Intelligent construction method for power customer relation graph based on multi-source data fusion

Abstract

The invention relates to an intelligent construction method for a power customer relationship graph based on multi-source data fusion, and the method comprises the following steps: S1, obtaining total data related to customers, and carrying out the preprocessing of the total data, and obtaining a structured standard data set; s2, based on the structured standard data set, identifying service entities and attribute information thereof, and obtaining a service entity list; s3, performing entity alignment and data fusion according to the business entity list to obtain a multi-source fusion data view; s4, on the basis of the multi-source fusion data view, entity data and a relation triple are obtained through knowledge extraction and relation discovery; s5, mapping the entities and the relationships into an operable form of a graph database, and constructing and optimizing a power customer relationship knowledge graph to obtain a complete power customer relationship knowledge graph; and S6, according to the complete power customer relationship knowledge graph, realizing graph visualization through Cytoscape. According to the method, the electric power customer relationship graph is intelligently constructed based on multi-source data fusion, and the digital management capability of an electric power enterprise is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph construction, and in particular to an intelligent construction method for a power customer relationship graph based on multi-source data fusion. Background Technology

[0002] During the digital transformation of power companies, a large amount of heterogeneous data is generated in multiple business processes such as customer management, equipment operation and maintenance, and incident response. This data is scattered across different business systems such as marketing, metering, operation and maintenance, and geographic information, exhibiting diverse formats, inconsistent structures, and ambiguous relationships. This results in the widespread existence of data silos, which cannot effectively support complex business insights and intelligent decision-making.

[0003] With the development of smart grids, the Internet of Things, and big data technologies, power companies urgently need to break down business process and data barriers to achieve the integration and sharing of multi-source heterogeneous data. Only by deeply integrating multi-dimensional data such as customer, equipment, region, and business events, and systematically mining the semantic relationships between various entities, can a power customer relationship knowledge graph covering the entire business chain be built, providing a solid data foundation for precision marketing, operation and maintenance decision-making, fault tracing, and value-added services.

[0004] However, existing data management methods have significant shortcomings in areas such as multi-source heterogeneous data integration, unified entity identification, and dynamic relationship modeling, making it difficult to meet the needs of intelligent and panoramic business analysis. Summary of the Invention

[0005] To address the aforementioned issues, the present invention aims to provide an intelligent construction method for power customer relationship graphs based on multi-source data fusion, which integrates heterogeneous data from various business systems, automatically identifies business entities, accurately extracts business relationships, and achieves efficient storage and visualization of relationship data.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The intelligent construction method for power customer relationship graphs based on multi-source data fusion includes the following steps:

[0008] S1: Obtain all customer-related data and preprocess it to obtain a structured standard dataset;

[0009] S2: Based on a structured standard dataset, identify business entities and their attribute information, and obtain a list of business entities, including the type of each entity and its attribute characteristics;

[0010] S3: Based on the business entity list, perform entity alignment and data fusion to obtain a multi-source fused data view, where each entity has a globally unique identifier;

[0011] S4: Based on a multi-source fusion data view, entity data and relation triples are obtained through knowledge extraction and relation discovery;

[0012] S5: Map entities and relationships to a graph database operable form, construct and optimize the power customer relationship knowledge graph to obtain a complete power customer relationship knowledge graph;

[0013] S6: Based on the complete knowledge graph of electricity customer relationships, visualize the graph using Cytoscape.

[0014] Furthermore, the full range of customer-related data includes basic customer information from the marketing system, electricity consumption data from the metering system, spatial location data from the geographic information system, and business records from the customer service system. By establishing a unified data interface and ETL (extract, transform, load) process, automated data collection from different database platforms is achieved, ensuring data integrity and timeliness.

[0015] Further preprocessing includes the following: A comprehensive quality assessment and cleaning process is performed on the collected raw data, including identifying and marking missing, null, and outlier values; unifying coding standards, time formats, and address specifications across different systems; simultaneously, a duplicate data detection algorithm is used to handle redundant records of the same customer in different systems; a master data management mechanism is established to ensure the principle of data uniqueness for each customer; based on power industry data standards and internal enterprise data specifications, the cleaned data undergoes standardization and structural reconstruction to establish a unified data model and dictionary, standardizing key identifiers such as customer numbers, equipment asset codes, and geographic location codes; unifying data formats, unit dimensions, and accuracy requirements; redesigning the data table structure, optimizing storage methods, and improving data access efficiency; ultimately forming a structured standard dataset covering multi-dimensional information on customer entities, electricity consumption behavior, geographical distribution, and business interactions.

[0016] Furthermore, based on the structured standard dataset, business entities and their attribute information are identified to obtain a list of business entities, as follows: Combining the field definitions and business semantics of the structured standard dataset, a method combining rule matching, feature extraction, and manual verification is used to identify and classify business entities. Entity types include, but are not limited to: customer entities, device entities, geographic entities, and business entities; and corresponding attribute information is further extracted for each type of entity.

[0017] After completing entity identification and classification, the attribute information of each business entity is uniformly modeled and standardized. By establishing an entity attribute dictionary, the name, data type, value range, and association rules with other entities of each entity attribute are clarified. At the same time, for entities with many attribute dimensions or complex hierarchical relationships, hierarchical mapping or nested structures are introduced to maintain the consistency and scalability of data hierarchy, and finally a list of business entities is generated.

[0018] Furthermore, by combining the field definitions and business semantics of the structured standard dataset, a method integrating rule matching, feature extraction, and manual verification is used to identify and classify business entities, as detailed below:

[0019] For a structured standard dataset D={f1,f2,…,f i ,…,f n For each field in}, perform semantic label mapping. For field f i Define its semantic tag set as S(f) i ):

[0020] S(f i )={s1,s2,…,s j ,…,s m},s j ∈Ontology Power ;

[0021] Among them, Ontology Power A predefined semantic dictionary for power business;

[0022] The matching between fields and semantic tags is determined by the following weighting function:

[0023] ;

[0024] Among them, Sim text Sim is a text similarity algorithm based on field names. context α1 and α2 are semantic similarity based on context table structure or field association; α1 and α2 are empirical weights.

[0025] Fields that match a specific semantic category are automatically included in the corresponding entity candidate set;

[0026] For the candidate field set, entity feature vectors are constructed using statistical distribution features and type features:

[0027] ;

[0028] Among them, t i For field data type; c i For non-empty rate or completeness; d i For data uniqueness; r i This is a correlation indicator between a field and other fields.

[0029] Using clustering algorithms to analyze x i Vectors are clustered to form clusters of fields with similar features, and the cluster centers are regarded as feature vectors μ of specific entity types. k ;

[0030]

[0031] Validation of μ through manual and rule-based definitions k The corresponding semantic type is used to determine the entity category;

[0032] After the entity type is determined, extract the corresponding attribute set A for each type of entity. e ={a1,a2,...,a n} is defined by the semantic rule base.

[0033] Furthermore, based on the business entity list, entity alignment and data fusion are performed to obtain a multi-source fused data view, where each entity has a globally unique identifier, as detailed below:

[0034] First, construct the field mapping matrix, assuming the source dataset is D1, D2, ..., D... n Their field sets are F1, F2, ..., F n Define the field similarity matrix:

[0035] ;

[0036] Where Sim is the comprehensive semantic similarity function: This represents the feature representation of the i-th entity in the p-th data source. This represents the feature representation of the j-th entity in the q-th data source;

[0037] ;

[0038] Among them, Sim text For name-based text similarity; Sim type For data type structure matching; Sim contex t represents the context relationship based on the field; β1, β2, and β3 are weight coefficients;

[0039] When M ij >τ f , τ f To set a threshold, fields are assumed to be mappable, forming a field-level mapping table M. F ;

[0040] After aligning the field structures, cross-source matching is performed on the actual data instance. Let E be the set of similar entities in the two data sources. (1) and E (2) ;

[0041] Define a cross-source entity similarity function:

[0042] ;

[0043] in, Sim represents the value of the k-th corresponding attribute for two entities; k Different algorithms are selected based on the attribute type; ω k The weight coefficient for the k-th attribute; Represents the i-th entity in the first data source. With the j-th entity in the second data source The overall attribute similarity between them; m represents the total number of attributes used for similarity calculation;

[0044] After entity matching is completed, multi-source attribute data is merged to resolve conflicts and inconsistencies. Let the attribute values ​​of the same entity E in different sources be {v1, v2, ..., v...}. n The fusion strategy selects based on attribute type, ultimately resulting in a unified set of entity attributes.

[0045] ;

[0046] ;

[0047] in, "a" refers to the fused set of entities; k This represents the name or identifier of the k-th attribute; This represents the final value of the k-th attribute after attribute fusion; This represents the attribute value after fusion;

[0048] Fusion strategy selected by attribute type:

[0049] Time-first method: Take the data with the latest timestamp;

[0050] ;

[0051] Credibility-weighted method: Synthesized based on the reliability weight of the data source;

[0052] ;

[0053] Majority voting: Select the value that appears most frequently;

[0054] ;

[0055] in, Indicates the attribute value v i Number of times it appears across all sources; This indicates taking the value that makes count(v) i The largest attribute value.

[0056] Furthermore, based on the multi-source fused data view, through knowledge extraction and relation discovery, entity data and relation triples are obtained, as follows:

[0057] Construct a knowledge ontology for the power sector, defining allowed combinations of entity types and sets of relationships:

[0058] ;

[0059] Among them, R allow The set of allowed relations, i.e., the types of relations that can be considered or extracted from the graph; r ab For entity E a and E b A specific relationship between them; E a E b These are different types of entities in the power knowledge graph; E represents a and E b Directed or undirected relationships can be established between them; Ontology Power This represents the set of relations defined in the power domain ontology;

[0060] When two types of entities are found to co-occur frequently in several instances within the fused data, their potential relationship is determined, and the formula for calculating the association strength is as follows:

[0061] ;

[0062] in, For entity E a and E b Relevance score under relation r; Under relation r, E a and E b The number of times they appear together; count(E) a ) is entity E a The total number of times something appears in all relations; count(E b ) is entity E b The total number of times it appears in all relations;

[0063] If RelScore>τ r , τ r If the preset threshold is met, then it is confirmed as a candidate relationship;

[0064] For business records, work order descriptions, or complaint texts, natural language processing (NLP) methods are used to extract potential relations, identify entities in the power sector from the text, and build a syntactic dependency tree to determine semantic subjects and objects, thereby deriving triples.

[0065] Calculate relational phrase similarity using a word vector model:

[0066] ;

[0067] Among them, Sim sem (r i ,r j ) represents relation r i and r j Semantic similarity between them; Relation embedding vector and Cosine similarity between them;

[0068] If Sim sem If the value is greater than 0.8, then semantically similar relations will be merged.

[0069] The fused relationship may contain noise and uncertainty, requiring the calculation of confidence weights ω, which are derived from multi-source evidence.

[0070] ;

[0071] Among them, P struct The reliability of structural relationships is determined based on the strength of field associations; P rule P represents the rule validation pass rate. text α3, α4, and α5 are the proportion of text mentions; α3, α4, and α5 are weighting coefficients; when ω > τ conf At that time, τ conf If a preset threshold is set, relation triples are retained; otherwise, they are discarded or temporarily designated as candidate relations. Finally, entity pairs and relation extraction results are uniformly mapped to a set of knowledge triples.

[0072] Furthermore, entities and relationships are mapped into a graph database operable form to construct a power customer relationship knowledge graph, as detailed below:

[0073] Based on the knowledge triples generated by S3, we map them to the node and edge structure in the graph database to obtain G=(V,E):

[0074] Where, V={v i |v i =f(E i )} represents the set of nodes; E={e ij =(v i ,r ij ,v j ,ω ij )} represents the set of edges; ω ij The confidence level of the relationship is used as the edge weight;

[0075] The node mapping function is defined as follows:

[0076] ;

[0077] Among them, GUID i A globally unique identifier for an entity; Type(E) i ) represents the node type; A i A collection of node attributes;

[0078] The edge mapping function is defined as:

[0079] ;

[0080] The output is a structured graph database object:

[0081] G=(V,E,Ω);

[0082] Where, Ω={ω ij} represents the confidence matrix for all edge weights.

[0083] Further optimization of the electricity customer relationship knowledge graph is as follows:

[0084] Create an index set for key query attributes:

[0085] ;

[0086] Where IndexSet represents the set of indices; IDX(a) represents the index corresponding to attribute a; Represents the set of key attributes of the i-th object or entity;

[0087] Using the confidence threshold τ ω Control plot density:

[0088] ;

[0089] in, This is the filtered set of edges;

[0090] Optimize customer group and device hierarchy structure using graph clustering algorithm, and compute node community tags:

[0091] ;

[0092] Where Q(c) is the scoring function corresponding to category c; and v is the node's scoring function. i Assigned community tags:

[0093] The final output is a knowledge graph of electricity customer relationships: ;

[0094] Where Σ is the definition of the graph structure pattern.

[0095] Furthermore, based on the complete knowledge graph of power customer relationships, graph visualization is achieved using Cytoscape, specifically as follows: First, a triplet dataset (h,r,t,ω) is exported from the graph database and converted to JSON or GraphML format supported by Cytoscape; each entity node contains a unique identifier, entity type, and key attributes; each relationship edge contains relationship type and confidence weight; then, visual encoding design is performed through attribute mapping functions; a hybrid model combining force-oriented layout and geographic mapping is adopted. The force-oriented layout is based on the spring-repulsion model, using attraction and repulsion functions to achieve spatial equilibrium of the network; while the geographic embedding layout maps nodes to real geographic space based on latitude and longitude information, achieving a unified display of business space and physical space.

[0096] The present invention has the following beneficial effects:

[0097] 1. This invention greatly reduces manual intervention by automating data acquisition, preprocessing, entity recognition, and attribute extraction, thereby improving the efficiency of relation graph construction. At the same time, multi-source data fusion and entity alignment technologies effectively eliminate redundancy and conflicts between various business data, ensuring data accuracy and consistency.

[0098] 2. This invention can unify the customer entity view under different business systems and realize the management of globally unique identifiers. It not only improves the flexibility of data fusion, but also helps to discover more implicit and potential customer value and business connections through knowledge extraction and relationship discovery, providing data support for marketing, service and risk management.

[0099] 3. This invention not only efficiently implements knowledge triples into graph database systems, enabling efficient retrieval and storage, but also leverages the Cytoscape visualization platform to support intuitive display and analysis of power customer relationship graphs, helping enterprises enhance their core competitiveness in customer relationship management, service innovation, and intelligent decision-making. Attached Figure Description

[0100] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0101] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0102] refer to Figure 1 In this embodiment, a method for intelligently constructing a power customer relationship graph based on multi-source data fusion is provided, including the following steps:

[0103] S1: Obtain all customer-related data and preprocess it to obtain a structured standard dataset;

[0104] S2: Based on a structured standard dataset, identify business entities and their attribute information, and obtain a list of business entities, including the type of each entity and its attribute characteristics;

[0105] S3: Based on the business entity list, perform entity alignment and data fusion to obtain a multi-source fused data view, where each entity has a globally unique identifier;

[0106] S4: Based on a multi-source fusion data view, entity data and relation triples are obtained through knowledge extraction and relation discovery;

[0107] S5: Map entities and relationships to a graph database operable form, construct and optimize the power customer relationship knowledge graph to obtain a complete power customer relationship knowledge graph;

[0108] S6: Based on the complete knowledge graph of electricity customer relationships, visualize the graph using Cytoscape.

[0109] In this embodiment, the full customer-related data includes basic customer information (name, address, contact information, electricity usage type, etc.) from the marketing system, electricity consumption data (electricity consumption, electricity bill, electricity consumption curve, peak, valley, and normal electricity consumption, etc.) from the metering system, spatial location data (latitude and longitude coordinates, administrative divisions, power supply area affiliation, etc.) from the geographic information system, and business records (installation applications, fault reports, complaint handling, payment records, etc.) from the customer service system. By establishing a unified data interface and ETL (extract, transform, load) process, automated data collection from different database platforms is achieved, ensuring data integrity and timeliness.

[0110] In this embodiment, preprocessing is performed as follows: The collected raw data undergoes comprehensive quality assessment and cleaning, including identifying and marking missing, null, and outlier values; unifying coding standards, time formats, and address specifications across different systems; simultaneously, a duplicate data detection algorithm is used to process redundant records of the same customer in different systems, establishing a master data management mechanism to ensure the principle of data uniqueness for each customer; based on power industry data standards and internal enterprise data specifications, the cleaned data undergoes standardization and structural reconstruction, establishing a unified data model and dictionary, standardizing customer numbers, equipment asset codes, and key identifiers such as geographic location codes; unifying data formats, unit dimensions, and precision requirements; redesigning the data table structure, optimizing storage methods, and improving data access efficiency, ultimately forming a structured standard dataset covering multi-dimensional information on customer entities, electricity consumption behavior, geographical distribution, and business interactions, laying a high-quality data foundation for subsequent entity identification, relationship extraction, and knowledge graph construction.

[0111] In this embodiment, based on a structured standard dataset, business entities and their attribute information are identified to obtain a list of business entities. Specifically, the following steps are taken: Combining the field definitions and business semantics of the structured standard dataset, a method integrating rule matching, feature extraction, and manual verification is used to identify and classify business entities. Entity types include, but are not limited to: customer entities (natural person customers, legal person customers, key customers, etc.), equipment entities (distribution transformers, substations, lines, metering devices, etc.), geographical entities (power supply areas, administrative divisions, geographical coordinates, etc.), and business entities (installation work orders, payment records, fault events, marketing plans, etc.). For each type of entity, corresponding attribute information is further extracted, such as the basic attributes of customer entities (customer number, customer name, electricity type, credit rating, etc.) and the technical attributes of equipment entities (equipment number, rated capacity, operating status, etc.), thereby forming a set of entity types with clear boundaries and definitions.

[0112] After completing entity identification and classification, the attribute information of each business entity is uniformly modeled and standardized. By establishing an entity attribute dictionary, the name, data type, value range, and association rules with other entities of each entity attribute are clarified. At the same time, for entities with many attribute dimensions or complex hierarchical relationships (such as multi-level power supply equipment or hierarchical geographical areas), hierarchical mapping or nested structures are introduced to maintain the consistency and scalability of data hierarchy. Finally, a business entity list is generated, which systematically lists all entity types, key attribute fields, and attribute feature descriptions, providing a precise and standardized business semantic foundation for entity alignment and data fusion.

[0113] In this embodiment, by combining the field definitions and business semantics of the structured standard dataset, a method combining rule matching, feature extraction, and manual verification is used to identify and classify business entities, as detailed below:

[0114] For a structured standard dataset D={f1,f2,…,f i ,…,f n For each field in}, perform semantic label mapping. For field f i Define its semantic tag set as S(f) i ):

[0115] S(f i )={s1,s2,…,s j ,…,s m},s j ∈Ontology Power ;

[0116] Among them, Ontology PowerFor a predefined semantic dictionary of power business, such as {customer number, device capacity, administrative division, application number, GIS coordinates, …};

[0117] The matching between fields and semantic tags is determined by the following weighting function:

[0118] ;

[0119] Among them, Sim text For text similarity based on field names (such as edit distance or cosine similarity); Sim context α1 and α2 are semantic similarity based on context table structure or field association; α1 and α2 are empirical weights.

[0120] Fields matching a specific semantic category are automatically assigned to the corresponding entity candidate set; for example: matching {customer ID, customer name} → customer entity candidate; matching {transformer capacity, line ID} → equipment entity candidate;

[0121] For the candidate field set, entity feature vectors are constructed using statistical distribution features and type features:

[0122] ;

[0123] Among them, t i For field data types (numeric, character, time, etc.); c i For non-empty rate or completeness; d i For data uniqueness (e.g., whether it is a primary key candidate); r i The correlation index between a field and other fields (such as mutual information MI or Pearson correlation coefficient);

[0124] Clustering algorithms (such as K-means or DBSCAN) are used to analyze x. i Vectors are clustered to form clusters of fields with similar features, and the cluster centers are regarded as feature vectors μ of specific entity types. k ;

[0125]

[0126] Validation of μ through manual and rule-based definitions k The corresponding semantic type determines the entity category: "Customer", "Equipment", "Geographic", "Business".

[0127] After the entity type is determined, extract the corresponding attribute set A for each type of entity. e ={a1,a2,...,a n}, defined by the semantic rule base (as shown in Table 1).

[0128] Table 1

[0129]

[0130] In this embodiment, entity alignment and data fusion are performed based on the business entity list to obtain a multi-source fused data view, wherein each entity has a globally unique identifier, as detailed below:

[0131] First, construct the field mapping matrix, assuming the source dataset is D1, D2, ..., D... n Their field sets are F1, F2, ..., F n Define the field similarity matrix:

[0132] ;

[0133] Where Sim is the comprehensive semantic similarity function: This represents the feature representation of the i-th entity in the p-th data source. This represents the feature representation of the j-th entity in the q-th data source;

[0134] ;

[0135] Among them, Sim text For name-based text similarity (e.g., cosine or Jaccard); Sim type For matching based on data type structure (e.g., int / string / date); Sim contex t represents the contextual relationship based on the field (e.g., belonging to the customer table and the equipment table); β1, β2, and β3 are weighting coefficients.

[0136] When M ij >τ f , τ f To set a threshold, fields are assumed to be mappable, forming a field-level mapping table M. F ;

[0137] After aligning the field structures, cross-source matching is performed on the actual data instance. Let E be the set of similar entities in the two data sources. (1) and E (2) ;

[0138] Define a cross-source entity similarity function:

[0139] ;

[0140] in, Sim represents the value of the k-th corresponding attribute for two entities; kDifferent algorithms are selected based on the attribute type (such as string edit distance, numerical difference, spatial distance, time difference, etc.); ω k The weight coefficient for the k-th attribute; Represents the i-th entity in the first data source. With the j-th entity in the second data source The overall attribute similarity between them; m represents the total number of attributes used for similarity calculation;

[0141] After entity matching is completed, multi-source attribute data is merged to resolve conflicts and inconsistencies. Let the attribute values ​​of the same entity E in different sources be {v1, v2, ..., v...}. n The fusion strategy selects based on attribute type, ultimately resulting in a unified set of entity attributes.

[0142] ;

[0143] ;

[0144] in, "a" refers to the fused set of entities; k This represents the name or identifier of the k-th attribute; This represents the final value of the k-th attribute after attribute fusion; This represents the attribute value after fusion;

[0145] Fusion strategy selected by attribute type:

[0146] Time-based algorithm: Retrieves data with the latest timestamp.

[0147] ;

[0148] Reliability-based method: Synthesized based on the reliability weights of the data source;

[0149] ;

[0150] Voting-based majority voting: The value that appears most frequently is selected.

[0151] ;

[0152] in, Indicates the attribute value v i Number of times it appears across all sources; This indicates taking the value that makes count(v) i The largest attribute value.

[0153] In this embodiment, based on a multi-source fused data view, entity data and relation triples are obtained through knowledge extraction and relation discovery, as follows:

[0154] Construct a knowledge ontology for the power sector, defining allowed combinations of entity types and sets of relationships:

[0155] ;

[0156] Among them, R allow The set of allowed relations, i.e., the types of relations that can be considered or extracted from the graph; r ab For entity E a and E b A specific relationship between them; E a E b For different types of entities in the power knowledge graph (such as customers, equipment, etc.); E represents a and E b Directed or undirected relationships can be established between them; Ontology Power This represents the set of relations defined in the power domain ontology;

[0157] When two types of entities are found to co-occur frequently in several instances within the fused data (e.g., a customer and power supply equipment belonging to the same distribution area), their potential relationship is determined, and the formula for calculating the association strength is as follows:

[0158] ;

[0159] in, For entity E a and E b Relevance score under relation r; Under relation r, E a and E b The number of times they co-occur (or are jointly linked); count(E) a ) is entity E a The total number of times something appears in all relations; count(E b ) is entity E b The total number of times it appears in all relations;

[0160] If RelScore>τ r , τ r If a preset threshold (such as 0.5) is set, then it is confirmed as a candidate relationship;

[0161] Example: High-frequency co-occurrence (customer, transformer area) → Relationship type "Power supply association"; Same address (customer, geographical location) → Relationship type "Geographical distribution";

[0162] For business records, work order descriptions, or complaint texts, natural language processing (NLP) methods are used to extract potential relations, identify entities in the power sector from the text, and build a syntactic dependency tree to determine semantic subjects and objects, thereby deriving triples.

[0163] Calculate relational phrase similarity using word vector models (such as Word2Vec, BERT):

[0164] ;

[0165] Among them, Sim sem (r i ,r j ) represents relation r i and r j Semantic similarity between them; Relation embedding vector and Cosine similarity between them;

[0166] If Sim sem If the value is greater than 0.8, semantically similar relationships (such as "connected to" and "belonging to") will be merged.

[0167] The fused relationship may contain noise and uncertainty, requiring the calculation of confidence weight ω, which is derived by integrating multi-source evidence (structured data support, rule support, text mention frequency, etc.).

[0168] ;

[0169] Among them, P struct The reliability of structural relationships is determined based on the strength of field associations; P rule P represents the rule validation pass rate. text α represents the percentage of text mentions; α3, α4, and α5 are weighting coefficients (usually α3>α4>α5); when ω>τ conf At that time, τ conf If a preset threshold is set, relation triples are retained; otherwise, they are discarded or temporarily designated as candidate relations. Finally, entity pairs and relation extraction results are uniformly mapped to a set of knowledge triples.

[0170] In this embodiment, entities and relationships are mapped into a graph database operable form to construct a knowledge graph of power customer relationships, as detailed below:

[0171] Based on the knowledge triples generated by S3, we map them to a node and edge structure in a graph database, resulting in G=(V,E):

[0172] Where, V={v i |v i =f(E i)} represents the set of nodes; E={e ij =(v i ,r ij ,v j ,ω ij )} represents the set of edges; ω ij The confidence level of the relationship is used as the edge weight;

[0173] The node mapping function is defined as follows:

[0174] ;

[0175] Among them, GUID i A globally unique identifier for an entity; Type(E) i ) represents the node type (customer, device, area, geographical location, etc.); A i A collection of node attributes;

[0176] The edge mapping function is defined as:

[0177] ;

[0178] The output is a structured graph database object:

[0179] G=(V,E,Ω);

[0180] Where, Ω={ω ij} represents the confidence matrix for all edge weights.

[0181] In this embodiment, the optimization of the power customer relationship knowledge graph is as follows:

[0182] Create an index set for key query attributes:

[0183] ;

[0184] Where IndexSet represents the set of indices; IDX(a) represents the index corresponding to attribute a; Represents the set of key attributes of the i-th object or entity;

[0185] Using the confidence threshold τ ω Control plot density:

[0186] ;

[0187] in, This is the filtered set of edges;

[0188] Optimize customer groups and device hierarchy using graph clustering algorithm (Louvain community detection), and compute node community tags:

[0189] ;

[0190] Where Q(c) is the scoring function corresponding to category c; and v is the node's scoring function. i Assigned community tags:

[0191] The final output is a knowledge graph of electricity customer relationships: ;

[0192] Where Σ is the definition of the graph structure pattern.

[0193] In this embodiment, based on the complete power customer relationship knowledge graph, graph visualization is achieved using Cytoscape, as follows: First, a triplet dataset (h,r,t,ω) is exported from a graph database (Neo4j, JanusGraph, etc.) and converted to JSON or GraphML format supported by Cytoscape; each entity node contains a unique identifier, entity type, and key attributes; each relationship edge contains relationship type and confidence weight; then, visual encoding design is performed through attribute mapping functions: node color is determined by entity type, for example, customers are blue, equipment is orange, transformer area is green, and events are red; node size is dynamically adjusted according to degree centrality D(vi), so that core nodes have higher visual weight; edge thickness, color, and transparency are determined by the relationship confidence ωω mapping, with high-confidence edges being more prominent; a hybrid model combining force-directed layout and geographic mapping is adopted. The force-directed layout is based on the spring-repulsion model, using attraction and repulsion functions to achieve spatial balance of the network; while the geographic embedding layout maps nodes to real geographic space based on latitude and longitude information, realizing a unified display of business space and physical space. Utilizing Cytoscape's programmable capabilities (PyCytoscape or Cytoscape.js API), the following functions are implemented: Node expansion and neighborhood retrieval: Clicking a node automatically loads its first-degree neighbors, dynamically displaying the customer-device-station link; Relationship filtering and cluster highlighting: Filtering the network based on relationship type or confidence level, applying the Louvain algorithm to achieve visual community clustering; Path analysis and centrality visualization: Calculating the shortest path to track event propagation chains, visualizing and highlighting key nodes using degree centrality and betweenness centrality indicators.

[0194] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0195] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0196] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0197] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0198] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for intelligent construction of power customer relationship graphs based on multi-source data fusion, characterized in that, Includes the following steps: S1: Obtain all customer-related data and preprocess it to obtain a structured standard dataset; S2: Based on a structured standard dataset, identify business entities and their attribute information, and obtain a list of business entities, including the type of each entity and its attribute characteristics; S3: Based on the business entity list, perform entity alignment and data fusion to obtain a multi-source fused data view, where each entity has a globally unique identifier; S4: Based on a multi-source fusion data view, entity data and relation triples are obtained through knowledge extraction and relation discovery; S5: Map entities and relationships to a graph database operable form, construct and optimize the power customer relationship knowledge graph to obtain a complete power customer relationship knowledge graph; S6: Based on the complete knowledge graph of electricity customer relationships, visualize the graph using Cytoscape.

2. The intelligent construction method for power customer relationship graph based on multi-source data fusion according to claim 1, characterized in that, The comprehensive customer-related data includes basic customer information from the marketing system, electricity consumption data from the metering system, spatial location data from the geographic information system, and business records from the customer service system. By establishing a unified data interface and ETL (extract, transform, load) process, the system enables automated data collection from different database platforms, ensuring data integrity and timeliness.

3. The intelligent construction method for power customer relationship graph based on multi-source data fusion according to claim 2, characterized in that, The preprocessing is as follows: The collected raw data undergoes comprehensive quality assessment and cleaning, including identifying and marking missing, null, and outlier values; unifying coding standards, time formats, and address specifications across different systems; simultaneously, a duplicate data detection algorithm is used to process redundant records of the same customer in different systems, establishing a master data management mechanism to ensure the principle of data uniqueness for each customer; based on power industry data standards and internal enterprise data specifications, the cleaned data undergoes standardization and structural reconstruction, establishing a unified data model and dictionary, standardizing customer numbers, equipment asset codes, and key identifiers for geographic location codes; unifying data formats, unit dimensions, and accuracy requirements; redesigning the data table structure, optimizing storage methods, and improving data access efficiency, ultimately forming a structured standard dataset covering multi-dimensional information on customer entities, electricity consumption behavior, geographical distribution, and business interactions.

4. The intelligent construction method for power customer relationship graph based on multi-source data fusion according to claim 2, characterized in that, The process of identifying business entities and their attribute information based on a structured standard dataset and obtaining a list of business entities is as follows: Combining the field definitions and business semantics of the structured standard dataset, a method combining rule matching, feature extraction, and manual verification is used to identify and classify business entities. Entity types include, but are not limited to: customer entities, device entities, geographic entities, and business entities; and corresponding attribute information is further extracted for each type of entity. After completing entity identification and classification, the attribute information of each business entity is uniformly modeled and standardized. By establishing an entity attribute dictionary, the name, data type, value range, and association rules with other entities of each entity attribute are clarified. At the same time, for entities with many attribute dimensions or complex hierarchical relationships, hierarchical mapping or nested structures are introduced to maintain the consistency and scalability of data hierarchy, and finally a list of business entities is generated.

5. The intelligent construction method for power customer relationship graph based on multi-source data fusion according to claim 4, characterized in that, The method combines field definitions and business semantics from a structured standard dataset, employing a combination of rule matching, feature extraction, and manual verification to identify and classify business entities, as detailed below: For a structured standard dataset D={f1,f2,…,f i ,…,f n For each field in}, perform semantic label mapping. For field f i Define its semantic tag set as S(f) i ): S(f i )={s1,s2,…,s j ,…,s m },s j ∈Ontology Power ; Among them, Ontology Power A predefined semantic dictionary for power business; The matching between fields and semantic tags is determined by the following weighting function: ； Among them, Sim text Sim is a text similarity algorithm based on field names. context α1 and α2 are semantic similarity based on context table structure or field association; α1 and α2 are empirical weights. Fields that match a specific semantic category are automatically included in the corresponding entity candidate set; For the candidate field set, entity feature vectors are constructed using statistical distribution features and type features: ； Among them, t i For field data type; c i For non-empty rate or completeness; d i For data uniqueness; r i This is a correlation indicator between a field and other fields. Using clustering algorithms to analyze x i Vectors are clustered to form clusters of fields with similar features, and the cluster centers are regarded as feature vectors μ of specific entity types. k ; ； Validation of μ through manual and rule-based definitions k The corresponding semantic type is used to determine the entity category; After the entity type is determined, extract the corresponding attribute set A for each type of entity. e ={a1,a2,...,a n } is defined by the semantic rule base.

6. The intelligent construction method for power customer relationship graph based on multi-source data fusion according to claim 1, characterized in that, The process involves aligning and fusing entities based on the business entity list to obtain a multi-source fused data view, where each entity has a globally unique identifier, as detailed below: First, construct the field mapping matrix, assuming the source dataset is D1, D2, ..., D... n Their field sets are F1, F2, ..., F n Define the field similarity matrix: ； Where Sim is the comprehensive semantic similarity function: This represents the feature representation of the i-th entity in the p-th data source. This represents the feature representation of the j-th entity in the q-th data source; ； Among them, Sim text For name-based text similarity; Sim type For data type structure matching; Sim contex t represents the context relationship based on the field; β1, β2, and β3 are weight coefficients; When M ij >τ f , τ f To set a threshold, fields are assumed to be mappable, forming a field-level mapping table M. F ; After aligning the field structures, cross-source matching is performed on the actual data instance. Let E be the set of similar entities in the two data sources. (1) and E (2) ; Define a cross-source entity similarity function: ； in, Sim represents the value of the k-th corresponding attribute for two entities; k Different algorithms are selected based on the attribute type; ω k The weight coefficient for the k-th attribute; Represents the i-th entity in the first data source. With the j-th entity in the second data source The overall attribute similarity between them; m represents the total number of attributes used for similarity calculation; After entity matching is completed, multi-source attribute data is merged to resolve conflicts and inconsistencies. Let the attribute values ​​of the same entity E in different sources be {v1, v2, ..., v...}. n The fusion strategy selects based on attribute type, ultimately resulting in a unified set of entity attributes. ； ； in, "a" refers to the fused set of entities; k This represents the name or identifier of the k-th attribute; This represents the final value of the k-th attribute after attribute fusion; This represents the attribute value after fusion; Fusion strategy selected by attribute type: Time-first method: Take the data with the latest timestamp; ； Credibility-weighted method: Synthesized based on the reliability weight of the data source; ； Majority voting: Select the value that appears most frequently; ； in, Indicates the attribute value v i Number of times it appears across all sources; This indicates taking the value that makes count(v) i The largest attribute value.

7. The intelligent construction method for power customer relationship graph based on multi-source data fusion according to claim 6, characterized in that, The multi-source fusion data view, through knowledge extraction and relation discovery, yields entity data and relation triples, as detailed below: Construct a knowledge ontology for the power sector, defining allowed combinations of entity types and sets of relationships: ； Among them, R allow The set of allowed relations, i.e., the types of relations that can be considered or extracted from the graph; r ab For entity E a and E b A specific relationship between them; E a E b These are different types of entities in the power knowledge graph; E represents a and E b Directed or undirected relationships can be established between them; Ontology Power This represents the set of relations defined in the power domain ontology; When two types of entities are found to co-occur frequently in several instances within the fused data, their potential relationship is determined, and the formula for calculating the association strength is as follows: ； in, For entity E a and E b Relevance score under relation r; Under relation r, E a and E b The number of times they appear together; count(E) a ) is entity E a The total number of times something appears in all relations; count(E b ) is entity E b The total number of times it appears in all relations; If RelScore>τ r , τ r If the preset threshold is met, then it is confirmed as a candidate relationship; For business records, work order descriptions, or complaint texts, natural language processing (NLP) methods are used to extract potential relations, identify entities in the power sector from the text, and build a syntactic dependency tree to determine semantic subjects and objects, thereby deriving triples. Calculate relational phrase similarity using a word vector model: ； Among them, Sim sem (r i ,r j ) represents relation r i and r j Semantic similarity between them; Relation embedding vector and Cosine similarity between them; If Sim sem If the value is greater than 0.8, then semantically similar relations will be merged. The fused relationship may contain noise and uncertainty, requiring the calculation of confidence weights ω, which are derived from multi-source evidence. ； Among them, P struct The reliability of structural relationships is determined based on the strength of field associations; P rule P represents the rule validation pass rate. text α3, α4, and α5 are the proportion of text mentions; α3, α4, and α5 are weighting coefficients; when ω > τ conf At that time, τ conf If a preset threshold is set, relation triples are retained; otherwise, they are discarded or temporarily designated as candidate relations. Finally, entity pairs and relation extraction results are uniformly mapped to a set of knowledge triples.

8. The intelligent construction method for power customer relationship graph based on multi-source data fusion according to claim 7, characterized in that, The process of mapping entities and relationships into a graph database-operable form to construct a power customer relationship knowledge graph is as follows: Based on the knowledge triples generated by S3, we map them to the node and edge structure in the graph database to obtain G=(V,E): Where, V={v i |v i =f(E i )} represents the set of nodes; E={e ij =(v i ,r ij ,v j ,ω ij )} represents the set of edges; ω ij The confidence level of the relationship is used as the edge weight; The node mapping function is defined as follows: ； Among them, GUID i A globally unique identifier for an entity; Type(E) i ) represents the node type; A i A collection of node attributes; The edge mapping function is defined as: ； The output is a structured graph database object: G=(V,E,Ω); Where, Ω={ω ij } represents the confidence matrix for all edge weights.

9. The intelligent construction method for power customer relationship graph based on multi-source data fusion according to claim 7, characterized in that, The optimization of the power customer relationship knowledge graph is as follows: Create an index set for key query attributes: ； Where IndexSet represents the set of indices; IDX(a) represents the index corresponding to attribute a; Represents the set of key attributes of the i-th object or entity; Using the confidence threshold τ ω Control plot density: ； in, This is the filtered set of edges; Optimize customer group and device hierarchy structure using graph clustering algorithm, and compute node community tags: ； Where Q(c) is the scoring function corresponding to category c; and v is the node's scoring function. i Assigned community tags: The final output is a knowledge graph of electricity customer relationships: ; Where Σ is the definition of the graph structure pattern.

10. The intelligent construction method for power customer relationship graph based on multi-source data fusion according to claim 1, characterized in that, The process involves visualizing a complete knowledge graph of power customer relationships using Cytoscape, as follows: First, a triplet dataset (h, r, t, ω) is exported from the graph database and converted to JSON or GraphML format supported by Cytoscape. Each entity node contains a unique identifier, entity type, and key attributes. Each relationship edge contains relationship type and confidence weight. Next, visual encoding design is performed using attribute mapping functions. A hybrid model combining force-directed layout and geographic embedding is adopted. The force-directed layout is based on the spring-repulsion model, using attraction and repulsion functions to achieve spatial equilibrium of the network. The geographic embedding layout maps nodes to real geographic space based on latitude and longitude information, achieving a unified display of business space and physical space.

Images