Modeling method and device of power business decision model, equipment and medium

Abstract

The invention relates to the technical field of mathematical modeling, and discloses a modeling method for a power business decision model, and the method comprises the steps: constructing a power corpus block library and a power vector knowledge library, and determining a power business problem text; performing mixed retrieval on the power business problem text by means of a power corpus block library and a power vector knowledge library to obtain a first similarity score corresponding to a candidate corpus block and a second similarity score corresponding to a candidate document representation vector; screening the candidate corpus blocks and the candidate document representation vectors based on the first similarity score and the second similarity score, and determining a target reference text; 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 comprises a description text of the power business problem, a target function and constraint conditions thereof. The method has the beneficial effects that the reliability and the accuracy of the target decision model are remarkably improved based on standard output of the hybrid retrieval and the large language model.

Description

Technical Field

[0001] The present application relates to the technical field of mathematical modeling, and in particular to a modeling method, apparatus, equipment, and medium for an electric power business decision-making model. Background Art

[0002] Mathematical modeling involves rationally abstracting and simplifying real-world problems, converting them into mathematical problems, and then using mathematical theories, methods, and tools to build models described in mathematical language. The models are then solved using mathematical methods, ultimately yielding mathematical results corresponding to the real-world problems, thereby providing solutions or decision-making basis for the problems. Traditional mathematical modeling often relies on manual labor, requiring high levels of professional knowledge and modeling experience, and failing to meet practical needs in terms of modeling efficiency and accuracy. With the development of large language model technology, mathematical modeling approaches assisted by large language models have gradually emerged, leveraging their ability to quickly analyze problems and provide modeling ideas. However, the performance of large language models is constrained by various factors, making them unable to 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] The present application provides a modeling method, apparatus, equipment and medium for an electric power business decision model, which significantly improves the reliability and accuracy of the target decision model by performing a mixed search in a corpus segmentation library and a vector knowledge library to screen target reference texts, and constructing a target decision model corresponding to the question text through a large language model.

[0005] In order to achieve the above objectives, the main technical solutions adopted in this application include: In a first aspect, an embodiment of the present application provides a method for modeling an electric power business decision model, the method comprising: Build a power corpus segmentation database and power vector knowledge base, and determine the power business problem text; A hybrid search is performed on the power business question text using the power corpus chunk library and the power vector knowledge base to obtain a first similarity score corresponding to the candidate corpus chunk 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 chunk in the power corpus chunk library; screening the candidate corpus segments and the candidate document representation vectors based on the first similarity score and the second similarity score, and determining a target reference text based on the screening result; The power business problem text and the target reference text are input into the target large language model to obtain a target decision model corresponding to the power business problem text; wherein the target decision model includes a description text of the power business problem, an objective function and constraints of the objective function.

[0006] The modeling method of the electric power business decision model proposed in the embodiment of the present application performs a hybrid search on the electric power business question text in the electric power corpus segmentation library and the electric power vector knowledge base, obtains a first similarity score corresponding to the candidate corpus segmentation and a second similarity score corresponding to the candidate document representation vector, and screens the target reference text according to the first similarity score and the second similarity score to be used as the input of the target large language model to construct a target decision model corresponding to the electric power business question text. Compared with the related art, the present application performs a hybrid search on the question text based on the corpus segmentation and the representation vector to understand the semantics of the question text from different perspectives, and obtains similarity scores that represent the degree of match of different semantic understanding results. This can integrate the advantages of multiple search methods, effectively improve the accuracy of the target reference text, and further facilitate the understanding and reasoning of the question text by the target large language model, thereby improving the reliability and accuracy of the obtained target decision model for solving electric power business problems. In addition, through the standardized output of the hybrid search and large language model, the present application realizes the automation of the semantic understanding and mathematical modeling process, improves the efficiency of decision model construction, promotes the promotion and application of the large language model in the electric power industry, and is conducive to the intelligent optimization and upgrading of the electric power industry.

[0007] Optionally, the hybrid search of the power business question text by means of the power corpus segmentation library and the power vector knowledge base to obtain a first similarity score corresponding to the candidate corpus segmentation and a second similarity score corresponding to the candidate document representation vector includes: Performing a sparse search in the power corpus chunk library based on the power business question text to obtain a plurality of candidate corpus chunks and a sparse similarity score corresponding to each candidate corpus chunk as the first similarity score; Based on the power business question text, a dense search is performed in the power vector knowledge base 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.

[0008] Optionally, the electric power corpus segmentation library contains target meta-information corresponding to the target reference text; and inputting the electric power business question text and the target reference text into a target large language model to obtain a target decision model corresponding to the electric power business question text includes: Using the target reference text to enhance the power business question text to obtain enhanced question prompt words; The enhanced question prompt words and the target meta-information are input into the target large language model for collaborative reasoning to obtain the target decision model.

[0009] Optionally, the power corpus segments are obtained by: Convert the initial document data to obtain target format document data; Cleaning and verifying the target format document data to obtain standardized document data; Adaptively segment the standardized document data to obtain the power corpus segments.

[0010] Optionally, converting the initial document data to obtain target format document data includes: Performing data conversion on the initial document data using a first type conversion tool to obtain a first conversion result; Performing data conversion on the initial document data using a second type conversion tool to obtain a second conversion result; The target format document data is obtained based on the collaborative complementation of the first conversion result and the second conversion result.

[0011] Optionally, the cleaning and verifying the target format document data to obtain standardized document data includes: Cleaning the target format document data, retaining preset core knowledge elements, and obtaining intermediate document data; wherein the preset core knowledge elements include problem description, objective function, and constraint conditions; In response to a verification operation on the intermediate document data, the standardized document data is obtained.

[0012] Optionally, the power vector knowledge base is obtained by: Convert the initial document data to obtain target format document data; Performing data cleaning and data verification on the target format document data to obtain standardized document data; Adaptively segmenting the standardized document data to obtain power corpus segments; The electric power corpus is cut into pieces and text embedded to obtain text representation vectors, and the electric power vector knowledge base is constructed based on the text representation vectors.

[0013] In a second aspect, an embodiment of the present application provides a modeling device for an electric power business decision model, characterized in that the device includes: The question text acquisition module is used to build the power corpus segmentation library and power vector knowledge base, and determine the power business question text; A hybrid search and scoring module is configured to perform a hybrid search on the power business question text using the power corpus chunk library and the power vector knowledge base to obtain a first similarity score corresponding to the candidate corpus chunk 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 chunk in the power corpus chunk library; a reference text screening module, configured to screen the candidate corpus segments and the candidate document representation vectors based on the first similarity score and the second similarity score, and determine a target reference text based on the screening result; A 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 a target decision model corresponding to the power business problem text; wherein, the target decision model includes a description text of the power business problem, an objective function and the constraints of the objective function.

[0014] In a third aspect, an embodiment of the present application provides a computer device comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, computer instructions are stored in the memory, and the processor executes the method described in any one of the above embodiments by executing the computer instructions.

[0015] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to enable a computer to execute any one of the methods in the above embodiments.

[0016] In a fifth aspect, an embodiment of the present application provides a computer program product, comprising computer instructions, wherein the computer instructions are used to enable a computer to execute any one of the methods described in the above embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the specific implementation methods or the description of the prior art. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0018] Figure 1 A diagram showing the steps of a modeling method for a power business decision model provided in an embodiment of the present application; Figure 2 This is a flow chart of the modeling method in the embodiment of the present application; Figure 3 A diagram showing the steps of hybrid search in an embodiment of the present application; Figure 4 This is a flow chart of the hybrid search in the embodiment of the present application; Figure 5 A diagram showing the steps for obtaining a target decision model in an embodiment of the present application; Figure 6 A flowchart of a target decision model obtained in an embodiment of the present application; Figure 7 A diagram showing the steps for obtaining electric power corpus segments in an embodiment of the present application; Figure 8 This is a flowchart for obtaining electric power corpus segmentation in an embodiment of the present application; Figure 9 A diagram showing the steps for obtaining document data in a target format according to an embodiment of the present application; Figure 10 A diagram showing the steps for obtaining standardized document data in an embodiment of the present application; Figure 11 A diagram showing the steps for constructing a power vector knowledge base in an embodiment of the present application; Figure 12 A flowchart of constructing a power vector knowledge base in an embodiment of the present application; Figure 13 A module diagram of a modeling device for a power business decision model provided in an embodiment of the present application; Figure 14 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] To make the purpose, technical solutions, and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of this application.

[0020] Mathematical modeling involves rationally abstracting and simplifying real-world problems, converting them into mathematical problems, and then using mathematical theories, methods, and tools to build models described in mathematical language. The models are then solved using mathematical methods, ultimately yielding mathematical results corresponding to the real-world problems, thereby providing solutions or decision-making basis for practical problems. Traditional mathematical modeling often relies on manual labor, requiring high levels of professional knowledge and modeling experience, and failing to meet practical needs in terms of modeling efficiency and accuracy. With the development of large language model technology, mathematical modeling approaches assisted by large language models have gradually emerged, leveraging their ability to quickly analyze problems and provide modeling ideas. This has significantly improved the efficiency of mathematical models.

[0021] However, in practice, the performance of large language models is constrained by various factors, making them unable to produce accurate conclusions when handling complex real-world problems. For one thing, large language models are subject to model hallucination, which can easily generate erroneous information that deviates from reality when building decision models. This reduces the reliability of the decision models and can lead to decision errors. Furthermore, large language models are trained based on historical data. When applied to fields with rapidly evolving knowledge, the training data may not be updated in a timely manner, resulting in the large language models failing to correctly understand the question text and, consequently, failing to produce reliable conclusions.

[0022] In addition, the general large language model in related technologies usually has extensive knowledge in multiple fields, but for some highly specialized specific fields, the general large language model may not have in-depth knowledge of the field, resulting in the general large language model being unable to accurately understand the question text, thereby affecting the accuracy of the obtained target decision model.

[0023] Based on the above problems, the present application provides a modeling method, device, equipment and medium for an electric power business decision model, the method including: constructing an electric power corpus segmentation library and an electric power vector knowledge base to determine the electric power business problem text; performing a mixed search on the electric power business problem text with the help of the electric power corpus segmentation library and the electric power vector knowledge base to obtain a first similarity score corresponding to the candidate corpus segmentation and a second similarity score corresponding to the candidate document representation vector; screening the candidate corpus segmentation and the candidate document representation vector based on the first similarity score and the second similarity score to determine the target reference text; inputting the electric power business problem text and the target reference text into the target large language model to obtain a target decision model; the target decision model includes a description text of the electric power business problem, an objective function and its constraints.

[0024] The modeling method of the electric power business decision model provided in the present application performs a mixed search in the electric power corpus segmentation library and the electric power vector knowledge base based on the electric power business problem text, obtains a first similarity score of the corresponding candidate corpus segmentation and a second similarity score of the corresponding candidate document representation vector, and screens the target reference text according to the first similarity score and the second similarity score to obtain the target reference text, which is used as the input of the target large language model to construct a target decision model corresponding to the electric power business problem text.

[0025] Compared with related technologies, this application performs hybrid retrieval of question texts based on corpus segmentation and representation vectors to understand the semantics of question texts from different perspectives, and obtains similarity scores that represent the degree of matching of different semantic understanding results. This can integrate the advantages of multiple retrieval methods and effectively improve the accuracy of the target reference text, which in turn is beneficial to the understanding and reasoning of the question text by the target large language model, and improves the reliability and accuracy of the obtained target decision model for solving power business problems. In addition, this application realizes the automation of semantic understanding and mathematical modeling processes through the standardized output of hybrid retrieval and large language models, improves the efficiency of building decision models, promotes the promotion and application of large language models in the power industry, and is beneficial to the intelligent optimization and upgrading of the power industry.

[0026] The modeling method for the electric power business decision model provided in this specification can be applied to construct decision models for business problems in the electric power sector, including but not limited to classic electric power business decision problems such as power dispatching, unit commitment, power flow numerical calculation, power generation control, and photovoltaic capacity. It is understood that the modeling method for the electric power business decision model provided in this specification, after adaptive modification, can also be used to construct decision models for business problems in other fields, including but not limited to transportation, manufacturing production, and financial management.

[0027] According to an embodiment of the present application, an embodiment of a modeling method for an electric power business decision model is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] In this embodiment, a modeling method for a power business decision model is provided, which can be used to build a decision model for business problems in the power field. Figure 1 As shown, the method includes: S100. Build a power corpus segmentation library and a power vector knowledge base, and determine the power business problem text.

[0029] S200. With the help of the power corpus segmentation library and the power vector knowledge base, a hybrid search is performed on the power business question text to obtain a first similarity score corresponding to the candidate corpus segmentation 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 segmentation in the power corpus segmentation library.

[0030] S300. Screen candidate corpus segments and candidate document representation vectors based on the first similarity score and the second similarity score, and determine a target reference text based on the screening result.

[0031] 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 target function and the constraints of the target function.

[0032] The power corpus segmentation database and the power vector knowledge base can be constructed based on knowledge related to the core areas of the power business, and used for hybrid searches based on power business question texts to determine target reference texts. The power corpus segmentation database can be constructed based on corpus segments of knowledge related to the core areas of the power business. The power corpus segmentation database stores corpus segments of relevant knowledge for use in searching power business question texts based on these corpus segments. Corpus segments can be obtained through semantic understanding and semantic segmentation of knowledge related to the core areas of the power business, with each corpus segment corresponding to a meaningful semantic unit. The power vector knowledge base can be constructed based on representation vectors of corpus segments of relevant knowledge. The power vector knowledge base stores representation vectors of corpus segments for use in searching power business question texts based on these representation vectors. The representation vectors can be obtained by vectorizing the power corpus segments in the power corpus segmentation database. Vectorization enables a deeper understanding of the relevant knowledge from the perspective of the representation vectors, providing a foundation for subsequent hybrid searches. It can be understood that the document representation vectors in the electric power vector knowledge base are obtained by vectorizing the electric power corpus segments in the electric power corpus segment library.

[0033] The power business question text may be text data representing a power business question, obtained based on a user inputting the power business question to be queried. The power business question may be a business problem 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 commitment, power flow numerical calculation, power generation control, and photovoltaic capacity. It is understood that by performing text embedding on the power business question text, a representation vector corresponding to the power business question text is obtained, which can be used for retrieval in the power vector knowledge base.

[0034] Specifically, based on the key business needs and technical pain points of the power industry, relevant knowledge from multiple core power business areas is collected and processed in text form to construct a power corpus and power vector knowledge base. In response to user input of a power business question, the text of the power business question is obtained and embedded into the text to obtain a representation vector corresponding to the power business question text.

[0035] Reference Figure 2As shown, a hybrid search is performed in the electric power business question text and its representation vector in the electric power corpus segmentation library and the electric power vector knowledge base, and based on the search results in multiple libraries, a first similarity score corresponding to the candidate corpus segmentation and a second similarity score corresponding to the candidate document representation vector are obtained. It should be noted that the hybrid search includes two or more search methods, wherein at least one search method is adopted in the electric power corpus segmentation library, and at least one search method is also adopted in the electric power vector knowledge base. According to the search method adopted in the electric power corpus segmentation library, multiple segmentation similarity scores corresponding to the candidate corpus segmentation are obtained respectively, and by fusing the above multiple segmentation similarity scores, the first similarity score corresponding to the candidate corpus segmentation is obtained. Similarly, according to the search method adopted in the electric power vector knowledge base, multiple vector similarity scores corresponding to the candidate document representation vector are obtained respectively, and by fusing the above multiple vector similarity scores, the second similarity score corresponding to the candidate document representation vector is obtained.

[0036] It should be noted that when searching for power business problem texts in the power corpus segmentation library, the power business problem texts can be understood from the perspective of corpus segmentation and searched by the similarity of text content, thereby intuitively and quickly obtaining candidate corpus segments related to the power business problem texts. When searching for the representation vectors corresponding to power business problem texts in the power vector knowledge base, the power business problem texts can be understood from the perspective of representation vectors and searched by the similarity of text semantics, thereby effectively capturing the deep semantic similarity implicit in the texts and improving the accuracy of the target reference texts. By hybrid retrieval and comprehensively considering the first similarity score and the second similarity score, the advantages of multiple retrieval methods are combined, not only ensuring the retrieval efficiency and retrieval scope of hybrid retrieval, but also improving the accuracy of retrieval results, significantly improving the retrieval quality of hybrid retrieval.

[0037] Furthermore, since the document representation vectors in the electric power vector knowledge base are obtained by vectorizing the electric power corpus segments in the electric power corpus segmentation library, the candidate document representation vectors obtained based on the electric power vector knowledge base correspond to the candidate corpus segments obtained based on the electric power corpus segmentation library, that is, the first similarity score and the second similarity score both correspond to the same knowledge text content. Taking the first similarity score and the second similarity score into comprehensive consideration, similarity matching screening is performed based on the candidate corpus segments and candidate document representation vectors corresponding to the same knowledge text content to obtain the target reference text related to the electric power business question text. Exemplarily, the way to fuse the first similarity score and the second similarity score can be weighted summation, and the knowledge text content whose similarity score meets the similarity requirement is selected as the target reference text. The method for selecting the target reference text can be the top-k method, which selects the k knowledge text contents with the highest degree of matching with the electric power business question text as the target reference text.

[0038] Furthermore, the power business problem text and the obtained target reference text are jointly input into the target large language model, and the target large language model is 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 for solving power business problems.

[0039] It should be noted that the target large language model can generate and output a target decision model based on the output template prompt words. The output template prompt words are used to normalize the output of the target large language model to ensure that the resulting target decision model has logical consistency and problem correspondence. It is understandable that the output template prompt words include the description text of the power business problem, the objective function and its constraints. The description text of the power business problem is used to describe the application scenarios and application requirements of the power business problem. The objective function and its constraints can be used to solve the power business problem through mathematical tools, providing users with solutions or decision-making basis.

[0040] The modeling method of the electric power business decision model provided in this embodiment performs a mixed search in the electric power corpus segmentation library and the electric power vector knowledge base based on the electric power business problem text, obtains a first similarity score corresponding to the candidate corpus segmentation and a second similarity score corresponding to the candidate document representation vector, and screens the target reference text according to the first similarity score and the second similarity score to obtain the target reference text, which is used as the input of the target large language model to construct a target decision model corresponding to the electric power business problem text.

[0041] Compared with related technologies, this application performs hybrid retrieval of question texts based on corpus segmentation and representation vectors to understand the semantics of question texts from different perspectives, and obtains similarity scores that represent the degree of matching of different semantic understanding results. This can integrate the advantages of multiple retrieval methods and effectively improve the accuracy of the target reference text, which in turn is beneficial to the understanding and reasoning of the question text by the target large language model, and improves the reliability and accuracy of the obtained target decision model for solving power business problems. In addition, this application realizes the automation of semantic understanding and mathematical modeling processes through the standardized output of hybrid retrieval and large language models, improves the efficiency of building decision models, promotes the promotion and application of large language models in the power industry, and is beneficial to the intelligent optimization and upgrading of the power industry.

[0042] Reference Figure 3 As shown, as an embodiment of the present application, a hybrid search is performed on the power business question text with the help of the power corpus segmentation library and the power vector knowledge base to obtain a first similarity score corresponding to the candidate corpus segmentation and a second similarity score corresponding to the candidate document representation vector, including: S210. Perform a sparse search in the electric power corpus segmentation library based on the electric power business question text to obtain multiple candidate corpus segments and a sparse similarity score corresponding to each candidate corpus segment as a first similarity score.

[0043] S220. Perform a dense search in the power vector knowledge base based on the power business question text to obtain multiple candidate document representation vectors and a dense similarity score corresponding to each candidate document representation vector as a second similarity score.

[0044] Reference Figure 4 As shown, in this embodiment, after text embedding is performed on the power business question text to obtain the corresponding representation vector, a sparse search is performed on the power business question text to obtain the first similarity score of the candidate corpus segment. A dense search is then performed on the representation vector corresponding to the power business question text to obtain the second similarity score of the candidate document representation vector. Finally, the first and second similarity scores are fused to determine the target reference text that matches the power business question text.

[0045] Specifically, sparse retrieval is used to search for power business question texts in the power corpus segmentation library based on the similarity of text content, so as to understand the power business question texts from the perspective of corpus segmentation. The steps of sparse retrieval may include: performing word segmentation preprocessing on the power business question text to obtain multiple question subwords; calculating the correlation score between each question subword and the words in each corpus segmentation in the power corpus segmentation library, determining the degree of match between the power business question text and each corpus segmentation, and obtaining multiple candidate corpus segments and their corresponding sparse similarity scores as the first similarity score. Exemplarily, the algorithm used for sparse retrieval may be the BM25 algorithm, which performs retrieval based on the idea of ​​word frequency-inverse document frequency.

[0046] It's important to note that sparse search primarily relies on word similarity, making it particularly effective for specific terms and proper nouns, enabling precise matches. Furthermore, sparse search relies solely on word frequency statistics, effectively reducing search time and improving the efficiency of hybrid searches.

[0047] Furthermore, dense search is used to search the power vector knowledge base for representation vectors corresponding to the power business question text based on text semantic similarity, thereby understanding the power business question text from the perspective of the representation vector. The dense search step may include: calculating the semantic similarity between the document representation vector and the representation vector corresponding to the power business question text based on the document representation vector and the representation vector corresponding to the power business question text, obtaining multiple candidate document representation vectors and a dense similarity score corresponding to each candidate document representation vector as a second similarity score.

[0048] It should be noted that dense retrieval is mainly based on the similarity between representation vectors. It can understand the overall semantics of the power business problem text, has strong semantic generalization, and can capture synonyms, near-synonyms and complex semantic relationships. It can show better robustness in complex query processes, effectively improve the accuracy of retrieval results, and enhance the retrieval quality of hybrid retrieval.

[0049] Reference Figure 5 As shown, as an embodiment of the present application, the electric power corpus segmentation library contains target meta-information corresponding to the target reference text; the electric power business question text and the target reference text are input into the target large language model to obtain the target decision model corresponding to the electric power business question text, including: S410. Use the target reference text to enhance the power business question text to obtain enhanced question prompt words.

[0050] S420. Input the enhanced question prompt words and target meta-information into the target large language model for collaborative reasoning to obtain the target decision model.

[0051] Among them, the target meta information can be the hierarchical key points of the target reference text, including but not limited to the 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 refined, which is conducive to the target large language model to understand the target reference text.

[0052] Reference Figure 6 As shown, the target reference text is the result of a hybrid search in the power corpus and the power vector knowledge base based on the power business question text. 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 with the target reference text may include: contextually combining the target reference text with the power business question text to semantically expand the content in the power business question text using the target reference text to further explain the power business question text; splicing the target reference text to a specific part of the power business question text to supplement the missing content in the power business question text using the target reference text to improve the comprehensiveness and accuracy of the power business question text; inserting the target reference text as a keyword into the power business question text to improve the semantic accuracy of the power business question text, which is conducive to the target large language model's understanding of the power business question text; and obtaining enhanced question prompt words by enhancing the power business question text.

[0053] Furthermore, the enhanced question prompt words and target meta-information are input into the target large language model. The target large language model is used to perform semantic understanding of the enhanced question prompt words based on the target meta-information, and collaborative reasoning is performed based on the understood content to construct a target decision model corresponding to the power business question text.

[0054] It should be noted that the target large language model can generate and output a target decision model based on the output template prompt words. The output template prompt words are used to normalize the output of the target large language model to ensure that the resulting target decision model has logical consistency and problem correspondence. It is understandable that the output template prompt words include the description text of the power business problem, the objective function and its constraints. The description text of the power business problem is used to describe the application scenarios and application requirements of the power business problem, and the objective function and its constraints can be used to solve the power business problem and provide users with solutions or decision-making basis.

[0055] Reference Figure 7 As shown, as an embodiment of the present application, the power corpus segments are obtained by the following method: S110. Convert the initial document data to obtain document data in a target format.

[0056] S120. Clean and verify the target format document data to obtain standardized document data.

[0057] S130. Adaptively segment the standardized document data to obtain power corpus segments.

[0058] Specifically, relevant knowledge in the core areas of the power industry can be in document form, and initial document data can be obtained based on this document-based knowledge. Initial document data can be obtained from authoritative academic databases, professional journal libraries, and university dissertation resource libraries. This includes, but is not limited to, power industry e-books, journal articles, master's and doctoral dissertations, and other relevant literature, covering key business needs and technical pain points in the power industry.

[0059] Reference Figure 8 As shown, the initial document data may have inconsistent formats. After obtaining the initial document data, data conversion is performed on the initial document data to convert the document format of the initial document data to the same format to obtain the target format document data. For example, the document format of the initial document data can be PDF, Word, or CAJ, and the document format of the target format document data can be Markdown.

[0060] Furthermore, the target format document data may contain redundant information that is irrelevant to the power business problem, and this redundant information may interfere with the large language model's understanding of the power business problem text. Therefore, it is necessary to clean the target format document data to remove redundant information and retain the important core information in the target format document data to improve the pertinence of the power corpus segmentation. It is understandable that during the cleaning process, core information may be removed or the data content of the target format document data may be greatly affected. At this time, the cleaning results need to be verified to ensure that the format of the cleaned target format document data is unified and there are no data content errors and logical problems, to ensure the data quality of the target format document data, and thus obtain standardized document data.

[0061] Furthermore, metadata is extracted from the standardized document data to obtain document metadata. The document metadata may be the hierarchical key points of the standardized document data, including but not limited to the chapter titles or core keywords in the document, so as to facilitate the grasp of the structure and hierarchy of the standardized document data. The document metadata is spliced ​​with the standardized document data to utilize the document metadata to perform structural division and semantic generalization on the standardized document data, thereby enabling the spliced ​​standardized document data to be adaptively segmented to obtain a plurality of power corpus segments. It is understandable that the content in each power corpus segment targets the same topic and has high pertinence and logic. By storing the power corpus segments in the knowledge base to form a power corpus segment library, the power business knowledge can be structuredly managed, which facilitates the updating and maintenance of the power business knowledge and improves the accuracy of hybrid retrieval.

[0062] Reference Figure 9 As shown, as an embodiment of the present application, the initial document data is converted to obtain the target format document data, including: S112. Use a first type conversion tool to perform data conversion on the initial document data to obtain a first conversion result.

[0063] S114. Use a second type conversion tool to perform data conversion on the initial document data to obtain a second conversion result.

[0064] S116. Perform collaborative complementation based on the first conversion result and the second conversion result to obtain target format document data.

[0065] Specifically, the initial document data may include first-type document data and second-type document data. The first-type document data may be large documents with a large amount of text content, such as master's or doctoral dissertations or e-books, while the second-type document data may be small documents with less text content, such as journals and conference papers. Given such a large amount of initial document data, using only a single type conversion tool may not achieve both efficient and accurate conversion.

[0066] Based on the above reasons, this embodiment adopts different first-type conversion tools and second-type conversion tools. The first-type conversion tool is used to perform data conversion on the initial document data to obtain a first conversion result, and the second-type conversion tool is used to perform data conversion on the initial document data to obtain a second conversion result. Exemplarily, the first-type conversion tool can be the Doubao web version, which has high accuracy and fidelity when processing large documents, but is prone to over-summarization problems and has low efficiency in processing document data. The second-type conversion tool can be MinerU software, which has batch processing capabilities and can quickly convert multiple document data, but it has the problem of a high formula conversion error rate.

[0067] Furthermore, after obtaining the first conversion result and the second conversion result, the target format document data is obtained by performing synergistic complementation based on the first conversion result and the second conversion result, so as to respectively utilize the respective advantages of the first type conversion tool and the second type conversion tool to improve the conversion quality of the target format document data. Specifically, the synergistic complementation process can be: comparing the first conversion result and the second conversion result, determining the document compression data in the first conversion result that over-compresses the initial document data, and the formula error data in the second conversion result that has a formula conversion error; determining the 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 the 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 based on the first synergistic result and the second synergistic result.

[0068] It should be noted that replacing the document compression data in the first conversion result with document conversion data can supplement the document data in the first conversion result that was over-summarized by the first type conversion tool, thereby improving the integrity of the first collaborative result. Replacing the formula error data in the second conversion result with formula conversion data can correct the formula content in the second conversion result that was caused by the second type conversion tool, thereby improving the accuracy of the second collaborative result. The target format document data is obtained based on the first collaborative result and the second collaborative result, thereby combining the respective advantages of the first type conversion tool and the second type conversion tool to improve the conversion quality of the target format document data.

[0069] Reference Figure 10 As shown, as an embodiment of the present application, the target format document data is cleaned and verified to obtain standardized document data, including: 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 constraint conditions.

[0070] S124. In response to the verification operation on the intermediate document data, standardized document data is obtained.

[0071] Specifically, redundant information irrelevant to the power business problem in the target format document data is removed by cleaning, and the preset core knowledge elements therein are retained. The preset core knowledge elements include problem description, objective function and constraints, where the problem description can be the descriptive text of the business problem to be solved in the target format document data, which is used to help the large language model understand the business problems in the relevant knowledge. The objective function can be the mathematical tool used by the target format document data to solve business problems, which is used to help the large language model learn the corresponding tool to solve the business. Constraints can be used to constrain the objective function in the target format document data to match the actual scenario corresponding to the power business problem text, thereby improving the accuracy and reliability of the target decision model. By retaining the preset core knowledge elements, the information quality of the standardized document data is improved, and the model building capability of the target large language model is enhanced.

[0072] It is understandable that during the cleaning process, the preset core knowledge elements may be removed or the data content of the target format document data may be greatly affected. At this time, the cleaning results need to be verified to ensure that the cleaned target format document retains the preset core knowledge elements, and the data format is unified, without content errors and logical problems, to ensure the data quality of the target format document data, and thus obtain standardized document data.

[0073] Reference Figure 11As shown, as an embodiment of the present application, the power vector knowledge base is obtained by the following method: S140. Convert the initial document data to obtain document data in a target format.

[0074] S150. Perform data cleaning and data verification on the target format document data to obtain standardized document data.

[0075] S160. Adaptively segment the standardized document data to obtain power corpus segments.

[0076] S170. Cut the electric power corpus into blocks and perform text embedding to obtain text representation vectors, and build an electric power vector knowledge base based on the text representation vectors.

[0077] Specifically, relevant knowledge in the core areas of the power industry can be in document form, and initial document data can be obtained based on this document-based knowledge. Initial document data can be obtained from authoritative academic databases, professional journal libraries, and university dissertation resource libraries. This includes, but is not limited to, power industry e-books, journal articles, master's and doctoral dissertations, and other relevant literature, covering key business needs and technical pain points in the power industry.

[0078] Reference Figure 12 As shown, the initial document data may have inconsistent formats. After obtaining the initial document data, data conversion is performed on the initial document data to convert the document format of the initial document data to the same format to obtain the target format document data. For example, the document format of the initial document data can be PDF, Word, or CAJ, and the document format of the target format document data can be Markdown.

[0079] Furthermore, the target format document data may contain redundant information that is irrelevant to the power business problem, and this redundant information may interfere with the large language model's understanding of the power business problem text. Therefore, it is necessary to clean the target format document data to remove redundant information and retain the important core information in the target format document data to improve the pertinence of the power corpus segmentation. It is understandable that during the cleaning process, core information may be removed or the data content of the target format document data may be greatly affected. At this time, the cleaning results need to be verified to ensure that the format of the cleaned target format document data is unified and there are no data content errors and logical problems, to ensure the data quality of the target format document data, and thus obtain standardized document data.

[0080] Furthermore, metadata is extracted from the standardized document data to obtain document metadata. The document metadata may be the hierarchical key points of the standardized document data, including but not limited to the chapter titles or core keywords in the document, so as to facilitate the grasp of the structure and hierarchy of the standardized document data. The document metadata is spliced ​​with the standardized document data to utilize the document metadata to perform structural division and semantic generalization on the standardized document data, thereby enabling the spliced ​​standardized document data to be adaptively segmented to obtain a plurality of power corpus segments. It is understandable that the content in each power corpus segment targets the same topic and has high pertinence and logic. By storing the power corpus segments in the knowledge base to form a power corpus segment library, the power business knowledge can be structuredly managed, which facilitates the updating and maintenance of the power business knowledge and improves the accuracy of hybrid retrieval.

[0081] Furthermore, the text content in the power corpus segments is mapped to a vector space, and a representation vector corresponding to the power corpus segments is generated based on the mapping result to obtain a text representation vector. The text representation vector is stored in a vector library, thereby constructing a power vector knowledge base. Exemplarily, 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 vector and also store it in the power vector knowledge base, thereby improving the efficiency of searching for document representation vectors.

[0082] The following example describes an implementation scenario for this method. To evaluate this method, this example collects and cleans 17 questions from the "Economic Operation Theory of Power Systems," covering classic power scenarios such as power dispatch, unit commitment, and power flow calculation. The decision model corresponding to each question serves as the standard answer, forming an evaluation dataset. This example uses the Retrieval Augmented Generation Assessment (RAGAs) framework for quantitative evaluation, employing the external large language model Qwen-Plus as the model referee and focusing on both retrieval quality and generation quality indicators.

[0083] Retrieval quality is primarily measured in three aspects: the relevance of the context to the query, the relevance of the context to the ground truth, and the relevance of the context to the model output. The main metric for measuring the relevance of context to the query is Context Relevance (CR). Low relevance indicates that the retriever has introduced noise irrelevant to the query. The main metrics for measuring the relevance of context to the ground truth include LLM Context Recall (LCR) and Context Precision (CP). LCR measures whether the retrieved text contains all the information necessary to generate the ground truth answer. Low recall means that the retriever has omitted key information. CP measures whether the retrieved text is semantically and accurately related to the ground truth answer. Low precision indicates that the retrieved context is poorly semantically related to the ground truth answer, meaning that the context lacks valid information. Metrics for measuring the relevance of context to the model output include Faithfulness (FF) and Factual Correctness (FC). These metrics assess whether the model-generated answer is based on the retrieved text and whether any fabricated information exacerbates the illusion.

[0084] Generation quality is primarily measured by the relevance of the model output to the query and the relevance of the model output to the ground truth answer. Regarding the relevance of the model output to the query, the AnswerRelevancy (AR) metric can be used to determine whether the answer directly addresses the question and is not irrelevant. Regarding the relevance of the model output to the ground truth answer, the AnswerCorrectness (AC) and AnswerSimilarity (AS) metrics can be used to assess the accuracy of the model's answer and its relevance to the standard answer.

[0085] In this example, the DeepSeek V3 model was quantitatively compared with this method. The performance of three search methods, dense, sparse, and hybrid, was also compared. The comparison results are shown in Table 1, where the numerical values ​​represent the degree of improvement of the corresponding method on the quantitative indicator. It should be noted that the target reference text corresponding to the DeepSeek V3 model was set to empty.

[0086] Table 1 Comparison results of quantitative indicators between this method and DeepSeek V3 model Table 1 shows that compared to the DeepSeek V3 model, this method significantly improves on the three generation quality metrics of AR, AC, and AS, demonstrating a significant improvement in decision model generation compared to conventional techniques. Furthermore, for power business questions, hybrid search yields reference texts that are more relevant to the query and more similar to the standard answer than dense or sparse search alone, resulting in the best generation quality.

[0087] Accordingly, please refer to Figure 13 , an embodiment of the present application provides a modeling device for an electric power business decision model, the device comprising: The question text acquisition module 1310 is used to construct an electric power corpus segmentation library and an electric power vector knowledge base, and determine the electric power business question text.

[0088] The hybrid retrieval scoring module 1320 is used to perform a hybrid search on the power business question text with the help of the power corpus segmentation library and the power vector knowledge base, and obtain a first similarity score corresponding to the candidate corpus segmentation 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 segmentation in the power corpus segmentation library.

[0089] The reference text screening module 1330 is configured to screen candidate corpus segments and candidate document representation vectors based on the first similarity score and the second similarity score, and determine a target reference text based on the screening result.

[0090] 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 target function and the constraints of the target function.

[0091] In some optional implementations, the hybrid search scoring module 1320 includes: The corpus segmentation sparse retrieval unit is used to perform sparse retrieval in the power corpus segmentation library based on the power business question text, and obtain multiple candidate corpus segments and a sparse similarity score corresponding to each candidate corpus segmentation as a first similarity score.

[0092] The representation vector dense retrieval unit is used to perform dense retrieval in the power vector knowledge base based on the power business question text, and obtain multiple candidate document representation vectors and a dense similarity score corresponding to each candidate document representation vector as a second similarity score.

[0093] In some optional implementations, the decision model building module 1340 includes: The question text enhancement unit is used to enhance the power business question text using the target reference text to obtain enhanced question prompt words.

[0094] The collaborative reasoning unit is used to input the enhanced question prompt words and target meta-information into the target large language model for collaborative reasoning to obtain the target decision model.

[0095] In some optional implementations, the question text acquisition module 1310 includes: The document data conversion unit is used to convert the initial document data to obtain document data in a target format.

[0096] The document cleaning and verification unit is used to clean and verify the target format document data to obtain standardized document data.

[0097] The document adaptive segmentation unit is used to adaptively segment the standardized document data to obtain power corpus segments.

[0098] In some optional implementations, the document data conversion unit includes: The first data conversion subunit is configured to perform data conversion on the initial document data using a first type conversion tool to obtain a first conversion result.

[0099] The second data conversion subunit is configured to perform data conversion on the initial document data using a second type conversion tool to obtain a second conversion result.

[0100] The data collaborative complementation subunit is used to perform collaborative complementation based on the first conversion result and the second conversion result to obtain document data in a target format.

[0101] In some optional implementations, the document cleaning and verification unit includes: 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; wherein the preset core knowledge elements include problem description, objective function and constraint conditions.

[0102] The data verification subunit is configured to obtain standardized document data in response to a verification operation on the intermediate document data.

[0103] In some optional implementations, the question text acquisition module 1310 further includes: The document data conversion unit is used to convert the initial document data to obtain document data in a target format.

[0104] The document cleaning and verification unit is used to clean and verify the target format document data to obtain standardized document data.

[0105] The document adaptive segmentation unit is used to adaptively segment the standardized document data to obtain power corpus segments.

[0106] The block text embedding unit is used to block the power corpus for text embedding, obtain text representation vectors, and build a power vector knowledge base based on the text representation vectors.

[0107] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0108] The modeling device of the power business decision model in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0109] See also Figure 14 , Figure 14 1 is a structural diagram of a computer device provided by an embodiment of the present application. As shown in the figure, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are connected to each other using different buses for communication and can be installed on a common mainboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to an interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 14 A processor 10 is taken as an example.

[0110] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0111] The memory 20 stores instructions that can be executed by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0112] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0113] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0114] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0115] The embodiments of the present application also provide a computer-readable storage medium. The above-mentioned method according to the embodiment of the present application can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0116] An embodiment of the present application provides a computer program product, which includes 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 a method according to any embodiment of the present application.

[0117] Although the embodiments of the present application have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations shall fall within the scope defined by the appended claims.

[0118] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0119] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0120] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0121] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0122] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1The function specified in one or more boxes.

[0123] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0124] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0125] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0126] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

[0127] Although the embodiments of the present application have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations shall fall within the scope defined by the appended claims.

Claims

1. A modeling method for an electric power business decision model, characterized in that: The method comprises: Build a power corpus segmentation database and power vector knowledge base, and determine the power business problem text; A hybrid search is performed on the power business question text using the power corpus chunk library and the power vector knowledge base to obtain a first similarity score corresponding to the candidate corpus chunk 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 chunk in the power corpus chunk library; screening the candidate corpus segments and the candidate document representation vectors based on the first similarity score and the second similarity score, and determining a target reference text based on the screening result; The power business problem text and the target reference text are input into the target large language model to obtain a target decision model corresponding to the power business problem text; wherein the target decision model includes a description text of the power business problem, an objective function and constraints of the objective function.

2. The method according to claim 1, characterized in that The hybrid search of the power business question text by means of the power corpus segmentation library and the power vector knowledge base to obtain a first similarity score corresponding to the candidate corpus segmentation and a second similarity score corresponding to the candidate document representation vector includes: Performing a sparse search in the power corpus chunk library based on the power business question text to obtain a plurality of candidate corpus chunks and a sparse similarity score corresponding to each candidate corpus chunk as the first similarity score; Based on the power business question text, a dense search is performed in the power vector knowledge base 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 according to claim 1, characterized in that The electric power corpus segmentation library contains target meta-information corresponding to the target reference text; The step of inputting the electric power business question text and the target reference text into a target large language model to obtain a target decision model corresponding to the electric power business question text includes: Using the target reference text to enhance the power business question text to obtain enhanced question prompt words; The enhanced question 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 according to claim 1, wherein The power corpus segments are obtained in the following way: Convert the initial document data to obtain target format document data; Cleaning and verifying the target format document data to obtain standardized document data; Adaptively segment the standardized document data to obtain the power corpus segments.

5. The method according to claim 4, characterized in that The step of converting the initial document data to obtain the target format document data includes: Performing data conversion on the initial document data using a first type conversion tool to obtain a first conversion result; Performing data conversion on the initial document data using a second type conversion tool to obtain a second conversion result; The target format document data is obtained based on the collaborative complementation of the first conversion result and the second conversion result.

6. The method according to claim 4, characterized in that The step of cleaning and verifying the target format document data to obtain standardized document data includes: Cleaning the target format document data, retaining preset core knowledge elements, and obtaining intermediate document data; wherein the preset core knowledge elements include problem description, objective function, and constraint conditions; In response to a verification operation on the intermediate document data, the standardized document data is obtained.

7. The method according to claim 1, characterized in that The power vector knowledge base is obtained by the following method: Convert the initial document data to obtain target format document data; Performing data cleaning and data verification on the target format document data to obtain standardized document data; Adaptively segmenting the standardized document data to obtain power corpus segments; The electric power corpus is cut into pieces and text embedded to obtain text representation vectors, and the electric power vector knowledge base is constructed based on the text representation vectors.

8. A modeling device for an electric power business decision model, characterized in that: The device comprises: The question text acquisition module is used to build the power corpus segmentation library and power vector knowledge base, and determine the power business question text; A hybrid search and scoring module is configured to perform a hybrid search on the power business question text using the power corpus chunk library and the power vector knowledge base to obtain a first similarity score corresponding to the candidate corpus chunk 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 chunk in the power corpus chunk library; a reference text screening module, configured to screen the candidate corpus segments and the candidate document representation vectors based on the first similarity score and the second similarity score, and determine a target reference text based on the screening result; A 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 a target decision model corresponding to the power business problem text; wherein, the target decision model includes a description text of the power business problem, an objective function and the constraints of the objective function.

9. A computer device, characterized in that: include: 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 method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images