Knowledge graph construction method for multi-source heterogeneous data of power transformation main equipment and related system

Abstract

The invention relates to the technical field of data processing, in particular to a knowledge graph construction method for multi-source heterogeneous data of power transformation main equipment and a related system.The method comprises the steps that multi-source heterogeneous data collection is carried out, and for structured database data, a database connection tool is adopted, and needed data is obtained by writing SQL query statements; analyzing the semi-structured data by using a corresponding analysis library to extract key information, and converting the key information into a structured data format; for unstructured text data and webpage data, a natural language processing technology and a web crawler technology are adopted; according to the knowledge graph construction method for the multi-source heterogeneous data of the power transformation main equipment, the data utilization rate is increased, and the data can be enhanced and supplemented by introducing a data enhancement algorithm in secondary data processing.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and specifically to a knowledge graph construction method and a related system for multi-source heterogeneous data of a main substation equipment. Background Art

[0002] A knowledge graph is a graphical structure used to represent knowledge, in which entities and relationships are represented as nodes and edges in the graph. The process of constructing a knowledge graph includes extracting entities and relationships from structured and unstructured data and organizing them into a meaningful graph. The construction of a knowledge graph for multi-source heterogeneous data of substation main equipment refers to the process of constructing a knowledge graph for data from different sources and with different structures related to the substation main equipment. By integrating multi-source heterogeneous data, a comprehensive understanding and unified management of substation equipment can be achieved, and the efficiency and accuracy of equipment operation and maintenance can be improved. The existing knowledge graph construction method for multi-source heterogeneous data of substation main equipment can only perform simple data cleaning, denoising, and other normalization operations based on multi-source heterogeneous data. In the case of poor data integrity, the constructed knowledge graph is not comprehensive enough. Summary of the invention

[0003] The purpose of the present invention is to provide a knowledge graph construction method and related system for multi-source heterogeneous data of substation main equipment to solve the problems raised in the above background technology.

[0004] To achieve the above object, the present invention provides the following technical solution: a method for constructing a knowledge graph of multi-source heterogeneous data of a substation main device, the method comprising the following steps: S1. Multi-source heterogeneous data collection: For structured database data, database connection tools are used to obtain required data by writing SQL query statements; for semi-structured data, corresponding parsing libraries are used to parse and extract key information, and then convert it into structured data format; for unstructured text data and web page data, natural language processing technology and web crawler technology are used. The web crawler uses the Scrapy framework to collect data according to the set target URL and data extraction rules. After unstructured text data is collected, it is stored in local files or databases and waits for subsequent processing; S2. Primary data processing, which includes structured data processing, semi-structured data processing and unstructured data processing; Structured data processing: perform data cleaning, remove duplicate records, compare key fields, identify and delete duplicate record rows by writing database query statements or using data processing tools; correct incorrect formats, and convert date formats that are not uniform to standard formats; for abnormal values ​​in numerical data, such as temperature values ​​that are obviously beyond the normal range, filter and correct them by setting reasonable threshold ranges; Data integration integrates relevant structured data obtained from different database tables or systems, associates and merges basic equipment information in the equipment management system with operating data in the monitoring system through key fields such as equipment number, and forms a data set containing various aspects of equipment information; Semi-structured data processing: format conversion. For maintenance reports in XML format, parse the XML node content, extract key information, and convert it into a structured table format; for device configuration files in JSON format, parse the key-value pair information, extract device parameter configuration information, and convert it into a structured format that is easy to process; Data normalization: normalize the extracted semi-structured data content; Unstructured data processing: text cleaning, removing punctuation marks, special characters, and meaningless character noise, using Chinese word segmentation tools to segment text, splitting continuous text into word sequences, and removing common stop words that have no practical meaning; Text annotation: For inspection records and fault description text data, manual annotation or supervised machine learning algorithm annotation is performed to identify the key information of equipment entities and status descriptions; S3, secondary data processing, introduces data enhancement algorithm, sets generator and discriminator. In multi-source heterogeneous data processing, the generator is used to generate new data similar to real data, and the discriminator is used to distinguish real data from generated data. Through the confrontation between the two, the generator continuously improves the quality of generated data, thereby achieving data enhancement and supplementing data sources with less data volume; S4, extract knowledge from the data obtained in S3, including entity extraction, relationship extraction and attribute extraction. Entity extraction identifies entities by formulating rule templates; relationship extraction builds a neural network model, using convolutional neural network CNN or recurrent neural network RNN ​​and its variants to build a relationship extraction model. The model input is a sentence text vector representation containing two entities. After the network layer learns the sentence semantic features, it outputs the relationship type result between entities; attribute extraction is divided into extraction from structured data and extraction from unstructured data. When extracting from structured data, the inherent attribute information of the device is directly obtained from the database table field. For the attribute information clearly marked in the semi-structured data, it is extracted according to the format analysis. When extracting from unstructured data, define the attribute extraction rule template. When the text meets the template rules, extract the corresponding attribute information. At the same time, combine natural language processing technology to perform semantic analysis to improve the accuracy of attribute extraction; S5. Perform knowledge fusion, wherein the knowledge fusion includes ontology construction, entity alignment, and conflict detection and resolution; S6. Formation of knowledge graph: storing the extracted and fused knowledge in a graph database, storing the attribute information of entities as nodes, and storing the attributes and direction information of relationships as edges to form a knowledge graph. The graph database includes Neo4j and JanusGraph. S7, optimize the highlight processing, through the optimization algorithm to assign different attention weights to the nodes according to the characteristics of the knowledge graph nodes and the information of the neighboring nodes. In the knowledge graph of substation equipment, some key equipment nodes and nodes with important connection relationships have an important impact on the operation and fault propagation of the entire system. These important nodes are identified through the algorithm and given higher attention weights; S8, application implementation, based on the knowledge graph after further optimization and highlighting in S7, equipment fault diagnosis, analysis, and operation and maintenance strategy optimization are realized.

[0005] Preferably, the data enhancement algorithm described in S3 is specifically as follows: Generator G receives random noise z and some substation equipment data features As input, generate new data samples , the discriminator D receives the real data Or generate data , output a probability value, indicating the probability that the input data is real data. The goal of the discriminator is to maximize the probability of correctly distinguishing real data from generated data, while the goal of the generator is to make the discriminator unable to distinguish between generated data and real data; Loss function of the discriminator And the loss function of the generator They are defined as follows:

[0006] in, is the distribution of the real data, It is the distribution of random noise. By alternately training the discriminator and the generator and continuously optimizing their parameters, the generated data can better simulate the distribution of real data and enhance the diversity and integrity of multi-source heterogeneous data.

[0007] Preferably, the ontology construction described in S5: first, define the concept hierarchy, determine the core concepts of substation equipment, including equipment, components, operating status, and operation, and build a conceptual hierarchical structure, define clear semantics and connotations for each concept, and ensure the accuracy and consistency of the concepts by combining natural language description and formal definition; define relationship types, determine the relationship types between entities, including the composition relationship of equipment, the relationship between equipment and the environment, and the operating status relationship of equipment, and define the name, semantics, and constraints for each relationship type; create an ontology model, use an ontology modeling tool to build an ontology model of substation equipment, and present the defined concepts and relationships in a visual and formalized manner. In the model, classes, attributes, and instance elements can be defined, and logical relationships between them can be established. The ontology model provides a unified framework and standard for subsequent knowledge fusion.

[0008] Preferably, the entity alignment in S5 calculates entity similarity in the following manner: For entity pairs from different data sources that may represent the same entity, the edit distance algorithm is used to calculate the name similarity, and the attribute similarity is combined to calculate the comprehensive similarity. and The names are and , the edit distance is , name similarity , For attribute similarity, if the attribute value is numerical, Euclidean distance is used for calculation; if it is textual, cosine similarity is used for calculation. The comprehensive similarity is the weighted sum of name similarity and attribute similarity. When the comprehensive similarity of an entity pair exceeds the set threshold, it is merged into one entity.

[0009] Preferably, the conflict detection and resolution in S5 includes attribute value conflict detection and relationship conflict detection. Attribute value conflict detection: when integrating knowledge from different sources, detect inconsistent attribute values ​​of the same entity, and perform conflict detection by comparing attribute values, data source reliability, and data collection time; relationship conflict detection: detect inconsistent relationships between entities, and perform relationship conflict detection by analyzing text semantics and related device knowledge; for conflicting data in multiple data sources, use voting to determine the final value or relationship. When the majority of data sources believe that a certain attribute value of a device is X, X is used as the final attribute value. At the same time, expert knowledge and domain rules are combined to assist in judgment to improve the accuracy of conflict resolution.

[0010] Preferably, the optimization algorithm described in S7 is specifically as follows: For each node in the knowledge graph , whose eigenvector is expressed as , the set of neighbor nodes is , first calculate the node With neighbor nodes The attention coefficient between :

[0011] in, is a learnable parameter vector, represents vector concatenation, is the activation function, and then the attention coefficient is normalized:

[0012] Finally, the node The new feature vector of By weighted summation of neighbor node feature vectors:

[0013] in, It is an activation function. After multiple cycles of calculation, it highlights the role of important nodes in the knowledge graph and allocates higher attention to the corresponding nodes in subsequent analysis and application.

[0014] A system includes a storage medium and a processor, wherein the storage medium is provided with a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The knowledge graph construction method of multi-source heterogeneous data of substation main equipment of the present invention improves data utilization. By introducing a data enhancement algorithm in the secondary data processing, the data can be enhanced and supplemented, the diversity and integrity of the multi-source heterogeneous data can be enhanced, and the construction of the knowledge graph is more comprehensive and accurate.

[0016] In the knowledge graph construction method of the present invention, by optimizing the highlight processing, different attention weights can be assigned to the nodes according to the characteristics of the knowledge graph nodes and the information of the neighboring nodes, and important nodes can be highlighted to give higher attention weights, so that the knowledge graph can be more easily noticed during use, thereby increasing work processing efficiency and reducing the probability of neglecting attention to important nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] See also Figure 1 The present invention provides a technical solution: a method for constructing a knowledge graph of multi-source heterogeneous data of a substation main device, the method comprising the following steps: S1. Multi-source heterogeneous data collection: For structured database data, database connection tools such as JDBC are used to obtain the required data by writing SQL query statements; For semi-structured data such as monitoring data in XML and JSON formats, the corresponding parsing library is used to parse and extract key information. For XML, Python's xml.etree.ElementTree library can be used and converted into a structured data format. For unstructured text data and web page data, such as operation and maintenance records and documents, natural language processing technology and web crawler technology are used. The web crawler uses the Scrapy framework to collect data according to the set target URL and data extraction rules. After the unstructured text data is collected, it is stored in a local file or database and awaits subsequent processing. S2. Primary data processing: Primary data processing is divided into structured data processing, semi-structured data processing and unstructured data processing; Structured data processing: perform data cleaning, remove duplicate records, compare key fields, identify and delete duplicate record rows by writing database query statements or using data processing tools; correct incorrect formats, and convert date formats that are not uniform to standard formats; for abnormal values ​​in numerical data, such as temperature values ​​that are obviously beyond the normal range, filter and correct them by setting reasonable threshold ranges; Data integration integrates relevant structured data obtained from different database tables or systems, associates and merges basic equipment information in the equipment management system with operating data in the monitoring system through key fields such as equipment number, and forms a data set containing various aspects of equipment information; Semi-structured data processing: format conversion. For maintenance reports in XML format, parse the XML node content, extract key information, and convert it into a structured table format; for device configuration files in JSON format, parse the key-value pair information, extract device parameter configuration information, and convert it into a structured format that is easy to process; Data normalization: normalize the extracted semi-structured data content, for example, unify the expression of the names of equipment components, such as unifying "switch contact" and "switch contact head" into "switch contact"; Unstructured data processing: text cleaning, removing punctuation marks, special characters, and meaningless character noise, using Chinese word segmentation tools to segment text, splitting continuous text into word sequences, and removing common stop words that have no practical meaning; Text annotation: For inspection records and fault description text data, manual annotation or supervised machine learning algorithm annotation is performed to identify the key information of equipment entities and status descriptions; S3, secondary data processing, introduces data enhancement algorithm, sets generator and discriminator. In multi-source heterogeneous data processing, the generator is used to generate new data similar to real data, and the discriminator is used to distinguish real data from generated data. Through the confrontation between the two, the generator continuously improves the quality of generated data, thereby achieving data enhancement and supplementing data sources with less data volume. The data enhancement algorithm is as follows: Generator G receives random noise z and some substation equipment data features As input, generate new data samples , the discriminator D receives the real data Or generate data , output a probability value, indicating the probability that the input data is real data. The goal of the discriminator is to maximize the probability of correctly distinguishing real data from generated data, while the goal of the generator is to make the discriminator unable to distinguish between generated data and real data; Loss function of the discriminator And the loss function of the generator They are defined as follows:

[0020] in, is the distribution of the real data, It is the distribution of random noise. By alternately training the discriminator and the generator and continuously optimizing their parameters, the generated data can better simulate the distribution of real data and enhance the diversity and integrity of multi-source heterogeneous data.

[0021] S4: Extract knowledge from the data obtained in S3, including entity extraction, relationship extraction, and attribute extraction. Entity extraction identifies entities by formulating rule templates. For example, define a rule: if the text contains a pattern such as "device model: specific model", then extract "specific model" as an entity. Formulate a series of rules for common entity types such as device name and manufacturer for extraction; Relation extraction is done by building a neural network model. The convolutional neural network (CNN) or recurrent neural network (RNN) and its variants are used to build the relationship extraction model. The model input is a sentence text vector representation containing two entities. After the network layer learns the semantic features of the sentence, it outputs the relationship type result between the entities. Attribute extraction is divided into extraction from structured data and extraction from unstructured data. When extracting from structured data, the inherent attribute information of the device is directly obtained from the database table field. For the attribute information clearly marked in the semi-structured data, it is extracted according to the format analysis. When extracting from unstructured data, the attribute extraction rule template is defined. When the text meets the template rules, the corresponding attribute information is extracted. At the same time, the semantic analysis is combined with natural language processing technology to improve the accuracy of attribute extraction. S5. Perform knowledge fusion, which includes ontology construction, entity alignment, and conflict detection and resolution; Ontology construction: First, define the concept hierarchy, determine the core concepts of substation equipment, including equipment, components, operating status, and operations, and build a conceptual hierarchical structure. Define clear semantics and connotations for each concept, and ensure the accuracy and consistency of the concepts by combining natural language description with formal definition; define relationship types, determine the relationship types between entities, including the composition relationship of equipment, the relationship between equipment and the environment, and the operating status relationship of equipment, and define the name, semantics, and constraints for each relationship type; create an ontology model, use the ontology modeling tool to build the substation equipment ontology model, and present the defined concepts and relationships in a visual and formalized way. In the model, you can define classes, attributes, and instance elements, and establish logical relationships between them. The ontology model provides a unified framework and standard for subsequent knowledge fusion.

[0022] Entity alignment: Entity similarity is calculated by: For entity pairs from different data sources that may represent the same entity, the edit distance algorithm is used to calculate the name similarity, and the attribute similarity is combined to calculate the comprehensive similarity. and The names are and , the edit distance is , name similarity For attribute similarity, if the attribute value is numerical, Euclidean distance is used for calculation; if it is textual, cosine similarity is used for calculation. The comprehensive similarity is the weighted sum of name similarity and attribute similarity. When the comprehensive similarity of an entity pair exceeds the set threshold, it is merged into one entity.

[0023] Conflict detection and resolution, including attribute value conflict detection and relationship conflict detection. Attribute value conflict detection: when integrating knowledge from different sources, detect inconsistent attribute values ​​of the same entity, and perform conflict detection by comparing attribute values, data source reliability, and data collection time; relationship conflict detection: detect inconsistent relationships between entities, such as device A and device B are in a parallel relationship in one data source, but in a series relationship in another data source, and perform relationship conflict detection by analyzing text semantics and related device knowledge; for conflicting data in multiple data sources, use voting to determine the final value or relationship. When the majority of data sources believe that a certain attribute value of a device is X, X is used as the final attribute value. At the same time, expert knowledge and domain rules are combined for auxiliary judgment to improve the accuracy of conflict resolution.

[0024] S6. Knowledge graph formation: The extracted and integrated knowledge is stored in a graph database. Entities are used as nodes to store their attribute information, and relationships are used as edges to store the attributes and direction information of relationships to form a knowledge graph. Graph databases include Neo4j and JanusGraph. Neo4j is easy to use and has high query performance, making it suitable for small and medium-sized knowledge graph storage and query. JanusGraph has distributed storage capabilities and good scalability, making it suitable for large-scale knowledge graph applications. Select a suitable graph database based on factors such as the amount of substation equipment data and application requirements.

[0025] S7, optimize the highlight processing, through the optimization algorithm to assign different attention weights to the nodes according to the characteristics of the knowledge graph nodes and the information of the neighboring nodes. In the knowledge graph of substation equipment, some key equipment nodes and nodes with important connection relationships have an important impact on the operation and fault propagation of the entire system. These important nodes are identified by the algorithm and given higher attention weights; the optimization algorithm is as follows: For each node in the knowledge graph , whose eigenvector is expressed as , the set of neighbor nodes is , first calculate the node With neighbor nodes The attention coefficient between :

[0026] in, is a learnable parameter vector, represents vector concatenation, is the activation function, and then the attention coefficient is normalized:

[0027] Finally, the node The new feature vector of By weighted summation of neighbor node feature vectors:

[0028] in, It is an activation function. After multiple cycles of calculation, it highlights the role of important nodes in the knowledge graph and allocates higher attention to the corresponding nodes in subsequent analysis and application.

[0029] S8, application implementation, based on the knowledge graph after further optimization and highlighting in S7, equipment fault diagnosis, analysis, and operation and maintenance strategy optimization are realized.

[0030] A system includes a storage medium and a processor. The storage medium is provided with a computer program. When the computer program is executed by the processor, the processor executes the steps of the above method.

[0031] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A knowledge graph construction method for multi-source heterogeneous data of substation main equipment, characterized in that: The method comprises the following steps: S1. Multi-source heterogeneous data collection: For structured database data, database connection tools are used to obtain required data by writing SQL query statements; for semi-structured data, corresponding parsing libraries are used to parse and extract key information, and then convert it into structured data format; for unstructured text data and web page data, natural language processing technology and web crawler technology are used. The web crawler uses the Scrapy framework to collect data according to the set target URL and data extraction rules. After unstructured text data is collected, it is stored in local files or databases and waits for subsequent processing; S2. Primary data processing, which includes structured data processing, semi-structured data processing and unstructured data processing; Structured data processing: perform data cleaning, remove duplicate records, compare key fields, identify and delete duplicate record rows by writing database query statements or using data processing tools; correct incorrect formats, and convert date formats that are not uniform to standard formats; for abnormal values ​​in numerical data, such as temperature values ​​that are obviously beyond the normal range, filter and correct them by setting reasonable threshold ranges; Data integration integrates relevant structured data obtained from different database tables or systems, associates and merges basic equipment information in the equipment management system with operating data in the monitoring system through key fields such as equipment number, and forms a data set containing various aspects of equipment information; Semi-structured data processing: format conversion. For maintenance reports in XML format, parse the XML node content, extract key information, and convert it into a structured table format; for device configuration files in JSON format, parse the key-value pair information, extract device parameter configuration information, and convert it into a structured format that is easy to process; Data normalization: normalize the extracted semi-structured data content; Unstructured data processing: text cleaning, removing punctuation marks, special characters, and meaningless character noise, using Chinese word segmentation tools to segment text, splitting continuous text into word sequences, and removing common stop words that have no practical meaning; Text annotation: For inspection records and fault description text data, manual annotation or supervised machine learning algorithm annotation is performed to identify the key information of equipment entities and status descriptions; S3, secondary data processing, introduces data enhancement algorithm, sets generator and discriminator. In multi-source heterogeneous data processing, the generator is used to generate new data similar to real data, and the discriminator is used to distinguish real data from generated data. Through the confrontation between the two, the generator continuously improves the quality of generated data, thereby achieving data enhancement and supplementing data sources with less data volume; S4, extract knowledge from the data obtained in S3, including entity extraction, relationship extraction and attribute extraction. Entity extraction identifies entities by formulating rule templates; relationship extraction builds a neural network model, using convolutional neural network CNN or recurrent neural network RNN ​​and its variants to build a relationship extraction model. The model input is a sentence text vector representation containing two entities. After the network layer learns the sentence semantic features, it outputs the relationship type result between entities; attribute extraction is divided into extraction from structured data and extraction from unstructured data. When extracting from structured data, the inherent attribute information of the device is directly obtained from the database table field. For the attribute information clearly marked in the semi-structured data, it is extracted according to the format analysis. When extracting from unstructured data, define the attribute extraction rule template. When the text meets the template rules, extract the corresponding attribute information. At the same time, combine natural language processing technology to perform semantic analysis to improve the accuracy of attribute extraction; S5. Perform knowledge fusion, wherein the knowledge fusion includes ontology construction, entity alignment, and conflict detection and resolution; S6. Formation of knowledge graph: storing the extracted and fused knowledge in a graph database, storing the attribute information of entities as nodes, and storing the attributes and direction information of relationships as edges to form a knowledge graph. The graph database includes Neo4j and JanusGraph. S7, optimize the highlight processing, through the optimization algorithm to assign different attention weights to the nodes according to the characteristics of the knowledge graph nodes and the information of the neighboring nodes. In the knowledge graph of substation equipment, some key equipment nodes and nodes with important connection relationships have an important impact on the operation and fault propagation of the entire system. These important nodes are identified through the algorithm and given higher attention weights; S8, application implementation, based on the knowledge graph after further optimization and highlighting in S7, equipment fault diagnosis, analysis, and operation and maintenance strategy optimization are realized.

2. The method for constructing a knowledge graph of multi-source heterogeneous data of a main substation equipment according to claim 1 is characterized in that: The data enhancement algorithm described in S3 is as follows: Generator G receives random noise z and some substation equipment data features As input, generate new data samples , the discriminator D receives the real data Or generate data , output a probability value, indicating the probability that the input data is real data. The goal of the discriminator is to maximize the probability of correctly distinguishing real data from generated data, while the goal of the generator is to make the discriminator unable to distinguish between generated data and real data; Loss function of the discriminator And the loss function of the generator They are defined as follows: ； in, is the distribution of the real data, It is the distribution of random noise. By alternately training the discriminator and the generator and continuously optimizing their parameters, the generated data can better simulate the distribution of real data and enhance the diversity and integrity of multi-source heterogeneous data.

3. The method for constructing a knowledge graph of multi-source heterogeneous data of a main substation equipment according to claim 1 is characterized in that: The ontology construction described in S5 is as follows: first, define the concept hierarchy, determine the core concepts of substation equipment, including equipment, components, operating status, and operation, and build a concept hierarchy structure, define clear semantics and connotations for each concept, and ensure the accuracy and consistency of the concepts by combining natural language description with formal definition; define relationship types, determine the relationship types between entities, including the composition relationship of equipment, the relationship between equipment and environment, and the operating status relationship of equipment, and define the name, semantics, and constraints for each relationship type; Create an ontology model and use ontology modeling tools to build an ontology model of substation equipment. The defined concepts and relationships can be presented in a visual and formal way. Classes, attributes, and instance elements can be defined in the model, and logical relationships between them can be established. The ontology model provides a unified framework and standard for subsequent knowledge fusion.

4. The method for constructing a knowledge graph of multi-source heterogeneous data of a main substation equipment according to claim 1 is characterized in that: The entity alignment described in S5 calculates entity similarity in the following way: For entity pairs from different data sources that may represent the same entity, the edit distance algorithm is used to calculate the name similarity, and the attribute similarity is combined to calculate the comprehensive similarity. and The names are and , the edit distance is , name similarity For attribute similarity, if the attribute value is numerical, Euclidean distance is used for calculation; if it is textual, cosine similarity is used for calculation. The comprehensive similarity is the weighted sum of name similarity and attribute similarity. When the comprehensive similarity of an entity pair exceeds the set threshold, it is merged into one entity.

5. The method for constructing a knowledge graph of multi-source heterogeneous data of a main substation equipment according to claim 1 is characterized in that: The conflict detection and resolution described in S5 includes attribute value conflict detection and relationship conflict detection. Attribute value conflict detection: when integrating knowledge from different sources, detect inconsistent attribute values ​​of the same entity, and perform conflict detection by comparing attribute values, data source reliability, and data collection time; relationship conflict detection: detect inconsistent relationships between entities, and perform relationship conflict detection by analyzing text semantics and related device knowledge; for conflicting data in multiple data sources, use voting to determine the final value or relationship. When most data sources believe that a certain attribute value of a device is X, X is used as the final attribute value. At the same time, expert knowledge and domain rules are combined to assist in judgment to improve the accuracy of conflict resolution.

6. The method for constructing a knowledge graph of multi-source heterogeneous data of a main substation equipment according to claim 1 is characterized by: The optimization algorithm described in S7 is as follows: For each node in the knowledge graph , Its eigenvector is expressed as , the set of neighbor nodes is , first calculate the node With neighbor nodes The attention coefficient between : ， in, is a learnable parameter vector, represents vector concatenation, is the activation function, and then the attention coefficient is normalized: ， Finally, the node The new feature vector of By weighted summation of neighbor node feature vectors: ， in, It is an activation function. After multiple cycles of calculation, it highlights the role of important nodes in the knowledge graph and allocates higher attention to the corresponding nodes in subsequent analysis and application.

7. A system, characterized in that: The method comprises a storage medium and a processor, wherein the storage medium is provided with a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 6.

Images