A modeling method, apparatus, equipment and medium for a power business decision model

Abstract

This application relates to the field of mathematical modeling technology and discloses a modeling method for a power business decision-making model. The method includes: constructing a power corpus segmentation library and a power vector knowledge base to determine the power business problem text; performing a hybrid retrieval of the power business problem text using the power corpus segmentation library and the power vector knowledge base to obtain a first similarity score corresponding to candidate corpus segments and a second similarity score corresponding to candidate document representation vectors; filtering the candidate corpus segments and candidate document representation vectors based on the first and second similarity scores to determine the target reference text; and inputting the power business problem text and the target reference text into a target large language model to obtain the target decision model. The target decision model includes a description of the power business problem, an objective function, and its constraints. Its beneficial effect is that, based on hybrid retrieval and the standardized output of the large language model, the reliability and accuracy of the target decision model are significantly improved.

Description

Technical Field

[0001] This application relates to the field of mathematical modeling technology, and in particular to a modeling method, apparatus, equipment and medium for a power business decision model. Background Technology

[0002] Mathematical modeling involves abstracting and simplifying real-world problems, transforming them into mathematical problems, and then using mathematical theories, methods, and tools to build models described in mathematical language. These models are then solved using mathematical methods to obtain mathematical results corresponding to the real-world problem, thus providing solutions or decision-making basis. Traditional mathematical modeling typically relies on manual labor, requiring high levels of professional knowledge and modeling experience, and often fails to meet practical needs in terms of efficiency and accuracy. With the development of large language modeling technology, leveraging its ability to quickly analyze problems and provide modeling ideas, large language model-assisted mathematical modeling has gradually emerged. However, the performance of large language models is constrained by various factors, and it cannot provide accurate conclusions when dealing with complex real-world problems.

[0003] Therefore, the reliability of mathematical models constructed with the assistance of large language models in related technologies still needs to be improved. Summary of the Invention

[0004] This application provides a modeling method, apparatus, equipment, and medium for a power business decision-making model. By performing a hybrid search in a corpus segmentation library and a vector knowledge base to filter target reference texts, and by constructing a target decision-making model corresponding to the question text through a large language model, the reliability and accuracy of the target decision-making model are significantly improved.

[0005] To achieve the above objectives, the main technical solutions adopted in this application include:

[0006] In a first aspect, embodiments of this application provide a modeling method for a power business decision-making model, the method comprising:

[0007] Construct a power corpus segmentation library and a power vector knowledge base, and identify the text of power business issues;

[0008] By using the power corpus segmentation library and the power vector knowledge base, a hybrid retrieval is performed on the text of the power business question to obtain a first similarity score corresponding to the candidate corpus segment and a second similarity score corresponding to the candidate document representation vector; wherein, the document representation vector in the power vector knowledge base is obtained by vectorizing the power corpus segments in the power corpus segmentation library;

[0009] The candidate corpus segments and candidate document representation vectors are filtered based on the first similarity score and the second similarity score, and the target reference text is determined based on the filtering results.

[0010] The power business problem text and the target reference text are input into the target large language model to obtain the target decision model corresponding to the power business problem text; wherein, the target decision model includes the description text of the power business problem, the objective function, and the constraints of the objective function.

[0011] The modeling method for the power business decision-making model proposed in this application involves a hybrid retrieval of the power business problem text in a power corpus segmentation library and a power vector knowledge base. This yields a first similarity score for the corresponding candidate corpus segments and a second similarity score for the corresponding candidate document representation vectors. The target reference text is then selected based on these first and second similarity scores and used as input to a target large language model to construct a target decision-making model corresponding to the power business problem text. Compared to related technologies, this application uses a hybrid retrieval of the problem text based on corpus segments and representation vectors to understand the semantics of the problem text from different perspectives. This results in similarity scores representing the matching degree of different semantic understanding results, thus integrating the advantages of multiple retrieval methods. This effectively improves the accuracy of the target reference text, which in turn facilitates the target large language model's understanding and reasoning of the problem text, enhancing the reliability and accuracy of the resulting target decision-making model in solving power business problems. Furthermore, this application automates the semantic understanding and mathematical modeling process through hybrid retrieval and the standardized output of the large language model, improving the efficiency of decision-making model construction and promoting the application of large language models in the power industry, which is beneficial for the intelligent optimization and upgrading of the power industry.

[0012] Optionally, the step of performing a hybrid retrieval of the power business question text using the power corpus segmentation library and the power vector knowledge base to obtain a first similarity score corresponding to the candidate corpus segment and a second similarity score corresponding to the candidate document representation vector includes:

[0013] Based on the power business issue text, a sparse search is performed in the power corpus segmentation library to obtain multiple candidate corpus segments and a sparse similarity score corresponding to each candidate corpus segment, which is used as the first similarity score;

[0014] Based on the power business issue text, a dense search is performed in the power vector knowledge base to obtain multiple candidate document representation vectors and a dense similarity score corresponding to each candidate document representation vector, which is used as the second similarity score.

[0015] Optionally, the power corpus segmentation library contains target metadata corresponding to the target reference text; the step of inputting the power business problem text and the target reference text into the target large language model to obtain the target decision model corresponding to the power business problem text includes:

[0016] The target reference text is used to enhance the power business issue text, resulting in enhanced issue prompt words;

[0017] The enhanced question prompts and the target meta-information are input into the target large language model for collaborative reasoning to obtain the target decision model.

[0018] Optionally, the power corpus segments are obtained in the following manner:

[0019] The initial document data is transformed to obtain the target format document data;

[0020] The target format document data is cleaned and validated to obtain standardized document data;

[0021] The standardized document data is adaptively segmented to obtain the power corpus segments.

[0022] Optionally, the step of converting the initial document data to obtain the target format document data includes:

[0023] The initial document data is converted using a first-type conversion tool to obtain a first conversion result;

[0024] The initial document data is transformed using a second type of conversion tool to obtain a second conversion result;

[0025] Based on the first conversion result and the second conversion result, the target format document data is obtained through synergistic complementarity.

[0026] Optionally, the step of cleaning and validating the target format document data to obtain standardized document data includes:

[0027] The target format document data is cleaned to retain preset core knowledge elements, resulting in intermediate document data; wherein, the preset core knowledge elements include problem description, objective function, and constraints.

[0028] In response to the validation operation on the intermediate document data, the standardized document data is obtained.

[0029] Optionally, the power vector knowledge base is obtained in the following way:

[0030] The initial document data is transformed to obtain the target format document data;

[0031] The target format document data is cleaned and validated to obtain standardized document data;

[0032] The standardized document data is adaptively segmented to obtain power corpus segments;

[0033] The power corpus is segmented and embedded into text to obtain text representation vectors, and the power vector knowledge base is constructed based on the text representation vectors.

[0034] Secondly, embodiments of this application provide a modeling apparatus for a power business decision-making model, characterized in that the apparatus comprises:

[0035] The question text acquisition module is used to build a power corpus segmentation library and a power vector knowledge base, and to identify power business question texts;

[0036] The hybrid retrieval and scoring module is used to perform hybrid retrieval of the power business question text using the power corpus segmentation library and the power vector knowledge base, to obtain a first similarity score corresponding to the candidate corpus segment and a second similarity score corresponding to the candidate document representation vector; wherein, the document representation vector in the power vector knowledge base is obtained by vectorizing the power corpus segments in the power corpus segmentation library;

[0037] The reference text filtering module is used to filter the candidate corpus segments and the candidate document representation vectors based on the first similarity score and the second similarity score, and to determine the target reference text based on the filtering results;

[0038] The decision model construction module is used to input the power business problem text and the target reference text into the target large language model to obtain the target decision model corresponding to the power business problem text; wherein, the target decision model includes the description text of the power business problem, the objective function, and the constraints of the objective function.

[0039] Thirdly, embodiments of this application provide a computer device, including: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method described in any of the above embodiments.

[0040] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer instructions, which are used to cause a computer to perform the method described in any one of the above embodiments.

[0041] Fifthly, embodiments of this application provide a computer program product, including computer instructions, which are used to cause a computer to perform the method described in any of the above embodiments. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0043] Figure 1 A step diagram illustrating the modeling method for the power business decision-making model provided in this application embodiment;

[0044] Figure 2 This is a flowchart of the modeling method in the embodiments of this application;

[0045] Figure 3 This is a flowchart illustrating the steps of the hybrid search in an embodiment of this application;

[0046] Figure 4 This is a flowchart of the hybrid search in the embodiments of this application;

[0047] Figure 5 This is a flowchart illustrating the steps involved in obtaining the target decision model in the embodiments of this application.

[0048] Figure 6 This is a flowchart of the target decision model obtained in the embodiments of this application;

[0049] Figure 7 This is a diagram illustrating the steps involved in obtaining power corpus segments in an embodiment of this application.

[0050] Figure 8 This is a flowchart illustrating the process of obtaining power corpus segments in this application embodiment;

[0051] Figure 9 This is a diagram illustrating the steps involved in obtaining the target format document data in an embodiment of this application.

[0052] Figure 10 This is a diagram illustrating the steps involved in obtaining standardized document data in an embodiment of this application.

[0053] Figure 11 This is a diagram illustrating the steps involved in constructing a power vector knowledge base in an embodiment of this application.

[0054] Figure 12 This is a flowchart illustrating the construction of the power vector knowledge base in this application embodiment;

[0055] Figure 13 A modeling device module diagram for the power business decision-making model provided in the embodiments of this application;

[0056] Figure 14 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0058] Mathematical modeling involves abstracting and simplifying real-world problems, transforming them into mathematical problems, and then using mathematical theories, methods, and tools to build models described in mathematical language. These models are then solved using mathematical methods to obtain mathematical results corresponding to the real-world problem, thus providing solutions or decision-making basis. Traditional mathematical modeling typically relies on manual labor, requiring high levels of professional knowledge and modeling experience, and often fails to meet practical needs in terms of efficiency and accuracy. With the development of large language modeling technology, leveraging its ability to quickly analyze problems and provide modeling ideas, large language model-assisted mathematical modeling methods have gradually emerged, significantly improving the efficiency of mathematical modeling.

[0059] However, in reality, the performance of large language models is constrained by various factors, making it unable to provide accurate conclusions when dealing with complex practical problems. On one hand, large language models suffer from model illusion, easily generating erroneous information that deviates from reality when constructing decision-making models, leading to reduced reliability and decision-making errors. On the other hand, large language models are trained on historical data; when applied to domains with rapid knowledge updates, the training data may not be updated in a timely manner, causing the large language model to fail to correctly understand the problem text and thus fail to output reliable conclusions.

[0060] In addition, while general-purpose large language models in related technologies typically possess extensive knowledge across multiple domains, they may lack in-depth knowledge of some highly specialized specific domains. This can lead to the general-purpose large language model being unable to accurately understand the problem text, thereby affecting the accuracy of the resulting target decision model.

[0061] To address the aforementioned issues, this application provides a modeling method, apparatus, equipment, and medium for a power business decision-making model. The method includes: constructing a power corpus segmentation library and a power vector knowledge base to determine the power business problem text; performing a hybrid search on the power business problem text using the power corpus segmentation library and the power vector knowledge base to obtain a first similarity score corresponding to candidate corpus segments and a second similarity score corresponding to candidate document representation vectors; filtering the candidate corpus segments and candidate document representation vectors based on the first and second similarity scores to determine the target reference text; and inputting the power business problem text and the target reference text into a target large language model to obtain a target decision model. The target decision model includes a description of the power business problem, an objective function, and its constraints.

[0062] The modeling method for the power business decision-making model provided in this application is based on a hybrid retrieval of power business problem text in a power corpus segmentation library and a power vector knowledge base to obtain the first similarity score of the corresponding candidate corpus segment and the second similarity score of the corresponding candidate document representation vector. The target reference text is then obtained by filtering based on the first and second similarity scores, and is used as the input of the target large language model to construct the target decision-making model corresponding to the power business problem text.

[0063] Compared with related technologies, this application uses a hybrid retrieval method based on corpus segmentation and representation vectors to understand the semantics of the question text from different perspectives. This yields similarity scores representing the matching degree of different semantic understanding results, thus integrating the advantages of multiple retrieval methods and effectively improving the accuracy of the target reference text. This, in turn, facilitates the understanding and reasoning of the question text by the target large language model, enhancing the reliability and accuracy of the resulting target decision model for solving power business problems. Furthermore, this application automates the semantic understanding and mathematical modeling processes through hybrid retrieval and the standardized output of the large language model, improving the efficiency of decision model construction and promoting the application of large language models in the power industry, which is beneficial for the intelligent optimization and upgrading of the power industry.

[0064] The modeling methods for power business decision-making models provided in this manual can be applied to construct decision-making models for business problems in the power sector, including but not limited to classic power business decision-making problems such as power dispatch, unit combination, power flow numerical calculation, generation control, and photovoltaic capacity. It is understood that the modeling methods for power business decision-making models provided in this manual, after adaptive modifications, can also be used to construct decision-making models for business problems in other fields, including but not limited to transportation, manufacturing, and financial management.

[0065] According to an embodiment of this application, a modeling method for a power business decision model is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0066] This embodiment provides a modeling method for a power business decision-making model, which can be used to construct a decision-making model for business problems in the power sector. (Refer to...) Figure 1 As shown, the method includes:

[0067] S100. Construct a power corpus segmentation library and a power vector knowledge base, and identify power business issue texts.

[0068] S200. By using the power corpus segmentation library and the power vector knowledge base, a hybrid retrieval of power business issue text is performed to obtain the first similarity score corresponding to the candidate corpus segment and the second similarity score corresponding to the candidate document representation vector; wherein, the document representation vector in the power vector knowledge base is obtained by vectorizing the power corpus segments in the power corpus segmentation library.

[0069] S300. Based on the first similarity score and the second similarity score, candidate corpus segments and candidate document representation vectors are filtered, and the target reference text is determined based on the filtering results.

[0070] S400. Input the power business problem text and the target reference text into the target large language model to obtain the target decision model corresponding to the power business problem text; wherein, the target decision model includes the description text of the power business problem, the objective function, and the constraints of the objective function.

[0071] The power corpus segmentation library and the power vector knowledge base are constructed based on relevant knowledge in the core domain of power business. They are used for hybrid retrieval based on power business issue texts to determine target reference texts. The power corpus segmentation library is constructed from corpus segments of relevant knowledge in the core domain of power business. It stores corpus segments of relevant knowledge for retrieval of power business issue texts based on these corpus segments. These corpus segments are obtained through semantic understanding and semantic segmentation of relevant knowledge in the core domain of power business, with each corpus segment corresponding to a meaningful semantic unit. The power vector knowledge base is constructed from the representation vectors of the relevant knowledge corpus segments. It stores the representation vectors of these corpus segments for retrieval of power business issue texts based on these representation vectors. These representation vectors are obtained by vectorizing the power corpus segments in the power corpus segmentation library. Vectorization allows for a deeper understanding of relevant knowledge from the perspective of representation vectors, providing a foundation for subsequent hybrid retrieval. It is understandable that the document representation vectors in the power vector knowledge base are obtained by vectorizing the power corpus segments in the power corpus segmentation library.

[0072] The power business issue text can be text data representing power business problems, obtained based on user input of the power business problem to be queried. The power business problem can be a business issue that needs to be solved in a power scenario, including but not limited to classic power business decision-making problems such as power dispatch, unit combination, power flow numerical calculation, generation control, and photovoltaic capacity. It can be understood that by embedding the power business problem text, a corresponding representation vector is obtained for retrieval in a power vector knowledge base.

[0073] Specifically, based on the key business needs and technical pain points of the power industry, relevant knowledge from multiple core areas of power business is collected, and the textual form of this knowledge is processed to construct a power corpus segmentation library and a power vector knowledge base. In response to user input regarding power business questions, the text of the power business question is obtained, and text embedding is performed on the power business question text to obtain a representation vector corresponding to the power business question text.

[0074] Reference Figure 2As shown, a hybrid retrieval is performed on the text of power business issues and its representation vectors in a power corpus segmentation database and a power vector knowledge base. Based on the retrieval results from multiple databases, a first similarity score corresponding to the candidate corpus segment and a second similarity score corresponding to the candidate document representation vector are obtained. It should be noted that the hybrid retrieval includes two or more retrieval methods, with at least one retrieval method used in both the power corpus segmentation database and the power vector knowledge base. Multiple segment similarity scores are obtained for each candidate corpus segment based on the retrieval method used in the power corpus segmentation database. These multiple segment similarity scores are then fused to obtain the first similarity score for each candidate corpus segment. Similarly, multiple vector similarity scores are obtained for each candidate document representation vector based on the retrieval method used in the power vector knowledge base. These multiple vector similarity scores are then fused to obtain the second similarity score for each candidate document representation vector.

[0075] It should be noted that when retrieving texts related to power business issues from the power corpus segmentation database, the text can be understood from the perspective of corpus segmentation. Searching based on text content similarity yields candidate corpus segments relevant to the power business issues text intuitively and quickly. Conversely, when retrieving representation vectors corresponding to power business issues texts from the power vector knowledge base, the text can be understood from the perspective of representation vectors. Searching based on text semantic similarity effectively captures the implicit deep semantic similarities of the texts, improving the accuracy of the target reference text. By combining multiple retrieval methods and comprehensively considering both the first and second similarity scores, the advantages of hybrid retrieval are combined. This not only ensures the efficiency and scope of hybrid retrieval but also improves the accuracy of the search results, significantly enhancing the overall quality of hybrid retrieval.

[0076] Furthermore, since the document representation vectors in the power vector knowledge base are obtained by vectorizing power corpus segments in the power corpus segmentation library, the candidate document representation vectors obtained based on the power vector knowledge base correspond to the candidate corpus segments obtained based on the power corpus segmentation library. That is, the first similarity score and the second similarity score both correspond to the same knowledge text content. Taking into account both the first and second similarity scores, similarity matching is performed based on the candidate corpus segments and candidate document representation vectors corresponding to the same knowledge text content to obtain target reference texts related to power business issues. For example, the method of fusing the first and second similarity scores can be a weighted summation, and knowledge text content that meets the similarity requirements is selected as the target reference text. The method for selecting target reference texts can be the top-k method, which selects the k knowledge text contents with the highest matching degree to the power business issue text as target reference texts.

[0077] Furthermore, the power business problem text and the obtained target reference text are jointly input into the target large language model. The target large language model is then used to perform collaborative reasoning on the power business problem text and the target reference text to generate a target decision model corresponding to the power business problem text, which is then used to solve the power business problem.

[0078] It should be noted that the target large language model can be generated and output as a target decision model based on the output template prompts. The output template prompts are used to standardize the output of the target large language model, ensuring that the resulting target decision model has logical consistency and problem relevance. Understandably, the output template prompts include a description of the power business problem, the objective function, and its constraints. The description of the power business problem describes its application scenario and requirements, while the objective function and its constraints can be used to solve the power business problem using mathematical tools, providing users with solutions or decision-making basis.

[0079] The modeling method for the power business decision-making model provided in this embodiment is based on a hybrid retrieval of power business problem text in a power corpus segmentation library and a power vector knowledge base to obtain the first similarity score of the corresponding candidate corpus segment and the second similarity score of the corresponding candidate document representation vector. The target reference text is then obtained by filtering based on the first and second similarity scores, and is used as the input of the target large language model to construct the target decision-making model corresponding to the power business problem text.

[0080] Compared with related technologies, this application uses a hybrid retrieval method based on corpus segmentation and representation vectors to understand the semantics of the question text from different perspectives. This yields similarity scores representing the matching degree of different semantic understanding results, thus integrating the advantages of multiple retrieval methods and effectively improving the accuracy of the target reference text. This, in turn, facilitates the understanding and reasoning of the question text by the target large language model, enhancing the reliability and accuracy of the resulting target decision model for solving power business problems. Furthermore, this application automates the semantic understanding and mathematical modeling processes through hybrid retrieval and the standardized output of the large language model, improving the efficiency of decision model construction and promoting the application of large language models in the power industry, which is beneficial for the intelligent optimization and upgrading of the power industry.

[0081] Reference Figure 3 As shown, in one embodiment of this application, a hybrid retrieval of power business issue text is performed using a power corpus segmentation library and a power vector knowledge base to obtain a first similarity score corresponding to the candidate corpus segment and a second similarity score corresponding to the candidate document representation vector, including:

[0082] S210. Based on the text of power business issues, perform sparse retrieval in the power corpus segmentation library to obtain multiple candidate corpus segments and the sparse similarity score corresponding to each candidate corpus segment, which is used as the first similarity score.

[0083] S220. Based on the text of power business issues, perform dense retrieval in the power vector knowledge base to obtain multiple candidate document representation vectors and the dense similarity score corresponding to each candidate document representation vector, which is used as the second similarity score.

[0084] Reference Figure 4 As shown, in this embodiment, after embedding the text of the power business issue to obtain the corresponding representation vector, sparse retrieval is performed on the power business issue text to obtain the first similarity score of the candidate corpus blocks, and dense retrieval is performed on the representation vectors corresponding to the power business issue text to obtain the second similarity score of the candidate document representation vectors. Finally, the first similarity score and the second similarity score are fused to determine the target reference text that matches the power business issue text.

[0085] Specifically, sparse retrieval is used to retrieve power business issue texts from a power corpus segmentation database based on the similarity of text content, in order to understand the power business issue texts from the perspective of corpus segmentation. The steps of sparse retrieval may include: preprocessing the power business issue text by word segmentation to obtain multiple issue sub-words; calculating the relevance score between each issue sub-word and words in each corpus segmentation database, determining the matching degree between the power business issue text and each corpus segmentation, obtaining multiple candidate corpus segments and their corresponding sparse similarity scores, which are used as the first similarity score. For example, the algorithm used for sparse retrieval may be the BM25 algorithm, which is based on the term frequency-inverse document frequency (IF-IVF) approach.

[0086] It should be noted that sparse search primarily relies on the similarity between words, offering good search results for specific terms and proper nouns, and enabling precise matching of specific words. Furthermore, sparse search depends solely on word frequency statistics, effectively reducing search time and improving the efficiency of hybrid search methods.

[0087] Furthermore, dense retrieval is used to retrieve the representation vectors corresponding to power business issue texts in the power vector knowledge base based on textual semantic similarity, in order to understand the power business issue texts from the perspective of representation vectors. The steps of dense retrieval may include: calculating the semantic similarity between the document representation vector and the power business issue text based on the document representation vector and the representation vector corresponding to the power business issue text, obtaining multiple candidate document representation vectors and the dense similarity score corresponding to each candidate document representation vector, as a second similarity score.

[0088] It should be noted that dense retrieval is mainly based on the similarity between representation vectors. It can understand the overall semantics of power business problem texts, has strong semantic generalization, and can capture synonyms, near-synonyms, and complex semantic relationships. It can exhibit better robustness in complex query processes, effectively improve the accuracy of retrieval results, and enhance the retrieval quality of hybrid retrieval.

[0089] Reference Figure 5 As shown, in one embodiment of this application, the power corpus segmentation library contains target metadata corresponding to the target reference text; the power business problem text and the target reference text are input into the target large language model to obtain the target decision model corresponding to the power business problem text, including:

[0090] S410. Enhance the power business issue text using the target reference text to obtain enhanced issue prompt words.

[0091] S420. Input the enhanced question prompts and target meta-information into the target large language model for collaborative reasoning to obtain the target decision model.

[0092] Among them, target meta-information can be the hierarchical key points of the target reference text, including but not limited to chapter titles or core keywords in the target reference text. By extracting the target meta-information corresponding to the target reference text, the core semantics and hierarchical structure of the target reference text can be effectively extracted, which is conducive to the target large language model's understanding of the target reference text.

[0093] Reference Figure 6 As shown, the target reference text is the result of a hybrid search of the power business question text in the power corpus chunk library and the power vector knowledge base. The target reference text can be used to supplement, expand, and enhance the power business question text. The process of enhancing the power business question text using the target reference text can include: combining the target reference text with the power business question text in context to semantically expand the content of the power business question text, thereby further explaining the power business question text; concatenating the target reference text into specific parts of the power business question text to supplement missing content and improve the comprehensiveness and accuracy of the power business question text; inserting the target reference text as keywords into the power business question text to improve the semantic accuracy of the power business question text, which is beneficial for the target large language model to understand the power business question text; and obtaining enhanced question prompt words through the enhancement of the power business question text.

[0094] Furthermore, the enhanced question prompts and target meta-information are input into the target large language model. The target large language model performs semantic understanding of the enhanced question prompts based on the target meta-information, and performs collaborative reasoning based on the understood content to construct a target decision model corresponding to the power business question text.

[0095] It should be noted that the target large language model can be generated and output as a target decision model based on the output template prompts. The output template prompts are used to standardize the output of the target large language model, ensuring that the resulting target decision model has logical consistency and problem relevance. Understandably, the output template prompts include a description of the power business problem, the objective function, and its constraints. The description of the power business problem describes its application scenario and requirements, while the objective function and its constraints provide users with solutions or decision-making basis for solving the power business problem.

[0096] Reference Figure 7 As shown, in one embodiment of this application, the power corpus segments are obtained in the following manner:

[0097] S110. Convert the initial document data to obtain the target format document data.

[0098] S120. Clean and validate the target format document data to obtain standardized document data.

[0099] S130. Adaptively segment the standardized document data to obtain power corpus segments.

[0100] Specifically, relevant knowledge in the core areas of power business can be in document form, and initial document data is obtained based on this document-based knowledge. This initial document data can be obtained from authoritative academic databases, professional journal databases, and university dissertation resource databases, including but not limited to relevant literature such as e-books on power business, journal articles, and master's and doctoral dissertations, covering the key business needs and technical pain points of the power industry.

[0101] Reference Figure 8 As shown, the initial document data may have inconsistent formats. After obtaining the initial document data, a data conversion process is performed to convert the document format of the initial document data to the same format, resulting in the target format document data. For example, the document format of the initial document data can be PDF, Word, or CAJ, while the document format of the target format document data can be Markdown.

[0102] Furthermore, the target format document data may contain redundant information unrelated to power business issues. This redundant information may interfere with the large language model's understanding of the text concerning power business issues. Therefore, it is necessary to clean the target format document data to remove redundant information while retaining the important core information, thereby improving the relevance of power-related corpus segmentation. Understandably, the cleaning process may result in the removal of core information or significant impact on the data content of the target format document data. In such cases, the cleaning results need to be validated to ensure that the cleaned target format document data has a consistent format and is free of data content errors and logical problems, thus ensuring the data quality of the target format document data and obtaining standardized document data.

[0103] Furthermore, metadata is extracted from the standardized document data to obtain document metadata. Document metadata can be the hierarchical key points of the standardized document data, including but not limited to chapter titles or core keywords, to facilitate understanding the structure and hierarchy of the standardized document data. The document metadata is then concatenated with the standardized document data to utilize the metadata for structural division and semantic summarization. This allows for adaptive segmentation of the concatenated standardized document data, resulting in multiple power-related corpus segments. It is understood that the content in each power-related corpus segment targets the same topic, exhibiting high relevance and logical consistency. By storing these power-related corpus segments in a knowledge base to form a power-related corpus segment library, structured management of power business knowledge becomes possible, facilitating updates and maintenance of power business knowledge while improving the accuracy of hybrid retrieval.

[0104] Reference Figure 9 As shown, in one embodiment of this application, the initial document data is converted to obtain target format document data, including:

[0105] S112. Use the first type conversion tool to convert the initial document data and obtain the first conversion result.

[0106] S114. Use the second type of conversion tool to perform data conversion on the initial document data to obtain the second conversion result.

[0107] S116. Based on the first and second conversion results, perform collaborative and complementary transformations to obtain the target format document data.

[0108] Specifically, the initial document data can include both Type I and Type II document data. Type I document data can be large documents with a lot of text content, such as master's or doctoral dissertations or e-books, while Type II document data can be small documents with less text content, such as papers from small journals or conferences. Faced with a large amount of initial document data, using only a single type conversion tool may not be able to balance conversion efficiency and accuracy.

[0109] For the reasons mentioned above, this embodiment employs different first-type and second-type conversion tools. The first-type conversion tool is used to convert the initial document data, yielding a first conversion result. Then, the second-type conversion tool is used to convert the initial document data, yielding a second conversion result. For example, the first-type conversion tool could be the Doubao web version, which has high accuracy and fidelity when processing large documents, but is prone to over-summarization and has low efficiency in processing document data. The second-type conversion tool could be the MinerU software, which has batch processing capabilities and can quickly convert multiple document data sets, but it suffers from a high error rate in formula conversion.

[0110] Furthermore, after obtaining the first and second conversion results, the first and second conversion results are synergistically complemented to obtain the target format document data. This leverages the respective strengths of the first and second type of conversion tools to improve the conversion quality of the target format document data. Specifically, the synergistic complementarity process can be as follows: comparing the first and second conversion results to identify over-compressed document data in the first conversion result and formula error data in the second conversion result; identifying document conversion data corresponding to the compressed document data in the second conversion result to replace the compressed document data in the first conversion result, thus obtaining the first synergistic result; identifying formula conversion data corresponding to the formula error data in the first conversion result to replace the formula error data in the second conversion result, thus obtaining the second synergistic result; and obtaining the target format document data based on the first and second synergistic results.

[0111] It should be noted that replacing the compressed document data in the first conversion result with document conversion data can supplement the document data that was over-summarized by the first type of conversion tool, improving the completeness of the resulting first collaborative result. Replacing the formula error data in the second conversion result with formula conversion data can correct the formula errors caused by the second type of conversion tool, improving the accuracy of the resulting second collaborative result. By combining the first and second collaborative results to obtain the target format document data, the respective advantages of the first and second type of conversion tools are utilized to improve the conversion quality of the target format document data.

[0112] Reference Figure 10 As shown, in one embodiment of this application, the target format document data is cleaned and validated to obtain standardized document data, including:

[0113] S122. Clean the target format document data, retain the preset core knowledge elements, and obtain intermediate document data; wherein, the preset core knowledge elements include problem description, objective function and constraints.

[0114] S124. In response to the validation operation on the intermediate document data, standardized document data is obtained.

[0115] Specifically, redundant information irrelevant to the power business problem is removed from the target format document data through cleaning, retaining the pre-defined core knowledge elements. These pre-defined core knowledge elements include a problem description, an objective function, and constraints. The problem description can be the descriptive text of the business problem to be solved within the target format document data, used to help the large language model understand the business problem within the relevant knowledge. The objective function can be the mathematical tool used in the target format document data to solve the business problem, used to help the large language model learn the corresponding tool for solving the business problem. Constraints can be used to constrain the objective function in the target format document data to match it with the actual scenario corresponding to the power business problem text, improving the accuracy and reliability of the target decision model. By retaining the pre-defined core knowledge elements, the information quality of the standardized document data is improved, enhancing the model building capabilities of the target large language model.

[0116] It is understandable that during the cleaning process, there may be situations where preset core knowledge elements are removed or the data content of the target format document data is significantly affected. In such cases, it is necessary to verify the cleaning results to ensure that the cleaned target format document retains the preset core knowledge elements, and that the data format is consistent, without content errors or logical problems, thereby ensuring the data quality of the target format document data and obtaining standardized document data.

[0117] Reference Figure 11As shown, in one embodiment of this application, the power vector knowledge base is obtained in the following manner:

[0118] S140. Convert the initial document data to obtain the target format document data.

[0119] S150. Perform data cleaning and data verification on the target format document data to obtain standardized document data.

[0120] S160. Adaptively segment the standardized document data to obtain power corpus segments.

[0121] S170. The power corpus is segmented and embedded into text to obtain text representation vectors, and a power vector knowledge base is constructed based on the text representation vectors.

[0122] Specifically, relevant knowledge in the core areas of power business can be in document form, and initial document data is obtained based on this document-based knowledge. This initial document data can be obtained from authoritative academic databases, professional journal databases, and university dissertation resource databases, including but not limited to relevant literature such as e-books on power business, journal articles, and master's and doctoral dissertations, covering the key business needs and technical pain points of the power industry.

[0123] Reference Figure 12 As shown, the initial document data may have inconsistent formats. After obtaining the initial document data, a data conversion process is performed to convert the document format of the initial document data to the same format, resulting in the target format document data. For example, the document format of the initial document data can be PDF, Word, or CAJ, while the document format of the target format document data can be Markdown.

[0124] Furthermore, the target format document data may contain redundant information unrelated to power business issues. This redundant information may interfere with the large language model's understanding of the text concerning power business issues. Therefore, it is necessary to clean the target format document data to remove redundant information while retaining the important core information, thereby improving the relevance of power-related corpus segmentation. Understandably, the cleaning process may result in the removal of core information or significant impact on the data content of the target format document data. In such cases, the cleaning results need to be validated to ensure that the cleaned target format document data has a consistent format and is free of data content errors and logical problems, thus ensuring the data quality of the target format document data and obtaining standardized document data.

[0125] Furthermore, metadata is extracted from the standardized document data to obtain document metadata. Document metadata can be the hierarchical key points of the standardized document data, including but not limited to chapter titles or core keywords, to facilitate understanding the structure and hierarchy of the standardized document data. The document metadata is then concatenated with the standardized document data to utilize the metadata for structural division and semantic summarization. This allows for adaptive segmentation of the concatenated standardized document data, resulting in multiple power-related corpus segments. It is understood that the content in each power-related corpus segment targets the same topic, exhibiting high relevance and logical consistency. By storing these power-related corpus segments in a knowledge base to form a power-related corpus segment library, structured management of power business knowledge becomes possible, facilitating updates and maintenance of power business knowledge while improving the accuracy of hybrid retrieval.

[0126] Furthermore, the text content in the power corpus segments is mapped to a vector space, and representation vectors corresponding to the power corpus segments are generated based on the mapping results, resulting in text representation vectors. These text representation vectors are stored in a vector library to construct a power vector knowledge base. For example, the text embedding method can be dense embedding to map the power corpus segments into dense feature vectors. In some embodiments, to further improve the efficiency of hybrid retrieval, this embodiment can also establish a vector index for the text representation vectors and store it in the power vector knowledge base, thereby improving the efficiency of retrieving document representation vectors.

[0127] The following example illustrates an implementation scenario of this method. To evaluate this method, this example collects and cleanses 17 problems from the "Economic Operation Theory of Power Systems," including classic power scenario problems such as power dispatch, unit combination, and power flow calculation. The decision model corresponding to each problem is used as the standard answer, thus forming an evaluation dataset. The quantitative evaluation framework adopted in this example is the Retrieval Augmented Generation Assessment (RAGAs) framework, using the external large language model Qwen-Plus as the model judge, and simultaneously focusing on retrieval quality indicators and generation quality indicators.

[0128] The quality of retrieval is primarily measured in three aspects: the relevance of the context to the query question, the relevance of the context to the true answer, and the relevance of the context to the model output. The main metric for measuring the relevance of the context to the query question is Context Relevance (CR). Low relevance indicates that the retrieval system has introduced noise irrelevant to the query question. The main metrics for measuring the relevance of the context to the true answer are LLM Context Recall (LCR, Large Language Model Context Recall) and Context Precision (CP). LCR measures whether the retrieved text contains all the information needed to generate the true answer; low recall means the retrieval system has missed key information. CP measures whether the retrieved text is semantically relevant and accurate to the true answer; low precision indicates poor semantic relevance between the retrieved context and the true answer, meaning the context lacks effective information. The metrics for measuring the relevance of the context to the model output include Faithfulness (FF) and Factual Correctness (FC), which measure whether the model-generated answer is based on the retrieved text and whether it contains fabricated false information that exacerbates the illusion.

[0129] The quality of the generated data is primarily measured by two aspects: the relevance of the model output to the query question and the relevance of the model output to the actual answer. Regarding the relevance of the model output to the query question, the AnswerRelevancy (AR) metric can be used to determine whether the answer directly and effectively addresses the question, rather than being irrelevant. Regarding the relevance of the model output to the actual answer, the AnswerCorrectness (AC) and AnswerSimilarity (AS) metrics can be used to evaluate the accuracy of the model's answer and its relevance to the standard answer.

[0130] This embodiment quantitatively compares the DeepSeek V3 model with the proposed method, and also compares the performance of three retrieval methods: dense retrieval, sparse retrieval, and hybrid retrieval. The comparison results are shown in Table 1, where the values ​​represent the improvement of the corresponding method on the quantitative indicator. It should be noted that the target reference text for the DeepSeek V3 model is set to empty.

[0131] Table 1. Comparison of quantitative indicators between this method and the DeepSeek V3 model.

[0132]

[0133] As shown in Table 1, compared with the DeepSeek V3 model, our method significantly improves in all three generation quality metrics: AR, AC, and AS, indicating that our method offers a substantial improvement in decision model generation compared to conventional techniques. Furthermore, for power business questions, compared to using dense or sparse retrieval alone, the reference text obtained through hybrid retrieval is more relevant to the query question and has a higher similarity to the standard answer, resulting in the best generation quality.

[0134] Accordingly, please refer to Figure 13 This application provides a modeling apparatus for a power business decision-making model, the apparatus comprising:

[0135] The problem text acquisition module 1310 is used to construct a power corpus segmentation library and a power vector knowledge base, and to determine the problem text of power business.

[0136] The hybrid retrieval and scoring module 1320 is used to perform hybrid retrieval of power business issue texts by means of the power corpus segmentation library and the power vector knowledge base, and to obtain the first similarity score corresponding to the candidate corpus segment and the second similarity score corresponding to the candidate document representation vector; wherein, the document representation vector in the power vector knowledge base is obtained by vectorizing the power corpus segments in the power corpus segmentation library.

[0137] The reference text filtering module 1330 is used to filter candidate corpus segments and candidate document representation vectors based on the first similarity score and the second similarity score, and to determine the target reference text based on the filtering results.

[0138] The decision model construction module 1340 is used to input the power business problem text and the target reference text into the target large language model to obtain the target decision model corresponding to the power business problem text; wherein, the target decision model includes the description text of the power business problem, the objective function, and the constraints of the objective function.

[0139] In some alternative implementations, the hybrid retrieval scoring module 1320 includes:

[0140] The sparse retrieval unit for corpus segments is used to perform sparse retrieval in the power corpus segmentation library based on power business issue text, and obtain multiple candidate corpus segments and the sparse similarity score corresponding to each candidate corpus segment, which is used as the first similarity score.

[0141] The representation vector dense retrieval unit is used to perform dense retrieval in the power vector knowledge base based on power business issue text, and obtain multiple candidate document representation vectors and the dense similarity score corresponding to each candidate document representation vector, which is used as the second similarity score.

[0142] In some alternative implementations, the decision model building module 1340 includes:

[0143] The problem text enhancement unit is used to enhance the power business problem text using the target reference text, resulting in enhanced problem prompt words.

[0144] The collaborative reasoning unit is used to input the enhanced question prompts and target meta-information into the target large language model for collaborative reasoning to obtain the target decision model.

[0145] In some optional implementations, the problem text acquisition module 1310 includes:

[0146] The document data conversion unit is used to convert the initial document data to obtain the target format document data.

[0147] The document cleaning and verification unit is used to clean and verify the target format document data to obtain standardized document data.

[0148] The document adaptive slicing unit is used to adaptively slice standardized document data to obtain power corpus slicing.

[0149] In some optional implementations, the document data conversion unit includes:

[0150] The first data conversion subunit is used to convert the initial document data using a first type of conversion tool to obtain the first conversion result.

[0151] The second data conversion subunit is used to convert the initial document data using a second type of conversion tool to obtain a second conversion result.

[0152] The data collaboration and complementarity subunit is used to collaborate and complement the first and second transformation results to obtain target format document data.

[0153] In some optional implementations, the document cleaning and verification unit includes:

[0154] The data cleaning and retention subunit is used to clean the target format document data, retain the preset core knowledge elements, and obtain intermediate document data; the preset core knowledge elements include problem description, objective function, and constraints.

[0155] The data validation subunit is used to obtain standardized document data in response to validation operations on intermediate document data.

[0156] In some optional implementations, the problem text acquisition module 1310 further includes:

[0157] The document data conversion unit is used to convert the initial document data to obtain the target format document data.

[0158] The document cleaning and verification unit is used to clean and verify the target format document data to obtain standardized document data.

[0159] The document adaptive slicing unit is used to adaptively slice standardized document data to obtain power corpus slicing.

[0160] The segmented text embedding unit is used to segment the power corpus into segments and embed the text to obtain text representation vectors, and to build a power vector knowledge base based on the text representation vectors.

[0161] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0162] In this embodiment, the modeling device for the power business decision-making model is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0163] Please see Figure 14 , Figure 14 This is a schematic diagram of a computer device according to an embodiment of this application. As shown in the figure, the computer device includes one or more processors 10, a memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components communicate with each other using different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 14 Take a processor 10 as an example.

[0164] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0165] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0166] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0167] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0168] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0169] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0170] This application provides a computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method of any embodiment of this application.

[0171] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

[0172] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0173] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0174] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0175] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0176] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0177] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0178] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0179] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0180] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0181] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A modeling method of an electric power business decision model, characterized by, The method comprises: The method comprises: The method comprises: The method comprises: The method comprises:

2. The method of claim 1, wherein, The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: perform sparse retrieval on the power corpus cutout library based on the power business problem text, to obtain a plurality of candidate corpus cutouts and a sparse similarity score corresponding to each candidate corpus cutout as the first similarity score; perform dense retrieval on the power vector knowledge base based on the power business problem text, to obtain a plurality of candidate document representation vectors and a dense similarity score corresponding to each candidate document representation vector as the second similarity score.

3. The method of claim 1, wherein, The power corpus cutout library has target meta-information corresponding to the target reference text; The power business problem text and the target reference text are input into a target large language model to obtain a target decision model corresponding to the power business problem text, including: The target reference text is used to enhance the power business problem text to obtain enhanced problem prompt words; The enhanced problem prompt words and the target meta-information are input into the target large language model for collaborative reasoning to obtain the target decision model.

4. The method of claim 1, wherein, The target format document data is cleaned and verified to obtain standardized document data, including: The target format document data is cleaned to retain preset core knowledge elements to obtain intermediate document data; wherein the preset core knowledge elements include problem description, objective function and constraint condition; In response to the verification operation on the intermediate document data, the standardized document data is obtained.

5. The method of claim 1, wherein, The power vector knowledge base is obtained in the following way: The initial document data is converted to obtain target format document data; The target format document data is cleaned and verified to obtain standardized document data; The standardized document data is adaptively cut to obtain power corpus cutouts; The power corpus cutouts are text embedded to obtain text representation vectors, and the power vector knowledge base is constructed based on the text representation vectors.

6. A modeling apparatus of a power business decision model, characterized by, The device includes: The problem text acquisition module is configured to construct an electric power corpus cutout library and an electric power vector knowledge base, and determine an electric power business problem text; wherein the electric power corpus cutout in the electric power corpus cutout library is obtained by: using a first type conversion tool to perform data conversion on initial document data to obtain a first conversion result; using a second type conversion tool to perform data conversion on the initial document data to obtain a second conversion result; performing synergistic complementation based on the first conversion result and the second conversion result to obtain target format document data; performing cleaning and verification on the target format document data to obtain standardized document data; and performing adaptive cutout on the standardized document data to obtain the electric power corpus cutout; the process of synergistic complementation includes: comparing the first conversion result and the second conversion result to determine document compression data in the first conversion result that is over-compressed on the initial document data, and formula error data in the second conversion result that has a formula conversion error; determining document conversion data corresponding to the document compression data in the second conversion result to replace the document compression data in the first conversion result to obtain a first synergistic result; determining formula conversion data corresponding to the formula error data in the first conversion result to replace the formula error data in the second conversion result to obtain a second synergistic result; and obtaining the target format document data according to the first synergistic result and the second synergistic result; The hybrid retrieval scoring module is configured to perform hybrid retrieval on the electric power business problem text by means of the electric power corpus cutout library and the electric power vector knowledge base to obtain a first similarity score corresponding to a candidate corpus cutout and a second similarity score corresponding to a candidate document representation vector; wherein the document representation vector in the electric power vector knowledge base is obtained by vectorizing the electric power corpus cutout in the electric power corpus cutout library; The reference text screening module is configured to screen the candidate corpus cutout and the candidate document representation vector based on the first similarity score and the second similarity score, and determine a target reference text based on a screening result; The decision model construction module is configured to input the electric power business problem text and the target reference text into a target large language model to obtain a target decision model corresponding to the electric power business problem text; wherein the target decision model includes a description text of an electric power business problem, a target function, and a constraint condition of the target function, the target function is a mathematical tool for solving a business problem, the constraint condition is used to constrain the target function to match an actual scenario corresponding to the electric power business problem text, and the target function and its constraint condition can be used to solve the electric power business problem by a mathematical tool to provide a solution or a decision basis for a user.

7. A computer device, comprising: The method comprises: A memory and a processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method of any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium has stored thereon computer instructions for causing a computer to perform the method of any one of claims 1 to 5.

Images