Power grid knowledge management method and system based on large model

Abstract

The invention relates to a power grid knowledge management method and system based on a large model, and belongs to the technical field of artificial intelligence. The method comprises the steps of obtaining a large model standardization criterion document, performing extraction according to the large model standardization criterion document to obtain an entity recognition result and a relation recognition result, and constructing a criterion knowledge graph according to the entity recognition result and the relation recognition result; obtaining a large-model standardized question, and obtaining a question entity sequence and intention data of the large-model standardized question; presetting a semantic slot template, filling the semantic slot template according to the question entity sequence and the intention data to obtain a Cypher statement, and retrieving the criterion knowledge graph according to the Cypher statement to obtain a question entity triple sequence; and constructing a sentence tree according to the question entity triple sequence, calculating to obtain an analysis vector through an embedded layer according to the sentence tree and the large model standardized question, and calculating to obtain an analysis result through a decoder according to the analysis vector. And intelligent management of power grid knowledge is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and particularly relates to a power grid knowledge management method and system based on a large model. Background Art

[0002] With the rapid development of artificial intelligence technology, large models are increasingly widely used in various fields. Especially in the power industry, the application of large models provides more intelligent and efficient services for the system. However, the application of large models also brings challenges to knowledge management. Traditional knowledge management methods often struggle to adapt to the massive and complex knowledge systems of large models, resulting in low efficiency in knowledge acquisition, integration, and application.

[0003] In the power industry, the system of pre-trained large models requires a comprehensive evaluation standard system to ensure its performance and quality. These standards should cover multiple aspects such as the basic platform, model functions, scenario applications, and security requirements to provide a multi-dimensional evaluation framework. However, existing evaluation methods often lack systematicness and comprehensiveness and are difficult to comprehensively evaluate the performance of large models.

[0004] In addition, large models in the power industry need to process a large amount of data, including various modalities of data such as voice and images. In the data processing process, a complete data processing framework is required to ensure the accuracy, integrity, and security of the data. However, existing data processing methods often have problems such as cumbersome processes and low efficiency and are difficult to meet the application requirements of large models.

[0005] Therefore, how to effectively manage the knowledge of large models, improve the efficiency of knowledge acquisition, integration, and application, and establish a comprehensive evaluation standard system and data processing framework has become an urgent problem to be solved in the power industry. Summary of the Invention

[0006] To solve the above problems existing in the prior art, the present invention provides a power grid knowledge management method and system based on a large model.

[0007] The object of the present invention can be achieved by the following technical solutions:

[0008] Obtain a large model standardization criterion document, calculate entity recognition results and relationship recognition results through extraction models according to the large model standardization criterion document, and construct a criterion knowledge graph according to the entity recognition results and the relationship recognition results;

[0009] Obtain large model standardization questions through an HTTP interface, calculate a question entity sequence through a recognition model according to the large model standardization questions, and calculate intention data through an extraction model according to the large model standardization questions;

[0010] A preset semantic slot template, filling the semantic slot template according to the problem entity sequence and the intention data to obtain a Cypher statement, and retrieving the problem entity triple sequence from the criterion knowledge graph according to the Cypher statement;

[0011] Constructing a sentence tree according to the problem entity triple sequence, calculating an analysis vector through an embedding layer according to the sentence tree and the standardized question sentence of the large model, calculating an analysis result through a decoder according to the analysis vector, and the intelligent terminal transmitting the analysis result to the human-computer interaction interface.

[0012] Specifically, the specific steps of the extraction model include:

[0013] Preprocessing the standardized criterion document of the large model to obtain a preprocessed document, calculating word embedding vectors through a SecBERT model according to the preprocessed document, and calculating word encodings through a Bi-LSTM model according to the word embedding vectors;

[0014] Calculating an entity recognition specific vector and a relation extraction specific vector through a multi-layer perceptron according to the word encoding, obtaining an upper triangular region matrix and a relative position matrix of the preprocessed document, and performing two-dimensional convolution on the upper triangular region matrix, the relative position matrix, the entity recognition specific vector, and the relation extraction specific vector to obtain a shared vector;

[0015] Obtaining an entity convolution and a relation convolution through dilated convolution according to the entity recognition specific vector and the relation extraction specific vector, performing feature extraction through a channel attention mechanism on the entity convolution and the relation convolution to obtain an entity feature vector and a relation feature vector, splicing the entity feature vector and the relation feature vector with the shared vector respectively to obtain an entity splicing vector and a relation splicing vector, calculating an entity splicing recognition probability through a multi-layer perceptron according to the entity splicing vector, calculating a relation splicing recognition probability through a multi-layer perceptron according to the relation splicing vector, calculating an entity dependency value and a relation dependency value through a Biaffine model according to the entity recognition specific vector and the relation extraction specific vector, adding the entity splicing recognition probability and the entity dependency value to obtain an entity prediction value, adding the relation splicing recognition probability and the relation dependency value to obtain a relation prediction value, and calculating the entity recognition result and the relation recognition result through a classifier according to the entity prediction value and the relation prediction value.

[0016] Specifically, constructing the criterion knowledge graph according to the entity recognition result and the relation recognition result specifically includes:

[0017] Calculate the entity similarity based on the entity recognition result, preset a similarity threshold, and determine whether the entity similarity is less than the similarity threshold. If so, do not process; otherwise, perform deduplication and fusion.

[0018] Construct a structural triple according to the entity recognition result and the relationship recognition result, and obtain the rule knowledge graph by indexing through Hash based on the structural triple.

[0019] Specifically, the specific calculation steps of the recognition model include:

[0020] Calculate the question word vector through the CBERT model according to the standardized question of the large model, calculate the global information through the BiLSTM model according to the question word vector, calculate the local information through the IDCNN model according to the question word vector, calculate the fused feature vector through the auto-decoder for the global information and the local information, decode the question entity recognition result through the CRF layer, and store the question entity recognition result to obtain the question entity sequence.

[0021] Specifically, the specific calculation steps of the extraction model include:

[0022] Calculate the question word vector through the CBERT model according to the standardized question of the large model, calculate the semantic feature vector through the TextCNN model according to the question word vector, perform pooling on the semantic feature vector through the MaxPooling method to obtain the fused feature, and calculate the intention data through the classifier for the fused feature.

[0023] Specifically, the specific steps of constructing a sentence tree according to the question entity triple sequence include:

[0024] Perform dependency tree analysis on the standardized question of the large model to obtain a sentence dependency tree, determine candidate paths according to the sentence dependency tree, and connect the question entity triple sequence according to the candidate paths to obtain the sentence tree.

[0025] Specifically, calculating the parsing vector through the embedding layer according to the sentence tree and the standardized question of the large model includes:

[0026] Obtain the entity embedding vector through entity information encoding according to the sentence tree, calculate the word embedding vector through the CBERT model according to the standardized question of the large model, obtain the position encoding matrix of the standardized question of the large model, add the even position encoding matrix to the word embedding vector to obtain the position encoding vector, and splice the position encoding vector with the entity embedding vector to obtain the parsing vector.

[0027] A power grid knowledge management system based on a large model, comprising: a knowledge graph construction module, a data extraction module, a query module, and an analysis module;

[0028] The knowledge graph construction module is used to obtain a large model standardization criterion document, calculate entity recognition results and relationship recognition results through an extraction model according to the large model standardization criterion document, and construct a criterion knowledge graph according to the entity recognition results and the relationship recognition results;

[0029] The data extraction module is used to obtain a large model standardized question through an HTTP interface, calculate a question entity sequence through a recognition model according to the large model standardized question, and calculate intention data through an extraction model according to the large model standardized question;

[0030] The query module is used to preset a semantic slot template, fill the semantic slot template according to the question entity sequence and the intention data to obtain a Cypher statement, and retrieve the criterion knowledge graph according to the Cypher statement to obtain a question entity triple sequence;

[0031] The analysis module is used to construct a sentence tree according to the question entity triple sequence, calculate an analysis vector through an embedding layer according to the sentence tree and the large model standardized question, calculate an analysis result through a decoder according to the analysis vector, and the intelligent terminal transmits the analysis result to the human-computer interaction interface.

[0032] An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, it implements the power grid knowledge management method based on a large model as described above.

[0033] A storage medium containing computer-executable instructions, characterized in that the computer-executable instructions are used to execute the power grid knowledge management method based on a large model as described above when executed by a computer processor.

[0034] The beneficial effects of the present invention are as follows:

[0035] (1) By obtaining a large model standardization criterion document and constructing a criterion knowledge graph, the present invention realizes the systematic and structured management of large model knowledge. This greatly improves the efficiency of knowledge acquisition, integration, and application, and provides strong support for the application of large models in the smart home industry.

[0036] (2) By setting up an identification model and an extraction model, it is used to obtain the question entity sequence and intention data of the standardized questions of the large model. A preset semantic slot template is filled with the question entity sequence and intention data to obtain a Cypher statement, and the knowledge graph is retrieved through Cypher to obtain the question entity triple sequence, realizing the intelligent parsing of user questions and accurately understanding the user's intention.

[0037] (3) A sentence tree is constructed according to the question entity triple sequence, and an analysis vector is calculated through the embedding layer based on the sentence tree and the standardized questions of the large model. The analysis result is calculated through the decoder according to the analysis vector, realizing the construction of the knowledge graph according to the standardized criterion document of the large model, intelligently answering user questions based on the knowledge graph, enabling users to more simply and quickly find solutions to their problems and the required industry criteria, and realizing the standardized execution of the large voice model integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings.

[0039] Figure 1 It is a schematic flowchart of a power grid knowledge management method based on a large model of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will describe in detail the specific embodiments, structures, features and their effects of the present invention with reference to the accompanying drawings and preferred embodiments.

[0041] Please refer to Figure 1 , a power grid knowledge management method based on a large model,

[0042] Obtain the standardized criterion document of the large model, calculate the entity recognition result and relationship recognition result through the extraction model according to the standardized criterion document of the large model, and construct a criterion knowledge graph according to the entity recognition result and the relationship recognition result;

[0043] Obtain the standardized questions of the large model through the HTTP interface, calculate the question entity sequence through the recognition model according to the standardized questions of the large model, and calculate the intention data through the extraction model according to the standardized questions of the large model;

[0044] Preset a semantic slot template, fill the semantic slot template with the question entity sequence and the intention data to obtain a Cypher statement, and retrieve the criterion knowledge graph according to the Cypher statement to obtain a question entity triple sequence;

[0045] Construct a sentence tree according to the sequence of the problem entity triples, calculate an analysis vector through an embedding layer according to the sentence tree and the standardized question sentence of the large model, calculate an analysis result through a decoder according to the analysis vector, and the intelligent terminal transmits the analysis result to the human-computer interaction interface.

[0046] In this embodiment, the large model standardization criterion document provides definitions, terms, frameworks, and general requirements for the system of the pre-trained large model for smart home industry applications. It also covers the basic platform, model functions, scenario applications, and security requirements, and establishes a comprehensive evaluation standard system for evaluating large speech models and their extensions to multi-modal models, providing a multi-dimensional evaluation framework for users, teams, and enterprises evaluating or using LLM technology or models. This standard not only describes a three-dimensional evaluation framework based on the principles of "versatility, intelligence, efficiency, and security", but also provides a framework for large-scale pre-trained deep learning, including methods, taxonomies, related roles and activities, software toolkits, key indicators of application capabilities, and evaluation means. It also provides a data processing framework for training large speech models and refines related terms and definitions, as well as the processes, methods, features, and performance evaluations for data processing in the pre-training and fine-tuning stages.

[0047] Specifically, the specific steps of the extraction model include:

[0048] Obtain a preprocessed document through preprocessing according to the large model standardization criterion document, calculate word embedding vectors through a SecBERT model according to the preprocessed document, and calculate word encodings through a Bi-LSTM model according to the word embedding vectors;

[0049] Calculate entity recognition specific vectors and relation extraction specific vectors through a multi-layer perceptron according to the word encodings, obtain the upper and lower triangular region matrix and the relative position matrix of the preprocessed document, and perform two-dimensional convolution on the upper and lower triangular region matrix, the relative position matrix, the entity recognition specific vectors, and the relation extraction specific vectors to obtain a shared vector;

[0050] Identifying specific vectors for the entities and extracting specific vectors for the relationships, obtaining entity convolutions and relationship convolutions through dilated convolutions, extracting entity feature vectors and relationship feature vectors through a channel attention mechanism based on the entity convolutions and the relationship convolutions, concatenating the entity feature vectors and the relationship feature vectors with the shared vectors respectively to obtain entity concatenated vectors and relationship concatenated vectors, calculating entity concatenated recognition probabilities based on the entity concatenated vectors through a multi-layer perceptron, calculating relationship concatenated recognition probabilities based on the relationship concatenated vectors through a multi-layer perceptron, calculating entity dependency values and relationship dependency values based on the entity identifying specific vectors and the relationship extracting specific vectors through a Biaffine model, adding the entity concatenated recognition probabilities and the entity dependency values to obtain entity prediction values, adding the relationship concatenated recognition probabilities and the relationship dependency values to obtain relationship prediction values, and calculating the entity recognition results and the relationship recognition results through a classifier based on the entity prediction values and the relationship prediction values.

[0051] Specifically, the preprocessing includes cleaning and formatting.

[0052] Specifically, the constructing of the criterion knowledge graph according to the entity recognition results and the relationship recognition results specifically includes:

[0053] Calculating entity similarity according to the entity recognition results, presetting a similarity threshold, and determining whether the entity similarity is less than the similarity threshold. If so, no processing is performed; otherwise, duplicate removal and fusion are carried out.

[0054] The formula for calculating the entity similarity is:

[0055] sim(e1, e2) = a * sim e ((e1, e2) + (1 - a) * sim s ((e1, e2),

[0056] where sim(e1, e2) is the entity similarity between the first entity recognition result and the second entity recognition result, e1 is the first entity recognition result, e2 is the second entity recognition result, a is the weight value, sim e (e1, e2) is the entity name similarity, and sim s (e1, e2) is the attribute similarity;

[0057] Constructing a structure triple according to the entity recognition results and the relationship recognition results, and indexing the obtained rule knowledge graph through Hash.

[0058] Specifically, the specific calculation steps of the recognition model include:

[0059] The standardized question according to the large model is used to calculate the question word vector through the CBERT model, the global information is calculated through the BiLSTM model according to the question word vector, the local information is calculated through the IDCNN model according to the question word vector, the global information and the local information are used to calculate the fusion feature vector through the auto-decoder, the problem entity recognition result is obtained through decoding by the CRF layer according to the fusion feature vector, and the problem entity recognition result is stored to obtain the problem entity sequence.

[0060] Specifically, the specific calculation steps of the extraction model include:

[0061] The standardized question according to the large model is used to calculate the question word vector through the CBERT model, the semantic feature vector is calculated through the TextCNN model according to the question word vector, the fusion feature is obtained through pooling by the MaxPooling method according to the semantic feature vector, and the intention data is obtained through calculation by the classifier according to the fusion feature.

[0062] Specifically, the specific steps of constructing the sentence tree according to the problem entity triple sequence include:

[0063] Dependency tree analysis is performed on the standardized question according to the large model to obtain the sentence dependency tree, candidate paths are determined according to the sentence dependency tree, and the problem entity triple sequence is connected according to the candidate paths to obtain the sentence tree.

[0064] Specifically, the parsing vector calculated through the embedding layer according to the sentence tree and the standardized question according to the large model includes:

[0065] The entity embedding vector is obtained through entity information encoding according to the sentence tree, the word embedding vector is calculated through the CBERT model according to the standardized question according to the large model, the position encoding matrix of the standardized question according to the large model is obtained, the position encoding vector is obtained by adding the even position encoding matrix and the word embedding vector, and the parsing vector is obtained by splicing the position encoding vector and the entity embedding vector.

[0066] A power grid knowledge management system based on a large model, comprising: a knowledge graph construction module, a data extraction module, a query module, and an analysis module;

[0067] The knowledge graph construction module is used to obtain the large model standardized criterion document, calculate the entity recognition result and the relationship recognition result through the extraction model according to the large model standardized criterion document, and construct the criterion knowledge graph according to the entity recognition result and the relationship recognition result;

[0068] The data extraction module is used to obtain the standardized questions of the large model through the HTTP interface, calculate the question entity sequence through the recognition model according to the standardized questions of the large model, and calculate the intent data through the extraction model according to the standardized questions of the large model.

[0069] The query module is used to preset a semantic slot template, fill the semantic slot template according to the question entity sequence and the intent data to obtain a Cypher statement, and retrieve the criterion knowledge graph according to the Cypher statement to obtain a question entity triple sequence.

[0070] The parsing module is used to construct a sentence tree according to the question entity triple sequence, calculate a parsing vector through the embedding layer according to the sentence tree and the standardized questions of the large model, calculate a parsing result through the decoder according to the parsing vector, and the intelligent terminal transmits the parsing result to the human-computer interaction interface.

[0071] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the grid knowledge management method based on the large model as described above.

[0072] A storage medium containing computer-executable instructions, characterized in that the computer-executable instructions are used to execute the grid knowledge management method based on the large model as described above when executed by a computer processor.

[0073] The computer storage medium of the embodiments of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0074] A computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device.

[0075] The program code contained on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical fiber cable, RF, etc., or any suitable combination of the foregoing. The computer program code for performing the operations of the present invention may be written in one or more programming languages or combinations thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and also including conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via an Internet service provider through the Internet).

[0076] As described above, the above are only the preferred embodiments of the present invention and do not impose any formal limitations on the present invention. Although the present invention has been disclosed above in preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art, without departing from the scope of the technical solution of the present invention, may make some modifications or refinements to the above-disclosed technical content to obtain equivalent embodiments with equivalent changes. However, any simple modification, equivalent change, and refinement made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A power grid knowledge management method based on a large model, characterized in that Including: Obtain the large model standardization criterion document, calculate the entity recognition result and the relationship recognition result through the extraction model according to the large model standardization criterion document, and construct a criterion knowledge graph according to the entity recognition result and the relationship recognition result; Obtain the large model standardization question through the HTTP interface, calculate the question entity sequence through the recognition model according to the large model standardization question, and calculate the intention data through the extraction model according to the large model standardization question; Preset a semantic slot template, fill the semantic slot template according to the question entity sequence and the intention data to obtain a Cypher statement, and retrieve the question entity triple sequence from the criterion knowledge graph according to the Cypher statement; Construct a sentence tree according to the question entity triple sequence, calculate an analysis vector through an embedding layer according to the sentence tree and the large model standardization question, calculate an analysis result through a decoder according to the analysis vector, and the intelligent terminal transmits the analysis result to the human-computer interaction interface.

2. The method for power grid knowledge management based on a large model according to claim 1, wherein The specific steps of the extraction model include: Obtain a preprocessed document through preprocessing the large model standardization criterion document, calculate word embedding vectors through a SecBERT model according to the preprocessed document, and calculate word encodings through a Bi-LSTM model according to the word embedding vectors; Calculate entity recognition specific vectors and relationship extraction specific vectors through a multi-layer perceptron according to the word encodings, obtain the upper and lower triangular region matrix and the relative position matrix of the preprocessed document, and perform two-dimensional convolution on the upper and lower triangular region matrix, the relative position matrix, the entity recognition specific vectors, and the relationship extraction specific vectors to obtain a shared vector; Obtain an entity convolution and a relationship convolution through dilated convolution according to the entity recognition specific vectors and the relationship extraction specific vectors, perform feature extraction through a channel attention mechanism on the entity convolution and the relationship convolution to obtain an entity feature vector and a relationship feature vector, splice the entity feature vector and the relationship feature vector with the shared vector respectively to obtain an entity splicing vector and a relationship splicing vector, calculate an entity splicing recognition probability through a multi-layer perceptron according to the entity splicing vector, calculate a relationship splicing recognition probability through a multi-layer perceptron according to the relationship splicing vector, calculate an entity dependency value and a relationship dependency value through a Biaffine model according to the entity recognition specific vectors and the relationship extraction specific vectors, add the entity splicing recognition probability to the entity dependency value to obtain an entity prediction value, add the relationship splicing recognition probability to the relationship dependency value to obtain a relationship prediction value, and calculate the entity recognition result and the relationship recognition result through a classifier according to the entity prediction value and the relationship prediction value.

3. The method for power grid knowledge management based on a large model according to claim 1, wherein The constructing of the criterion knowledge graph according to the entity recognition result and the relationship recognition result specifically includes: Calculate entity similarity based on the entity recognition result, preset a similarity threshold, and determine whether the entity similarity is less than the similarity threshold. If yes, do not process; if no, perform deduplication and fusion. Construct a structural triple based on the entity recognition result and the relationship recognition result, and index the criterion knowledge graph through Hash based on the structural triple.

4. The method for power grid knowledge management based on a large model according to claim 1, wherein, The specific calculation steps of the recognition model include: Calculate the question word vector through the CBERT model based on the standardized question of the large model, calculate the global information through the BiLSTM model based on the question word vector, calculate the local information through the IDCNN model based on the question word vector, calculate the fused feature vector through the autoencoder based on the global information and the local information, decode the question entity recognition result through the CRF layer based on the fused feature vector, and store the question entity recognition result to obtain the question entity sequence.

5. The method for power grid knowledge management based on a large model according to claim 1, wherein The specific calculation steps of the extraction model include: Calculate the question word vector through the CBERT model based on the standardized question of the large model, calculate the semantic feature vector through the TextCNN model based on the question word vector, pool the semantic feature vector through the MaxPooling method to obtain the fused feature, and calculate the intention data through the classifier based on the fused feature.

6. The method for power grid knowledge management based on large models according to claim 1, wherein The specific steps for constructing a sentence tree based on the question entity triple sequence include: Perform dependency tree analysis on the standardized question of the large model to obtain a sentence dependency tree, determine candidate paths based on the sentence dependency tree, and connect the question entity triple sequence based on the candidate paths to obtain the sentence tree.

7. The method for power grid knowledge management based on large models according to claim 1, wherein The calculation of the parsing vector based on the sentence tree and the standardized question of the large model includes: Obtain the entity embedding vector through entity information encoding based on the sentence tree, calculate the word embedding vector through the CBERT model based on the standardized question of the large model, obtain the position encoding matrix of the standardized question of the large model, add the even position encoding matrix to the word embedding vector to obtain the position encoding vector, and splice the position encoding vector with the entity embedding vector to obtain the parsing vector.

8. A power grid knowledge management system based on a large model, characterized in that, Include: A knowledge graph construction module, a data extraction module, a query module, and an analysis module; The knowledge graph construction module is used to obtain the standardized criterion document of the large model, calculate the entity recognition result and the relationship recognition result through the extraction model based on the standardized criterion document of the large model, and construct the criterion knowledge graph based on the entity recognition result and the relationship recognition result; The data extraction module is used to obtain the standardized question of the large model through the HTTP interface, calculate the question entity sequence through the recognition model based on the standardized question of the large model, and calculate the intention data through the extraction model based on the standardized question of the large model; The query module is used to preset a semantic slot template, fill the semantic slot template according to the question entity sequence and the intent data to obtain a Cypher statement, and retrieve the criterion knowledge graph according to the Cypher statement to obtain a question entity triple sequence; The parsing module is used to construct a sentence tree according to the question entity triple sequence, calculate a parsing vector through an embedding layer according to the sentence tree and the standardized question sentence of the large model, calculate a parsing result through a decoder according to the parsing vector, and the intelligent terminal transmits the parsing result to the human-computer interaction interface.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the large model-based power grid knowledge management method according to any one of claims 1-7.

10. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions are used to execute the large model-based power grid knowledge management method according to any one of claims 1-7 when executed by a computer processor.

Images