A substation work ticket knowledge graph construction method based on a large language model

Abstract

The application discloses a substation work ticket knowledge graph construction method based on a large language model, and belongs to the field of energy power and artificial intelligence. The method comprises the following steps: based on standardized electrical drawing annotation data, independent knowledge graphs are generated through data reorganization preprocessing, rough knowledge graph construction and attribute supplement reasoning; all independent knowledge graphs are divided into framework layer graphs and detail layer graphs according to voltage levels, and hierarchical fusion strategies are adopted to complete full graph fusion; breakpoint detection is carried out by using an isolated point detection algorithm based on degree centrality, connectivity analysis is carried out through depth-first search, and non-connected subgraphs are re-fused; abnormal nodes are screened based on electrical connection rules, a large language model is input to automatically fill in Agent to generate correction suggestions, and graph optimization is completed. Through the cooperation of the large model Agent and multiple algorithms, the application improves the efficiency and accuracy of electrical data structuring, and provides knowledge network support for substation automation operation and fault diagnosis.

Description

Technical Field

[0001] This invention belongs to the fields of energy, power and artificial intelligence, and specifically relates to a method for constructing a knowledge graph of substation work orders based on a large language model. Background Technology

[0002] In the construction and operation and maintenance of substation electrical systems, the interrelationships between internal equipment are often highly complex. Various primary and secondary equipment, power lines, and transformers are interconnected, forming a vast and intricate network. This complex connection logic primarily relies on electrical system diagrams for intuitive representation and organization. However, with the development of industrial automation and intelligence, to achieve automated control, intelligent fault diagnosis, and efficient scheduling of substation electrical processes, traditional unstructured drawings are no longer sufficient to meet the demands of data-driven technologies. Standardized structured data must serve as the core support.

[0003] Although machine vision technology and some structured data modeling methods such as linked lists and adjacency lists have been attempted to be applied to the automatic recognition and information extraction of electrical system diagrams, their recognition accuracy and efficiency still have significant bottlenecks due to problems such as inconsistent drawing formats, diverse component symbols, and strong concealment of connection logic, making it difficult to stably output usable structured information.

[0004] Against this backdrop, using advanced structured data formats such as knowledge graphs to characterize the equipment relationships and logical constraints of electrical systems can transform scattered electrical relationships into a computable and reasonable knowledge network, laying a solid data foundation for the automated operation and intelligent management of electrical systems. This has become a preferred solution for overcoming current technical difficulties and improving the digitalization level of electrical systems. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies, provide a structured data foundation for substation work order issuance, and offer a method for constructing a substation work order knowledge graph based on a large language model. This invention can structure the original unstructured and complex circuit diagrams of substations into a standard knowledge graph, providing strong data support for the subsequent automated and intelligent issuance of work orders.

[0006] The specific technical solution adopted in this invention is as follows:

[0007] Firstly, this invention provides a method for constructing a knowledge graph of substation work orders based on a large language model, as detailed below:

[0008] S1, Construct a substation knowledge graph blueprint: Define the attribute categories of electrical entities, the types of relationships between entities, and the numerical types of attribute values ​​to construct a substation knowledge graph blueprint (Schema).

[0009] S2, Design the annotation data format and annotation method based on substation electrical drawings: Based on the substation knowledge graph blueprint obtained in S1, obtain the electrical entity information with annotations, and generate standardized electrical drawing annotation data containing annotation text, labels, location coordinates and unique codes;

[0010] In practical use, this step is based on the substation knowledge graph blueprint obtained in S1, specifies the format and method of annotation data, provides annotation data based on the substation electrical drawings, and ensures that the annotation data meets the needs of graph construction.

[0011] S3, Construct an independent knowledge graph for the substation: Based on the standardized electrical drawing annotation data obtained in S2, an independent knowledge graph is generated through data recombination preprocessing, coarse knowledge graph construction, and attribute supplementation reasoning; wherein, the attribute supplementation reasoning includes source information supplementation and voltage level reasoning;

[0012] In practical use, this step is based on the standardized electrical drawing annotation data obtained from S2, and uses a knowledge graph construction algorithm to transform the raw data of a single electrical drawing into a knowledge graph description language.

[0013] S4, Electrical Graph Layered Fusion: All independent knowledge graphs obtained in S3 are divided into frame layer graphs and detail layer graphs according to voltage level, and a layered fusion strategy is adopted to complete the full graph fusion of frame layer graphs and detail layer graphs.

[0014] In practical use, this step is based on the multiple knowledge graphs obtained by S3, which are classified according to voltage attributes into frame layer graphs and detail layer graphs. A hierarchical fusion strategy is used to fuse the graphs to obtain the equipment association knowledge graph of the entire substation.

[0015] S5, Multi-step verification and correction based on graph structure: Based on the full graph fused in S4, breakpoint detection is performed using an outlier detection algorithm based on degree centrality, connectivity analysis is conducted through depth-first search, and disconnected subgraphs are re-fused; abnormal nodes are screened based on electrical connection rules (e.g., transformer-connection-circuit breaker, circuit breaker-connection-grounding switch), and the large language model is input to automatically complete the agent to generate correction suggestions, thus completing the graph optimization.

[0016] In practical use, this step uses an outlier detection algorithm based on degree centrality to perform breakpoint detection and connectivity analysis on the knowledge graph of all substation equipment obtained in S4, detects isolated entities in the graph, and uses the results of connectivity analysis to perform secondary fusion of the graph. Abnormal nodes are then automatically filled in by the large language model to generate correction suggestions, which are then submitted to manual review.

[0017] As a preferred embodiment, the data reconstruction preprocessing described in S3 employs an information reconstruction algorithm, the specific method of which is as follows:

[0018] For standardized electrical drawing annotation data on a single drawing, two core data types are distinguished: entity annotation data and relation annotation data. The entity annotation data is used to extract unique codes, text content, and tag information, which are then stored in association. Abnormal entity numbers without valid text or tags are filtered out. For the relation annotation data, the relationship initiator, receiver, and relationship type are extracted, and a mapping relationship is established. Finally, all annotation text in the standardized electrical drawing annotation data is cleaned, outliers are removed, and duplicate data is overwritten to ensure data integrity and uniqueness.

[0019] As a preferred embodiment, the voltage level inference described in S3 employs a voltage level propagation algorithm, the specific method of which is as follows:

[0020] First, the voltage levels of core electrical equipment in the substation, including circuit breakers, busbars, and disconnectors, are obtained through a large language model inference agent. Then, starting from the core electrical equipment in the substation, the voltage level attribute is propagated to related equipment through graph traversal, excluding transformer nodes during the propagation process. For the core electrical equipment in the substation that is not covered, it is uniformly marked as having no voltage level attribute, thereby realizing the systematic assignment of voltage levels for all equipment in the substation.

[0021] Preferably, the framework layer diagram in S4 includes a busbar principle wiring diagram and a DC principle wiring diagram; the framework layer diagram fusion adopts an electrical entity alignment algorithm, the specific method of which is as follows:

[0022] For the multiple independent knowledge graphs already constructed in S3, when merging them in pairs, the two sets of electrical entity lists to be merged are extracted. First, the core features of the entity, such as the number, equipment type, and voltage level, are extracted and normalized. After verifying the consistency of the features, the optimal matching entity is selected by calculating the semantic similarity of the entity names. The matching rule is limited to one-to-one matching to generate entity alignment results, so as to ensure accurate matching of the same entity in different framework graphs.

[0023] Preferably, the detailed layer diagrams in S4 include bus configuration wiring diagrams, UPS wiring diagrams, and DC sub-panel configuration wiring diagrams; the detailed layer diagram fusion method is as follows:

[0024] Based on the skeleton map formed after the fusion of the framework layer maps, differentiated fusion rules are adopted for different types of detail layer maps. Among them, the bus configuration wiring diagram focuses on the construction of the connection relationship between load equipment and power supply bus, while the UPS wiring diagram and DC split screen configuration wiring diagram focus on the construction of the subordinate and inclusion relationship between load equipment and DC power supply equipment. When fusion of the framework layer maps, an overlay strategy is adopted to retain all information of the fused entities to the traceability attribute, ensuring data traceability.

[0025] As a preferred embodiment, the outlier detection algorithm based on degree centrality described in S5 is as follows:

[0026] The degree centrality index is used to quantify the connection strength of each node in the entire graph. A threshold is set based on the average connection strength of the graph to filter out isolated nodes with connection strength far below the benchmark. At the same time, a depth-first search is used to traverse the nodes of the entire graph, identify independent subgraphs and trigger secondary fusion to ensure the connectivity of the graph.

[0027] As a preferred embodiment, the specific method for automatically completing the Agent in the input large language model described in S5 is as follows:

[0028] The system receives information on abnormal nodes in the graph and electrical connection rules of the substation. Based on electrical wiring specifications and equipment association logic, it generates suggestions for relationship supplementation and connection correction of abnormal nodes. After review and confirmation, the suggestions are updated to the entire graph, achieving accurate correction of the entire graph.

[0029] In a second aspect, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement the substation work order knowledge graph construction method based on a large language model as described in any of the first aspects.

[0030] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the substation work ticket knowledge graph construction method based on a large language model as described in any of the first aspects.

[0031] Fourthly, the present invention provides a computer electronic device, including a memory and a processor;

[0032] The memory is used to store computer programs;

[0033] The processor is configured to, when executing the computer program, implement the method for constructing a substation work order knowledge graph based on a large language model as described in any of the first aspects.

[0034] Compared with the prior art, the present invention has the following advantages:

[0035] A structured method for the relationship of electrical equipment in substations that conforms to the power standard specification (IEC-61970) is proposed. Based on the traditional manual annotation of data, a multi-step algorithm and large model agent collaboration are introduced to improve the efficiency of data structuring. The data is organized in the form of a knowledge graph, which further enhances the structure of the data. Subsequently, knowledge graph retrieval algorithms can be combined to quickly screen the circuit topology and related equipment. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention.

[0037] Figure 2 This is a schematic diagram of a knowledge graph specific to substations.

[0038] Figure 3 This is a schematic diagram of the information reorganization algorithm.

[0039] Figure 4 This is a schematic diagram of the complete voltage level propagation algorithm.

[0040] Figure 5 This is a flowchart for aligning two electrical entities. Detailed Implementation

[0041] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. Technical features in various embodiments of the present invention can be combined accordingly without mutual conflict.

[0042] like Figure 1 As shown in the figure, this embodiment provides a method for constructing a knowledge graph of substation work orders based on a large language model. It aims to solve the problem of automated management and control caused by the unstructured electrical system diagram and complex equipment associations in substations, and to provide structured data support for the intelligent issuance of work orders.

[0043] This method mainly includes five core steps: First, a substation-specific knowledge graph schema is constructed, clarifying entity attributes (voltage level, equipment type, traceability information) and relationship types (connection, containment); second, based on the Label-Studio platform, annotation rules and data formats are designed to annotate entities, relationships, and drawing names for five types of electrical drawings, generating standardized annotation data; then, through data reorganization preprocessing, coarse knowledge graph construction, and attribute reasoning agent supplementation of traceability information and voltage level, an independent knowledge graph for each drawing is generated; then, the graph is divided into a framework layer and a detail layer according to voltage level, and a layered fusion strategy and electrical entity alignment algorithm are adopted to achieve full-station graph fusion; finally, degree centrality outlier detection, connectivity analysis, and large language model automatic completion agent are used for graph verification, combined with manual review to optimize abnormal nodes.

[0044] The method of the present invention specifically includes the following steps:

[0045] S1: Define the attribute categories of electrical entities, the types of relationships between entities, and the numerical types of attribute values ​​to construct a substation knowledge graph blueprint (Schema).

[0046] In this embodiment, the specific steps are as follows:

[0047] Schema content as follows Figure 2 As shown, electrical entities contain three attributes: voltage level, equipment type, and traceability information. The voltage level attribute corresponds to the actual voltage of the substation's electrical equipment; the equipment type attribute specifies the type of electrical entity; and the traceability information attribute specifies which drawing the diagram information originates from, facilitating subsequent verification. Relationships between electrical entities are categorized into two types: connection and inclusion. Connection relationships are physical connections, while inclusion relationships specify the subordinate relationship of the equipment.

[0048] S2: Based on the substation knowledge graph blueprint obtained in S1, obtain the labeled electrical entity information and generate standardized electrical drawing annotation data containing annotation text, labels, location coordinates and unique codes.

[0049] In other words, based on the substation knowledge graph blueprint obtained from S1, the format and method of annotation data are specified. Annotation data is provided on the basis of the substation electrical drawings, ensuring that the annotation data meets the needs of graph construction. For five types of drawings (busbar principle wiring diagram, busbar configuration wiring diagram, UPS wiring diagram, DC principle wiring diagram, and DC substation configuration wiring diagram), different types of data need to be annotated, mainly including electrical entities, electrical entity relationships, and drawing names.

[0050] In this embodiment, the Label-Studio annotation platform is used in this step. Electrical entities need to be labeled with their names and equipment types. To simplify the annotation process, the equipment type here is a mapping of the actual electrical equipment type. The mapping relationship is shown in Table 1.

[0051] Table 1

[0052]

[0053] Electrical entity relationship annotations are based on the entity annotation results, and connection / inclusion relationships are annotated for them in Label-Studio; the drawing name is annotated separately, and its equipment type is specified as "drawing name". All annotation elements are shown in Table 2.

[0054] Table 2

[0055]

[0056] Finally, the above annotation method is used to generate field-content mapping relationships as shown in Table 3, which will be used for subsequent knowledge graph construction.

[0057] Table 3

[0058]

[0059] Let data be the result set of annotations for a single electrical icon using label-studio. As shown in Table 3, the annotation text content of the nth data can be found by searching data[annotations][0][result][n][value][text][0].

[0060] S3: Based on the standardized electrical drawing annotation data obtained in S2, an independent knowledge graph is generated through data recombination preprocessing, coarse knowledge graph construction, and attribute supplementation reasoning; among which, attribute supplementation reasoning includes source information supplementation and voltage level reasoning.

[0061] In this embodiment, the first step is to reorganize and preprocess the data, initially assembling the annotation information of each entity and cleaning up data with missing annotations and abnormal data formats; secondly, a rough knowledge graph is initially constructed, transforming the annotated data after the previous reorganization process into knowledge graph entities and triples; finally, using the schema pattern specified in S1, the attribute reasoning agent based on electrical rules is used to supplement and reason about the attributes, and finally an independent knowledge graph is constructed for each electrical structure diagram.

[0062] This step specifically includes the following steps:

[0063] S31 Data Reorganization and Preprocessing: This step reorganizes and preprocesses the JSON data blocks obtained in S2. The reorganization steps are as follows: For the annotation data result set `data` of each electrical drawing, iterate through the `data[annotations][0][result]` list. Each annotation data may be annotation text, annotation label, or entity relationship. Distinguish and reorganize the data according to the following algorithm:

[0064] Iterate through `data[annotations][0][result]`. For each `data[annotations][0][result][n]`, check its `value` field. If it exists, read its `id` field. Continue checking if there is a `text` or `labels` field under the `value` field. If there is, read its content, associate it with the `id`, and iterate to the next item. If there is no `text` or `labels` field, or if a field has no content, mark the `id`, move it to the problem entity list, and iterate to the next item until overflow occurs. The algorithm pseudocode is shown in Table 4, and the flowchart is shown in [link to flowchart]. Figure 3 .

[0065] Table 4

[0066]

[0067] The text portion of each field is then processed as follows: spaces are cleaned and outliers are removed. In addition, since the data block contains data from different labeled versions, duplicate data overwriting is also required to ensure that the data block is clean and not redundant.

[0068] The S32 coarse knowledge graph construction uses the unique coded ID generated during annotation to encode each annotated entity, and uses the electrical entity list (entity_list) and entity relationship list (relation_list) generated in S31 to map them to knowledge graph entity names, entity attributes and relationships. The mapping relationship is shown in Table 5.

[0069] Table 5

[0070]

[0071] S33 Attribute Reasoning and Supplementation: Based on the graph mapping relationship shown in Table 5, a knowledge graph containing equipment naming and equipment type can be constructed for a single electrical diagram. According to the knowledge graph schema designed in S1, traceability information and voltage level need to be supplemented.

[0072] Specifically, it includes the following steps:

[0073] S331 Source Information Supplement: Source information is used to ensure information traceability. Specifically, it represents the specific page number and location of the entity from the substation electrical diagram. Therefore, it is necessary to extract the above information from the annotation information. The specific extraction method is to obtain the page number from data[inner_id], construct the label position rectangle from data[annotations][0][result][n][value] based on the four labels x, y, width, and height, and store it. Here, x, y, width, and height are the x-coordinate and y-coordinate of the top left corner vertex of the annotation box, the width of the annotation box, and the height of the annotation box, respectively.

[0074] S332 Voltage Level Inference: Voltage level is a crucial indicator for electrical equipment. Current is conducted through multiple stages via transformer voltage reduction. Different equipment has different voltage levels, commonly 500kV, 10kV, etc. Obtaining the voltage level of each electrical device is essential for subsequent graph fusion. The inference logic is as follows: For a single electrical graph, search for nodes with the equipment type "drawing name" and extract their names; search for nodes with the equipment types "circuit breaker," "busbar," and "disconnector," extract their entity information, integrate them into the voltage level inference agent based on the large model, obtain the voltage levels of entities with the equipment types "circuit breaker," "busbar," and "disconnector," and use a breadth-first graph traversal algorithm to infer the voltage levels of other types of electrical equipment. The inference algorithm is shown in Table 6, and the complete voltage level propagation algorithm flowchart is shown below. Figure 4 As shown.

[0075] Table 6

[0076]

[0077] S4. All independent knowledge graphs obtained in S3 are divided into frame layer graphs and detail layer graphs according to voltage level. A layered fusion strategy is then used to complete the full graph fusion of the frame layer graphs and detail layer graphs to obtain the equipment association knowledge graph of the entire substation (i.e., the full graph).

[0078] In this embodiment, this step specifically includes the following steps:

[0079] S41 Graph Classification and Layering: Based on the voltage level attributes and drawing names supplemented to electrical entities in S332, and utilizing the different naming of different drawings and the different types of equipment they contain, the multiple electrical knowledge graphs constructed in S3 can be divided into five types: busbar principle wiring diagram, busbar configuration wiring diagram, UPS wiring diagram, DC principle wiring diagram, and DC sub-panel configuration wiring diagram. Among them, the busbar principle wiring diagram and the DC principle wiring diagram are used as the framework layer, and the other three types of drawings are used as the detail layer.

[0080] S42 seed graph generation extracts the knowledge graphs of all framework layers, distinguishes the voltage levels of these graphs using the voltage level attribute, selects the framework graph with the highest voltage level as the seed graph to start fusion, and then merges the framework graphs with lower voltage levels into the seed graph, and finally merges the electrical graphs of the detail layers.

[0081] S43 Frame Graph Fusion: After generating the seed graph in S42, the remaining frame graphs are sequentially fused into the seed graph according to voltage level from high to low. The fused graph is then used as the new seed graph for further frame graph fusion. Graph fusion involves aligning electrical entities in two independent knowledge graphs. Since frame graphs at different voltage levels may contain common equipment such as transformers and busbars, an electrical entity alignment algorithm is needed. The overall alignment algorithm logic is shown in Table 7, and the process of aligning the two entities is as follows: Figure 5 .

[0082] Table 7

[0083]

[0084] After obtaining the entity alignment results of the two framework diagrams, entity overlay can be performed. All the information of the overlaid entity is loaded into the traceability information attribute of the overlaid entity to ensure that the entity can still be traced after fusion, which facilitates subsequent verification.

[0085] S44 full-map fusion, after completing the entire S43 process, merges all framework diagrams into a single skeleton diagram, and then sequentially stitches together all the detailed diagrams. Different fusion methods are used depending on the type of detailed diagram. For example, the bus configuration wiring diagram needs to establish a connection between the load equipment and the power supply bus; the UPS wiring diagram and DC sub-panel configuration wiring diagram need to establish an inclusion relationship with the load equipment to illustrate the inclusion relationship between the load equipment and the DC power supply equipment.

[0086] S5, based on the full graph fused from S4, uses an outlier detection algorithm based on degree centrality to detect breakpoints, performs connectivity analysis through depth-first search, and re-fused disconnected subgraphs; it filters out abnormal nodes based on electrical connection rules, inputs a large language model to automatically complete the Agent to generate correction suggestions, and completes the graph optimization after manual review.

[0087] In this embodiment, this step specifically includes the following steps:

[0088] S51 Outlier Detection and Connectivity Analysis: Degree centrality is a fundamental metric for measuring node importance, representing the number of direct connections between a node and other nodes. Outliers are nodes in the graph with very few connections and are disconnected from the core network. By calculating the degree centrality of all nodes and setting a threshold, nodes with degree values ​​far below the graph average are identified as outliers. Connectivity analysis uses depth-first search to traverse the graph nodes, checking if they can be decomposed into multiple subgraphs. If they can be decomposed into multiple subgraphs, they are decomposed into multiple knowledge graphs, and then the graphs are fused again.

[0089] The S52 large language model agent completes the graph. After multiple re-fusions of the graph, it uses predefined expert rules to filter the nodes in the graph, filtering out nodes with abnormal connection counts or connections. These nodes are then input into the large language model-based auto-completion agent along with the corresponding expert rules. Finally, the agent generates modification suggestions.

[0090] This invention improves the efficiency and accuracy of electrical data structuring by using a large model agent and multiple algorithms, providing reliable knowledge network support for substation automated operation and fault diagnosis.

[0091] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A method for constructing a knowledge graph of substation work orders based on a large language model, characterized in that, Specifically as follows: S1: Define the attribute categories of electrical entities, the types of relationships between entities, and the numerical types of attribute values ​​to construct the substation knowledge graph ontology; S2: Based on the substation knowledge graph ontology obtained in S1, obtain the electrical entity information with annotations, and generate standardized electrical drawing annotation data containing annotation text, tags, location coordinates and unique codes; S3: Based on the standardized electrical drawing annotation data obtained in S2, an independent knowledge graph is generated through data recombination preprocessing, coarse knowledge graph construction, and attribute supplementation reasoning; wherein, the attribute supplementation reasoning includes source information supplementation and voltage level reasoning; The coarse knowledge graph construction method is as follows: each labeled entity is encoded using the unique coded ID generated during annotation, and it is mapped to knowledge graph entity names, entity attributes and relationships using a list and an entity relationship list; The traceability information indicates that the entity originates from the specific page number and location of the substation electrical diagram. S4: Divide all independent knowledge graphs obtained in S3 into frame layer graphs and detail layer graphs according to voltage level, and use a layered fusion strategy to complete the full graph fusion of frame layer graphs and detail layer graphs; The framework layer diagram includes a busbar principle wiring diagram and a DC principle wiring diagram; the framework layer diagram fusion adopts an electrical entity alignment algorithm, the specific method of which is as follows: For the multiple independent knowledge graphs already constructed in S3, when merging them in pairs, the two sets of electrical entity lists to be merged are extracted. First, the core features of the entity, such as the number, equipment type, and voltage level, are extracted and normalized. After verifying the consistency of the features, the optimal matching entity is selected by calculating the semantic similarity of the entity names. The matching rule is limited to one-to-one matching to generate entity alignment results, so as to ensure accurate matching of the same entity in different framework graphs. The detailed layer diagrams include bus configuration wiring diagrams, UPS wiring diagrams, and DC substation configuration wiring diagrams; the detailed layer diagram fusion method is as follows: Based on the skeleton map formed after the fusion of the framework layer maps, differentiated fusion rules are adopted for different types of detail layer maps. Among them, the bus configuration wiring diagram focuses on the construction of the connection relationship between load equipment and power supply bus, while the UPS wiring diagram and DC split screen configuration wiring diagram focus on the construction of the subordinate and inclusion relationship between load equipment and DC power supply equipment. When fusioning the framework layer maps, an overlay strategy is adopted to retain all information of the fused entities to the traceability attribute, ensuring data traceability. S5: Based on the full graph fused from S4, breakpoint detection is performed using an outlier detection algorithm based on degree centrality. Connectivity analysis is conducted through depth-first search, and disconnected subgraphs are refused. Abnormal nodes are screened based on electrical connection rules, and the large language model is used to automatically complete the Agent and generate correction suggestions to complete the graph optimization.

2. The method for constructing a substation work order knowledge graph based on a large language model according to claim 1, characterized in that, The data reconstruction preprocessing described in S3 employs an information reconstruction algorithm, the specific method of which is as follows: For standardized electrical drawing annotation data on a single drawing, two core data types are distinguished: entity annotation data and relation annotation data. The entity annotation data is used to extract unique codes, text content, and tag information, which are then stored in association. Abnormal entity numbers without valid text or tags are filtered out. For the relation annotation data, the relationship initiator, receiver, and relationship type are extracted, and a mapping relationship is established. Finally, all annotation text in the standardized electrical drawing annotation data is cleaned, outliers are removed, and duplicate data is overwritten to ensure data integrity and uniqueness.

3. The method for constructing a substation work order knowledge graph based on a large language model according to claim 1, characterized in that, The voltage level inference described in S3 uses a voltage level propagation algorithm, the specific method of which is as follows: First, the voltage levels of core electrical equipment in the substation, including circuit breakers, busbars, and disconnectors, are obtained through a large language model inference agent. Then, starting from the core electrical equipment in the substation, the voltage level attribute is propagated to related equipment through graph traversal, excluding transformer nodes during the propagation process. For the core electrical equipment in the substation that is not covered, it is uniformly marked as having no voltage level attribute, thereby realizing the systematic assignment of voltage levels for all equipment in the substation.

4. The method for constructing a substation work order knowledge graph based on a large language model according to claim 1, characterized in that, The outlier detection algorithm based on degree centrality described in S5 is as follows: The degree centrality index is used to quantify the connection strength of each node in the entire graph. A threshold is set based on the average connection strength of the graph to filter out isolated nodes with connection strength less than the preset threshold. At the same time, a depth-first search is used to traverse the nodes of the entire graph, identify independent subgraphs and trigger secondary fusion to ensure the connectivity of the graph.

5. The method for constructing a substation work order knowledge graph based on a large language model according to claim 1, characterized in that, The specific method for automatically completing the Agent in the input large language model described in S5 is as follows: The system receives information on abnormal nodes in the graph and electrical connection rules of the substation. Based on electrical wiring specifications and equipment association logic, it generates suggestions for relationship supplementation and connection correction of abnormal nodes. After review and confirmation, the suggestions are updated to the entire graph, achieving accurate correction of the entire graph.

6. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it can realize the substation work ticket knowledge graph construction method based on a large language model as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the substation work ticket knowledge graph construction method based on a large language model as described in any one of claims 1 to 5.

8. A computer electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to, when executing the computer program, implement the substation work ticket knowledge graph construction method based on a large language model as described in any one of claims 1 to 5.

Images