Method for constructing multi-source heterogeneous data knowledge graph of power distribution network

Abstract

The present invention relates to the technical field of data metering management for power distribution networks, and in particular, to a method for constructing a multi-source heterogeneous data knowledge graph of a power distribution network. The method comprises: model layer construction: summarizing and analyzing a power transmission network knowledge content system framework, analyzing multi-source heterogeneous data of a power distribution network, defining an entity type, an attribute type, and an entity relationship, and constructing an organizational framework of knowledge of the power distribution network; data layer construction: multi-source heterogeneous data access and fusion, data preprocessing, knowledge extraction, knowledge integration, and knowledge storage and updating; and metering management application: real-time monitoring and early-warning, abnormal data troubleshooting, data quality management, and statistical analysis and decision support. According to the design of the present invention, data fusion and application integration can be performed with various service systems. The multi-source heterogeneous data of the power distribution network can be effectively integrated, the organization and management of the knowledge of the power distribution network are implemented, and the sustainable development of a power grid is promoted. By means of interactive displaying of a knowledge graph and use of a data quality management platform, abnormal metering data is automatically troubleshot.

Description

A method for constructing knowledge graph of multi-source heterogeneous data in distribution network Technical Field

[0001] The present invention relates to the technical field of distribution network data metering management, and in particular to a method for constructing a knowledge graph of multi-source heterogeneous data in a distribution network. Background Art

[0002] The distribution network is part of the power system. Its primary function is to convert electricity transmitted by high-voltage transmission lines into low-voltage electricity suitable for households, industry, and commercial users, and then distribute this electricity to various end users to meet their electricity needs. The distribution network, typically consisting of substations, distribution transformers, cables, switchgear, and other components, is a key component of urban power supply. With urbanization and increasing electricity demand, the scale and complexity of the distribution network are increasing, placing higher demands on its operation and management.

[0003] The operation and management of distribution networks generates a vast amount of data from diverse sources, such as sensors, monitoring equipment, and operational records. This data, due to its diverse sources and types, is referred to as multi-source heterogeneous data. This data can be structured, such as measured values ​​like voltage and current, or unstructured, such as text reports and log files. The presence of multi-source heterogeneous data complicates comprehensive analysis and decision-making regarding distribution networks, necessitating the effective integration and processing of this data to extract valuable information and knowledge.

[0004] A knowledge graph is a graphical structure for representing knowledge, in which entities and relationships are represented as nodes and edges. The construction of a knowledge graph involves extracting entities and relationships from structured and unstructured data and organizing them into a meaningful graph. Knowledge graphs help people understand and reason about the connections between knowledge, thereby supporting various tasks such as data query, information retrieval, and intelligent recommendation.

[0005] In the field of distribution networks, specialized knowledge graphs, called domain knowledge graphs, can be constructed. These domain knowledge graphs focus specifically on distribution network-related knowledge and relationships, such as equipment information, power operating status, and fault diagnosis. Domain knowledge graphs can help distribution network operators and managers better understand the status and operation of the distribution network, enabling them to make more informed decisions and optimize distribution network operations and management.

[0006] Metering management refers to the process of measuring and monitoring energy and resources in public utilities or industrial production, such as electricity, water, and natural gas. When applying the knowledge graph of multi-source, heterogeneous data in distribution networks to metering management, real-time data integrated into the knowledge graph can be used to monitor metering equipment in real time, identify anomalies, and issue early warnings. Furthermore, the interactive display of the knowledge graph allows for the rapid location of abnormal metering data, assisting data managers in data quality management and troubleshooting, providing statistical analysis and decision support to management departments, and ultimately optimizing the operational management of distribution networks.

[0007] In summary, distribution networks, as the lifeblood of urban power supply, face increasing complexity and challenges. The presence of multi-source, heterogeneous data complicates comprehensive analysis and decision-making regarding distribution networks, necessitating an effective method to integrate and process this data. Knowledge graphs, as graphical structures for representing knowledge, can help us understand and reason about relationships between knowledge, providing support for the operation and management of distribution networks. To this end, we propose a method for constructing a knowledge graph for multi-source, heterogeneous data in distribution networks.

[0008] Summary of the Invention

[0009] The purpose of the present invention is to provide a method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network to solve the problems raised in the above background technology.

[0010] To solve the above technical problems, one of the objectives of the present invention is to provide a method for constructing a knowledge graph of multi-source heterogeneous data in a distribution network, comprising the following steps:

[0011] S1. Model layer construction: Model layer construction is one of the key steps in the construction of the knowledge graph of multi-source heterogeneous data in the distribution network. It summarizes and analyzes the knowledge content system framework of the transmission network in a top-down manner. By analyzing the multi-source heterogeneous data of the distribution network in detail and starting from the transmission layer, substation layer, distribution network layer and user layer, it accurately defines the entity type, attribute type and entity relationship, and constructs the organizational framework of the distribution network knowledge. Specifically:

[0012] S1.1. Summarization and Analysis of the Knowledge Content Framework of the Transmission Network: First, summarize and analyze the knowledge content in the transmission network field to understand the important concepts, key attributes, and relationships between entities in this field;

[0013] S1.2. Detailed analysis of multi-source heterogeneous data on the distribution network: After understanding the knowledge content of the transmission network, a detailed analysis of multi-source heterogeneous data on the distribution network is conducted. This includes various data obtained from different sources, such as monitoring data, equipment information, and operation records.

[0014] S1.3. Define entity types, attribute types, and entity relationships: Based on the analysis of multi-source heterogeneous data, determine the entity types, attribute types, and entity relationships of the distribution network knowledge graph;

[0015] S1.4. Constructing an organizational framework for distribution network knowledge: After determining entity types, attribute types, and entity relationships, organize them to construct an organizational framework for distribution network knowledge.

[0016] S2. Data layer construction: The knowledge graph of multi-source heterogeneous data of the distribution network can be represented by G = (E, R, S), where E is the entity set in the knowledge base, R is the relationship set, and S is the knowledge set of "entity-relationship-entity" triples. The data layer construction is to extract the required entities and relationships from the multi-source heterogeneous data of the distribution network under the guidance of the model layer organization framework. The key technologies of data layer construction include multi-source heterogeneous data fusion, data preprocessing, knowledge extraction, knowledge fusion, and knowledge storage and update. Specifically:

[0017] S2.1. Multi-source heterogeneous data access and fusion: Obtain raw data and generate structured, semi-structured, and unstructured data by accessing multi-source heterogeneous data from different types of distribution networks. Then, use multi-source heterogeneous data fusion technology to aggregate, correlate, and integrate various data from different data sources in the distribution network.

[0018] S2.2. Data Preprocessing: Preprocess the fused multi-source heterogeneous data, including cleaning, denoising, and normalizing structured and semi-structured data to ensure data quality and consistency. For unstructured data, such as large amounts of text in Word and PDF formats, perform structured conversion, Chinese word segmentation, text annotation, and construct word vector models.

[0019] S2.3. Knowledge Extraction: Use rule-based and deep learning-based knowledge extraction to extract the required entity and relationship information from the preprocessed data. The goal of knowledge extraction is to transform the data into "entity-relationship-entity" triples, thereby forming the basic structure of the knowledge graph.

[0020] S2.4, Knowledge Fusion: Fusion of knowledge from different data sources and different knowledge extraction methods to eliminate duplication and conflict and form a consistent knowledge representation. Knowledge fusion helps improve the integrity and consistency of the knowledge graph.

[0021] S2.5. Knowledge Storage and Update: Finally, the extracted and integrated knowledge is stored using knowledge storage technologies such as the Neo4j graph database to form a knowledge graph of multi-source heterogeneous data in the distribution network. At the same time, the knowledge graph is regularly updated to maintain its consistency and accuracy with the actual distribution network data.

[0022] S3. Metering Management Application: Using the Neo4j graph database to store distribution network knowledge can form a complete and structured knowledge graph that contains information on various entities, attributes, and relationships in the distribution network field. Neo4j is a graph database with important application value in metering management business scenarios. Specifically:

[0023] S3.1. Real-time monitoring and early warning: Utilize the multi-source data of the distribution network integrated in the knowledge graph to monitor the status of metering equipment in real time;

[0024] S3.2. Abnormal data investigation: Quickly locate abnormal measurement data through interactive display of knowledge graph;

[0025] S3.3 Data Quality Management: Ensure data quality and consistency through data preprocessing and knowledge extraction during the data layer construction process. The data in the knowledge graph can be used as a reference for the data quality management platform.

[0026] S3.4. Statistical analysis and decision support: The knowledge graph of multi-source heterogeneous data in the distribution network can integrate data from different systems and conduct comprehensive statistics and analysis on the metering data of public transformers, special transformers, and low-voltage users.

[0027] As a further improvement of this technical solution, in step S1.3, entity types, attribute types and entity relationships are defined in the model layer construction; specifically:

[0028] Entity types refer to various specific objects in the distribution network, such as cables, transformers, switches, etc.

[0029] Attribute types are the characteristics and properties of these entities, such as voltage, current, power, etc.

[0030] Entity relationships represent the association relationships between entities, such as connection, belonging, etc.

[0031] As a further improvement of the present technical solution, in step S2.1, when accessing multi-source heterogeneous data in the data layer construction, the multi-source heterogeneous data can be divided from different dimensions, including the time dimension, the data structure dimension, and the business domain dimension; specifically, the following:

[0032] In the time dimension, real-time data: data collected immediately, reflecting the real-time status of distribution network equipment, lines, loads, etc.; historical data: data collected over a period of time, used to analyze the operation history and trends of the distribution network; forecast data: data predicted based on historical data and models, used to predict future power demand and load conditions;

[0033] In the data structure dimension, structured data: data stored in a fixed format and schema, such as table data in a database, with clear fields and relationships; semi-structured data: data with a certain structure but not conforming to the traditional relational database table format, such as XML, JSON, etc.; unstructured data: data without a fixed format and schema, such as text, images, audio, video, etc.

[0034] The business area dimensions include: substation data, distribution equipment data, power line data, user electricity consumption data, power quality data, operation and maintenance data, etc.

[0035] As a further improvement of this technical solution, in step S2.1, when fusion of multi-source heterogeneous data in data layer construction includes the following three key aspects: data association, multi-source heterogeneous data estimation, and data source management; specifically, including:

[0036] First, data association technology is used to associate distribution network data from different data sources, identify their relationships, and ensure that they reflect the same distribution network entity or event, thus avoiding data redundancy and duplication.

[0037] Secondly, a multi-source data estimation technique is used to integrate information from different data sources to improve the estimation of distribution network targets;

[0038] In addition, data source management technology also plays an important role in the data fusion process. By giving the environmental status of the data source, the data collection and processing sources are reasonably allocated to achieve the optimization of operation costs;

[0039] For multi-source data in the distribution network, tasks can be flexibly assigned based on factors such as the location and performance of different sensors to improve the efficiency and performance of data collection and processing.

[0040] As a further improvement of the present technical solution, in step S2.2, data preprocessing in data layer construction includes the following steps:

[0041] First, for structured data: use data cleaning tools to clean the data to remove duplicate data, missing data, and erroneous data to ensure data accuracy and completeness; use filtering and other technologies to process noisy data and perform data denoising; and convert the data into a unified format to perform data normalization;

[0042] Second, for unstructured data: use OCR technology, Python's docx2txt, PyPDF2 library, or text extraction tools to convert unstructured data into a processable text format;

[0043] The characters of the data are annotated according to the BIO annotation method, and the annotated corpus is used as the input raw corpus of the BERT word vector conversion layer;

[0044] The pre-trained model BERT is used to map the annotated original corpus into dynamic word vectors; this step serves as the input of the next layer of the network;

[0045] Finally, use the Chinese word segmentation tool to segment the Chinese text and divide the continuous Chinese text into meaningful words.

[0046] As a further improvement of this technical solution, in step S2.3, the knowledge extraction in the data layer construction includes entity extraction, relationship extraction, and event extraction; specifically, the following steps are included:

[0047] Step 1: Use rule-based and deep learning-based knowledge extraction methods. Rule-based key information extraction has the advantages of precise scope and high accuracy. Leveraging this advantage, specific rules are written to match a small number of precise objects for subsequent corpus import.

[0048] Step 2: This step will enable sequence annotation to continue extracting key information, and part of the results output in Step 1 can be used as training corpus to replace the process of manually injecting corpus;

[0049] Step 3: Use the training corpus output in Step 2 to perform knowledge graph modeling based on the BERT-BiLSTM-CRF algorithm;

[0050] Step 4: Use the model from Step 3 to judge the remaining corpus from Step 2. If the model judgment result shows that the model does not meet the standard, return to Step 1 to add more corpus. Continue until the model automatically determines that it meets the standard, exit the iteration, and use the most recently generated model as the final model.

[0051] As a further improvement of this technical solution, in step S2.4, the knowledge fusion in the data layer construction is to represent multiple entities with the same meaning with a single entity, that is, A = {a1, a2, a3, ..., a m}, where a1~a m Represents m entities with the same meaning, and A is the fused entity. Use the text clustering algorithm to perform knowledge fusion, that is, traverse each entity, calculate the text similarity between the entity and the rest of the entities, and merge the entities with similarity higher than the threshold. The calculation formula of entity semantic similarity T is shown in the formula:

[0052] Where A i 、B i The word frequency vectors of the two entity names A and B are obtained after word segmentation, and n is the number of word segmentations. The higher the semantic similarity T, the higher the similarity between the two entities.

[0053] Finally, by extracting the "entity-relationship-entity" triples, these knowledge elements are effectively encoded and represented, thereby constructing a knowledge graph of multi-source heterogeneous data of the distribution network; this knowledge graph organically organizes the distribution network information and presents it in the form of a graphical structure; the constructed knowledge graph is stored in the Neo4j graph database to facilitate the storage and query of knowledge.

[0054] As a further improvement of this technical solution, in step S3.1, the real-time monitoring and early warning in the metering management application specifically includes:

[0055] First, it is necessary to integrate heterogeneous data from multiple distribution network sources, originating from various systems and data sources. These data sources may include marketing management systems, metering automation systems, intelligent turnover cabinets, and error prevention and control systems. Data preprocessing and fusion are required to convert the data into a unified data format.

[0056] The accessed data needs to be transmitted and processed in real time; data stream processing technology is used to ensure that the data can be transmitted to the knowledge graph system in a timely manner and processed in real time;

[0057] In the knowledge graph, the logic of real-time monitoring of the status of metering equipment is implemented by defining rules and query statements.

[0058] As a further improvement of the present technical solution, in step S3.2, when checking abnormal data in the metering management application, the following steps are included:

[0059] During the model layer construction phase, define the types of abnormal data in metering management; this may include meter offline, voltage abnormality, current abnormality, meter reading abnormality, and arrears.

[0060] Determine the characteristics and attributes of abnormal data for subsequent knowledge extraction and identification.

[0061] The second object of the present invention is to provide a platform device for constructing a multi-source heterogeneous data knowledge graph, including a processor, a memory, and a computer program stored in the memory and running on the processor, and the processor is used to implement the steps of the above-mentioned method for constructing a multi-source heterogeneous data knowledge graph of a distribution network when executing the computer program.

[0062] The third object of the present invention is to provide a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the steps of the method for constructing the knowledge graph of multi-source heterogeneous data of the distribution network.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. This method for constructing a knowledge graph of multi-source heterogeneous data in a distribution network provides an efficient and intelligent way to integrate and manage multi-source heterogeneous data in the distribution network, providing more comprehensive, accurate, and intelligent support for the operation and management of the distribution network. It can also integrate data and applications with various business systems. It can effectively integrate multi-source heterogeneous data in the distribution network and construct a structured knowledge graph. In terms of metering management business, the knowledge graph of multi-source heterogeneous data in the distribution network has significant applications, ultimately realizing the organization and management of distribution network knowledge and promoting the sustainable development of the power grid.

[0065] 2. In the construction method of the knowledge graph of multi-source heterogeneous data in the distribution network, by combining this knowledge graph with the metering management business scenario, the following optimizations can be achieved: automatic troubleshooting of metering abnormal data, improved abnormal data processing efficiency, optimized decision support, data integration, and application integration. Through the interactive display of the knowledge graph and the application of the data quality management platform, it is possible to automatically troubleshoot metering abnormal data, intuitively display abnormal analysis results and issue warnings, and automatically and synchronously send overdue warning information to the corresponding work leaders.

[0066] 3. The construction method of the distribution network multi-source heterogeneous data knowledge graph has important application value in the field of metering management. It provides strong data support and decision-making basis for distribution network management, and promotes the optimization and development of the entire distribution network system. At the same time, through the automated processing of abnormal data and early warning functions, it effectively reduces the workload of the metering management department, improves management efficiency and data processing speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] FIG1 is a flowchart of an exemplary method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network according to the present invention;

[0068] FIG2 is an exemplary diagram of multi-source heterogeneous data types in a distribution network according to the present invention;

[0069] FIG3 is a schematic diagram of an exemplary method of generating text word vectors using the BERT model in the present invention;

[0070] FIG4 is a schematic diagram of an exemplary method for constructing a knowledge graph using the BERT-BiLSTM-CRF model in the present invention;

[0071] FIG5 is a structural diagram of an exemplary electronic computer platform device in the present invention. DETAILED DESCRIPTION

[0072] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0073] Example 1

[0074] As shown in Figures 1-4, this embodiment provides a method for constructing a knowledge graph of multi-source heterogeneous data for distribution networks. This method has important application value in the field of metering management, providing powerful data support and decision-making basis for distribution network management, and promoting the optimization and development of the entire distribution network system. At the same time, through automated processing of abnormal data and early warning functions, it effectively reduces the workload of metering management departments, improves management efficiency and data processing speed. As shown in Figure 1, this method specifically includes three parts: model layer construction, data layer construction, and metering management application.

[0075] Part I S1, Model Layer Construction: Model layer construction is one of the key steps in the construction of the knowledge graph of multi-source heterogeneous data in the distribution network. It summarizes and analyzes the knowledge content system framework of the transmission network in a top-down manner. By analyzing the multi-source heterogeneous data of the distribution network in detail and starting from the transmission layer, substation layer, distribution network layer and user layer, it accurately defines the entity type, attribute type and entity relationship, and constructs the organizational framework of the distribution network knowledge. Specifically:

[0076] S1.1. Summarization and Analysis of the Knowledge Content Framework of the Transmission Network: First, summarize and analyze the knowledge content in the transmission network field to understand the important concepts, key attributes, and relationships between entities in this field;

[0077] S1.2. Detailed analysis of multi-source heterogeneous data in the distribution network: After understanding the knowledge content of the transmission network, a detailed analysis of the multi-source heterogeneous data in the distribution network is performed. As shown in Figure 2, this includes various data obtained from different sources, such as monitoring data, equipment information, and operation records.

[0078] S1.3. Define entity types, attribute types, and entity relationships: Based on the analysis of multi-source heterogeneous data, determine the entity types, attribute types, and entity relationships of the distribution network knowledge graph;

[0079] In this step, the entity types, attribute types, and entity relationships are defined in the model layer construction; specifically:

[0080] Entity types refer to various specific objects in the distribution network, such as cables, transformers, switches, etc.

[0081] Attribute types are the characteristics and properties of these entities, such as voltage, current, power, etc.

[0082] Entity relationships represent the association relationships between entities, such as connection, belonging, etc.

[0083] S1.4. Construct an organizational framework for distribution network knowledge: After determining the entity type, attribute type, and entity relationship, organize them to construct an organizational framework for distribution network knowledge.

[0084] Part 2, S2, Data Layer Construction: The knowledge graph for the distribution network's multi-source heterogeneous data can be represented by G = (E, R, S), where E is the entity set in the knowledge base, R is the relationship set, and S is the knowledge set of "entity-relationship-entity" triples. Data layer construction involves extracting the required entities and relationships from the distribution network's multi-source heterogeneous data under the guidance of the model layer's organizational framework. Key technologies for data layer construction include multi-source heterogeneous data fusion, data preprocessing, knowledge extraction, knowledge fusion, and knowledge storage and updating. Specifically:

[0085] S2.1. Multi-source heterogeneous data access and fusion: Obtain raw data and generate structured, semi-structured, and unstructured data by accessing multi-source heterogeneous data from different types of distribution networks. Then, use multi-source heterogeneous data fusion technology to aggregate, correlate, and integrate various data from different data sources in the distribution network.

[0086] In this step, when accessing multi-source heterogeneous data in the data layer construction, the multi-source heterogeneous data can be divided according to different dimensions, including time dimension, data structure dimension, and business domain dimension; specifically, including:

[0087] In the time dimension, real-time data: data collected immediately, reflecting the real-time status of distribution network equipment, lines, loads, etc.; historical data: data collected over a period of time, used to analyze the operation history and trends of the distribution network; forecast data: data predicted based on historical data and models, used to predict future power demand and load conditions;

[0088] In the data structure dimension, structured data: data stored in a fixed format and schema, such as table data in a database, with clear fields and relationships; semi-structured data: data with a certain structure but not conforming to the traditional relational database table format, such as XML, JSON, etc.; unstructured data: data without a fixed format and schema, such as text, images, audio, video, etc.

[0089] The business area dimensions include: substation data, distribution equipment data, power line data, user electricity consumption data, power quality data, operation and maintenance data, etc.

[0090] Furthermore, when building a data layer, the fusion of multi-source heterogeneous data includes the following three key aspects: data association, multi-source heterogeneous data estimation, and data source management. Specifically,

[0091] First, data association technology is used to associate distribution network data from different data sources, identify their relationships, and ensure that they reflect the same distribution network entity or event, thus avoiding data redundancy and duplication.

[0092] Secondly, multi-source data estimation technology is used to integrate information from different data sources to improve the estimation of distribution network targets. For example, current, voltage and other data from different sensors can be combined to obtain more accurate grid state estimation results, improving the accuracy and reliability of distribution network data.

[0093] In addition, data source management technology also plays an important role in the data fusion process. By giving the environmental status of the data source, the data collection and processing sources are reasonably allocated to achieve the optimization of operation costs;

[0094] For multi-source data in the distribution network, tasks can be flexibly assigned based on factors such as the location and performance of different sensors to improve the efficiency and performance of data collection and processing.

[0095] S2.2. Data Preprocessing: Preprocess the fused multi-source heterogeneous data, including cleaning, denoising, and normalizing structured and semi-structured data to ensure data quality and consistency. For unstructured data, such as large amounts of text in Word and PDF formats, perform structured conversion, Chinese word segmentation, text annotation, and construct word vector models.

[0096] In this step, data preprocessing in data layer construction includes the following steps:

[0097] First, for structured data: Use data cleaning tools (such as the pandas library in Python) to clean the data to remove duplicate data, missing data, and erroneous data to ensure data accuracy and completeness; use filtering and other techniques to process noisy data and perform data denoising; and convert the data into a unified format to perform data normalization;

[0098] Second, for unstructured data: use OCR technology, Python's docx2txt, PyPDF2 library, or text extraction tools (such as Tesseract, PDFMiner, etc.) to convert unstructured data into a processable text format;

[0099] The characters of the data are annotated according to the BIO annotation method, and the annotated corpus is used as the input raw corpus of the BERT word vector conversion layer;

[0100] The pre-trained model BERT is used to map the annotated original corpus into dynamic word vectors, as shown in Figure 3. This step serves as the input of the next layer of the network.

[0101] Finally, use the Chinese word segmentation tool to segment the Chinese text and divide the continuous Chinese text into meaningful words.

[0102] S2.3. Knowledge Extraction: Use rule-based and deep learning-based knowledge extraction to extract the required entity and relationship information from the preprocessed data. The goal of knowledge extraction is to transform the data into "entity-relationship-entity" triples, thereby forming the basic structure of the knowledge graph.

[0103] In this step, knowledge extraction in the data layer construction includes entity extraction, relationship extraction, and event extraction; specifically, the following steps are included:

[0104] Step 1: Use rule-based and deep learning-based knowledge extraction methods. Rule-based key information extraction has the advantages of precise scope and high accuracy. Leveraging this advantage, specific rules are written to match a small number of precise objects for subsequent corpus import.

[0105] Step 2: This step will enable sequence tagging to continue extracting key information, and part of the results output in Step 1 (such as 70%) can be used as training corpus to replace the process of manually injecting corpus;

[0106] Step 3: Use the training corpus output in Step 2 to perform knowledge graph modeling based on the BERT-BiLSTM-CRF algorithm, as shown in Figure 4.

[0107] Step 4: Use the model from Step 3 to judge the remaining corpus from Step 2. If the model judgment result shows that the model does not meet the standard, return to Step 1 to add more corpus. Continue until the model automatically determines that it meets the standard, exit the iteration, and use the most recently generated model as the final model.

[0108] S2.4, Knowledge Fusion: Fusion of knowledge from different data sources and different knowledge extraction methods to eliminate duplication and conflict and form a consistent knowledge representation. Knowledge fusion helps improve the integrity and consistency of the knowledge graph.

[0109] In this step, knowledge fusion in data layer construction is to represent multiple entities with the same meaning with a single entity, that is, A = {a1, a2, a3, ..., a m}, where a1~a mRepresents m entities with the same meaning, and A is the fused entity. Use the text clustering algorithm to perform knowledge fusion, that is, traverse each entity, calculate the text similarity between the entity and the rest of the entities, and merge the entities with similarity higher than the threshold. The calculation formula of entity semantic similarity T is shown in the formula:

[0110] Where A i 、B i The word frequency vectors of the two entity names A and B are obtained after word segmentation, and n is the number of word segmentations. The higher the semantic similarity T, the higher the similarity between the two entities.

[0111] Ultimately, by extracting "entity-relationship-entity" triples and effectively encoding and representing these knowledge elements, we constructed a knowledge graph for the distribution network's multi-source, heterogeneous data. This knowledge graph organically organizes distribution network information and presents it in a graphical structure. To facilitate knowledge storage and query, we stored the constructed knowledge graph in a Neo4j graph database. This database allows for efficient retrieval, updating, and expansion of distribution network knowledge, providing more convenient and accurate support for its operation and management.

[0112] S2.5. Knowledge storage and update: Finally, the extracted and integrated knowledge is stored using knowledge storage technologies such as the Neo4j graph database to form a knowledge graph of multi-source heterogeneous data in the distribution network. At the same time, the knowledge graph is updated regularly to maintain its consistency and accuracy with the actual distribution network data.

[0113] Part 3 S3, Metering Management Application: Using the Neo4j graph database to store distribution network knowledge can form a complete and structured knowledge graph that contains information on various entities, attributes, and relationships in the distribution network field.

[0114] Neo4j is a graph database that stores data in a graph format, where nodes represent entities and edges represent relationships between entities. This data storage method is well-suited for expressing complex relationships and is therefore highly suitable for storing knowledge graphs of multi-source heterogeneous data in distribution networks. It has significant application value in metering management scenarios. It can integrate data from different business systems, providing managers with comprehensive and accurate decision support and business analysis, helping to optimize distribution network operations and management, and promoting the optimized development of the entire distribution network system. Specifically:

[0115] S3.1. Real-time monitoring and early warning: Utilize the multi-source data of the distribution network integrated in the knowledge graph to monitor the status of metering equipment in real time;

[0116] In this step, real-time monitoring and early warning in the measurement management application specifically include:

[0117] First, it is necessary to integrate heterogeneous data from multiple distribution network sources, originating from various systems and data sources. These data sources may include marketing management systems, metering automation systems, intelligent turnover cabinets, and error prevention and control systems. Data preprocessing and fusion are required to convert the data into a unified data format.

[0118] The accessed data needs to be transmitted and processed in real time; data stream processing technology is used to ensure that the data can be transmitted to the knowledge graph system in a timely manner and processed in real time;

[0119] In the knowledge graph, the logic for real-time monitoring of metering equipment status is implemented through defined rules and query statements. For example, rules can be set to monitor whether the current of metering equipment exceeds a preset threshold or whether a meter displays abnormal data. When the warning conditions are met, the system automatically triggers the warning mechanism and sends a warning message to the relevant personnel, notifying them to promptly address the abnormal situation.

[0120] S3.2. Abnormal data investigation: Quickly locate abnormal measurement data through interactive display of knowledge graph;

[0121] In this step, abnormal data troubleshooting in the metering management application includes:

[0122] During the model layer construction phase, define the types of abnormal data in metering management; this may include meter offline, voltage abnormality, current abnormality, meter reading abnormality, and arrears.

[0123] Determine the characteristics and attributes of abnormal data for subsequent knowledge extraction and identification.

[0124] S3.3. Data quality management: Ensure data quality and consistency through data preprocessing and knowledge extraction during the data layer construction process; the data in the knowledge graph can be used as a reference for the data quality management platform.

[0125] S3.4. Statistical analysis and decision support: The knowledge graph of multi-source heterogeneous data in the distribution network can integrate data from different systems and conduct comprehensive statistics and analysis on the metering data of public transformers, special transformers, and low-voltage users.

[0126] In this embodiment, during the implementation of the above method, entities are first labeled and the extracted entities are associated with corresponding entity types. For example, the extracted cable entity is labeled as a "cable" type.

[0127] Secondly, the relationships are labeled and the extracted relationships are associated with the corresponding relationship types. For example, the identified connection relationship is labeled as the "connection" type.

[0128] Then, using natural language processing technology, we perform entity extraction on structured and semi-structured data. For example, we can extract entities (such as cables and transformers) from the data by identifying specific names and attribute information.

[0129] Thirdly, a relation extraction algorithm is used to identify the relationships between entities from the data. For example, by analyzing keywords such as connection and belonging in the data, the association relationship between entities can be determined.

[0130] Then, the knowledge obtained from different data sources and different knowledge extraction methods is integrated to eliminate duplication and conflict and form a consistent knowledge representation. Knowledge integration helps to improve the integrity and consistency of the knowledge graph.

[0131] Finally, the labeled and represented entity-relationship-entity triples are stored in a graph database, such as Neo4j. The advantage of a graph database is that it can efficiently store entities and their relationships, presenting them in a graphical structure for easy query and analysis.

[0132] Furthermore, entities and relationships are imported into the graph database to form an initial knowledge graph. At this stage, the graph may not be complete and needs further improvement and expansion.

[0133] The graph is expanded and updated based on actual needs. Regular data collection and processing can be used to continuously update the entity and relationship information in the graph to ensure consistency and accuracy with actual distribution network data.

[0134] Leveraging the query and analysis capabilities of graph databases, we can explore and mine knowledge graphs. By querying the relationships between specific entities, we can analyze the topology of the distribution network and identify potential problems and optimization solutions.

[0135] Use visualization tools to display the knowledge graph in graphical form. Through graphical display, you can more intuitively understand the entities and relationships in the distribution network, helping decision makers make more informed decisions.

[0136] Specifically, this embodiment is applied to metering management. Through the interactive display of the knowledge graph and the application of the data quality management platform, it can automatically troubleshoot metering data anomalies. This includes issues such as offline meters, voltage and current anomalies, etc. The system can automatically identify abnormal data and intuitively display anomaly analysis results and early warning information, allowing metering managers to discover and resolve problems in a timely manner.

[0137] Furthermore, data fusion and knowledge extraction technologies within the knowledge graph of multi-source heterogeneous data in the distribution network make it efficient and accurate to extract the information required for metering management from multiple data sources. This will help metering management departments process large amounts of data more efficiently and reduce data processing errors.

[0138] Finally, the construction of a knowledge graph for multi-source, heterogeneous data in the distribution network provides detailed analysis of the transmission, substation, distribution, and user layers, accurately defining entity types, attribute types, and entity relationships. This provides comprehensive and accurate decision support for metering management departments, enabling more precise identification and statistical analysis of meter pressure loss, current loss, phase failure, reverse polarity, meter reading anomalies, arrears, and metering file errors at public and specialized transformers, and low-voltage users.

[0139] At the same time, through data fusion and application integration with various business systems, the distribution network's multi-source heterogeneous data knowledge graph can better share and interact with marketing management systems, metering automation systems, intelligent turnover cabinets, error prevention and control systems, data quality management platforms, etc., further improving data quality management efficiency.

[0140] In summary, the present invention realizes the effective organization and management of distribution network knowledge by deeply analyzing the multi-source heterogeneous data of the distribution network and applying the knowledge graph construction method. Through the construction of the model layer, from the transmission network to the user layer, the distribution network knowledge is summarized and analyzed, the entity type, attribute type and entity relationship are determined, and the organizational framework of the distribution network knowledge is constructed. In the construction of the data layer, the multi-source heterogeneous data fusion technology is used to associate and integrate the distribution network data from different data sources to form the basis of the knowledge graph. Subsequently, through data preprocessing, the multi-source heterogeneous data is cleaned, denoised and normalized to ensure data quality and consistency. Using knowledge extraction technology, entity and relationship information is extracted from the preprocessed data and converted into the "entity-relationship-entity" triple form to form the basic structure of the knowledge graph. The knowledge fusion process eliminates the duplication and conflict of different data sources and extraction methods, and improves the integrity and consistency of the knowledge graph.

[0141] Through the above construction process, the present invention obtains a knowledge graph of multi-source heterogeneous data of the distribution network. The knowledge graph presents the distribution network information in the form of a graphical structure and is stored in a graph database such as Neo4j. This makes the storage, query and update of knowledge efficient and convenient, providing strong support for the operation and management of the distribution network. The application of the knowledge graph can explore and mine the knowledge of the distribution network, support intelligent decision-making and optimization, provide intelligent user services, and quickly troubleshoot and respond to faults, thereby realizing the intelligence and optimization of distribution network management. At the same time, the maintenance and update of the graph ensures the continued effectiveness of the knowledge graph, continuously optimizes its structure and performance, and maintains consistency and accuracy with the actual data.

[0142] In terms of metering management, the solution's knowledge graph for multi-source heterogeneous data in the distribution network has significant applications. Through the interactive display of the knowledge graph and the application of the data quality management platform, it is possible to automatically troubleshoot abnormal metering data, intuitively display abnormal analysis results and issue warnings, and automatically and synchronously send overdue warning information to the corresponding work person in charge. At the same time, by integrating system data such as the marketing management system, metering automation system, intelligent turnover cabinet, error prevention and control system, and data quality management platform, it is possible to identify and statistically analyze situations such as loss of pressure, loss of current, phase failure, reverse polarity, meter reading anomalies, arrears, and metering file errors in public transformers, special transformers, and low-voltage users. These functions and applications provide comprehensive and accurate decision support and business analysis for metering management departments, help optimize distribution network operation management, and improve the efficiency of data quality management.

[0143] Overall, this solution constructs a complete, efficient, and accurate knowledge graph by deeply analyzing multi-source heterogeneous data of the distribution network, providing all-round support and guidance for the management and operation of the distribution network, and has broad application prospects and economic benefits.

[0144] As shown in FIG5 , this embodiment also provides a platform device for constructing a multi-source heterogeneous data knowledge graph, which includes a processor, a memory, and a computer program stored in the memory and running on the processor.

[0145] The processor includes one or more processing cores, and the processor is connected to the memory through a bus. The memory is used to store program instructions. When the processor executes the program instructions in the memory, the steps of the method for constructing the knowledge graph of multi-source heterogeneous data of the distribution network are implemented.

[0146] Alternatively, the memory may be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0147] In addition, the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the above-mentioned method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network.

[0148] Optionally, the present invention also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the steps of the method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network according to the above aspects.

[0149] Those skilled in the art will appreciate that the process of implementing all or part of the steps of the above embodiments may be accomplished by hardware, or may be accomplished by instructing the relevant hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk, or an optical disk, etc.

[0150] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a knowledge graph of multi-source heterogeneous data in a distribution network, characterized in that: The steps include: S1. Model layer construction: By adopting a top-down approach, the knowledge content system framework of the transmission network is summarized and analyzed; by analyzing the multi-source heterogeneous data of the distribution network in detail, starting from the transmission layer, substation layer, distribution network layer and user layer, the entity type, attribute type and entity relationship are accurately defined, and the organizational framework of the distribution network knowledge is constructed; Specifically: S1.

1. Summarization and analysis of the knowledge content system framework of the power transmission network: First, summarize and analyze the knowledge content in the power transmission network field to understand the important concepts, key attributes and relationships between entities in this field; S1.

2. Detailed analysis of multi-source heterogeneous data of distribution network: After understanding the knowledge content of transmission network, detailed analysis of multi-source heterogeneous data of distribution network is carried out; S1.

3. Define entity types, attribute types and entity relationships: Based on the analysis of multi-source heterogeneous data, determine the entity types, attribute types and entity relationships of the distribution network knowledge graph; S1.

4. Construct the organizational framework of distribution network knowledge: After determining the entity type, attribute type and entity relationship, construct the organizational framework of distribution network knowledge by organizing them; S2. Data layer construction: The knowledge graph of multi-source heterogeneous data of distribution network can be represented by G = (E, R, S), where E is the entity set in the knowledge base, R is the relationship set, and S is the "entity-relationship-entity" triple knowledge set; data layer construction is to extract the required entities and relationships from the multi-source heterogeneous data of distribution network under the guidance of the model layer organization framework; the key technologies of data layer construction include multi-source heterogeneous data fusion, data preprocessing, knowledge extraction, knowledge fusion and knowledge storage and update; specifically: S2.

1. Multi-source heterogeneous data access and fusion: Obtain the original data, and generate structured data, semi-structured data, and unstructured data by accessing multi-source heterogeneous data of different types of distribution networks; then, use multi-source heterogeneous data fusion technology to aggregate, associate, and integrate various data from different data sources of the distribution network; S2.2, Data preprocessing: Preprocess the fused multi-source heterogeneous data, including cleaning, denoising, and normalizing structured and semi-structured data to ensure data quality and consistency; perform structured conversion, Chinese word segmentation, text annotation, and word vector model construction on unstructured data; S2.3, Knowledge Extraction: Use rules and deep learning-based knowledge extraction to extract knowledge from preprocessed data Required entity and relationship information; S2.4, Knowledge Fusion: Fusion of knowledge obtained from different data sources and different knowledge extraction methods, eliminating duplication and conflict, and forming a consistent knowledge representation; S2.5, Knowledge storage and update: Finally, the knowledge storage technology of Neo4j graph database is used to store the extracted and integrated knowledge to form a knowledge graph of multi-source heterogeneous data of distribution network; at the same time, the knowledge graph is updated regularly; S3. Metering management application: Using the Neo4j graph database to store distribution network knowledge can form a complete and structured knowledge graph, which contains information on various entities, attributes and relationships in the distribution network field; specifically: S3.

1. Real-time monitoring and early warning: Using the multi-source data of the distribution network integrated in the knowledge graph, real-time data monitoring of the status of metering equipment; S3.

2. Abnormal data investigation: Through the interactive display of knowledge graph, abnormal measurement data can be quickly located; S3.3, Data quality management: Ensure data quality and consistency through data preprocessing and knowledge extraction in the process of data layer construction; the data in the knowledge graph can be used as a reference for the data quality management platform; S3.

4. Statistical analysis and decision support: The knowledge graph of multi-source heterogeneous data in the distribution network can integrate data from different systems and conduct comprehensive statistics and analysis on the metering data of public transformers, special transformers, and low-voltage users.

2. The method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network according to claim 1 is characterized by: In step S1.3, entity types, attribute types and entity relationships are defined in the model layer construction; specifically: Entity types refer to various specific objects in the distribution network; Attribute types are the characteristics and properties of these entities; Entity relationships represent the associations between entities.

3. The method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network according to claim 1, characterized in that: In step S2.1, when multi-source heterogeneous data is accessed in the data layer construction, the multi-source heterogeneous data can be divided from different dimensions, including time dimension, data structure dimension and business field dimension; specifically including: In the time dimension, real-time data: data collected immediately, reflecting the real-time status of distribution network equipment, lines, and loads; historical data: data collected over a period of time in the past, used to analyze the operation history and trends of the distribution network; forecast data: data predicted based on historical data and models, used to predict future power demand and load conditions; In the data structure dimension, structured data: data stored in a fixed format and mode; semi-structured data: data with a certain structure but not in the form of a traditional relational database table; unstructured data: data without a fixed format and mode; The business area dimensions include: substation data, distribution equipment data, power line data, user electricity consumption data, power quality data, and operation and maintenance data.

4. The method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network according to claim 3 is characterized by: In step S2.1, when merging multi-source heterogeneous data in the data layer construction, the following three key aspects are included: data association, multi-source heterogeneous data estimation, and data source management; specifically, the following: First, through data association technology, the distribution network data from different data sources are associated to find out the relationship between them, ensuring that they reflect the same distribution network entity or event, avoiding data redundancy and duplication; Secondly, multi-source data estimation technology is used to integrate information from different data sources to improve the estimation of distribution network targets; In addition, data source management technology optimizes operating costs by reasonably allocating data collection and processing sources based on the environmental status of a given data source; For multi-source data in the distribution network, tasks can be flexibly assigned based on the location and performance factors of different sensors to improve the efficiency and performance of data collection and processing.

5. The method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network according to claim 1, characterized in that: In step S2.2, data preprocessing in data layer construction includes the following steps: First, for structured data: use data cleaning tools to clean the data to remove duplicate data, missing data and erroneous data to ensure the accuracy and completeness of the data; use filtering technology to process noisy data and perform data denoising; And convert the data into a unified format to normalize the data; Second, for unstructured data: use OCR technology, docx2txt in Python, PyPDF2 library or text extraction tools to convert unstructured data into a processable text format; Label each character of the data according to the BIO labeling method, and use the labeled corpus as the input raw corpus of the BERT word vector conversion layer; And use the pre-trained model BERT to map the annotated original corpus into dynamic word vectors; Finally, use the Chinese word segmentation tool to segment the Chinese text and divide the continuous Chinese text into meaningful words.

6. The method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network according to claim 1, characterized in that: In step S2.3, knowledge extraction in data layer construction includes entity extraction, relationship extraction, and event extraction; specifically, the following steps are included: Step 1: Use rules and deep learning-based knowledge extraction methods. Rule-based key information extraction has the advantages of precise range and high accuracy. Using this advantage, write certain rules to match a small number of precise objects as subsequent corpus import; Step 2: Enable sequence annotation to continue extracting key information, and part of the results output in Step 1 can be used as training corpus to replace the process of manually injecting corpus; Step 3: Use the training corpus output in Step 2 to perform knowledge graph modeling based on the BERT-BiLSTM-CRF algorithm; Step 4: Use the model in Step 3 to judge the remaining corpus in Step 2. If the model judgment result shows that the model does not meet the standard, return to Step 1 to add more corpus. Until the model automatically determines that it meets the standard, exit the iteration and use the most recently generated model as the final model.

7. The method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network according to claim 1, characterized in that: In step S2.4, knowledge fusion in data layer construction is to represent multiple entities with the same meaning with a single entity, that is, A = {a1, a2, a3, ..., a m }, where a1~a m Represents m entities with the same meaning, and A is the fused entity. Use the text clustering algorithm to perform knowledge fusion, that is, traverse each entity, calculate the text similarity between the entity and the rest of the entities, and merge the entities with similarity higher than the threshold. The calculation formula of entity semantic similarity T is shown in the formula: In the formula, A i , B i are the word frequency vectors obtained after the two entity names A and B are segmented, and n is the number of segmented words; the higher the semantic similarity T, the higher the similarity between the two entities; Finally, by extracting the "entity-relationship-entity" triples, these knowledge elements are effectively encoded and Representation, thus constructing the knowledge graph of multi-source heterogeneous data of distribution network; This knowledge graph organically organizes the distribution network information and presents it in the form of a graphical structure; the constructed knowledge graph is stored in the Neo4j graph database to facilitate the storage and query of knowledge.

8. The method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network according to claim 1, characterized in that: In step S3.1, real-time monitoring and early warning in the metering management application specifically include: First, it is necessary to connect the multi-source heterogeneous data of the distribution network from different systems and data sources; it is necessary to pre-process and integrate the data to convert it into a unified data format; The accessed data needs to be transmitted and processed in real time; data stream processing technology is used to ensure that the data can be transmitted to the knowledge graph system in a timely manner and processed in real time; In the knowledge graph, the logic of real-time monitoring of the status of metering equipment is implemented by defining rules and query statements.

9. The method for constructing a knowledge graph of multi-source heterogeneous data of a distribution network according to claim 1, characterized in that: In the step S3.2, when checking abnormal data in the metering management application, it includes: In the model layer construction phase, define the abnormal data types in metering management; Determine the characteristics and attributes of abnormal data for subsequent knowledge extraction and identification.

Images