A method for constructing knowledge graph of power grid equipment based on power automation point meter information

Abstract

This invention discloses a method for constructing a knowledge graph for power grid equipment based on power automation point table information, comprising: S1. constructing a seed knowledge graph for power grid equipment; S2. improving the seed knowledge graph to obtain a refined knowledge graph; and S3. performing entity fusion processing on the refined knowledge graph to obtain a fused knowledge graph. This method ensures the timeliness and accuracy of data and improves the high precision of general domain knowledge extraction.

Description

Technical Field

[0001] The present invention relates to the field of power grid equipment, and in particular to a method for constructing a knowledge graph of power grid equipment based on power automation point table information. Background Art

[0002] Entering "point meter data" into the database is a significant and time-consuming task during automated commissioning. This data entry directly determines the correspondence between the "substation commissioning site" and the "dispatcher monitoring interface," making it crucial for the safe operation and real-time accuracy of power grid equipment.

[0003] Currently, knowledge graphs for power grid equipment are primarily constructed using structured data from existing systems within each power grid company as the foundation for cold start operations. Because power grid construction organizations and substation operators are separate entities, there's often a short lag in updating equipment data within the operational system. This leads to a large amount of "old data" in knowledge graphs constructed using this approach, and also results in low accuracy in extracting general domain knowledge.

[0004] Therefore, a method for constructing a knowledge graph of power grid equipment based on power automation point table information is needed to solve the above problems. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to overcome the defects in the prior art and provide a method for constructing a knowledge graph of power grid equipment based on power automation point table information, which can ensure the timeliness and accuracy of data and improve the high precision of general field knowledge extraction.

[0006] The method for constructing a knowledge graph of power grid equipment based on power automation point table information of the present invention comprises the following steps:

[0007] S1. Build a seed knowledge graph for power grid equipment;

[0008] S2. Improve the seed knowledge graph to obtain an improved knowledge graph;

[0009] S3. Perform entity fusion processing on the improved knowledge graph to obtain a fused knowledge graph.

[0010] Furthermore, we construct a seed knowledge graph for power grid equipment, specifically including:

[0011] S11. Collecting structured data of power grid equipment and obtaining initial unstructured data of power grid equipment; the structured data includes power automation point table information;

[0012] S12. Designing a knowledge graph data model based on the initial unstructured data;

[0013] S13. Based on the knowledge graph data model, the structured data is mapped to obtain a seed knowledge graph of power grid equipment.

[0014] Furthermore, the step S2 specifically includes:

[0015] S21. Obtaining an unstructured data set; the unstructured data set includes power dispatching operation procedures, power safety accident handling procedures, power equipment manuals, and website page data related to power equipment;

[0016] S22. Randomly select a% of the data from the unstructured data set as target data, and annotate the target data to obtain annotated data;

[0017] S23. Import the labeled data into the seed knowledge graph to obtain a preliminary improved knowledge graph;

[0018] S24. Constructing an improved pre-training model, and constructing an information extraction model based on the improved pre-training model;

[0019] S25. Use the information extraction model to extract all knowledge from the unstructured dataset to obtain the entities and relationships of the power grid equipment knowledge graph;

[0020] S26. Import the entities and relationships into the preliminarily improved knowledge graph to obtain an improved knowledge graph.

[0021] Furthermore, an improved pre-training model is constructed, specifically including:

[0022] Based on the pre-trained model Ernie, a prefinetune stage is added between the pretrain stage and the finetune stage. In the prefinetune stage, the parameters of the lower 8 layers of the Ernie model are frozen, and only the parameters of the upper 4 layers are updated to form an improved pre-trained model.

[0023] Furthermore, the step S3 specifically includes:

[0024] S31. Perform similarity calculation on all entities in the improved knowledge graph to obtain a similarity calculation result set;

[0025] S32. The two entities whose similarity calculation results in the similarity calculation result set are greater than the similarity threshold are taken as target entity pairs, and several target entity pairs form a target entity pair set;

[0026] S33. Perform entity fusion on each target entity pair in the target entity pair set to obtain a fused knowledge graph.

[0027] Furthermore, the method further includes: updating the fused knowledge graph to obtain an updated knowledge graph, specifically including:

[0028] The updated power grid equipment point table information is mapped to generate incremental knowledge graph data, and the incremental knowledge graph data is updated to the fused knowledge graph.

[0029] Furthermore, it also includes: manually reviewing the fused knowledge graph to obtain manual review results, and evaluating the quality of the knowledge graph based on the manual review results.

[0030] The beneficial effects of the present invention are as follows: the present invention discloses a method for constructing a knowledge graph of power grid equipment based on power automation point table information, which can ensure the timeliness and accuracy of data by using point table data to construct a knowledge graph, and also provide a guarantee for the accuracy of auxiliary decision-making; the information extraction model is used to extract the entities and relationships of power equipment, thereby ensuring the high precision of general domain knowledge extraction; by constructing an improved pre-training model, the time of the model fine-tuning stage is significantly shortened, and the precision and recall rate of knowledge extraction in the field of power equipment are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0032] Figure 1 Constructing a flow chart for the power grid equipment knowledge graph of the present invention;

[0033] Figure 2 Schematic diagram of the overall framework of the improved pre-training model of the present invention;

[0034] Figure 3 Schematic diagram of the knowledge graph application framework of the present invention. DETAILED DESCRIPTION

[0035] The present invention is further described below with reference to the accompanying drawings, as shown in the drawings:

[0036] The method for constructing a knowledge graph of power grid equipment based on power automation point table information of the present invention comprises the following steps:

[0037] S1. Build a seed knowledge graph for power grid equipment;

[0038] S2. Improve the seed knowledge graph to obtain an improved knowledge graph;

[0039] S3. Perform entity fusion processing on the improved knowledge graph to obtain a fused knowledge graph.

[0040] In this embodiment, in step S1, constructing a seed knowledge graph of power grid equipment specifically includes:

[0041] S11. Collect structured data of power grid equipment and obtain initial unstructured data of power grid equipment; wherein, the initial unstructured data includes auxiliary decision-making planning information for power dispatch fault processing; the structured data includes power automation point table information; the power automation point table information includes the name of each substation, signal information of equipment in each substation (such as each circuit breaker, disconnector, capacitor, etc.), and feeder information of each substation.

[0042] The existing structured data of power grid equipment can be exported through the running point meter monitoring system (such as the D5000 system). The point meter information of power equipment comes from the on-site equipment data. The method of using the point meter information to construct a knowledge graph can ensure the timeliness and accuracy of the data, and also provide a guarantee for the accuracy of auxiliary decision-making.

[0043] S12. Design a knowledge graph data model based on the initial unstructured data; wherein, the schema layer ontology is completed by extracting the power-related unstructured data from the bottom up, combining the information characteristics of the power point table and the auxiliary decision-making use planning of power dispatch fault processing to form a knowledge graph data model.

[0044] S13. Based on the knowledge graph data model, the structured data is mapped to obtain a seed knowledge graph of power grid equipment, wherein the mapping is performed using a D2R technology mapping formula.

[0045] After completing the above steps, the seed knowledge graph data is saved to the neo4j graph database.

[0046] In this embodiment, step S2 specifically includes:

[0047] S21. Obtain an unstructured data set; the unstructured data set includes power dispatching operation procedures, power safety accident handling procedures, power equipment manuals, and website page data related to power equipment; wherein, the Python Scrapy framework can be used to crawl encyclopedia website power equipment related page data;

[0048] S22. Randomly select a% of the data from the unstructured data set as the target data, and annotate the target data to obtain annotated data; wherein the a% can be set to 10%, and DOCCANO can be used for the annotation process;

[0049] S23. Import the annotated data into the seed knowledge graph to obtain a preliminarily improved knowledge graph. Through the above steps, the first improvement and update of the knowledge graph is completed.

[0050] S24. Build an improved pre-trained model and, based on it, build an information extraction model; the information extraction model is based on the improved pre-trained model. Existing pre-trained models, such as the Ernie 3.0 model, have high accuracy in extracting knowledge from general domains. However, in specialized domains, particularly the power sector, the pretrain + fine-tune model suffers from a certain degree of catastrophic forgetting. That is, when training a model on a new dataset, the knowledge learned on the old data is forgotten, resulting in significant dropouts when testing on the old data.

[0051] By building an improved pre-training model, the knowledge graph of power equipment formed in the early stage was added to the Ernie3.0 base, replacing the knowledge graph of the general knowledge field in the original Ernie3.0 model, and improving the precision F1 and recall rate recall of information extraction models (such as UIE) in the field of power equipment.

[0052] S25. Use the information extraction model to extract all knowledge from the unstructured dataset to obtain the entities and relationships of the power grid equipment knowledge graph;

[0053] When processing unstructured data sources, existing knowledge extraction in the power industry relies on rule-based templates. However, rule-based templates require a significant amount of manual work, require high-level rule-setting, and are unsuitable for building and maintaining medium- to large-scale knowledge graphs. The processing method described in this invention enables fast and efficient full knowledge extraction from unstructured datasets.

[0054] S26. Import the entities and relationships into the initially improved knowledge graph to obtain an improved knowledge graph. The improved knowledge graph is still saved in the neo4j graph database, thereby completing the second improvement and update of the knowledge graph.

[0055] In this embodiment, constructing an improved pre-training model specifically includes:

[0056] like Figure 2 As shown, based on the pre-trained model Ernie, for example, the existing Ernie3.0 is used, a prefinetune stage is added between the pretrain stage and the finetune stage, and in the prefinetune stage, the parameters of the lower 8 layers of the Ernie model are frozen, and only the parameters of the upper 4 layers are updated to form an improved pre-trained model.

[0057] Ernie uses 12 transformer layers. The parameters learned by the upper layers of the Ernie model tend to be domain-specific, while the parameters learned by the lower layers tend to be more general. Updating these 260 billion parameters during the fine-tuning phase was extremely time-consuming. Updating Ernie's underlying general knowledge graph to a knowledge graph specific to power equipment further increased the project's time.

[0058] Therefore, through the above improvements, the time of the fine-tune stage in the original Ernie model is significantly shortened, thereby improving the efficiency of constructing the knowledge graph of power equipment.

[0059] In this embodiment, step S3 specifically includes:

[0060] S31. Perform traversal similarity operations on all entities in the improved knowledge graph to obtain a similarity calculation result set; among them, the Jaccard graph calculation method provided by neo4j can be used to perform traversal similarity operations. The Jaccard similarity (coefficient) is used to measure the similarity between sets. Assuming that sets A and B are two entities stored in neo4j, the Jaccard similarity is defined as the size of the intersection divided by the size of the union of the two sets.

[0061] S32. Two entities in the similarity calculation result set whose similarity calculation results are greater than the similarity threshold are taken as target entity pairs, and several target entity pairs form a target entity pair set; wherein, a similarity calculation result of 1 indicates that they are exactly the same, and the similarity threshold is 0.75; the larger the Jaccard similarity calculation result value, the higher the sample similarity. When the result is greater than the set similarity threshold, it can be judged as the same entity and entity fusion can be performed.

[0062] S33. Perform entity fusion on each target entity pair in the target entity pair set to obtain a fused knowledge graph. The entity fusion is performed using entity merging. Through the above steps, the third improvement and update of the knowledge graph is completed.

[0063] In this embodiment, the method for constructing a knowledge graph for power grid equipment of the present invention further includes: updating the fused knowledge graph to obtain an updated knowledge graph, specifically including:

[0064] The updated grid equipment point table information is mapped to generate incremental knowledge graph data, and the incremental knowledge graph data is updated to the fused knowledge graph. The mapping process is performed using the D2R technology mapping method.

[0065] In this embodiment, the method for constructing a knowledge graph for power grid equipment of the present invention further includes: manually reviewing the fused knowledge graph to obtain manual review results, and evaluating the quality of the knowledge graph based on the manual review results. Due to the relatively large amount of knowledge graph data, manual review generally involves random sampling, with a total sampling of approximately 200 nodes, their attributes, and relationships. Based on the sampling review results, it is determined whether to conduct an expanded sampling review. The significance of manual review is to evaluate the quality of the knowledge graph and identify common issues and solutions.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for constructing a knowledge graph of power grid equipment based on power automation point table information, characterized by: The steps include: S1. Build a seed knowledge graph for power grid equipment; S2. Improve the seed knowledge graph to obtain an improved knowledge graph; The step S2 specifically includes: S21. Obtaining an unstructured data set; the unstructured data set includes power dispatching operation procedures, power safety accident handling procedures, power equipment manuals, and website page data related to power equipment; S22. Randomly select a% of the data from the unstructured data set as target data, and annotate the target data to obtain annotated data; S23. Import the labeled data into the seed knowledge graph to obtain a preliminary improved knowledge graph; S24. Constructing an improved pre-training model, and constructing an information extraction model based on the improved pre-training model; S25. Use the information extraction model to extract all knowledge from the unstructured dataset to obtain the entities and relationships of the power grid equipment knowledge graph; S26. Importing the entities and relationships into the initially improved knowledge graph to obtain an improved knowledge graph; Among them, building an improved pre-training model specifically includes: Based on the pre-trained model Ernie, a prefinetune phase is added between the pretrain phase and the finetune phase. In the prefinetune phase, the parameters of the lower 8 layers of the Ernie model are frozen, and only the parameters of the upper 4 layers are updated, forming an improved pre-trained model. S3. Perform entity fusion processing on the improved knowledge graph to obtain a fused knowledge graph.

2. The method for constructing a knowledge graph of power grid equipment based on power automation point table information according to claim 1, characterized in that: Constructing a seed knowledge graph for power grid equipment, specifically including: S11. Collecting structured data of power grid equipment and obtaining initial unstructured data of power grid equipment; the structured data includes power automation point table information; S12. Designing a knowledge graph data model based on the initial unstructured data; S13. Based on the knowledge graph data model, the structured data is mapped to obtain a seed knowledge graph of power grid equipment.

3. The method for constructing a knowledge graph of power grid equipment based on power automation point table information according to claim 1, characterized in that: The step S3 specifically includes: S31. Perform similarity calculation on all entities in the improved knowledge graph to obtain a similarity calculation result set; S32. The two entities whose similarity calculation results in the similarity calculation result set are greater than the similarity threshold are taken as target entity pairs, and several target entity pairs form a target entity pair set; S33. Perform entity fusion on each target entity pair in the target entity pair set to obtain a fused knowledge graph.

4. The method for constructing a knowledge graph of power grid equipment based on power automation point table information according to claim 1, characterized in that: Also includes: The fused knowledge graph is updated to obtain an updated knowledge graph, which specifically includes: The updated power grid equipment point table information is mapped to generate incremental knowledge graph data, and the incremental knowledge graph data is updated to the fused knowledge graph.

5. The method for constructing a knowledge graph of power grid equipment based on power automation point table information according to claim 1 is characterized in that: Also includes: The fused knowledge graph is manually reviewed to obtain manual review results, and the quality of the knowledge graph is evaluated based on the manual review results.

Images