this embodiment, the dynamic operation data of the transformer mainly includes the dissolved gas content in the transformer oil, real-time operation parameters, offline monitoring parameters, operation events related to the transformer, and environmental data. Specifically, the collection of dynamic operation data mainly includes the following aspects.

[0151] 1) Online monitoring system: By installing a special online monitoring system, the dissolved gas content in the oil is collected in real time, usually including the concentration of hydrogen (H2), methane (CH4), acetylene (C2H2), ethylene (C2H4), ethane (C2H6), carbon monoxide (CO) and carbon dioxide (CO2). The sampling frequency of the dissolved gas content in the oil can be set according to the needs, usually once an hour or once a day.

[0152] 2) SCADA system integration: Integrate with the substation's supervisory control and data acquisition (SCADA) system to obtain the transformer's real-time operating parameters, such as load current, oil temperature, winding temperature, tap position, etc. The transformer's real-time operating parameters are usually collected at a high frequency (e.g., every minute).

[0153] 3) Regular offline testing: For some offline monitoring parameters that cannot be monitored online in real time, such as insulation resistance, dielectric loss factor, winding DC resistance, etc., the corresponding data can be obtained through regular (such as annually or semi-annually) offline testing. After obtaining these offline monitoring parameters, they need to be entered into the system in a timely manner.

[0154] 4) Environmental data collection: Install environmental monitoring equipment to collect environmental data around the transformer, such as ambient temperature, humidity, and atmospheric pressure, etc. These factors may affect the operating status and life of the transformer.

[0155] 5) Event recording: Through the interface with the substation automation system, real-time recording of transformer-related operation events, such as switch operation, protection action, etc. These event data are crucial for analyzing the cause and development process of the fault.

[0156] Fault and maintenance data mainly include fault and maintenance reports.

[0157] Based on the above description, it can be seen that since the dynamic operation data, fault and maintenance data of the transformer change at any time, it is necessary to set the corresponding sampling frequency according to the change of the dynamic operation data of different transformers, and periodically collect the corresponding dynamic operation data of the transformer. The collected dynamic operation data, fault and maintenance data of the transformer are preprocessed.

[0158] In this embodiment, the dynamic operation data, fault and maintenance data of the transformer are preprocessed, including: data cleaning, data standardization and abnormal value processing.

[0159] It should be noted that by complete data collection and preprocessing of dynamic operation data, high-quality, standardized preprocessed dynamic operation data can be obtained. These preprocessed dynamic operation data are easier to identify, extract, and map to the entities, attributes, and relationships of the transformer fault diagnosis knowledge graph, laying a solid foundation for the subsequent knowledge graph construction.

[0160] For the pre-processed static data, dynamic operation data, fault and maintenance data, entities, attributes and relationships can be extracted in the following way to obtain the corresponding data extraction results. According to the actual situation, the extraction methods such as direct extraction, rule extraction and neural network extraction can be adopted to realize the extraction of entities, attributes and relationships. After the data is extracted, it can be represented by knowledge, that is, the extracted entities, attributes and relationships are converted into the graph structure in the knowledge graph to obtain the corresponding extraction results. The knowledge representation of static data is carried out in the form of triples, and the entities, attributes and relationships are converted into the graph structure in the knowledge graph to obtain the corresponding extraction results. The knowledge representation of dynamic operation data, fault and maintenance data is carried out in the form of quadruples, and the entities, attributes and relationships are converted into the graph structure in the knowledge graph.

[0161] Specifically, for static data, triples are used, which are represented as: entity-attribute-attribute value. Specific form: (subject, predicate, object). Examples: (transformer A, model, ONAN-500 / 220), (transformer A, production date, 2020-01-15), (transformer A, rated capacity, 500MVA), (transformer A, high-voltage side rated voltage, 500kV), (transformer A, low-voltage side rated voltage, 220kV). Exemplarily, a property graph database (such as Neo4j) can be used for storage. Each triple is represented as a node (subject) connected to another node or attribute value (object) through a directed edge (predicate).

[0162] Specifically, for dynamic operation data, four-tuple representation is used to effectively capture the timing characteristics. Four-tuple is used to represent dynamic operation data that changes over time, namely: entity-attribute-attribute value-time. Specific form: (subject, predicate, object, time). Examples: (transformer A, oil temperature, 75℃, 2023-06-01 10:00:00), (transformer A, load current, 1000A, 2023-06-01 10:00:00), (transformer A, H2 content, 150ppm, 2023-06-01 10:00:00), (transformer A, CH4 content, 80ppm, 2023-06-01 10:00:00), (transformer A, C2H2 content, 5ppm, 2023-06-01 10:00:00). During the specific implementation process, in the graph database corresponding to the knowledge graph, entities, attributes, attribute values, and time are all represented by nodes, and attributes are represented by relationships / edges. At the same time, in the data layer, there is only one time node at the same time, that is, the attribute values ​​of different entities at the same time are associated with the same time node. This embodiment uses dynamic data mapping and special processing procedures stored in the knowledge graph to reflect the improvements made in the process of storing dynamic data in the knowledge graph that is difficult to achieve in the prior art. (1) Uniqueness of time node: that is, the same time is only saved once in the knowledge graph, and dynamic data is associated with the time node through relationship lines. This method can avoid a large amount of redundancy of time nodes and accelerate the time retrieval speed. (2) Granularity relationship of time: that is, the same time is hierarchically represented by year, month, day, hour, minute, and second, and quarters and weeks are connected as separate nodes through relationship lines and associated with specific months and days. In this way, the time retrieval speed is accelerated, and it is convenient to count and reduce according to different time granularities.

[0163] Specifically, for different subject domains, the corresponding mapping relationships are described as follows.

[0164] 1) Transformer subject domain

[0165] Ontology layer definition: defines the entities, attributes and relationships of transformers.

[0166] Data layer implementation: a) Entity and attribute extraction: Extract specific transformer instances, such as "Transformer A", from the equipment ledger and parameter table. b) Attribute filling: Fill static attributes (such as model, rated capacity) and dynamic attributes (such as oil temperature, load current) with actual data. c) Relationship establishment: Create specific associations between transformer instances and their components, parameters, and states.

[0167] 2) Fault subject domain

[0168] Ontology layer definition: conceptual structure describing fault types, symptoms and causes.

[0169] Data layer implementation: a) Entity and attribute extraction: Extract specific fault instances from historical fault records and diagnostic reports. b) Attribute filling: Add specific attribute values ​​such as severity and frequency of occurrence to each fault instance. c) Relationship establishment: Associate fault instances with specific transformer instances, symptoms, and causes.

[0170] 3) Overhaul the subject domain

[0171] Ontology layer definition: Define the conceptual framework of maintenance activities, solutions and tools.

[0172] Data layer implementation: a) Entity and attribute extraction: Extract specific maintenance activity instances from maintenance records and work orders. b) Attribute filling: Add actual maintenance cycle, time consumption, cost and other data. c) Relationship establishment: Associate maintenance activities with specific transformer instances, tools used, and targeted faults.

[0173] 4) Algorithm subject domain

[0174] Ontology layer definition: conceptual structure describing diagnostic algorithms, parameters and results.

[0175] Data layer implementation: a) Entity and attribute extraction: Take the specific algorithm implementation (such as "Rogers ratio method") as an entity. b) Attribute filling: Add the actual accuracy of the algorithm, applicable conditions and other attributes. c) Relationship establishment: Associate the algorithm instance with its input data, output results, and applicable fault types.

[0176] Through the above detailed knowledge graph construction process, the complex knowledge system in the field of transformer fault diagnosis can be transformed into a structured graphical representation. This representation method can not only effectively organize and store a large amount of heterogeneous data, but also provide a strong semantic basis for subsequent intelligent diagnosis and decision support. The design of the ontology layer ensures the consistency and scalability of the knowledge system, while the construction process of the data layer realizes the conversion from raw data to structured knowledge. This representation method combining static triples and dynamic quadruples is particularly suitable for processing time-varying data during transformer operation, providing strong support for real-time fault diagnosis and predictive maintenance.

[0177] In addition, for fault diagnosis algorithms, they can be mapped and stored under the algorithm subject domain in the following way: Algorithms are represented as entities in the knowledge graph, and their characteristics, applicable conditions, and performance indicators are encoded into the graph structure to facilitate subsequent intelligent selection and application.

[0178] (1) Algorithm materialization

[0179] An independent algorithm node is created for each fault diagnosis algorithm. By analyzing the data set and algorithm logic of the corresponding fault diagnosis algorithm, the entity of the corresponding fault diagnosis algorithm is identified, and basic attributes are added to each algorithm node to realize the algorithm entity.

[0180] Taking the dissolved gas analysis algorithm as an example, various dissolved gas analysis algorithms in oil are represented as entities in the knowledge graph:

[0181] a) Algorithm node creation:

[0182] Create a separate node for each analysis algorithm (fault diagnosis algorithm), such as "Rogers Ratio Method", "Duval Triangle Plot Method", "IEC Ratio Method", etc.

[0183] Add basic properties to each algorithm node, such as algorithm name, developer, publication year, etc.

[0184] b) Algorithm classification:

[0185] Create algorithm category nodes, such as "Gas Ratio Method", "Graph Method", "Artificial Intelligence Method", etc.

[0186] Establish a "belongs to" relationship between the algorithm node and the corresponding category node to implement classification management of the algorithm.

[0187] c) Algorithm version control:

[0188] For algorithms with multiple versions, create version subnodes, such as "Rogers Ratio Method_1975 Version" and "Rogers Ratio Method_1978 Version".

[0189] The main algorithm node and the version sub-node are connected through the "version" relationship to facilitate tracking the evolution history of the algorithm.

[0190] (2) Algorithm feature description

[0191] In order to achieve intelligent algorithm selection, it is necessary to describe in detail the characteristics and applicable conditions of each algorithm:

[0192] a) Input parameter definition:

[0193] Create an "Input Parameter" node to describe the gas data required by the algorithm, such as H2, CH4, C2H2, etc.

[0194] Connect the algorithm node to the input parameter node through a "requires" relationship.

[0195] b) Description of applicable conditions:

[0196] Define the applicable conditions of the algorithm, such as gas content range, transformer type, operating environment, etc.

[0197] Create an Applicable to node and connect it to the algorithm node through the Applicable to relationship.

[0198] c) Performance index records:

[0199] Add performance metrics of the algorithm, such as accuracy, sensitivity, specificity, etc.

[0200] These indicators can be derived based on historical data statistics and stored as attributes of algorithm nodes.

[0201] d) Diagnostic capability means:

[0202] Create "Fault Type" nodes, such as "Partial Discharge", "Overheating", "Arc Discharge", etc.

[0203] An algorithm node is connected to the fault type nodes that it can diagnose through a "diagnosable" relationship.

[0204] (3) Algorithm integration

[0205] Integrate the specific implementation of the algorithm with the knowledge graph to support real-time diagnosis:

[0206] a) Algorithm encapsulation:

[0207] Encapsulate each analysis algorithm as a separate function or class.

[0208] Implement a unified interface, including input data format, output result format, etc.

[0209] b) Algorithm registration:

[0210] Add an "implementation path" attribute to each algorithm node in the knowledge graph to point to the specific implementation of the algorithm.

[0211] Design an algorithm registration mechanism to allow new algorithm implementations to be added dynamically.

[0212] c) Parameter mapping:

[0213] Establish a mapping relationship between the input parameter nodes in the knowledge graph and the actual algorithm input.

[0214] Implement data preprocessing function to automatically convert input data format according to the mapping relationship.

[0215] Based on the above algorithm feature description and algorithm integration, the attributes of the corresponding fault diagnosis algorithm can be identified, and the relationship between entities and between entities and attributes in the fault diagnosis algorithm can be described.

[0216] (4) Algorithmic relationship modeling

[0217] Describing the relationship between different algorithms helps to understand the algorithm system and choose the appropriate algorithm combination:

[0218] a) Inheritance relationship:

[0219] Use the "inherits from" relationship to indicate the improvement and derivation of an algorithm, such as "Improved Duval Triangle Method" inherits from "Duval Triangle Method".

[0220] b) Complementary relationship:

[0221] Algorithms that complement each other in diagnostic capabilities are connected through "complementary" relationships, such as the combination of ratio and profiling methods.

[0222] c) Integration relationship:

[0223] For a composite algorithm that integrates multiple basic algorithms, the "integration" relationship is used to represent its composition.

[0224] d) Contrast relationship:

[0225] Record the comparison results of different algorithms in specific scenarios as a reference for algorithm selection.

[0226] Thus, the relationship of the corresponding fault diagnosis algorithm is identified.

[0227] Through the integration process of the above-mentioned dissolved gas analysis algorithms, a comprehensive, flexible and scalable algorithm management framework can be realized. This framework not only integrates various traditional and modern dissolved gas analysis algorithms in oil into the knowledge graph, but also establishes the association between algorithms, providing a powerful analysis tool library for the intelligent diagnosis system. The integrated dissolved gas analysis algorithm is the core tool for subsequent knowledge reasoning and fault diagnosis.

[0228] The schematic diagram of the transformer fault diagnosis knowledge graph is as follows: Figure 2 As shown. Based on the transformer fault diagnosis knowledge graph established in this embodiment, knowledge reasoning and fault diagnosis can be performed. An example is given below.

[0229] a) Dissolved gas analysis algorithm application

[0230] Algorithm selection: Based on the current transformer operating status, historical data, and fault characteristics, the most suitable dissolved gas analysis algorithm or algorithm combination is selected from the knowledge graph.

[0231] Multi-algorithm fusion: Run multiple selected algorithms simultaneously, such as Rogers ratio method, Duval triangle diagram method and IEC ratio method, and comprehensively compare their diagnostic results.

[0232] Result consistency analysis: Evaluate the consistency of diagnostic results of different algorithms to improve the credibility of diagnosis.

[0233] Dynamic weight allocation: Dynamically adjust the weights of different algorithms based on historical diagnostic accuracy and current scenarios.

[0234] b) Feature extraction and pattern matching

[0235] Extract key features from raw data, such as the rate of change and trend of gas content.

[0236] The extracted features are matched with the predefined fault modes in the knowledge graph, and a comprehensive judgment is made based on the results of the dissolved gas analysis algorithm.

[0237] c) Anomaly Detection

[0238] Implement a knowledge graph-based anomaly detection mechanism to identify abnormal patterns that deviate from normal operating conditions.

[0239] Combining expert-defined threshold rules with the output of dissolved gas analysis algorithms improves the accuracy and interpretability of anomaly detection.

[0240] The main innovations of this embodiment are as follows:

[0241] (1) Innovative data representation methods:

[0242] Static data is represented using traditional triples (entity-attribute-attribute value).

[0243] Dynamic operation data is represented by a quaternary (entity-attribute-attribute value-time) to effectively capture the time series characteristics. The dynamic operation data that changes over time is represented by a quaternary, namely: entity-attribute-attribute value-time. And it is equipped with a corresponding knowledge storage method.

[0244] (2) Integration of dissolved gas analysis algorithms in oil:

[0245] Encapsulate various analysis algorithms (such as Rogers ratio method, Duval triangle graph method, etc.) as algorithm entities in the knowledge graph.

[0246] Define the algorithm's applicable conditions, evaluation indicators, inheritance relationships and other properties.

[0247] Implement a scenario-based automatic algorithm selection mechanism.

[0248] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0249] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for constructing a transformer fault diagnosis knowledge graph, characterized in that: The method comprises: The ontology layer and data layer of the transformer fault diagnosis knowledge graph are constructed separately; the ontology layer is the class of the knowledge graph, which consists of four subject domains, including: transformer, fault, maintenance and algorithm. Each subject domain includes entities, attributes and relationships; the data layer is an instance of the knowledge graph, which is used to store the instance data of the ontology layer; Collect data related to transformer fault diagnosis, and select a matching processing method according to the type of the collected data to map the collected data to the corresponding subject domain of the data layer; so as to construct a transformer fault diagnosis knowledge graph; the data related to transformer fault diagnosis includes dynamic operation data, fault and maintenance data; For dynamic operation data, fault and maintenance data, data extraction and time aggregation are performed, and the extraction results of dynamic operation data, fault and maintenance data are stored in the form of four-tuples under the fault and maintenance subject domain in the data layer, and are associated with time nodes through relationship lines; the four-tuple form is expressed as: entity-attribute-attribute value-time; for the same time, the year, month, day, hour, minute, and second are used for hierarchical representation, and the quarter and week are used as separate nodes through relationship lines; The data related to transformer fault diagnosis also includes a fault diagnosis algorithm; for the fault diagnosis algorithm, the following steps are performed: algorithm instantiation, algorithm feature description, algorithm integration and algorithm relationship modeling are performed on the fault diagnosis algorithm in sequence to realize entity, relationship and attribute extraction based on expert experience knowledge, and the extraction results of the algorithm are stored under the subject domain of the algorithm in the data layer.

2. The method for constructing a transformer fault diagnosis knowledge graph according to claim 1, characterized in that: For dynamic running data, execute: The dynamic operation data of the transformer is collected periodically and preprocessed, and entities, attributes and relationships are extracted from the preprocessed dynamic operation data to obtain corresponding extraction results; after adding time nodes to each group of entities, attributes and relationships extracted from the dynamic operation data, they are stored in the form of quadruple under the fault and maintenance subject domain in the data layer; multiple quadruple groups corresponding to the dynamic operation data of the same time node are associated with the only time node through relationship lines.

3. The method for constructing a transformer fault diagnosis knowledge graph according to claim 2, characterized in that: For fault and repair data, execute: The fault and maintenance data of the fault and maintenance nodes are collected and preprocessed, and the entities, attributes and relationships of the preprocessed fault and maintenance data are extracted to obtain the corresponding extraction results; after adding a unique time node to each group of entities, attributes and relationships extracted from the fault and maintenance data, they are stored in the form of quadruple under the fault and maintenance subject domain in the data layer; multiple quadruple groups corresponding to the dynamic operation data, fault and maintenance data of the same time node are associated with the unique time node through relationship lines.

4. The method for constructing a transformer fault diagnosis knowledge graph according to any one of claims 1 to 3, characterized in that: The fault diagnosis algorithm is sequentially subjected to algorithm instantiation, algorithm feature description, algorithm integration and algorithm relationship modeling, and executed: Create independent algorithm nodes for each fault diagnosis algorithm, identify the entities of the corresponding fault diagnosis algorithm by analyzing the data set and algorithm logic of the corresponding fault diagnosis algorithm, and add basic attributes to each algorithm node to realize the algorithm entity; Describe the characteristics and applicable conditions of the fault diagnosis algorithm and implement the algorithm feature description; Encapsulate, register, and map parameters of fault diagnosis algorithms to achieve algorithm integration; identify the attributes of the corresponding fault diagnosis algorithms based on algorithm feature description and algorithm integration, and describe the relationships between entities and between entities and attributes in the fault diagnosis algorithms; Describe the relationship between different fault diagnosis algorithms, implement algorithm relationship modeling, and identify the relationship between corresponding fault diagnosis algorithms.

5. The method for constructing a transformer fault diagnosis knowledge graph according to claim 4, characterized in that: The data related to transformer fault diagnosis also includes static data; For static data, preprocessing, entity, attribute and relationship extraction are performed in sequence, and the data extraction results are stored in the form of triples in the transformer subject domain in the data layer; The triple form is expressed as: entity-attribute-attribute value.

6. The method for constructing a transformer fault diagnosis knowledge graph according to claim 5, characterized in that: The transformer subject domain is used to describe the basic characteristics and operating status of the transformer; The subject domain of faults is used to describe various fault types and characteristics that may occur in transformers; The subject domain of maintenance is used to describe the maintenance and preventive measures for transformer failures; The subject domain of algorithms is used to describe various analysis algorithms for transformer fault diagnosis.

7. The method for constructing a transformer fault diagnosis knowledge graph according to claim 6, characterized in that: Each subject domain includes entities, attributes and relationships of the current subject domain; among them, in the transformer subject domain, Entities include: transformers, components, attributes and states; Attributes include: static attributes and dynamic attributes; Relationships include: containment relationship, attribute relationship and status relationship; Under the Fault subject area, Entities include: fault type, fault symptoms and fault cause; Attributes include: fault severity, frequency of occurrence, typical characteristics, and time attributes; Relationships include: causing relationships, expressing relationships, and associating relationships; Under the Overhaul subject area, The entities include: maintenance activities, maintenance types, maintenance plans and maintenance tools; Attributes include: maintenance cycle, maintenance time, maintenance cost and time attributes; Relationships include: targeting relationships, using relationships, and preventing relationships; Under the Algorithms subject area, Entities include: diagnostic algorithms, algorithm parameters, and diagnostic results; Attributes include: algorithm rules, applicable conditions and accuracy; Relationships include: input relationships, output relationships, optimization relationships and combination relationships.

Images