Power grid project knowledge base construction method, system and equipment based on large model and medium

Abstract

The invention discloses a power grid project knowledge base construction method, system and device based on a large model, and a medium, and belongs to the technical field of power grid knowledge base construction, and the method comprises the steps: obtaining and preprocessing a power grid science and technology project document, and generating a structured text; extracting structured knowledge points and calculating important scoring indexes; screening high-value knowledge points from the structured text; calculating the association strength between any two high-value knowledge points and establishing an association edge to form a knowledge network and store the knowledge network into a knowledge base; when a new document is input, knowledge points of the new document are extracted, an updating value index is calculated, and whether the knowledge points in the new document are updated to the knowledge base or not is judged. According to the method, by evaluating the importance of the knowledge points, it is ensured that only high-value knowledge points are included into the association network, and the accuracy of the knowledge base is improved; the update necessity of the new document is judged by calculating the association strength and introducing the update value index, so that blind update or delayed update is avoided.

Description

Technical Field

[0001] This invention relates to the field of power grid knowledge base construction technology, specifically to a method, system, equipment, and medium for constructing a power grid project knowledge base based on a large model. Background Technology

[0002] Power grid technology projects are technological innovation projects undertaken by the power industry to address energy transition and improve system efficiency and reliability. Their main purpose is to upgrade the traditional power grid to a modern energy system through intelligent and digital means. These projects involve technological breakthroughs across all stages of power generation, transmission, transformation, distribution, and consumption, and also encompass solutions in areas such as renewable energy integration, energy storage applications, and equipment automation. These projects typically aim to improve energy efficiency, reduce pollution emissions, and ensure power supply security. For example, smart grid construction projects integrate sensors, smart meters, and automated equipment to achieve real-time monitoring of power flow and demand forecasting. This optimizes resource allocation while supporting renewable energy integration, thereby promoting a green transformation of the energy structure while improving power supply reliability. Furthermore, the research and application of energy storage technologies are also an important part of power grid technology projects. Addressing the instability of renewable energy generation, technologies such as battery storage, compressed air storage, and pumped hydro storage balance power supply and demand, solving the volatility problem of renewable energy grid integration and providing technical support for large-scale clean energy integration.

[0003] The advancement of power grid technology projects generates a large amount of technical documents and professional knowledge. These documents and knowledge are stored in the power grid technology project knowledge base and updated regularly. However, the update frequency of traditional power grid technology project knowledge bases is often fixed, such as monthly or quarterly. This can easily lead to updates even when the knowledge content has not changed or has changed very little, resulting in a waste of computing and storage resources. Alternatively, the knowledge base may not be updated in a timely manner, and the knowledge base may not incorporate the latest important information, resulting in users obtaining outdated or even incorrect knowledge, which affects decision-making efficiency. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that the update frequency of existing power grid technology project knowledge bases is often fixed, such as monthly or quarterly updates. This can easily lead to updates even when the knowledge content has not changed or has changed very little, resulting in a waste of computing and storage resources. Alternatively, the knowledge base information may not be updated in a timely manner, and the knowledge base may not be able to incorporate the latest important information in a timely manner, resulting in users obtaining outdated or even incorrect knowledge, which affects decision-making efficiency.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for constructing a power grid project knowledge base based on a large model, comprising the following steps: Obtain power grid technology project documents and preprocess them to generate structured text; Extract structured knowledge points from structured text and calculate the important scoring index of structured knowledge points; High-value knowledge points are selected from structured text based on important scoring indices; Calculate the semantic similarity matrix and association strength of any two high-value knowledge points to form a knowledge point pair, establish association edges, and form a knowledge network to be stored in the knowledge base. When a new document is input, extract the knowledge points from the new document, calculate the update value index, and determine whether to update the knowledge points in the new document to the knowledge base.

[0007] As a preferred embodiment of the method for constructing a power grid project knowledge base based on a large model as described in this invention, the steps of extracting structured knowledge points from structured text and calculating the importance score index of the structured knowledge points include: Extract structured knowledge points from structured text by identifying large models; The importance score index is calculated based on the parameter information of the structured knowledge points.

[0008] As a preferred embodiment of the method for constructing a power grid project knowledge base based on a large model as described in this invention, the step of selecting high-value knowledge points from structured text based on an importance scoring index includes: Set an important rating threshold and compare the important rating threshold with the important rating index; If the importance rating index is greater than the importance rating threshold, then the structured knowledge point is marked as a high-value knowledge point. If the importance rating index is less than or equal to the importance rating threshold, then the structured knowledge point is marked as a low-value knowledge point.

[0009] The beneficial effects of this preferred technical solution are as follows: by recognizing large model entities, structured knowledge points are extracted from documents, and the importance of knowledge points is evaluated through an important scoring index, ensuring that only high-value knowledge points are included in the association network, thereby improving the accuracy of the knowledge base.

[0010] As a preferred embodiment of the power grid project knowledge base construction method based on a large model described in this invention, the step of calculating the semantic similarity matrix and association strength of any two high-value knowledge points and establishing association edges to form a knowledge network and storing it in the knowledge base includes: Semantic vectors are generated from high-value knowledge points using a large model; Calculate the semantic similarity matrix based on the semantic vectors; The semantic similarity between knowledge points is determined based on the semantic similarity matrix, and the association strength between knowledge points is calculated. Set a threshold for association strength, compare the association strength with the threshold, and if the association strength is greater than the threshold, then establish an association edge between the two knowledge points corresponding to the association strength. If the association strength is less than or equal to the association strength threshold, then there is no need to establish an association edge between the two knowledge points corresponding to the association strength. A knowledge network is formed by establishing connections between knowledge points, and the knowledge network is stored in a knowledge base.

[0011] The beneficial effects of this preferred technical solution are as follows: by combining semantic analysis and time weighting, the logical connections between knowledge points are constructed and the connection strength is calculated, which enhances the coherence and timeliness of the knowledge network and improves the efficiency of user decision-making.

[0012] As a preferred embodiment of the method for constructing a power grid project knowledge base based on a large model as described in this invention, the steps of extracting knowledge points from the new document, calculating the update value index, and determining whether to update the knowledge points in the new document to the knowledge base when a new document is input include: The new document is preprocessed to generate structured text, and the structured knowledge points of the new document are extracted from the structured text. Calculate the importance score index of the new document based on the parameter information of the structured knowledge points; The updated value index is calculated based on the new document importance rating index; Set an update value threshold, compare the update value index with the update value threshold, and if the update value index is greater than the update value threshold, then update the knowledge points in the new document to the knowledge base. If the update value index is less than or equal to the update value threshold, then there is no need to update the knowledge base.

[0013] The beneficial effects of this preferred technical solution are as follows: by introducing an update value index to determine the necessity of updating new documents, blind or delayed updates are avoided, ensuring that the knowledge base always covers the latest and most important information and guaranteeing the freshness of knowledge in the knowledge base.

[0014] As a preferred embodiment of the method for constructing a power grid project knowledge base based on a large model as described in this invention, the formula for calculating the correlation strength between knowledge points is: ; in, For knowledge points With knowledge points The strength of the association, For knowledge points With knowledge points semantic similarity, Subtract one from the total number of knowledge points in the knowledge base; For knowledge points Average semantic similarity with all other knowledge points; The weighting coefficients are preset similarity coefficients. The preset weighting coefficient for time decay. The weight coefficients are preset average semantic similarity, and .

[0015] As a preferred embodiment of the method for constructing a power grid project knowledge base based on a large model as described in this invention, the formula for calculating the updated value index is: ; in, To update the value index, The number of knowledge points extracted from the new document. For the first document in the new document The key scoring index for each knowledge point For the first document in the new document The similarity between each knowledge point and the knowledge base. As the authority index of a new document, The preset weighting coefficients for the similarity of new documents. This is the preset weighting coefficient for the authority of new documents.

[0016] This invention provides a system for constructing a knowledge base for power grid projects based on a large model.

[0017] To address the aforementioned technical problems, the present invention further provides the following technical solution: a power grid project knowledge base construction system based on a large model, comprising: Project Acquisition Unit: Acquires and preprocesses power grid technology project documents to generate structured text; Knowledge scoring unit: Extracts structured knowledge points from structured text and calculates the importance scoring index of the structured knowledge points; selects high-value knowledge points from the structured text based on the importance scoring index; Logical association unit: Calculate the semantic similarity matrix and association strength of any two high-value knowledge points to form a knowledge point pair, establish association edges, and form a knowledge network to be stored in the knowledge base; Data storage unit: stores knowledge base information; Update Analysis Unit: When a new document is input, extract the knowledge points of the new document, calculate the update value index, and determine whether to update the knowledge points in the new document to the knowledge base.

[0018] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the method for constructing a power grid project knowledge base based on a large model.

[0019] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the method for constructing a power grid project knowledge base based on a large model.

[0020] The beneficial effects of this invention are as follows: This invention extracts structured knowledge points from documents through large-scale model entity recognition and evaluates the importance of these knowledge points using an importance rating index, ensuring that only high-value knowledge points are included in the association network, thus improving the accuracy of the knowledge base. By combining semantic analysis and time weighting, it constructs logical connections between knowledge points and calculates the strength of these connections, enhancing the coherence and timeliness of the knowledge network and improving user decision-making efficiency. Furthermore, by introducing an update value index to determine the necessity of updating new documents, it avoids blind or delayed updates, ensuring that the knowledge base always covers the latest and most critical information, thus guaranteeing the freshness of the knowledge in the knowledge base. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating an overall method for constructing a power grid project knowledge base based on a large model, as provided in one embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of a system flow diagram of a method for constructing a power grid project knowledge base based on a large model, provided as an embodiment of the present invention. Detailed Implementation

[0024] 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. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0025] Example 1, referring to Figure 1This is the first embodiment of the present invention, which provides a method for constructing a power grid project knowledge base based on a large model, including: S100: Obtain power grid technology project documents and preprocess them to generate structured text; S200: Extract structured knowledge points from structured text and calculate the important scoring index of structured knowledge points; S300: Select high-value knowledge points from structured text based on important scoring indices; S400: Calculate the semantic similarity matrix and association strength of any two high-value knowledge points as knowledge point pairs based on high-value knowledge points, establish association edges, form a knowledge network, and store it in the knowledge base. S500: When a new document is input, extract the knowledge points of the new document, calculate the update value index, and determine whether to update the knowledge points in the new document to the knowledge base.

[0026] It should be noted that a large amount of technical documents and professional knowledge are generated during the advancement of power grid technology projects. These documents and knowledge are stored in the power grid technology project knowledge base and updated regularly. However, the update frequency of traditional power grid technology project knowledge bases is often fixed, such as monthly or quarterly. This can easily lead to updates even when the knowledge content has not changed or has changed very little, resulting in a waste of computing and storage resources. Alternatively, the knowledge base may not be updated in a timely manner, and the knowledge base may not incorporate the latest important information, resulting in users obtaining outdated or even incorrect knowledge, which affects decision-making efficiency. Therefore, it is very important to propose a method for constructing a power grid project knowledge base that can be updated.

[0027] Therefore, to address the aforementioned problems in updating the knowledge base of power grid technology projects, a method for constructing a power grid project knowledge base based on a large model is developed through steps S100-S500. This method extracts structured knowledge points from documents through entity recognition within the large model and assesses the importance of these knowledge points using an importance rating index, ensuring that only high-value knowledge points are included in the association network, thus improving the accuracy of the knowledge base. By combining semantic analysis and time weighting, logical connections between knowledge points are constructed and their strength calculated, enhancing the coherence and timeliness of the knowledge network and improving user decision-making efficiency. Furthermore, by introducing an update value index to determine the necessity of updating new documents, blind or delayed updates are avoided, ensuring that the knowledge base always covers the latest and most critical information, thus guaranteeing the freshness of the knowledge in the knowledge base.

[0028] Example 2, refer to Figure 1 and Figure 2 This is the second embodiment of the present invention, which provides a method for constructing a power grid project knowledge base based on a large model.

[0029] In this embodiment of the application, step S100, which involves obtaining and preprocessing power grid technology project documents to generate structured text, includes: New documents with updated dates that have changed are retrieved daily from a designated power grid technology project database via a pre-defined API interface. The retrieved documents are then cleaned, including removing blank pages, duplicate paragraphs, headers and footers, and irrelevant special characters. The cleaned text is then segmented, sentence-segmented, and normalized using the natural language processing tool NLTK to generate structured text in JSON format, providing a standardized data foundation for subsequent knowledge extraction and analysis.

[0030] In this embodiment of the application, step S200, which involves extracting structured knowledge points from structured text and calculating the importance score index of the structured knowledge points, includes the following steps A1-A2: A1: Extract structured knowledge points from structured text by recognizing large models; In this embodiment, the identification model adopts the BERT model. The BERT model is pre-trained on a large-scale general corpus for bidirectional language representation. The pre-trained BERT model is used to perform sequence labeling and relation extraction on structured text to achieve entity recognition and relation extraction. Entity recognition locates entities in the text, and relation extraction extracts the semantic relationships between these entities. The identified entities and their corresponding semantic relationships are encapsulated into structured knowledge points, and the structured knowledge points are associated with parameters such as the frequency of occurrence of keywords and terms, document authority index, and time importance index. A2: Calculate the importance scoring index based on the parameter information of the structured knowledge points.

[0031] In this embodiment of the application, the formula for calculating the importance scoring index based on the parameter information of structured knowledge points is as follows: ; in, As an important scoring index, The frequency of keyword terms appearing in structured knowledge points. A time importance index for structured knowledge points. This represents the authority index of the document to which the structured knowledge points belong. The standard frequency of the preset keyword terms. The preset frequency weighting coefficients, The preset time weighting coefficient, The preset weighting coefficient for document authority, and ; It should be noted that the formula for calculating the importance score index comprehensively considers the local prominence of structured knowledge points in the document, the freshness of the knowledge, and the credibility of the information source. The frequency of keywords and terms appearing in the structured knowledge points reflects their local prominence in the document, and is normalized using preset standard frequencies to ensure comparability. The time importance index of the structured knowledge points is calculated based on the document's publication time to ensure the freshness and timeliness of the knowledge. The authority index of the document to which the structured knowledge points belong is set according to the authority level of the publishing institution to measure the credibility of the information source. Preset weighting coefficients α, β, and γ are assigned to these three dimensions to adjust the contribution ratio of each dimension, and the final weighted sum is obtained to obtain the importance score index. By adopting the formula for calculating the importance score index, the importance assessment is transformed from subjective experience to objective calculation, providing a scientific and configurable basis for subsequent selection of high-value knowledge points.

[0032] In an alternative implementation, the importance score index can also be calculated using a machine learning model. A Transformer-based learning model is designed, whose main task is to predict a comprehensive importance score between 0 and 1. The structured text descriptions of knowledge points and their context paragraphs in historical samples are taken as input, and the importance score index in historical samples is taken as output. The learning model is trained, and after training, the structured text is input into the learning model to output the comprehensive importance score, which is then used as the importance score index.

[0033] In another alternative implementation, the importance rating index can be calculated through dynamic adversarial evaluation. An adversarial network framework containing a rater and a discriminator is constructed. During the training phase, the rater receives features of knowledge points and outputs an initial importance score. Simultaneously, the discriminator is trained to distinguish between two types of input: scores given by the rater for ordinary knowledge points and expert ratings pre-labeled as truly high-value knowledge points. Through repeated adversarial training, the discriminator continuously improves its ability to distinguish between genuine and false ratings, while the rater continuously optimizes its internal parameters. When the training process converges and the discriminator can no longer effectively distinguish between the two, the adversarial training ends. Structured text is then input into the adversarial network framework, and the importance rating index is output through the rater.

[0034] In this embodiment of the application, step S300, which involves filtering high-value knowledge points from structured text based on an importance scoring index, includes the following steps B1-B2: B1: Set an important scoring threshold and compare the important scoring threshold with the important scoring index; if the important scoring index is greater than the important scoring threshold, mark the structured knowledge point as a high-value knowledge point. In this embodiment, the importance scoring index of knowledge points in historical data is calculated, and the 70th percentile of the importance scoring index of knowledge points in historical data is used as the importance scoring threshold. The importance scoring threshold can be appropriately adjusted according to the growth of the knowledge base and the content update. The importance scoring threshold is set to 0.75. The importance scoring threshold is compared with the importance scoring index. If the importance scoring index calculated in step S200 is 0.81, the importance scoring index is greater than the importance scoring threshold. Then, the structured knowledge point corresponding to the importance scoring index is marked as a high-value knowledge point and the high-value knowledge point is retained for subsequent knowledge network construction. B2: If the importance rating index is less than or equal to the importance rating threshold, then the structured knowledge point is marked as a low-value knowledge point.

[0035] In this embodiment of the application, the importance scoring threshold is set to 0.75. The importance scoring threshold is compared with the importance scoring index. If the importance scoring index calculated in step S200 is 0.72, and the importance scoring index is less than the importance scoring threshold, then the structured knowledge point corresponding to the importance scoring index is marked as a low-value knowledge point. Low-value knowledge points do not participate in the subsequent construction of the knowledge network, thereby effectively filtering out knowledge points with low importance and improving the quality of the knowledge base.

[0036] In this embodiment of the application, step S400, which calculates the semantic similarity matrix and association strength of any two high-value knowledge points and establishes association edges to form a knowledge network and stores it in the knowledge base, includes the following steps C1-C2: C1: Generate semantic vectors based on high-value knowledge points using a large model; In this embodiment, the large model adopts the BERT model. By using the pre-trained BERT model as the encoder and the text description of high-value knowledge points as the input of the BERT model, the input text is bidirectionally encoded through the multi-layer self-attention mechanism inside the BERT model to capture the semantic relationships between words. The hidden state vectors corresponding to the CLS tags output by the BERT model are extracted as semantic vectors. The dimension of the semantic vector is usually consistent with the dimension of the hidden layer of the selected BERT model, thereby transforming unstructured text knowledge into machine-processable and quantifiable numerical features. C2: Calculate the semantic similarity matrix based on the semantic vectors; In this embodiment of the application, for a set of semantic vectors of multiple high-value knowledge points that have been generated, the formula for calculating the semantic similarity between the semantic vectors of any two high-value knowledge points is as follows: ; in, The semantic similarity between high-value knowledge points i and j and These are the semantic vectors of high-value knowledge points i and j, respectively; All the calculated semantic similarity values ​​are combined into a semantic similarity matrix. The element in the i-th row and j-th column of the semantic similarity matrix is ​​the semantic similarity between high-value knowledge points i and j. In an optional implementation, the semantic similarity matrix can also be calculated by combining multiple similarity metrics. The core entities in each knowledge point are aligned with their corresponding entities in the background knowledge graph. For each pair of knowledge points, their path similarity, graph embedding similarity, and information content similarity in the background knowledge graph are calculated. Path similarity is calculated by weighting all relational paths between the two entities as information content similarity; shorter paths have higher weights. Graph embedding similarity uses a graph neural network model to perform embedding learning on the knowledge graph and then calculates the reciprocal of the cosine or Euclidean distance between the two entities in the embedding space as graph embedding similarity. Information content similarity utilizes the hierarchical relationships of entities in the graph to calculate the information content of their nearest common ancestor as information content similarity. The path similarity, graph embedding similarity, and information content similarity are summed to obtain the semantic similarity. All calculated semantic similarity values ​​are then combined to form a semantic similarity matrix.

[0037] In another alternative implementation, the semantic similarity matrix can also be calculated using tree edit distance. A dependency parser is used to analyze the text description of each knowledge point, generating a dependency syntax tree. Nodes in the tree represent words, and edges represent grammatical relationships between words. For any two syntax trees, the minimum number of operations required to transform one tree into the other is calculated; this number is called the tree edit distance. The tree edit distance is then converted into semantic similarity, and all calculated semantic similarity values ​​are combined to form a semantic similarity matrix.

[0038] C3: Determine the semantic similarity between knowledge points based on the semantic similarity matrix and calculate the association strength between knowledge points; In this embodiment of the application, the semantic similarity between any two knowledge points is obtained from the semantic similarity matrix constructed in step C2, and then the association strength between the knowledge points is calculated based on the semantic similarity between any two knowledge points. The formula for calculating the association strength between knowledge points is as follows: ; in, For knowledge points With knowledge points The strength of the association, For knowledge points With knowledge points semantic similarity, Subtract one from the total number of knowledge points in the knowledge base; For knowledge points With knowledge points The time decay factor between; For knowledge points Average semantic similarity with all other knowledge points; The weighting coefficients are preset similarity coefficients. The preset weighting coefficient for time decay. The weight coefficients are preset average semantic similarity, and ; Knowledge Points With knowledge points The formula for calculating the time decay factor between them is: ; in, The time decay coefficient, For knowledge points With knowledge points The time difference between creation; It should be noted that in the formula for calculating the association strength between knowledge points, the first term represents the direct similarity of two knowledge points in semantic content; the second term, the time decay factor, makes knowledge points with closer creation times more associated, thus introducing a timeliness dimension into the knowledge network; the third term represents the average semantic similarity between knowledge point i and all other knowledge points in the knowledge base, reflecting the global centrality or connectivity of knowledge point i in the entire knowledge network; and then, by weighting and summing the above three factors using three preset weight coefficients, a comprehensive association strength value combining direct semantic association, time proximity, and global connectivity information is obtained. This comprehensive association strength value is the association strength between knowledge points. In an optional implementation, the strength of association between knowledge points can also be calculated through Bayesian fusion of multi-source evidence. Three evidence sources are defined: semantic evidence, co-occurrence evidence, and meta-path evidence. Each evidence source provides the likelihood of a pair of knowledge points being associated. The likelihood mapped to the 0-1 interval based on semantic similarity using a sigmoid function is used as semantic evidence. The co-occurrence frequency of two knowledge points at the document, paragraph, and sentence levels is calculated using a Poisson distribution model to determine the likelihood of association when this co-occurrence frequency is observed, serving as co-occurrence evidence. The connection strength between knowledge points is calculated within the constructed knowledge network subgraph. With knowledge points The number and weight of multiple meta-paths are used as meta-path evidence, and the structural likelihood calculated by the path ranking algorithm is used as the meta-path evidence. Then, the semantic evidence, co-occurrence evidence and meta-path evidence are used to calculate the posterior probability of the association based on Bayes' theorem, which is the association strength.

[0039] In another alternative implementation, the correlation strength between knowledge points can also be calculated through knowledge distillation. A general relation teacher model is constructed and pre-trained on a massive general text corpus to learn to recognize various general semantic relations. Simultaneously, a small-scale but high-quality power grid domain relation annotation dataset is prepared, which contains the relation types and correlation strengths between knowledge point pairs annotated by experts. The textual descriptions of the knowledge point pairs are input into the general teacher model to obtain a general relation strength vector. A student model composed of a small neural network is then trained. The input during the training process is the semantic vector of the knowledge point pair, the general relation strength vector, and the power grid-specific meta-features. The supervision signal is the annotation dataset. After training, the student model can learn the general knowledge of the teacher model, combine its own semantic features and domain meta-features, and output a correlation strength value that conforms to the characteristics of the power grid domain, which is the correlation strength.

[0040] C4: Set a threshold for association strength, compare the association strength with the threshold, and if the association strength is greater than the threshold, establish an association edge between the two knowledge points corresponding to the association strength. In this embodiment of the application, the association strength threshold is set according to the distribution of all calculated association strengths. The association strength threshold is set to 0.6. The association strength between each pair of knowledge points is compared with the association strength threshold. If the association strength of a pair of knowledge points calculated by step C3 is 0.73, the association strength is greater than the association strength threshold. Then, an association edge is established between the two knowledge points corresponding to the association strength. C5: If the association strength is less than or equal to the association strength threshold, then there is no need to establish an association edge between the two knowledge points corresponding to the association strength. In this embodiment of the application, the association strength threshold is set to 0.6. The association strength between each pair of knowledge points is compared with the association strength threshold. If the association strength of a pair of knowledge points calculated by step C3 is 0.54, the association strength is less than the association strength threshold. Therefore, it is not necessary to establish an association edge between the two knowledge points corresponding to the association strength. C6: The knowledge network is formed by the associated edges established based on the knowledge points, and the knowledge network is stored in the knowledge base.

[0041] In this embodiment, based on all the associated edges established through steps C4 and C5 and their corresponding high-value knowledge points, each high-value knowledge point is treated as a node in the graph. Each associated edge that meets the association strength threshold is treated as an edge connecting two nodes, and the calculated association strength is used as the weight attribute of the edge. This quantifies the tightness of the connection between nodes, forming a complete and machine-understandable knowledge network. The knowledge network is stored in a knowledge base in JSON format, where the nodes and edges arrays in JSON format describe the attributes of all nodes and edges, respectively, thereby supporting subsequent path queries and traversals of the knowledge network.

[0042] In this embodiment of the application, when a new document is input in step S500, extracting the knowledge points of the new document, calculating the update value index, and determining whether to update the knowledge points in the new document to the knowledge base includes the following steps D1-D4: D1: Preprocess the new document to generate structured text, extract structured knowledge points from the structured text, and calculate the importance score index based on the parameter information of the structured knowledge points. In this embodiment of the application, when a new document is input, the data cleaning, word segmentation, sentence segmentation and text normalization operations in step A1 are performed on the new document to generate structured text data of the new document and extract the structured knowledge points of the new document; then, the importance score index of the new document is calculated by using the formula for calculating the importance score index in step A2 based on the parameter information of the extracted structured knowledge points of the new document. D2: Calculate the updated value index based on the new document importance rating index; In this embodiment of the application, the importance score index of the new document calculated based on the new document in step D1 is obtained, and the update value index is calculated based on the importance score index of the new document. The formula for calculating the update value index is as follows: ; in, To update the value index, The number of knowledge points extracted from the new document. For the first document in the new document The key scoring index for each knowledge point For the first document in the new document The similarity between each knowledge point and the knowledge base. As the authority index of a new document, The preset weighting coefficients for the similarity of new documents. This is a preset weighting coefficient for the authority of new documents; It should be noted that the formula for calculating the update value index combines the average importance of new knowledge points and the source authority of new documents in the numerator, and adjusts the contribution through weighting coefficients to ensure that high-value and high-authority new documents receive higher update potential benchmark scores. The denominator introduces knowledge novelty assessment, which affects the final update value index by the similarity between the knowledge points in the new document and the knowledge base. If the content of the new document is highly similar to the knowledge base, the denominator increases, thereby reducing the update value index and suppressing the updating of redundant and duplicate knowledge. This achieves a fundamental shift from fixed-period updates to updates driven by the update value index, enabling the determination of when to update to incorporate high-value new knowledge and when not to update to avoid resource waste, thus improving resource utilization efficiency while ensuring the timeliness of the knowledge base. D3: Set an update value threshold, compare the update value index with the update value threshold, and if the update value index is greater than the update value threshold, then update the knowledge points in the new document to the knowledge base. In this embodiment, the update value threshold is set based on the distribution of the update value index of newly calculated documents over a period of time. As the update interval becomes longer, the update value threshold is appropriately reduced until an update action occurs, at which point the update value threshold is reset to its initial value. Set the update value threshold to 0.5, compare the update value index with the update value threshold. If the update value index of a new document calculated in step D2 is 0.6, and the update value index is greater than the update value threshold, then update the knowledge points in the new document corresponding to the update value index to the knowledge base. D4: If the update value index is less than or equal to the update value threshold, then there is no need to update the knowledge base.

[0043] Set the update value threshold to 0.5, compare the update value index with the update value threshold. If the update value index of a new document calculated in step D2 is 0.41, then the update value index is less than the update value threshold, and there is no need to update the knowledge base.

[0044] In an optional implementation, determining whether to update the knowledge points in a new document to the knowledge base can also be achieved through a dynamic priority queue. A multi-dimensional comprehensive priority score is calculated for the new document, which includes semantic novelty score, authority score, user demand matching degree, and technical urgency. These scores are fused through a learnable weighted network to obtain the comprehensive priority score. The new documents are sorted into the update queue according to the comprehensive priority score. At set intervals, the document with the highest average score is selected from the update queue and updated to the knowledge base. At the same time, a sliding time window mechanism is adopted so that documents that exceed a certain time limit will be removed from the update queue, even if they have a high priority, so that high-priority documents are processed in a timely manner.

[0045] In another alternative implementation, determining whether to update the knowledge points in a new document to the knowledge base can also be achieved through knowledge graph completion verification. A knowledge graph based on descriptive logic is constructed, where each knowledge point is associated with a confidence score and a completeness identifier. When a new document is input, a consistency check is first performed, formalizing the new knowledge point into a logical axiom and checking for logical conflicts with the existing knowledge graph. Then, evidence chain verification is performed, requiring the new knowledge point to form a verifiable evidence chain with existing knowledge points from at least two independent sources. Finally, completion opportunity identification is used to check whether the new knowledge point can complete missing links or attributes in the knowledge graph. The system maintains a buffer for verification. New knowledge points that fail verification but have potential value are stored in the buffer and manually reviewed. Documents that pass the review are assigned a confidence score. Only when the confidence score exceeds a threshold is it allowed to update the knowledge points in the new document to the knowledge base.

[0046] This invention extracts structured knowledge points from documents through large-scale entity recognition and evaluates the importance of these knowledge points using an importance rating index, ensuring that only high-value knowledge points are included in the association network, thus improving the accuracy of the knowledge base. By combining semantic analysis and time weighting, it constructs logical connections between knowledge points and calculates the strength of these connections, enhancing the coherence and timeliness of the knowledge network and improving user decision-making efficiency. By introducing an update value index to determine the necessity of updating new documents, it avoids blind or delayed updates, ensuring that the knowledge base always covers the latest and most critical information, thus guaranteeing the freshness of the knowledge in the knowledge base.

[0047] Example 3, referring to Figure 1 and Figure 2 This is the third embodiment of the present invention, which provides a power grid project knowledge base construction system based on a large model, including: Project Acquisition Unit: Acquires and preprocesses power grid technology project documents to generate structured text; Knowledge scoring unit: Extracts structured knowledge points from structured text and calculates the importance scoring index of the structured knowledge points; selects high-value knowledge points from the structured text based on the importance scoring index; Logical association unit: Calculate the semantic similarity matrix and association strength of any two high-value knowledge points to form a knowledge point pair, establish association edges, and form a knowledge network to be stored in the knowledge base; Data storage unit: stores knowledge base information; Update Analysis Unit: When a new document is input, extract the knowledge points of the new document, calculate the update value index, and determine whether to update the knowledge points in the new document to the knowledge base.

[0048] Example 4, the fourth embodiment of the present invention, differs from the previous three embodiments in that: if the function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0049] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0050] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0051] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0052] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for constructing a power grid project knowledge base based on a large model, characterized in that, include, Obtain power grid technology project documents and preprocess them to generate structured text; Extract structured knowledge points from structured text and calculate the important scoring index of structured knowledge points; High-value knowledge points are selected from structured text based on important scoring indices; Calculate the semantic similarity matrix and association strength of any two high-value knowledge points to form a knowledge point pair, establish association edges, and form a knowledge network to be stored in the knowledge base. When a new document is input, extract the knowledge points from the new document, calculate the update value index, and determine whether to update the knowledge points in the new document to the knowledge base.

2. The method for constructing a power grid project knowledge base based on a large model as described in claim 1, characterized in that, The steps for extracting structured knowledge points from structured text and calculating the key score index of these knowledge points include: Extract structured knowledge points from structured text by identifying large models; The importance score index is calculated based on the parameter information of the structured knowledge points.

3. The method for constructing a power grid project knowledge base based on a large model as described in claim 2, characterized in that, The steps for filtering high-value knowledge points from structured text based on importance scoring indices include: Set an important rating threshold and compare the important rating threshold with the important rating index; If the importance rating index is greater than the importance rating threshold, then the structured knowledge point is marked as a high-value knowledge point. If the importance rating index is less than or equal to the importance rating threshold, then the structured knowledge point is marked as a low-value knowledge point.

4. The method for constructing a power grid project knowledge base based on a large model as described in claim 3, characterized in that, The steps involved in creating and storing a knowledge network include: Semantic vectors are generated from high-value knowledge points using a large model; Calculate the semantic similarity matrix based on the semantic vectors; The semantic similarity between knowledge points is determined based on the semantic similarity matrix, and the association strength between knowledge points is calculated. Set a threshold for association strength, compare the association strength with the threshold, and if the association strength is greater than the threshold, then establish an association edge between the two knowledge points corresponding to the association strength. If the association strength is less than or equal to the association strength threshold, then there is no need to establish an association edge between the two knowledge points corresponding to the association strength. A knowledge network is formed by establishing connections between knowledge points, and the knowledge network is stored in a knowledge base.

5. The method for constructing a power grid project knowledge base based on a large model as described in claim 4, characterized in that, When a new document is input, the steps for extracting the knowledge points from the new document, calculating the update value index, and determining whether to update the knowledge points in the new document to the knowledge base include: The new document is preprocessed to generate structured text, and the structured knowledge points of the new document are extracted from the structured text. Calculate the importance score index of the new document based on the parameter information of the structured knowledge points; The updated value index is calculated based on the new document importance rating index; Set an update value threshold, compare the update value index with the update value threshold, and if the update value index is greater than the update value threshold, then update the knowledge points in the new document to the knowledge base. If the update value index is less than or equal to the update value threshold, then there is no need to update the knowledge base.

6. The method for constructing a power grid project knowledge base based on a large model as described in claim 4, characterized in that, The formula for calculating the strength of the association between knowledge points is: ； in, For knowledge points With knowledge points The strength of the association, For knowledge points With knowledge points semantic similarity, Subtract one from the total number of knowledge points in the knowledge base; For knowledge points Average semantic similarity with all other knowledge points; The weighting coefficients for the preset similarity. The preset weighting coefficient for time decay. The weight coefficients are preset average semantic similarity, and .

7. The method for constructing a power grid project knowledge base based on a large model as described in claim 6, characterized in that, The formula for calculating the updated value index is: ； in, To update the value index, The number of knowledge points extracted from the new document. For the first document in the new document The key scoring index for each knowledge point For the first document in the new document The similarity between each knowledge point and the knowledge base. As the authority index of a new document, The preset weighting coefficients for the similarity of new documents. This is the preset weighting coefficient for the authority of new documents.

8. A power grid project knowledge base construction system based on a large model, employing the power grid project knowledge base construction method based on a large model as described in any one of claims 1 to 7, characterized in that, include: Project Acquisition Unit: Acquires and preprocesses power grid technology project documents to generate structured text; Knowledge scoring unit: Extracts structured knowledge points from structured text and calculates the importance scoring index of the structured knowledge points; selects high-value knowledge points from the structured text based on the importance scoring index; Logical association unit: Calculate the semantic similarity matrix and association strength of any two high-value knowledge points to form a knowledge point pair, establish association edges, and form a knowledge network to be stored in the knowledge base; Data storage unit: stores knowledge base information; Update Analysis Unit: When a new document is input, extract the knowledge points of the new document, calculate the update value index, and determine whether to update the knowledge points in the new document to the knowledge base.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for constructing a power grid project knowledge base based on a large model, as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for constructing a power grid project knowledge base based on a large model, as described in any one of claims 1 to 7.

Images