Knowledge graph construction method and equipment of power distribution network material and medium

Abstract

The invention provides a knowledge graph construction method and device for a power distribution network material and a medium, and belongs to the field of intelligent material management and data governance of a power distribution network, and the method comprises the following steps: carrying out the preprocessing of multi-source power distribution network material data; defining a label dictionary, and performing label labeling on the text segments of the multi-source power distribution network material data according to a preset labeling rule; knowledge extraction is carried out according to the labels of the multi-source power distribution network material data, and a knowledge triple of a subject-predicate-relation object structure is generated; carrying out power distribution network material knowledge fusion on the knowledge triad based on rule constraints and coding constraints; and generating a power distribution network material knowledge graph according to a power distribution network material knowledge fusion result. The power distribution network material knowledge graph construction method based on the natural language processing technology is constructed, and the problems of material identification conflict and expression redundancy are solved through a material entity alignment method with double constraints of rules and codes.

Description

Technical Field

[0001] This invention belongs to the field of intelligent material management and data governance of power distribution networks, and specifically relates to a method, equipment and medium for constructing a knowledge graph of power distribution network materials. Background Technology

[0002] Distribution network materials refer to various materials and supplies used in the construction, maintenance and operation of power systems. They are key elements to ensure the safe, reliable and efficient operation of power systems. However, distribution network projects are large in scale and have a wide variety of materials. There are common problems of material identification conflicts and redundant descriptions in multi-source distribution network material data. Therefore, current multi-source distribution network material data needs to be processed manually. Chinese invention patent application document with publication number CN118657469A discloses a power grid material group management system based on large model technology. It improves the efficiency of power distribution network engineering material management and engineering response capability through large model technology, but does not optimize the problems of power distribution network material identification conflict and redundancy. Therefore, there is an urgent need for a knowledge graph that can eliminate material identification conflicts and redundant descriptions in the power distribution network, thereby improving the level of intelligence in power distribution network material management. Summary of the Invention

[0003] To address the problems in the prior art, this invention proposes a method, device, and medium for constructing a knowledge graph of distribution network materials. It constructs a method for constructing a knowledge graph of distribution network materials based on natural language processing technology. Through a material entity alignment method with dual constraints of rules and coding, it solves the problems of material identification conflicts and redundancy in expression.

[0004] The technical solution of the present invention is as follows: In a first aspect, the present invention provides a method for constructing a knowledge graph for materials in a power distribution network, comprising the following steps: Preprocessing of material data for multi-source power distribution networks; Define a tag dictionary, including three tag types: entity tags, attribute tags, and relationship tags, and tag text fragments of multi-source distribution network material data according to preset tagging rules; Knowledge is extracted from the tags of multi-source power distribution network material data to generate knowledge triples with a subject-predicate-relational object structure. Based on rule constraints and coding constraints, knowledge triples are used to integrate material knowledge of the power distribution network; The rule constraints are obtained by grouping the globally unique identifiers of distribution network material entities, checking for conflicts in distribution network material attributes, and fusing information within the same globally unique identifier group to obtain standardized knowledge triples. The coding constraints process distribution network material entities without valid globally unique identifiers and not included in the rule constraints through semantic coding and semantic vector similarity matching, and summarize the coding constraint results to obtain standardized knowledge triples. A distribution network material knowledge graph is generated based on the fusion results of distribution network material knowledge.

[0005] Furthermore, the grouping of materials through the globally unique identifier of the distribution network entity includes the following steps: Extract the associated identifiers of material entities in the power distribution network, and distinguish between globally unique identifiers and invalid identifiers; Distribution network material entities associated with the same globally unique identifier are grouped together using an identifier key-value hash aggregation algorithm; The identifier key-value hash aggregation algorithm uses the valid globally unique identifier of the distribution network as the key and the list of material entities in the distribution network as the value to initialize the hash container; Iterate through all power distribution network material entities, extract the globally unique identifier field value of each material entity, and perform validity verification of the identifier value through regular expression matching; Calculate the hash value of the globally unique identifier that passes the validity check, and map the corresponding power distribution network material entity to the entity list with the same key in the hash container to complete the aggregation and grouping.

[0006] Furthermore, the encoding constraints generate semantic vectors of power distribution network material entities using the all-MiniLM-L6-v2 model.

[0007] Furthermore, the semantic vector similarity is calculated using the cosine similarity algorithm, and the calculation formula is as follows: ; in, For distribution network material entities a semantic vector; For distribution network material entities b semantic vector; semantic vector V a The i Dimensional components; semantic vector V b The i Dimensional components.

[0008] Furthermore, the process of handling power distribution network material entities without valid identifiers and not included in rule constraints through semantic vector similarity matching specifically includes: Select power distribution network material entities with semantic vector similarity ≥ preset value and perform core attribute verification. If there is no conflict in the core attributes, they are grouped into the same power distribution network material entity alignment group. If there is a conflict in the core attributes, it is determined that they are different power distribution network material entities.

[0009] Furthermore, the preprocessing includes: Integrity verification and missing tagging of material identification data fields in power distribution networks; Regular expression matching and unified text formatting standards; Remove redundant spaces and duplicate content; Split unstructured data and filter out phrases with no real semantic meaning; The synonym text is unified based on the thesaurus of materials in the power distribution network.

[0010] Furthermore, the preset annotation rules include two types of annotation rules; The first annotation rule directly maps fixed fields to corresponding labels; The second method involves using regular expressions to match attribute-based text fragments based on keywords.

[0011] Furthermore, the knowledge graph is stored using the Neo4j database, which is based on a graph data model. The standardized knowledge triples and power distribution network material data are imported into the Neo4j database for storage.

[0012] In a second aspect, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the computer program is executed by the processor to perform the method described in any of the preceding claims.

[0013] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the preceding claims.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves efficient alignment and unified modeling of multi-source distribution network material data through a knowledge fusion method that combines rules and semantic encoding. It effectively eliminates material identification conflicts and redundancy in descriptions, constructs a structurally standardized and semantically consistent distribution network material knowledge graph, and significantly improves the quality of material data and the level of intelligence in distribution network material management. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the knowledge graph structure for materials in a power distribution network. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1 This embodiment provides a method for constructing a knowledge graph for materials in a power distribution network, specifically including the following steps: S1. Preprocess the material data of the multi-source distribution network, specifically: Check for missing data in the power distribution network materials field by field and record by record, and mark missing data in the identification fields; Use regular expression matching and text replacement to unify text format, remove redundant spaces and duplicate content, and standardize the material description fields of the power distribution network. Unstructured data in the power distribution network material data is split and phrases without actual semantic meaning, such as "none" and "see attachment for details", are filtered out. Establish a thesaurus of material synonyms for the power distribution network to unify the synonym representations of unstructured data splitting.

[0018] S2. Tag and label text segments of material data from multi-source power distribution networks, specifically: Define a tag dictionary, as shown in Table 1, which includes three tag types: entity tags, attribute tags, and relationship tags; Two types of labeling rules are set to achieve automatic labeling; The first method directly maps fixed fields to corresponding tags; The second method involves using regular expressions to match attribute-based text fragments based on keywords.

[0019]

[0020] S3. Knowledge extraction is performed based on the tags of multi-source power distribution network material data. Specifically: The system iterates through the tag annotation results record by record, locates the text of the [Material Entity] tag as the subject of the knowledge triple, and extracts the text of the [Classification Entity], [Scene Entity] and [Identifier Entity] tags as the relation objects corresponding to the subject. Based on the predicate mapping relationship between the preset attribute tags and relation tags, the system generates knowledge triples. During the generation process, duplicate knowledge triples are removed, and invalid records with empty relation objects or missing predicate mappings are filtered out.

[0021] S4. Integrate the generated knowledge triples with power distribution network material knowledge, specifically: The system employs a dual constraint mechanism of rules and coding. On one hand, it integrates related entities with the same identifier by grouping them with globally unique identifiers (such as material numbers and fixed IDs), checking for attribute conflicts, and fusing information within the group, thus balancing processing efficiency and accuracy. On the other hand, it uses semantic coding similarity matching as the core to fill in fuzzy scenarios without strong identifier associations, improve the coverage of entity alignment, and ultimately form a distribution network material knowledge unit that is free of redundancy and conflict and has unified attributes and relationships. The rule constraints are based on rule-driven knowledge fusion and include the following steps: The algorithm extracts the associated identifiers of material entities, distinguishes between globally unique identifiers and invalid identifiers, and uses an identifier key-value hash aggregation algorithm to group entities associated with the same globally unique identifier. First, it defines the valid globally unique identifier of the distribution network as the key and the list of material entities as the value, and initializes the hash container. Then, it iterates through all material entities, extracts the globally unique identifier field value of each material entity, and performs validity verification of the identifier value using regular expression matching. For globally unique identifiers that pass the validity verification, it calculates the hash value and maps the corresponding entity to the list of entities with the same identifier key in the hash container, thus achieving the aggregation and grouping of entities associated with the same globally unique identifier. This significantly improves efficiency compared to traditional nested traversal techniques. Extract all entity names within the group, select industry standard expressions as standard names, and use the rest as aliases; when there are no industry standard expressions, use the most frequently used entity name within the group as a temporary standard name, and use the rest as aliases; Compare the core attributes of all entities within the group, as shown in Table 2. Merge conflict-free core attributes and remove duplicate core attributes; if core attributes conflict, split the group. Summarize all valid globally unique identifiers of entities within the group, organize them in the order of valid globally unique identifier - data source, and update the [Identify Entity] label; Update the knowledge triples within the group, remove duplicates according to the "subject-predicate-relational object" structure, retain the unique valid record, and obtain the standardized knowledge triples.

[0022] The encoding constraints are based on encoding-driven knowledge fusion and include the following steps: For entities without valid identifiers and not included in rule-driven fusion, standardized feature text is generated according to the structure format of material category-core attribute; The lightweight all-MiniLM-L6-v2 model is used to directly generate entity semantic vectors. The model first splits continuous text into the smallest semantic units (tokens) through the built-in BERT tokenizer, and then maps each smallest semantic unit to the corresponding vocabulary ID. The vocabulary ID is encoded with integers. The sequence of vocabulary IDs after tokenization is input into the model's 6-layer Transformer encoder. The self-attention mechanism captures the semantic associations of each attribute in the text. Each layer of the encoder performs linear transformation and attention weighting on the input features, extracting deep semantic features of the text layer by layer, and outputting a high-dimensional feature vector corresponding to each smallest semantic unit. L2 normalization is performed on high-dimensional feature vectors to avoid affecting the accuracy of similarity calculation due to differences in vector scale. The cosine similarity algorithm is used to calculate the pairwise semantic vector similarity between all entities to be aligned. The calculation formula is as follows: ; in, For distribution network material entities a semantic vector; For distribution network material entities b semantic vector; semantic vector V a The i Dimensional components; semantic vector V b The i Dimensional components; Entity pairs with semantic vector similarity ≥ 0.85 are selected for core attribute verification. If there is no conflict in the core attributes, they are classified into the same alignment group; otherwise, they are judged as different entities. Summarize all entities within the group, organize them in the order of standardized feature text - data source, and update the [identify entity] label; Update the knowledge triples within the group, remove duplicates according to the "subject-predicate-relational object" structure, and record different material numbers and different fixed IDs as associated identifiers to obtain standardized knowledge triples.

[0023]

[0024] S5. Construct a knowledge graph framework for distribution network materials. Based on the integration of distribution network business specifications and prior knowledge, entity tags are defined, relationship tags between entities and attribute tags of entities are clarified, and a knowledge graph structure framework for distribution network materials is designed by combining entity uniqueness constraints and attribute necessity constraints. The framework is divided into an ontology layer, an entity layer, and a relationship layer, such as... Figure 1 As shown; Using the Neo4j database, which is based on a graph data model, as a carrier, the standardized knowledge triples and entity data are imported into the database in batches to adapt to the structured knowledge storage system for power distribution network business applications.

[0025] Example 2 This embodiment proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in Embodiment 1.

[0026] Example 3 This embodiment proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0027] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

Claims

1. A method for constructing a knowledge graph for materials in a power distribution network, characterized in that, Includes the following steps: Preprocessing of material data for multi-source power distribution networks; Define a tag dictionary, including three tag types: entity tags, attribute tags, and relationship tags, and tag text fragments of multi-source distribution network material data according to preset tagging rules; Knowledge is extracted from the tags of multi-source power distribution network material data to generate knowledge triples with a subject-predicate-relational object structure. Based on rule constraints and coding constraints, knowledge triples are used to integrate material knowledge of the power distribution network; The rule constraints are obtained by grouping the globally unique identifiers of distribution network material entities, checking for conflicts in distribution network material attributes, and fusing information within the same globally unique identifier group to obtain standardized knowledge triples. The coding constraints process distribution network material entities without valid globally unique identifiers and not included in the rule constraints through semantic coding and semantic vector similarity matching, and summarize the coding constraint results to obtain standardized knowledge triples. A distribution network material knowledge graph is generated based on the fusion results of distribution network material knowledge.

2. The method for constructing a knowledge graph for distribution network materials according to claim 1, characterized in that, The process of grouping materials using globally unique identifiers in the power distribution network includes the following steps: Extract the associated identifiers of material entities in the power distribution network, and distinguish between globally unique identifiers and invalid identifiers; Distribution network material entities associated with the same globally unique identifier are grouped together using an identifier key-value hash aggregation algorithm; The identifier key-value hash aggregation algorithm uses the valid globally unique identifier of the distribution network as the key and the list of material entities in the distribution network as the value to initialize the hash container; Iterate through all power distribution network material entities, extract the globally unique identifier field value of each material entity, and perform validity verification of the identifier value through regular expression matching; Calculate the hash value of the globally unique identifier that passes the validity check, and map the corresponding power distribution network material entity to the entity list with the same key in the hash container to complete the aggregation and grouping.

3. The method for constructing a knowledge graph for distribution network materials according to claim 1, characterized in that, The encoding constraints generate semantic vectors of power distribution network material entities using the all-MiniLM-L6-v2 model.

4. The method for constructing a knowledge graph for distribution network materials according to claim 1, characterized in that, The semantic vector similarity is calculated using the cosine similarity algorithm, and the calculation formula is as follows: ； in, For distribution network material entities a semantic vector; For distribution network material entities b semantic vector; semantic vector V a The i Dimensional components; semantic vector V b The i Dimensional components.

5. The method for constructing a knowledge graph for power distribution network materials according to claim 4, characterized in that, The process of handling distribution network material entities without valid identifiers and not included in rule constraints through semantic vector similarity matching specifically includes: Select power distribution network material entities with semantic vector similarity ≥ preset value and perform core attribute verification. If there is no conflict in the core attributes, they are grouped into the same power distribution network material entity alignment group. If there is a conflict in the core attributes, it is determined that they are different power distribution network material entities.

6. The method for constructing a knowledge graph for distribution network materials according to claim 1, characterized in that, The preprocessing includes: Integrity verification and missing tagging of material identification data fields in power distribution networks; Regular expression matching and unified text formatting standards; Remove redundant spaces and duplicate content; Split unstructured data and filter out phrases with no real semantic meaning; The synonym text is unified based on the thesaurus of materials in the power distribution network.

7. The method for constructing a knowledge graph for distribution network materials according to claim 1, characterized in that, The preset annotation rules include two types of annotation rules; The first annotation rule directly maps fixed fields to corresponding labels; The second method involves using regular expressions to match attribute-based text fragments based on keywords.

8. The method for constructing a knowledge graph for distribution network materials according to claim 1, characterized in that, The knowledge graph is stored using the Neo4j database, which is based on a graph data model. The standardized knowledge triples and power distribution network material data are imported into the Neo4j database for storage.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and running on the processor, wherein the computer program is executed by the processor to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images