Knowledge graph platform supporting data integration analysis and visualization of power system

Abstract

The invention discloses a knowledge graph platform supporting data integration analysis and visualization of an electric power system, and the platform comprises an electric power data preprocessing module which is used for extracting key information from documents and records of the electric power system through a natural language processing technology; the knowledge extraction module is used for extracting entities, relationships and attributes from the data source, constructing a triple and storing the triple into a knowledge graph; the data integration module is used for calculating the similarity among the attributes, the entities and the sub-graphs by using a model according to a preset requirement, identifying the ontologies needing to be aligned, and completing the integration of the entities; the graph construction module is used for constructing a knowledge graph in a dragging manner or from a public ontology library; the graph representation module is used for integrating a natural language processing function, providing a query service for a user in a natural language form, analyzing and querying through a natural language understanding technology, identifying a key entity and a user intention, and calling a graph algorithm to perform data retrieval and analysis; the user interface module is used for providing a user interface and providing user access and platform management, statistics and integration functions; the operation efficiency and the management level of the power system are improved.

Description

Technical Field

[0001] The present invention relates to a knowledge graph platform, and in particular to a knowledge graph platform that supports power system data integration analysis and visualization, belonging to the technical field of power grid regulation. Background Art

[0002] As the complexity of power systems increases, the demand for data processing and knowledge management of power systems is increasing. Traditional data processing methods, such as rule-based data processing and simple statistical analysis, can no longer meet the needs of modern power systems. Knowledge graph is a method for representing and organizing knowledge, which models entities, concepts and their relationships through a graphical structure. In the field of power systems, knowledge graphs can be used to represent power grid topology, equipment, parameters and events, etc., so as to achieve a comprehensive and multi-dimensional understanding and analysis of power systems. In power systems, data from different sources, such as monitoring data, operating data, equipment parameters, etc., are integrated and uniformly managed, and complex power grid operation status and analysis results are displayed in an intuitive way through visualization technology, which is convenient for researchers and engineers to observe and make decisions.

[0003] In general, the application of knowledge graph technology in the field of power grid regulation has achieved initial results, but at this stage, the application results of knowledge graphs in the field of power system regulation are relatively scattered, resource sharing is poor, algorithm models, data, and experience are difficult to share globally in the regulation field, and the depth and breadth of application results are insufficient. The scope and degree of support for regulation business are still insufficient. It is urgent to combine business needs and technical maturity to coordinate system design and research and application from the aspects of models, samples, platforms, operating mechanisms, and open ecosystems. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a knowledge graph platform that supports power system data integration analysis and visualization, which can improve the operating efficiency and management level of the power system.

[0005] Technical solution: The knowledge graph platform supporting power system data integration analysis and visualization described in the present invention includes:

[0006] Power data preprocessing module, which uses natural language processing technology to extract key information from power system documents and records;

[0007] The knowledge extraction module is used to extract entities, relations and attributes from the data source, construct triples and store them in the knowledge graph;

[0008] The data integration module is used to calculate the similarity between attributes, entities and subgraphs using the model according to the preset requirements, identify the entities to be aligned, and the user verifies the identification results. The platform merges or links the verified results to complete the entity integration.

[0009] Graph building module, used to build knowledge graphs by dragging and dropping or from public ontology libraries;

[0010] The graph representation module is used to integrate natural language processing functions, provide users with services for querying in natural language, parse queries through natural language understanding technology, identify key entities and user intent, and call graph algorithms for data retrieval and analysis;

[0011] The user interface module is used to provide a user interface, allowing users to access and utilize platform management, statistics and integration functions; the user interface module specifically includes: knowledge graph platform management interface function, knowledge graph platform statistics interface function, and knowledge graph platform graph integration function.

[0012] Furthermore, the power data preprocessing module is specifically implemented as follows:

[0013] S1.1: Add the professional terms, equipment names and technical vocabulary in the power field corresponding to the power data to the custom dictionary of the word segmentation tool, and select the word segmentation algorithm that supports Chinese and English for word segmentation;

[0014] S1.2: Use predefined rule sets to perform part-of-speech tagging, and then automatically identify the part-of-speech through the training model. Combining the predefined rule set and the training model method can improve the accuracy and robustness of part-of-speech tagging;

[0015] S1.3: Use a time series data analysis model based on deep learning to learn the time dependencies in the data and predict future data trends;

[0016] S1.4: Use visual text annotation tools to annotate the entities, relationships, and attributes to be extracted from the text file to be annotated, obtain the annotation results, and use the active learning model to learn the annotation rules.

[0017] Furthermore, the annotation result file in step S1.4 is in ann format, and step S1.4 specifically includes:

[0018] S1.4.1: Upload the text file to be annotated. The text file to be annotated is uploaded by the user. The file format supports txt format and the file encoding method is UTF-8. When the uploaded file is a new file, a corresponding blank ann file is automatically generated. The blank ann file encoding method is UTF-8.

[0019] S1.4.2: Perform text annotation, wherein the annotation method includes manual text annotation and automatic text annotation;

[0020] The manual text annotation method is to select the type or attribute of the text annotation entity on the page, and to drag from one entity tag to another entity tag on the page to annotate the relationship that meets the constraints, including:

[0021] Entity and attribute annotation: When a text paragraph on a page is selected, an annotation dialog box will automatically appear, which is used to give entity annotation or attribute annotation to the selected text; after the annotation is completed, the text paragraph will be highlighted in a preset color and accompanied by a corresponding annotation label; the annotation label can be edited or deleted;

[0022] Relationship annotation: When a user selects an annotated entity and drags it to another annotated entity, it is determined whether there is a preset relationship between the two entities. If there is a preset relationship, a connection line will be immediately presented, and a relationship annotation dialog box will pop up for relationship annotation. The annotation result can be triggered by the connection line. If there is no preset relationship between the two entities, a prompt will be given; the annotation result can be edited or deleted;

[0023] The automatic text annotation method is to first perform fine word segmentation on the text to split the text into multiple words or phrases; then, using similarity matching technology, each word or phrase in the text is compared with the existing annotation data to determine its corresponding entity and relationship; during the annotation process, the system automatically generates a mark for each annotation result for user confirmation; the annotation information confirmed by the user is added as new annotation data to the ann file that has been annotated;

[0024] S1.4.3: Check annotations according to preset standards; use preset colors to mark incorrect annotations identified according to the standards for user identification and correction; and use different colors to distinguish annotation contents according to the entity category to which they belong;

[0025] The standards include:

[0026] The annotated entity categories and attributes are all reflected in the established configuration and there is no missing;

[0027] The relationship type of the annotation needs to be defined in the configuration;

[0028] The annotated relationships follow the entity types specified in the configuration, and any behavior that does not conform to or changes the configuration without authorization is defined as an error;

[0029] S1.4.4: Download of annotated datasets: Set up a dataset access page to provide users with a download function. According to the user's download requirements, the source text files of the entire dataset and the ann files after annotation are automatically packaged, or the specified source text files and ann files after annotation are selected according to the user's needs, and the option of generating a json format file is provided for the user to download.

[0030] Furthermore, the data source of the knowledge extraction module includes structured data source, semi-structured data source and unstructured data source;

[0031] The extraction of structured data sources supports users to build graphs by configuring mapping rules; by parsing the file content and combining the mapping configuration, the information in the file is converted into concepts and relationships in the graph; specifically, it includes:

[0032] S2.1.1: Set mapping rules: The structured data includes files, rows and columns. The configuration rules of entities, relationships and attributes are written according to the field names of the columns. All values ​​in the mapping process are treated as strings. When the file name does not contain a hyphen, the file name is used as the name of the entity type, and the field name in the file is used as the attribute name of the entity. When the file name contains a hyphen, the file name is decomposed into triples according to the hyphen, which are used as the head entity type, the relationship type and the tail entity type respectively. In the file, the three fields are used as the head entity instance, the relationship instance and the tail entity instance of the triple respectively.

[0033] S2.1.2: Concept mapping: Use mapping rules to map the fields in the file to the core concepts of the graph ontology and the specific types of its nodes; specifically, the user specifies the file type and data format in the file selection area, and selects the target file to be mapped in the opened file browser. Then, when the platform receives the user's command, it triggers the automatic mapping process;

[0034] S2.1.3: Relationship mapping: Use mapping rules to select file type and data format, select the file to be mapped from the file browser; issue a mapping command to complete the mapping;

[0035] The extraction of the semi-structured data source provides a knowledge extraction function according to a template, allowing the user to define a template containing parts of speech to extract entities and relationships in the text; the template is specifically a regular expression written according to the content of the text of the knowledge to be extracted;

[0036] The unstructured extraction, based on text recognition in images using OCR technology, is associated with the dispatching automation system equipment to form equipment parameter knowledge; specifically includes:

[0037] S2.3.1: Image preprocessing: grayscale the image to make any pixel element of the color image satisfy the relationship R=G=B; then use binarization to make the grayscale value of each element in the pixel matrix of the image have two different values; use the Hough transform method to realize automatic image tilt and rotation correction; use the Layout-Parser text layout analysis toolkit to identify four types of blocks from the limit image file: title, text paragraph, table and picture, extract the table area from the power grid knowledge data, and generate a preprocessing result image file that meets the image recognition requirements;

[0038] S2.3.2: Image semantic segmentation: Through semantic segmentation, the precise shape and position of each object in the image are obtained, and the main equipment objects in the photo are extracted; specifically, the encoder gradually reduces the spatial size of the feature map to capture contextual information, and the decoder restores the spatial size of the feature map for pixel-level classification; introduces the hole convolution mechanism to capture contextual information by increasing the receptive field of the convolution kernel; uses the loss function CTC of the semantic segmentation task, inputs the preprocessed data into the network model, and adjusts the network parameters through the back-propagation algorithm to minimize the loss function; during the training process, uses the validation set to monitor the performance of the model and adjusts the parameters as needed;

[0039] S2.3.3: Equipment association: Use optical character recognition (OCR) technology to process the image and extract text information from the equipment nameplate; convert the text information into numbered text and data format; retrieve the unique identification number of the equipment from the power equipment knowledge graph; compare and match the nameplate information recognized by OCR with the equipment number in the graph; when the match is successful, the platform automatically associates the nameplate information with the corresponding equipment record in the graph, and synchronously updates all relevant information about the nameplate in the image to the properties of the graph equipment.

[0040] Furthermore, the data integration module is specifically implemented as follows:

[0041] S3.1: Range screening: Filter entity data with similar fusion identifiers based on the entity type and attribute range configured by the user;

[0042] S3.2: Determine similar entities: Determine entity similarity based on the attribute similarity between entities; the determination method, when the entities belong to the same graph, uses similarity determination based on representation vectors; when the entities belong to different graphs, uses edit distance determination; specifically:

[0043] S3.2.1: Similarity calculation based on representation learning; First, the entities to be aligned are converted into feature vectors, and then the cosine similarity between vectors is used as the similarity of the entities; for two vectors A(x 11 ,x 12 ,…,x1n ) and B(x 21 ,x 22 ,…,x 2n ), their cosine similarity calculation formula is as follows:

[0044]

[0045] The value range of cosine similarity is defined as [-1, 1]. When the cosine value is closer to 1, it means that the two vectors are more similar in direction; when the cosine value is closer to -1, it means that the two vectors are less similar in direction.

[0046] S3.2.2: Similarity based on edit distance. For two strings a and b, with lengths |a| and |b| respectively, their edit distances are as follows:

[0047]

[0048] S3.3: Data integration: merge the graph data whose attribute similarity meets the threshold conditions, specifically:

[0049] S3.3.1: Set a threshold p to determine whether two attributes are similar enough to be fused;

[0050] S3.3.2: For each entity in the graph data to be fused, perform attribute matching and similarity calculation; traverse the entities and their attributes in the two graphs; for each attribute, calculate the similarity of the same type of attribute in the other graph; when an attribute with a similarity higher than a preset threshold is found, the two attributes are matched;

[0051] S3.3.3: For matching attributes, merge the corresponding data into a unified graph. For numeric attributes, take the average of the values; for text attributes, select a primary value; for relational attributes, merge the graph structure of the relationship.

[0052] Furthermore, the steps of dragging and dropping the graph construction module to construct the knowledge graph are as follows:

[0053] S4.1.1: Set up a quick creation page, follow the page instructions, click to create concept nodes, design concepts in the knowledge graph model layer, and edit concept information and its attribute information;

[0054] S4.1.2: In the quick creation page, follow the page instructions to create relationships between concept nodes by dragging and clicking, and edit the relationship information and its attribute information;

[0055] The steps of constructing a knowledge graph from a public ontology library in the graph construction module are as follows:

[0056] S4.2.1: In the Create from Public Ontology Library page, select the required public ontology according to the page guide. The public ontology includes several ontology concepts and ontology relationships as the basis of the pattern layer of the new application;

[0057] S4.2.2: After selecting a public ontology, preview the ontology concepts and ontology relationships contained therein, and copy the contents of the public ontology into the private application;

[0058] S4.2.3: Confirm your selection and add a new private knowledge graph application.

[0059] Furthermore, the graph representation module provides representations for graph elements of different granularities by providing node encoders, edge encoders, and pseudo-sentence encoders;

[0060] The node encoder representation: uses the GraphSAGE algorithm, specifies the encoding dimension as d, samples neighbor nodes and aggregates features to learn the vector representation of the node, including:

[0061] S5.1.1: Sampling neighbors: For each target node, randomly sample a fixed number m of neighbor nodes from its neighbors;

[0062] S5.1.2: Aggregate neighbor information: Use the pooling aggregator to aggregate the features of the sampled m neighbor nodes. When the number of neighbors is less than m, all of them are used;

[0063] S5.1.3: Update node embedding: Update the node embedding by taking the weighted sum of the aggregated neighbor features and the target node’s own features, and passing it through a neural network layer;

[0064] S5.1.4: Repeat and optimize: Repeat steps S5.1.1 to S5.1.3 for all nodes in the graph according to the preset number of iterations, and train the model parameters through the optimization method of back propagation and gradient descent;

[0065] The edge encoder representation: constructs a dual graph of the original graph, converts the edges of the original graph into nodes of the new graph, converts the nodes of the original graph into edges of the new graph, and uses the GraphSAGE algorithm to obtain the encoding vector of each node on the new graph, specifically including:

[0066] S5.2.1: Collect the representations of nodes and edges in the original graph. Let the original graph G = (V, E), where V is the set of nodes and E is the set of edges. Specify the encoding dimension as d.

[0067] S5.2.2: Define the nodes and edges of the new dual graph: A node is an edge in the original graph that has a corresponding node in the dual graph. When two edges e1 and e2 in the original graph share a node v, there is an edge connecting e1 and e2 in the dual graph.

[0068] S5.2.3: Use the GraphSAGE algorithm to obtain the encoding vector of each node in the dual graph as the encoding of the edge in the corresponding original graph;

[0069] The triple encoder representation: for each triple in the graph, provide its corresponding encoding vector, specify the encoding dimension as d, specifically including:

[0070] S5.3.1: For the triple {h, r, t} in the graph, use the node encoder and edge encoder to map {h, r, t} into vectors (v_h, v_r, v_t) with dimensions (d_h, d_r, d_t) respectively;

[0071] S5.3.2: Construct an autoencoder with an encoder input dimension of (d_h+d_r+d_t) and an output dimension of d; a decoder with an input dimension of d and an output dimension of (d_h+d_r+d_t);

[0072] S5.3.3: Use the full-image triplet data to train the autoencoder network in step S5.3.2 to obtain its parameters;

[0073] S5.3.4: For each triplet {h, r, t}, use the output of the encoder network as its encoding vector.

[0074] Furthermore, the knowledge graph platform management interface function of the user interface module is used to provide users with concept and relationship definition, editing, import and export management functions, and to construct a knowledge graph model layer through a process-based guided drag-and-drop editing method; specifically, it includes:

[0075] Application definition unit: defines the namespace of the entire model layer, which serves as isolation between different knowledge graphs;

[0076] Concept definition unit: used to define the type of entity. After the application definition is completed, a new concept is created, and the necessary elements of the concept are edited, and the attributes of the concept are defined simultaneously; the definition and attributes of the concept can be edited;

[0077] Relationship definition unit: define the relationship between concepts, generate a connection line from the concept node, drag it to another concept node or itself to form a relationship between concepts, edit the necessary elements of the relationship, and choose whether to define the attributes of the relationship synchronously;

[0078] Attribute definition unit: defines the attributes of concepts and relationships, and configures the characteristics of attributes;

[0079] Import and export unit: supports file mapping and preset pattern library, supports importing ontology from csv, xlsx, json format files, supports importing multiple files; when importing concepts, specifies the fields of concepts, automatically detects each concept, configures attribute fields, attribute types, and characteristics for each concept, and supports exporting the defined ontology into json format files;

[0080] Visualization unit: Displays the hierarchical and associative relationships between ontologies in a variety of visual ways, supports visualization and drag-and-drop construction of ontologies; supports the creation, editing, copying and deletion of concepts; the concepts include concept names, icons and attributes; attributes can be configured with single-value and data types.

[0081] Furthermore, the knowledge graph platform statistics interface function of the user interface module is used to provide statistics for the application graph analysis of the distribution of existing ontologies, instances and degrees; specifically, it includes:

[0082] Overall statistical unit: used to count the number of ontology concept types, relationship types, instance nodes, and instance relationships, and displayed in a bar chart;

[0083] Instance type distribution unit: used to count the number of instances by type. When the number of instance types exceeds the preset threshold, the statistics are divided into buckets and displayed in a bar chart.

[0084] Relationship type distribution unit: used to count the number of relationships by type. When the number of relationship types exceeds the preset threshold, the statistics are divided into buckets and displayed in a bar chart.

[0085] Degree distribution unit: The degree is the number of edges on a node, which is counted by degree distribution density and displayed in a bar graph;

[0086] Drill-down function unit: used to display the node and relationship instances of the column statistics.

[0087] Furthermore, the knowledge graph platform graph integration function of the user interface module is used to integrate the selected knowledge graph through the knowledge graph platform configuration and graph algorithm. The specific integration process is as follows:

[0088] S6.3.1: Configure alignment items, including selecting the graphs to be integrated, the attributes of the graph concepts to be aligned, the attribute integration method of the entities in the instance, and the integration method of the relationships in the instance;

[0089] S6.3.2: Query graph data. Enter the search content in the search box, search for relevant instance data in multiple knowledge graphs to be aligned, and return them for display;

[0090] S6.3.3: Graph integration, call the data integration module, merge the instance data in multiple graphs according to the integration method in the configuration, and return the integrated graph.

[0091] Beneficial effects: Compared with the existing technology, the present invention seizes the fact that the knowledge graph platform can integrate various data sources in the power system, including power grid operation data, equipment monitoring data, energy market data, etc., to achieve comprehensive coverage and in-depth mining of knowledge in the power field; it helps power companies to have a more comprehensive understanding of the system operation status, discover potential problems, and formulate corresponding solutions; the present invention ensures that the integrated data remains consistent in format and quality through pre-processing operations such as data cleaning, deduplication, and format conversion, provides a unified data source for subsequent analysis and visualization, reduces the complexity and error rate of data processing, and improves the quality and availability of data; the present invention uses advanced technologies such as machine learning and data mining to conduct in-depth analysis of power data to discover hidden patterns and correlations; by mining and utilizing this information, it can achieve intelligent prediction of the power system operation status and fault warning, and provide strong support for the decision-making of power companies. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 Schematic diagram of the system structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0093] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The present invention adopts advanced natural language processing technology and graph algorithms to effectively integrate and manage various data and knowledge in the power system.

[0094] As attached Figure 1 As shown, the knowledge graph platform supporting power system data integration analysis and visualization in this embodiment includes:

[0095] Power data preprocessing module, which uses natural language processing technology to extract key information from power system documents and records;

[0096] The knowledge extraction module is used to extract entities, relations and attributes from the data source, construct triples and store them in the knowledge graph;

[0097] The data integration module is used to calculate the similarity between attributes, entities and subgraphs using the model according to the preset requirements, identify the entities to be aligned, and the user verifies the identification results. The platform merges or links the verified results to complete the entity integration.

[0098] Graph building module, used to build knowledge graphs by dragging and dropping or from public ontology libraries;

[0099] The graph representation module is used to integrate natural language processing functions, provide users with services for querying in natural language, parse queries through natural language understanding technology, identify key entities and user intent, and call graph algorithms for data retrieval and analysis;

[0100] The user interface module is used to provide a user interface, allowing users to access and utilize platform management, statistics and integration functions; the user interface module specifically includes: knowledge graph platform management interface function, knowledge graph platform statistics interface function, and knowledge graph platform graph integration function.

[0101] Specifically, 1. Power data preprocessing module: This module uses natural language processing technology, including text segmentation, part-of-speech tagging, and entity recognition. Through these steps, key information is extracted from power system-related documents and records, including equipment parameters, operation records, etc. The specific steps are:

[0102] S1.1 Text segmentation: For power data, add professional terms in the power field, equipment names, technical vocabulary, etc. to the custom dictionary of the word segmentation tool, and select a word segmentation algorithm that supports Chinese and English for word segmentation.

[0103] S1.2 Part-of-speech tagging: Use predefined rule sets to perform part-of-speech tagging, and then automatically identify the part of speech through training models. Combine the predefined rule sets and training model methods to improve the accuracy and robustness of part-of-speech tagging.

[0104] S1.3 Entity Recognition: Select a deep learning model to pre-process the power data. The data in the power system has time series and correlation. Use a time series data analysis model based on deep learning to learn the time dependency in the data and predict future data trends.

[0105] S1.4 Text annotation function: The knowledge graph platform provides a visual text annotation tool for users to annotate the entities, relationships, and attributes they want to extract on the page, and uses active learning models to learn annotation rules and reduce manual repetitive work.

[0106] Using the text annotation function includes the following steps:

[0107] Step S1.4.1: Text annotation includes the original text file to be annotated and the annotation result file, and the annotation result file is in ann format. The ann file contains detailed information on the annotation of the text in the txt file, and each line corresponds to an annotation item. T represents the annotation type, such as entity or relationship, ID is the unique identifier of the annotation, Span is the text range of the annotation, the format is the start offset and the end offset, and Text is the annotated text content.

[0108] Step S1.4.2: Upload the text file to be annotated. The original annotated text is uploaded by the user. The user creates a new text annotation dataset and fills in the dataset name and dataset description. Click the dataset card and upload the text file to be annotated in the new page. The file format supports txt format and the file encoding is UTF-8. When the uploaded file is a new file, the corresponding blank ann file is automatically generated, also encoded in UTF-8.

[0109] Step S1.4.3: Manual text annotation. Mark the type or attribute of the text annotation entity on the page, and drag from one entity tag to another entity tag on the page to mark the relationship that meets the constraints. The specific steps are as follows;

[0110] Entity and attribute annotation: When a text paragraph on a page is selected, the system will automatically present an annotation dialog box. Users can assign entity annotations or attribute annotations to the selected text through mouse operations or specific shortcut keys. After the annotation is completed, the text paragraph will be highlighted in a preset color and accompanied by the corresponding annotation label. If you need to edit or remove the annotation, the user can double-click the entity label on the page, and the system will immediately display a label dialog box, allowing the user to modify or delete the original annotation.

[0111] Relationship annotation: On the page, when the user selects an annotated entity and drags it to another annotated entity, the system will first determine whether there is a preset relationship between the two entity types. If there is a preset relationship, the connection line will be immediately presented, and the relationship annotation dialog box will be triggered to pop up for further operation by the user. However, if there is no preset relationship between the two entity types, the system will display a floating prompt to clearly inform the user of the result of the current operation.

[0112] In addition, users can also double-click an existing relationship line on the page to trigger the pop-up of the relationship annotation dialog box. In this dialog box, users have the authority to modify the original relationship or delete the relationship to meet the needs of different scenarios.

[0113] Step S1.4.4: Automatic text annotation. Automatic text annotation provides an efficient tagging method based on existing annotated data. The method first performs fine word segmentation on the text to split the text into multiple words or phrases. Then, using similarity matching technology, each word or phrase in the text is compared with the existing annotated data to determine its corresponding entity and relationship. During the annotation process, the system automatically generates a tag for each annotation result for user confirmation. The annotation information confirmed by the user will be added to the annotated ann file as new annotation data.

[0114] Step S1.4.5: Annotation check. When opening an annotated source file, the rationality of the annotation content must be strictly examined. Specifically, the following aspects should be ensured to meet the preset standards: 1) The annotated entity categories and attributes are reflected in the established configuration without any missing; 2) The annotated relationship types must also be defined in the configuration; 3) For annotated relationships, the entity types specified in the configuration must be strictly followed, and any non-compliance or unauthorized changes to the configuration will be considered an error. For identified erroneous annotation results, the system will use a striking red mark to facilitate user identification and correction. At the same time, in order to enhance the user experience, the correct annotation content will be distinguished by different colors according to the entity category to which it belongs, making the annotated text clearer and easier to read.

[0115] Step S1.4.6: Download the labeled dataset. After accessing the dataset page, the user can click the "Download dataset" button, and the system will automatically package the source text files and the labeled ann files in the entire dataset, and provide the option to generate a json format file for the user to download. In addition, if the user only needs to download part of the data, he can select the specified source text file in the dataset file list and click the "Download" button after the file. The system will download the text file and its corresponding ann file separately, and provide the option to generate and download the json format file.

[0116] 2. Knowledge extraction module: The knowledge graph platform has the ability to extract entities, relationships, and attributes from structured, semi-structured, and unstructured data sources, and then construct the required triples and store them in the knowledge graph. The knowledge extraction module of the platform includes three main steps: structured knowledge mapping, semi-structured knowledge extraction, and unstructured knowledge extraction. The specific execution process is as follows:

[0117] Step S2.1 Structured knowledge mapping: For structured data sources, such as CSV files and two-dimensional tables in relational databases, users are supported to build graphs by configuring mapping rules. By parsing the file content and combining the mapping configuration, the information in the file can be accurately converted into concepts and relationships in the graph. The specific steps are as follows:

[0118] Step S2.1.1 Mapping rules: Structured data includes files, rows and columns. The configuration rules of entities, relationships and attributes can be written according to the field names of the columns. During the mapping process, all values ​​will be treated as strings. If the file name does not contain the "-" connection symbol, the file name will be used as the name of the entity type, and the field name in the file will be used as the attribute name of the entity; if the file name contains the "-" connection symbol, the file name will be decomposed into triples according to the "-", which will be used as the head entity type, relationship type, and tail entity type respectively. In the file, the three fields will be used as the head entity instance, relationship instance, and tail entity instance of the triple.

[0119] Step S2.1.2 Concept Mapping: Use mapping rules to accurately map the fields in the file to the core concepts of the graph ontology and the specific types of its nodes. The user must first specify the file type and data format in the file selection area, and then accurately select the target file to be mapped in the opened file browser. Once the file is selected, the user can trigger the automatic mapping process by clicking the "Start Mapping" button.

[0120] Step S2.1.3 Relationship mapping: Use the mapping rules to select the file type and data format in the file selection area in the knowledge graph platform, and select the file to be mapped in the opened file browser. Click the "Start Mapping" button to complete the mapping.

[0121] Step S2.2 Semi-structured knowledge extraction: For semi-structured data such as logs and manuals written or generated according to a certain paradigm, a knowledge extraction function based on templates is provided, allowing users to define templates containing parts of speech to extract entities and relationships in the text. The specific steps are as follows;

[0122] Step S2.2.1: Write a template: Write a regular expression based on the content of the text to be extracted. For example, to extract the operator and switch object in the operation command, you can write a regular expression:

[0123] "(\w+)Disconnect (\d+kV system)\d+) switch"

[0124] The knowledge triple {"network control on-duty operator", "disconnect", "110kV system 2345 switch"} can be extracted from the text "network control on-duty operator disconnects 110kV system 2345 switch"

[0125] Step S2.3 Unstructured knowledge extraction: The platform provides text recognition in images based on OCR technology, and can be associated with dispatch automation system equipment to form equipment parameter knowledge. The specific steps include:

[0126] Step S2.3.1: Image preprocessing: grayscale the image to satisfy the relationship R=G=B for any pixel element of the color image; then use binarization to make the grayscale value of each element in the pixel matrix of the image 0 (black) or 255 (white); use the Hough transform method to achieve automatic image tilt and rotation correction; use the Layout-Parser text layout analysis toolkit to identify the four major types of blocks, namely, title, text paragraph, table and picture, from the limit image file, extract the table area from the power grid knowledge data, and generate a preprocessing result image file that meets the requirements of image recognition;

[0127] Step S2.3.2: Image semantic segmentation: Through semantic segmentation, the precise shape and position of each object in the image are obtained, thereby extracting the main device objects in the photo. First, the encoder gradually reduces the spatial size of the feature map to capture contextual information, and then the decoder gradually restores the spatial size of the feature map for pixel-level classification; introduces the hole convolution mechanism to capture more contextual information by increasing the receptive field of the convolution kernel; then uses the loss function CTC of the semantic segmentation task, inputs the preprocessed data into the network model, and adjusts the network parameters through the back propagation algorithm to minimize the loss function; during the training process, the validation set is used to monitor the performance of the model and adjust the parameters as needed.

[0128] Step S2.3.3: Equipment association: First, use OCR (optical character recognition) technology to accurately process the image and extract text information from the equipment nameplate. Subsequently, the text information will be converted into numbered text and data formats that are convenient for the system to process. Next, retrieve the unique identification number of the equipment from the power equipment knowledge graph, such as the equipment number or asset number. Subsequently, the system accurately compares and matches the nameplate information recognized by OCR with the equipment number in the graph. Once the match is successful, the system will automatically associate the nameplate information with the corresponding equipment record in the graph. In addition, other information related to the nameplate in the image will also be synchronously updated to the properties of the graph equipment to ensure the accuracy and completeness of the information.

[0129] 3. Data integration module: The knowledge graph platform pre-sets the entity types and attribute ranges to be integrated through the system configuration interface. Subsequently, the advanced model built into the platform will automatically calculate the similarities between attributes, entities, and subgraphs to accurately identify the entities that need to be aligned. After obtaining preliminary results, the system will display them to the user for confirmation. Once the user confirms that they are correct, the platform will automatically perform the merge or link operation to ensure that the aligned entities are properly integrated.

[0130] The specific steps are as follows:

[0131] Step S3.1: Range screening. Preliminary screening of entity data with similar fusion identifiers based on the entity type and attribute range configured by the user;

[0132] Step S3.2: Determine similar entities. Determine whether the entities can be integrated based on the attribute similarity between the entities. For entities in the same graph, the similarity based on the representation vector is used for judgment; for entities in different graphs, the edit distance is used for judgment.

[0133] Step S3.2.1: Similarity calculation based on representation learning. First, the entities to be aligned are converted into feature vectors, and then the cosine similarity between the vectors is used as the similarity of the entities.11 ,x 12 ,…,x 1n ) and B(x 21 ,x 22 ,…,x 2n ), their cosine similarity calculation formula is as follows:

[0134]

[0135] The value range of cosine similarity is defined as [-1,1], where the closer the cosine value is to 1, the more similar the two vectors are in direction; conversely, the closer the cosine value is to -1, the less similar the two vectors are in direction.

[0136] Step S3.2.2: Similarity based on edit distance.

[0137] The similarity based on the edit distance algorithm is the minimum number of single-character edits (insertion, deletion, or substitution) required to convert one word into another. The smaller the edit distance, the greater the similarity between the two strings. For two strings a and b, with lengths of |a| and |b| respectively, their edit distances are as follows:

[0138]

[0139] Step S3.3: Data integration: The graph data whose attribute similarities meet the threshold conditions are fused.

[0140] Step S3.3.1 sets a threshold p to determine whether two attributes are similar enough to be fused.

[0141] Step S3.3.2: For each entity in the graph data to be fused, perform attribute matching and similarity calculation. Traverse the entities and their attributes in the two graphs. For each attribute, calculate the similarity of the same type of attribute in the other graph. If an attribute with a similarity higher than a preset threshold is found, the two attributes are considered to be matched.

[0142] Step S3.3.3 Once the matching attribute pairs are determined, the data are merged into a unified graph, where, for numeric attributes, the average of their values ​​is taken; for text attributes, a primary value is selected; and for relational attributes, the graph structure of the relationship is merged.

[0143] 4. Graph construction module:

[0144] The knowledge graph platform that supports power system data integration analysis and visualization provides two graph construction methods: drag-and-drop construction and construction from a public ontology library. The steps are as follows:

[0145] Step S4.1 Drag and drop to build a map:

[0146] Step S4.1.1: In the quick creation page, according to the page guide, by clicking to create a concept node, you can design the concept in the knowledge graph model layer and edit the information of the concept and its attributes;

[0147] Step S4.1.2: In the quick creation page, follow the page instructions to create the relationship between concept nodes by dragging and clicking. You can edit the relationship information and its attribute information.

[0148] Step S4.2: Constructing a graph from a public ontology library:

[0149] Step S4.2.1: In the Create from Public Ontology Library page, select the required public ontology according to the page guidance. The public ontology already contains several ontology concepts and ontology relationships, which serve as the basis for the model layer of the new application;

[0150] Step S4.2.2: After selecting a public ontology, you can preview the ontology concepts and ontology relationships contained in it. Select "Import to private ontology library" to copy the content of the public ontology into the private application;

[0151] Step S4.2.3: Click “OK” to add a private knowledge graph application.

[0152] 5. Graph representation module: This module now integrates natural language processing functions, allowing users to ask queries in natural language. The system parses queries through natural language understanding technology, identifies key entities and user intent, and then calls graph algorithms for data retrieval and analysis. Graph algorithms include graph representation calculation and graph path exploration algorithms. Node encoders, edge encoders, and pseudo-sentence encoders are provided to represent graph elements of different granularities.

[0153] Step S5.1 Node Encoder: Use the GraphSAGE algorithm, specify the encoding dimension as d, and learn the vector representation of the node by sampling neighbor nodes and aggregating their features. The specific steps are as follows:

[0154] Step S5.1.1: Sampling neighbors: For each target node, a fixed number m of neighbor nodes are randomly sampled from its neighbors.

[0155] Step S5.1.2: Aggregate neighbor information: Use the pooling aggregator to aggregate the features of the sampled m neighbor nodes. When the number of neighbors is less than m, all of them are adopted;

[0156] Step S5.1.3: Update node embedding: weighted sum of aggregated neighbor features and target node’s own features, and then update the node’s embedding through a neural network layer;

[0157] Step S5.1.4: Repeat and optimize: Repeat the above steps for all nodes in the graph and train the model parameters through the optimization method of back propagation and gradient descent.

[0158] Step S5.2 Edge Encoder: Construct a dual graph of the original graph, convert the edges of the original graph into nodes of the new graph, and convert the nodes of the original graph into edges of the new graph, and then use the GraphSAGE algorithm to obtain the encoding vector of each node on the new graph, which is the encoding vector of the edge of the original graph.

[0159] Step S5.2.1: Collect the representations of nodes and edges in the original graph, assuming the original graph (G = (V, E)), where (V) is the node set and (E) is the edge set. Specify the encoding dimension as d.

[0160] Step S5.2.2: Definition of nodes and edges in the new dual graph: A node is an edge in the original graph that has a corresponding node in the dual graph. If two edges e1 and e2 in the original graph share a node v, then there will be an edge connecting e1 and e2 in the dual graph.

[0161] Step S5.2.3: Use the GraphSAGE algorithm to obtain the encoding vector of each node on the dual graph as the encoding of the edge in the corresponding original graph.

[0162] Step S5.3 Triple Encoder: For each triple in the graph, provide its corresponding encoding vector and specify the encoding dimension as d. The specific steps are:

[0163] Step S5.3.1: For the triple {h, r, t} in the graph, the node encoder and edge encoder described above can be used to map {h, r, t} into vectors (v_h, v_r, v_t) with dimensions (d_h, d_r, d_t) respectively.

[0164] Step S5.3.2: Construct an autoencoder, the encoder input dimension is (d_h+d_r+d_t), and the output dimension is d; the decoder input dimension is d, and the output dimension is (d_h+d_r+d_t);

[0165] Step S5.3.3: Use the full-image triplet data to train the autoencoder network in step S5.3.2 to obtain its parameters;

[0166] Step S5.3.4: For each triplet {h, r, t}, use the output of the encoder network as its encoding vector.

[0167] 6. User interface module: Provides an intuitive user interface so that non-expert users can easily access and utilize platform management, statistics and integration functions. The interface supports multiple views, such as graphical view, list view and timeline view, to meet the needs of different users. The following is an introduction to the design invention of the knowledge graph function module:

[0168] Step S6.1 Knowledge graph platform management interface function: The ontology management module provides users of the knowledge graph platform with management functions such as concept and relationship definition, editing, import and export, and builds their own knowledge graph model layer through a process-based guided drag-and-drop editing method. Ontology management is the basic function of knowledge graph construction and the pre-function of knowledge extraction. The specific module functions are as follows:

[0169] Step S6.1.1 Application definition: Define the namespace (domain) of the entire model layer. When you first create a model layer, you need to create a namespace for the domain of the model layer. The namespace serves as an isolation between different knowledge graphs;

[0170] Step S6.1.2 Concept Definition: Define the type of entity. After the domain definition is completed, you can enter the "Concept Definition" step. You can right-click on the blank space of the canvas and select "New Concept". Then edit the necessary elements of the concept on the right. The attributes of the concept are defined simultaneously. The concept can also have no attributes. Left-click on different concepts in the canvas, and the corresponding definitions and attributes will be displayed on the right. You can edit them; right-click on the concept, select "New Relationship" and jump to the "Relationship Definition" step;

[0171] Step S6.1.3 Relationship definition: define the relationship between concepts. After the "Concept definition" is completed, you can enter the "Relationship definition" step, right-click on the concept node and select "New relationship", generate a connection line from the concept node, drag it to another concept node (or itself) to form a relationship between concepts, and then edit the necessary elements of the relationship on the right. The attributes of the relationship are defined simultaneously, and the relationship can also have no attributes;

[0172] Step S6.1.4 Attribute definition: define the attributes of concepts and relationships, and configure the characteristics of attributes;

[0173] Step S6.1.5 Import and export: supports file mapping and preset pattern library, supports importing ontology from csv, xlsx, json format files, and supports importing multiple files. When importing concepts, you need to specify the fields of the concepts. After automatically detecting each concept, configure the attribute fields, attribute types, characteristics, etc. for each concept, and support exporting the defined ontology into a json format file;

[0174] Step S6.1.6 Visualization: Display the hierarchical relationship and association relationship between ontologies in a variety of visual ways, and support visualization and drag-and-drop construction of ontologies. Support creation, editing, copying, and deletion of concepts. Concepts include "concept name", "icon", and "attribute". Attributes can be configured as single-valued or data types.

[0175] Step S6.2 Knowledge graph platform statistical interface function: Provide statistical functions for applying graph analysis to the distribution of existing ontologies, instances and degrees. The statistical function includes the following modules:

[0176] Step S6.2.1 Overall statistics: Count the number of ontology concept types, relationship types, instance nodes, and instance relationships, and display them in a bar chart;

[0177] Step S6.2.2 Instance type distribution: Count the number of instances by type. When there are many instance types, count them by buckets and display them in a bar chart;

[0178] Step S6.2.3 Relationship type distribution: Count the number of relationships by type. When there are many relationship types, divide the statistics into buckets and display them in a bar chart;

[0179] Step S6.2.4 Degree distribution: Degree is the number of edges on a node. Degree statistics are helpful to examine the density characteristics of the graph. Degree distribution density statistics are displayed in a bar chart.

[0180] Step S6.2.5 Drill-down function: When you click a column in the column chart, the node and relationship instances of the column statistics are displayed.

[0181] Step S6.3 Knowledge graph platform graph integration function: Through the knowledge graph platform configuration and graph algorithm, the selected knowledge graph is integrated. The specific integration process is as follows:

[0182] Step S6.3.1: Configure alignment items. The configuration items include selecting the graph to be integrated, the attributes of the graph concepts to be aligned, the attribute integration method of the entities in the instance, and the integration method of the relationships in the instance.

[0183] Step S6.3.2: Query graph data. Enter the search content in the search box, search for relevant instance data in multiple knowledge graphs to be aligned and return them for display;

[0184] Step S6.3.3: Graph Integration: Click the "Graph Integration" button to merge the instance data in multiple graphs according to the integration method in the configuration, and return the integrated graph.

[0185] Through the collaborative work of the above modules, the power system knowledge graph platform of the present invention can significantly improve the operating efficiency and management level of the power system, and has high practicality and broad application prospects.

[0186] The present invention designs a knowledge graph platform that supports integrated analysis and visualization of power system data. The platform uses graph database technology to effectively manage and query large-scale power system knowledge data. Graph database technology supports efficient data association query and is suitable for processing complex network structure data, such as the relationship between various devices, sensors and user data in the power system.

[0187] In addition, the platform also integrates machine learning algorithms, including text comparison, relationship extraction, and graph neural networks, for deep analysis and pattern recognition of data. These algorithms can automatically learn and extract useful information from data to support intelligent decision-making in the power system. In particular, the graph neural network model used in this platform contains three encoders, including a node encoder, an edge encoder, and a triple encoder, which can process and analyze power knowledge data at different levels. Through the application of these technologies, the present invention can provide a comprehensive, efficient, and intelligent power system management solution, significantly improving the operating efficiency and management level of the power system.

Claims

1. A knowledge graph platform that supports integrated analysis and visualization of power system data, characterized in that: include: Power data preprocessing module, which uses natural language processing technology to extract key information from power system documents and records; The knowledge extraction module is used to extract entities, relations and attributes from the data source, construct triples and store them in the knowledge graph; The data integration module is used to calculate the similarity between attributes, entities and subgraphs using the model according to the preset requirements, identify the entities to be aligned, and the user verifies the identification results. The platform merges or links the verified results to complete the entity integration. Graph building module, used to build knowledge graphs by dragging and dropping or from public ontology libraries; The graph representation module is used to integrate natural language processing functions, provide users with services for querying in natural language, parse queries through natural language understanding technology, identify key entities and user intent, and call graph algorithms for data retrieval and analysis; The user interface module is used to provide a user interface, allowing users to access and utilize platform management, statistics and integration functions; the user interface module specifically includes: knowledge graph platform management interface function, knowledge graph platform statistics interface function, and knowledge graph platform graph integration function.

2. The knowledge graph platform supporting power system data integration analysis and visualization according to claim 1 is characterized in that: The power data preprocessing module is specifically implemented as follows: S1.1: Add the professional terms, equipment names and technical vocabulary in the power field corresponding to the power data to the custom dictionary of the word segmentation tool, and select the word segmentation algorithm that supports Chinese and English for word segmentation; S1.2: Use predefined rule sets to perform part-of-speech tagging, and then automatically identify the part-of-speech through the training model. Combining the predefined rule set and the training model method can improve the accuracy and robustness of part-of-speech tagging; S1.3: Use a time series data analysis model based on deep learning to learn the time dependencies in the data and predict future data trends; S1.4: Use visual text annotation tools to annotate the entities, relationships, and attributes to be extracted from the text file to be annotated, obtain the annotation results, and use the active learning model to learn the annotation rules.

3. The knowledge graph platform supporting integrated analysis and visualization of power system data according to claim 2, characterized in that: The annotation result file in step S1.4 is in ann format, and step S1.4 specifically includes: S1.4.1: Upload the text file to be annotated. The text file to be annotated is uploaded by the user. The file format supports txt format and the file encoding method is UTF-8. When the uploaded file is a new file, a corresponding blank ann file is automatically generated. The blank ann file encoding method is UTF-8. S1.4.2: Perform text annotation, wherein the annotation method includes manual text annotation and automatic text annotation; The manual text annotation method is to select the type or attribute of the text annotation entity on the page, and to drag from one entity tag to another entity tag on the page to annotate the relationship that meets the constraints, including: Entity and attribute annotation: When a text paragraph on a page is selected, an annotation dialog box will automatically appear, which is used to give entity annotation or attribute annotation to the selected text; after the annotation is completed, the text paragraph will be highlighted in a preset color and accompanied by a corresponding annotation label; the annotation label can be edited or deleted; Relationship annotation: When a user selects an annotated entity and drags it to another annotated entity, it is determined whether there is a preset relationship between the two entities. If there is a preset relationship, a connection line will be immediately presented, and a relationship annotation dialog box will pop up for relationship annotation. The annotation result can be triggered by the connection line. If there is no preset relationship between the two entities, a prompt will be given; the annotation result can be edited or deleted; The automatic text annotation method is to first perform fine word segmentation on the text to split the text into multiple words or phrases; then, using similarity matching technology, each word or phrase in the text is compared with the existing annotation data to determine its corresponding entity and relationship; during the annotation process, the system automatically generates a mark for each annotation result for user confirmation; the annotation information confirmed by the user is added as new annotation data to the ann file that has been annotated; S1.4.3: Check annotations according to preset standards; use preset colors to mark incorrect annotations identified according to the standards for user identification and correction; and use different colors to distinguish annotation contents according to the entity category to which they belong; The standards include: The annotated entity categories and attributes are all reflected in the established configuration and there is no missing; The relationship type of the annotation needs to be defined in the configuration; The annotated relationships follow the entity types specified in the configuration, and any behavior that does not conform to or changes the configuration without authorization is defined as an error; S1.4.4: Download of annotated datasets: Set up a dataset access page to provide users with a download function. According to the user's download requirements, the source text files of the entire dataset and the ann files after annotation are automatically packaged, or the specified source text files and ann files after annotation are selected according to the user's needs, and the option of generating a json format file is provided for the user to download.

4. The knowledge graph platform supporting power system data integration analysis and visualization according to claim 1, characterized in that: The data source of the knowledge extraction module includes structured data source, semi-structured data source and unstructured data source; The extraction of structured data sources supports users to build graphs by configuring mapping rules; by parsing the file content and combining the mapping configuration, the information in the file is converted into concepts and relationships in the graph; specifically, it includes: S2.1.1: Set mapping rules: The structured data includes files, rows and columns. The configuration rules of entities, relationships and attributes are written according to the field names of the columns. All values ​​in the mapping process are treated as strings. When the file name does not contain a hyphen, the file name is used as the name of the entity type, and the field name in the file is used as the attribute name of the entity. When the file name contains a hyphen, the file name is decomposed into triples according to the hyphen, which are used as the head entity type, the relationship type and the tail entity type respectively. In the file, the three fields are used as the head entity instance, the relationship instance and the tail entity instance of the triple respectively. S2.1.2: Concept mapping: Use mapping rules to map the fields in the file to the core concepts of the graph ontology and the specific types of its nodes; specifically, the user specifies the file type and data format in the file selection area, and selects the target file to be mapped in the opened file browser. Then, when the platform receives the user's command, it triggers the automatic mapping process; S2.1.3: Relationship mapping: Use mapping rules to select file type and data format, select the file to be mapped from the file browser; issue a mapping command to complete the mapping; The extraction of the semi-structured data source provides a knowledge extraction function according to a template, allowing the user to define a template containing parts of speech to extract entities and relationships in the text; the template is specifically a regular expression written according to the content of the text of the knowledge to be extracted; The unstructured extraction, based on text recognition in images using OCR technology, is associated with the dispatching automation system equipment to form equipment parameter knowledge; specifically includes: S2.3.1: Image preprocessing: grayscale the image to make any pixel element of the color image satisfy the relationship R=G=B; then use binarization to make the grayscale value of each element in the pixel matrix of the image have two different values; use the Hough transform method to realize automatic image tilt and rotation correction; use the Layout-Parser text layout analysis toolkit to identify four types of blocks from the limit image file: title, text paragraph, table and picture, extract the table area from the power grid knowledge data, and generate a preprocessing result image file that meets the image recognition requirements; S2.3.2: Image semantic segmentation: Through semantic segmentation, the precise shape and position of each object in the image are obtained, and the main equipment objects in the photo are extracted; specifically, the encoder gradually reduces the spatial size of the feature map to capture contextual information, and the decoder restores the spatial size of the feature map for pixel-level classification; introduces the hole convolution mechanism to capture contextual information by increasing the receptive field of the convolution kernel; uses the loss function CTC of the semantic segmentation task, inputs the preprocessed data into the network model, and adjusts the network parameters through the back-propagation algorithm to minimize the loss function; during the training process, uses the validation set to monitor the performance of the model and adjusts the parameters as needed; S2.3.3: Equipment association: Use optical character recognition (OCR) technology to process the image and extract text information from the equipment nameplate; convert the text information into numbered text and data format; retrieve the unique identification number of the equipment from the power equipment knowledge graph; compare and match the nameplate information recognized by OCR with the equipment number in the graph; when the match is successful, the platform automatically associates the nameplate information with the corresponding equipment record in the graph, and synchronously updates all relevant information about the nameplate in the image to the properties of the graph equipment.

5. The knowledge graph platform supporting power system data integration analysis and visualization according to claim 1, characterized in that: The data integration module is specifically implemented as follows: S3.1: Range screening: Filter entity data with similar fusion identifiers based on the entity type and attribute range configured by the user; S3.2: Determine similar entities: Determine entity similarity based on the attribute similarity between entities; the determination method, when the entities belong to the same graph, uses similarity determination based on representation vectors; when the entities belong to different graphs, uses edit distance determination; specifically: S3.2.1: Similarity calculation based on representation learning; First, the entities to be aligned are converted into feature vectors, and then the cosine similarity between vectors is used as the similarity of the entities; for two vectors A(x 11 ,x 12 ,…,x 1n ) and B(x 21 ,x 22 ,…,x 2n ), their cosine similarity calculation formula is as follows: The value range of cosine similarity is defined as [-1, 1]. When the cosine value is closer to 1, it means that the two vectors are more similar in direction; when the cosine value is closer to -1, it means that the two vectors are less similar in direction. S3.2.2: Similarity based on edit distance. For two strings a and b, with lengths |a| and |b| respectively, their edit distances are as follows: S3.3: Data integration: merge the graph data whose attribute similarity meets the threshold conditions, specifically: S3.3.1: Set a threshold p to determine whether two attributes are similar enough to be fused; S3.3.2: For each entity in the graph data to be fused, perform attribute matching and similarity calculation; traverse the entities and their attributes in the two graphs; for each attribute, calculate the similarity of the same type of attribute in the other graph; when an attribute with a similarity higher than a preset threshold is found, the two attributes are matched; S3.3.3: For matching attributes, merge the corresponding data into a unified graph. For numeric attributes, take the average of the values; for text attributes, select a primary value; for relational attributes, merge the graph structure of the relationship.

6. The knowledge graph platform supporting power system data integration analysis and visualization according to claim 1, characterized in that: The steps of dragging and dropping the graph construction module to build a knowledge graph are as follows: S4.1.1: Set up a quick creation page, follow the page instructions, click to create concept nodes, design concepts in the knowledge graph model layer, and edit concept information and its attribute information; S4.1.2: In the quick creation page, follow the page instructions to create relationships between concept nodes by dragging and clicking, and edit the relationship information and its attribute information; The steps of constructing a knowledge graph from a public ontology library in the graph construction module are as follows: S4.2.1: In the Create from Public Ontology Library page, select the required public ontology according to the page guide. The public ontology includes several ontology concepts and ontology relationships as the basis of the pattern layer of the new application; S4.2.2: After selecting a public ontology, preview the ontology concepts and ontology relationships contained therein, and copy the contents of the public ontology into the private application; S4.2.3: Confirm your selection and add a new private knowledge graph application.

7. The knowledge graph platform supporting power system data integration analysis and visualization according to claim 1, characterized in that: The graph representation module provides representations for graph elements of different granularities by providing node encoders, edge encoders, and pseudo-sentence encoders; The node encoder representation: uses the GraphSAGE algorithm, specifies the encoding dimension as d, samples neighbor nodes and aggregates features to learn the vector representation of the node, including: S5.1.1: Sampling neighbors: For each target node, randomly sample a fixed number m of neighbor nodes from its neighbors; S5.1.2: Aggregate neighbor information: Use the pooling aggregator to aggregate the features of the sampled m neighbor nodes. When the number of neighbors is less than m, all of them are used; S5.1.3: Update node embedding: Update the node embedding by taking the weighted sum of the aggregated neighbor features and the target node’s own features, and passing it through a neural network layer; S5.1.4: Repeat and optimize: Repeat steps S5.1.1 to S5.1.3 for all nodes in the graph according to the preset number of iterations, and train the model parameters through the optimization method of back propagation and gradient descent; The edge encoder representation: constructs a dual graph of the original graph, converts the edges of the original graph into nodes of the new graph, converts the nodes of the original graph into edges of the new graph, and uses the GraphSAGE algorithm to obtain the encoding vector of each node on the new graph, specifically including: S5.2.1: Collect the representations of nodes and edges in the original graph. Let the original graph G = (V, E), where V is the set of nodes and E is the set of edges. Specify the encoding dimension as d. S5.2.2: Define the nodes and edges of the new dual graph: A node is an edge in the original graph that has a corresponding node in the dual graph. When two edges e1 and e2 in the original graph share a node v, there is an edge connecting e1 and e2 in the dual graph. S5.2.3: Use the GraphSAGE algorithm to obtain the encoding vector of each node in the dual graph as the encoding of the edge in the corresponding original graph; The triple encoder representation: for each triple in the graph, provide its corresponding encoding vector, specify the encoding dimension as d, specifically including: S5.3.1: For the triple {h, r, t} in the graph, use the node encoder and edge encoder to map {h, r, t} into vectors (v_h, v_r, v_t) with dimensions (d_h, d_r, d_t) respectively; S5.3.2: Construct an autoencoder with an encoder input dimension of (d_h+d_r+d_t) and an output dimension of d; a decoder with an input dimension of d and an output dimension of (d_h+d_r+d_t); S5.3.3: Use the full-image triplet data to train the autoencoder network in step S5.3.2 to obtain its parameters; S5.3.4: For each triplet {h, r, t}, use the output of the encoder network as its encoding vector.

8. The knowledge graph platform supporting power system data integration analysis and visualization according to claim 1, characterized in that: The knowledge graph platform management interface function of the user interface module is used to provide users with concept and relationship definition, editing, import and export management functions, and to build a knowledge graph model layer through a process-based guided drag-and-drop editing method; specifically, it includes: Application definition unit: defines the namespace of the entire model layer, which serves as isolation between different knowledge graphs; Concept definition unit: used to define the type of entity. After the application definition is completed, a new concept is created, and the necessary elements of the concept are edited, and the attributes of the concept are defined simultaneously; the definition and attributes of the concept can be edited; Relationship definition unit: define the relationship between concepts, generate a connection line from the concept node, drag it to another concept node or itself to form a relationship between concepts, edit the necessary elements of the relationship, and choose whether to define the attributes of the relationship synchronously; Attribute definition unit: defines the attributes of concepts and relationships, and configures the characteristics of attributes; Import and export unit: supports file mapping and preset pattern library, supports importing ontology from csv, xlsx, json format files, supports importing multiple files; when importing concepts, specifies the fields of concepts, automatically detects each concept, configures attribute fields, attribute types, and characteristics for each concept, and supports exporting the defined ontology into json format files; Visualization unit: Displays the hierarchical and associative relationships between ontologies in a variety of visual ways, supports visualization and drag-and-drop construction of ontologies; supports the creation, editing, copying and deletion of concepts; the concepts include concept names, icons and attributes; attributes can be configured with single-value and data types.

9. The knowledge graph platform supporting power system data integration analysis and visualization according to claim 1, characterized in that: The knowledge graph platform statistics interface function of the user interface module is used to provide statistics for the application graph analysis of the distribution of existing ontologies, instances and degrees; specifically, it includes: Overall statistical unit: used to count the number of ontology concept types, relationship types, instance nodes, and instance relationships, and displayed in a bar chart; Instance type distribution unit: used to count the number of instances by type. When the number of instance types exceeds the preset threshold, the statistics are divided into buckets and displayed in a bar chart. Relationship type distribution unit: used to count the number of relationships by type. When the number of relationship types exceeds the preset threshold, the statistics are divided into buckets and displayed in a bar chart. Degree distribution unit: The degree is the number of edges on a node, which is counted by degree distribution density and displayed in a bar graph; Drill-down function unit: used to display the node and relationship instances of the column statistics.

10. The knowledge graph platform supporting power system data integration analysis and visualization according to claim 1, characterized in that: The knowledge graph platform graph integration function of the user interface module is used to integrate the selected knowledge graph through the knowledge graph platform configuration and graph algorithm. The specific integration process is as follows: S6.3.1: Configure alignment items, including selecting the graphs to be integrated, the attributes of the graph concepts to be aligned, the attribute integration method of the entities in the instance, and the integration method of the relationships in the instance; S6.3.2: Query graph data. Enter the search content in the search box, search for relevant instance data in multiple knowledge graphs to be aligned, and return them for display; S6.3.3: Graph integration, call the data integration module, merge the instance data in multiple graphs according to the integration method in the configuration, and return the integrated graph.

Images