Method, device and equipment for constructing power line loss analysis dataset based on large model

Abstract

The application relates to a power line loss analysis dataset construction method, device and equipment based on a large model. The method comprises the following steps: obtaining line loss related data derived by each power system in a power distribution network, inputting the line loss related data into a pre-constructed first-level multi-modal large model, generating a summary synthesis prompt word based on a structure analysis prompt word in the first-level multi-modal large model and the line loss related data, obtaining a data description summary corresponding to the line loss related data based on the summary synthesis prompt word, retrieving a power line loss domain knowledge base based on the data description summary, obtaining line loss domain knowledge corresponding to the line loss related data, and then generating a line loss related data extraction instruction according to the line loss domain knowledge, inputting the line loss related data extraction instruction and the line loss related data into a pre-constructed second-level multi-modal large model, and obtaining a power line loss analysis dataset. The method improves the accuracy of the finally constructed dataset.

Description

Technical Field

[0001] This application relates to the field of power distribution network line loss analysis technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for constructing a power line loss analysis dataset based on a large model. Background Technology

[0002] The core challenge in building a high-quality dataset for theoretical line loss analysis of distribution networks lies in how to automatically process and integrate multi-source heterogeneous tabular data from different business systems.

[0003] Currently, there are two main ways to construct relevant datasets: one is manual work that relies entirely on expert experience, where data is stitched together through manual comparison and operation; the other is a dedicated script based on hard-coded rules, where fixed programs are written for specific data sources to achieve automated extraction.

[0004] However, both of these current methods suffer from insufficient accuracy in the constructed datasets. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for constructing a power line loss analysis dataset based on a large model, which can improve the accuracy of the constructed dataset, in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a method for constructing a power line loss analysis dataset based on a large model, including:

[0007] The system acquires line loss-related data exported from each power system in the distribution network and inputs the line loss-related data into a pre-constructed first-level multimodal large model. Based on the structural parsing prompts in the first-level multimodal large model, the system parses the line loss-related data to obtain the metadata of the line loss-related data. Based on the metadata, the system generates summary synthesis prompts corresponding to the line loss-related data and obtains the data description summary corresponding to the line loss-related data based on the summary synthesis prompts.

[0008] Based on the data description summary, the knowledge base in the field of power line loss is retrieved to obtain the standardized field definitions, historical alias sets and business rule fragments corresponding to line loss related data.

[0009] Based on standardized field definitions, historical alias sets, and business rule fragments, generate instructions for extracting line loss-related data.

[0010] The command to extract line loss-related data and the line loss-related data are input into a pre-built second-level multimodal large model, and the power line loss analysis dataset is output through the second-level multimodal large model.

[0011] In one embodiment, a data description summary corresponding to line loss-related data is obtained based on summary-synthesized prompt words, including:

[0012] Re-input the summary synthesis prompts into the first-level multimodal large model to obtain the original data description summary corresponding to the line loss related data;

[0013] The original data description summary is filled into the pre-configured classification and discrimination prompt word template to obtain the data description summary corresponding to the line loss related data.

[0014] In one embodiment, a data extraction instruction related to line loss is generated based on standardized field definitions, a set of historical aliases, and fragments of business rules, including:

[0015] Obtain a pre-configured machine-readable structured instruction template;

[0016] The standardized field definitions, historical alias sets, and business rule fragments are filled into the machine-readable structured instruction template to obtain the line loss related data extraction instruction.

[0017] In an exemplary embodiment, the line loss related data extraction instruction and the line loss related data are input into a pre-constructed second-level multimodal large model, and the second-level multimodal large model outputs a power line loss analysis dataset, including:

[0018] Perform semantic understanding on line loss-related data to obtain semantically understood line loss-related data;

[0019] The semantically understood line loss-related data is matched with the standardized fields in the line loss-related data extraction instruction, and the corresponding standardized data is extracted from the semantically understood line loss-related data in JSON format.

[0020] By combining the standardized data, a power line loss analysis dataset is obtained.

[0021] In one embodiment, the power line loss knowledge base includes a business type definition table, a standard field master table, and a historical mapping relationship table; the historical mapping relationship table is used to store the mapping relationship from non-standard line loss descriptions to standard line loss descriptions; the data description summary includes the line loss business type corresponding to the line loss related data;

[0022] Based on the data description summary, a knowledge base for the power line loss domain is retrieved to obtain the corresponding standardized field definitions, historical alias sets, and business rule fragments, including:

[0023] By retrieving the business type definition table using the line loss business type, the core standard field ID corresponding to the line loss business type can be obtained;

[0024] Retrieve the main table of standard fields based on the core standard field ID to obtain the core standard field corresponding to the core standard field ID;

[0025] Search the historical mapping table according to the data description summary to obtain the mapping relationship corresponding to the line loss related data;

[0026] Based on the mapping relationship and core standard fields, we obtain the standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss related data.

[0027] In one embodiment, a historical mapping table is retrieved according to the data description summary to obtain the mapping relationship corresponding to the line loss related data, including:

[0028] Construct the corresponding semantic retrieval query vector based on the data description summary;

[0029] The pre-built vector database is retrieved according to the semantic retrieval query vector to obtain the non-standard line loss description corresponding to the semantic retrieval query vector; the vector database is used to store the non-standard line loss descriptions in the historical mapping relationship table;

[0030] The corresponding mapping relationship is determined from the historical mapping relationship table based on the non-standard line loss description.

[0031] In an exemplary embodiment, based on the mapping relationship and core standard fields, standardized field definitions, historical alias sets, and business rule fragments corresponding to line loss-related data are obtained, including:

[0032] Map the core standard fields according to the mapping relationship;

[0033] If the mapping is successful, the standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss-related data are extracted from the standard field master table based on the core standard fields.

[0034] Secondly, this application also provides a device for constructing a power line loss analysis dataset based on a large model, including:

[0035] The acquisition module is used to acquire line loss related data exported from various power systems in the distribution network, input the line loss related data into a pre-constructed first-level multimodal large model, parse the line loss related data based on the structural parsing prompt words in the first-level multimodal large model, obtain the metadata of the line loss related data, and generate summary synthesis prompt words corresponding to the line loss related data based on the metadata, and obtain the data description summary corresponding to the line loss related data based on the summary synthesis prompt words;

[0036] The retrieval module is used to retrieve the knowledge base in the field of power line loss based on the data description summary, and obtain the standardized field definitions, historical alias sets and business rule fragments corresponding to the line loss related data;

[0037] The generation module is used to generate data extraction instructions related to line loss based on standardized field definitions, historical alias sets, and business rule fragments.

[0038] The module is used to input line loss related data extraction instructions and line loss related data into a pre-built second-level multimodal large model, and output a power line loss analysis dataset through the second-level multimodal large model.

[0039] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0040] The system acquires line loss-related data exported from each power system in the distribution network and inputs the line loss-related data into a pre-constructed first-level multimodal large model. Based on the structural parsing prompts in the first-level multimodal large model, the system parses the line loss-related data to obtain the metadata of the line loss-related data. Based on the metadata, the system generates summary synthesis prompts corresponding to the line loss-related data and obtains the data description summary corresponding to the line loss-related data based on the summary synthesis prompts.

[0041] Based on the data description summary, the knowledge base in the field of power line loss is retrieved to obtain the standardized field definitions, historical alias sets and business rule fragments corresponding to line loss related data.

[0042] Based on standardized field definitions, historical alias sets, and business rule fragments, generate instructions for extracting line loss-related data.

[0043] The command to extract line loss-related data and the line loss-related data are input into a pre-built second-level multimodal large model, and the power line loss analysis dataset is output through the second-level multimodal large model.

[0044] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0045] The system acquires line loss-related data exported from each power system in the distribution network and inputs the line loss-related data into a pre-constructed first-level multimodal large model. Based on the structural parsing prompts in the first-level multimodal large model, the system parses the line loss-related data to obtain the metadata of the line loss-related data. Based on the metadata, the system generates summary synthesis prompts corresponding to the line loss-related data and obtains the data description summary corresponding to the line loss-related data based on the summary synthesis prompts.

[0046] Based on the data description summary, the knowledge base in the field of power line loss is retrieved to obtain the standardized field definitions, historical alias sets and business rule fragments corresponding to line loss related data.

[0047] Based on standardized field definitions, historical alias sets, and business rule fragments, generate instructions for extracting line loss-related data.

[0048] The command to extract line loss-related data and the line loss-related data are input into a pre-built second-level multimodal large model, and the power line loss analysis dataset is output through the second-level multimodal large model.

[0049] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0050] The system acquires line loss-related data exported from each power system in the distribution network and inputs the line loss-related data into a pre-constructed first-level multimodal large model. Based on the structural parsing prompts in the first-level multimodal large model, the system parses the line loss-related data to obtain the metadata of the line loss-related data. Based on the metadata, the system generates summary synthesis prompts corresponding to the line loss-related data and obtains the data description summary corresponding to the line loss-related data based on the summary synthesis prompts.

[0051] Based on the data description summary, the knowledge base in the field of power line loss is retrieved to obtain the standardized field definitions, historical alias sets and business rule fragments corresponding to line loss related data.

[0052] Based on standardized field definitions, historical alias sets, and business rule fragments, generate instructions for extracting line loss-related data.

[0053] The command to extract line loss-related data and the line loss-related data are input into a pre-built second-level multimodal large model, and the power line loss analysis dataset is output through the second-level multimodal large model.

[0054] The aforementioned method, apparatus, computer equipment, computer-readable storage medium, and computer program product for constructing power line loss analysis datasets acquire line loss-related data exported from various power systems in the distribution network. This data is then input into a pre-constructed first-level multimodal large-scale model. Based on structural parsing prompts in the first-level multimodal large-scale model, the line loss-related data is parsed to obtain metadata. Based on this metadata, summary synthesis prompts corresponding to the line loss-related data are generated. Data description summaries corresponding to the line loss-related data are obtained based on these summary synthesis prompts. A knowledge base in the power line loss domain is retrieved based on the data description summaries to obtain standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss-related data. Then, based on these standardized field definitions, historical alias sets, and business rule fragments, a line loss-related data extraction instruction is generated. This instruction and the line loss-related data are input into a pre-constructed second-level multimodal large-scale model, which outputs the power line loss analysis dataset. By introducing first- and second-level multimodal large models to deeply analyze the relevant data, the actual meaning of the data is clearly revealed, and the accuracy of the final dataset is improved through a progressive approach. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a diagram illustrating the application environment of a power line loss analysis dataset construction method in one embodiment.

[0057] Figure 2 This is a flowchart illustrating a method for constructing a power line loss analysis dataset in one embodiment;

[0058] Figure 3 This is a flowchart illustrating the steps involved in constructing a power line loss analysis dataset in one embodiment.

[0059] Figure 4 This is a flowchart illustrating the method for constructing a power line loss analysis dataset in another embodiment;

[0060] Figure 5 This is a structural block diagram of a power line loss analysis dataset construction device in one embodiment;

[0061] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0063] The method for constructing a power line loss analysis dataset based on a large model provided in this application can be applied to, for example... Figure 1 The application environment shown depicts a power distribution network where various power systems communicate with server 102 via a network. A data storage system can store the data that server 102 needs to process. This data storage system can be integrated onto server 102 or hosted on the cloud or other network servers. Server 102 acquires line loss-related data from at least one of the following: daily photovoltaic power generation of distribution transformers, monthly line loss of distribution transformers, and line power, derived from various power systems in the distribution network. This data is then input into a pre-constructed first-level multimodal large-scale model. Based on the structural parsing prompts in the first-level multimodal large-scale model, the line loss-related data is parsed to obtain metadata. Based on this metadata, summary synthesis prompts corresponding to the line loss-related data are generated. A data description summary corresponding to the line loss-related data is obtained based on the summary synthesis prompts. A knowledge base in the field of power line loss is retrieved based on the data description summary to obtain standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss-related data. Then, based on the standardized field definitions, historical alias sets, and business rule fragments, a line loss-related data extraction instruction is generated. This instruction and the line loss-related data are input into a pre-constructed second-level multimodal large-scale model, which outputs a power line loss analysis dataset. Server 102 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0064] In one exemplary embodiment, such as Figure 2 As shown, a method for constructing a power line loss analysis dataset based on a large model is presented, and this method is applied to... Figure 1 Taking server 102 as an example, the explanation includes the following steps S201 to S204. Wherein:

[0065] Step S201: Obtain line loss related data exported from each power system in the distribution network, and input the line loss related data into the pre-constructed first-level multimodal large model. Based on the structural parsing prompt words in the first-level multimodal large model, parse the line loss related data to obtain the metadata of the line loss related data, and generate the summary synthesis prompt words corresponding to the line loss related data according to each metadata. Based on the summary synthesis prompt words, obtain the data description summary corresponding to the line loss related data.

[0066] Among them, line loss related data can be understood as structured data such as power supply, electricity sales, line loss, and transformer loss; the first-level multimodal large model can be understood as a reinforcement learning model used to perform semantic understanding and summarization of line loss related data; structure parsing prompts can be understood as instructions used to guide the actions of the first-level multimodal large model, which may include data extraction assistants; metadata can be understood as the basic data items that make up the line loss related data, which may include header rows, partial data samples, and the data types of each column (such as text, numerical, and date); summary synthesis prompts can be understood as instructions used to guide the actions of the first-level multimodal large model; and data description summaries can be understood as a brief text description of the line loss related data, the content of which includes the line loss business type corresponding to the line loss related data.

[0067] For example, server 102 obtains line loss-related data exported from various power systems in the distribution network, and inputs the line loss-related data into a pre-constructed first-level multimodal large model. With the structural parsing prompt words in the first-level multimodal large model, the first-level multimodal large model is clearly defined as a "data extraction assistant" to extract the corresponding metadata from the line loss-related data. Based on each metadata, it generates summary synthesis prompt words corresponding to the line loss-related data. The summary synthesis prompt words are then re-inputted into the first-level multimodal large model to generate the original data description summary corresponding to the line loss-related data. The original data description summary is then input into a pre-configured classification and discrimination prompt word template to obtain the data description summary corresponding to the line loss-related data.

[0068] Step S202: Based on the data description summary, retrieve the knowledge base in the field of power line loss to obtain the standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss related data.

[0069] Among them, the power line loss domain knowledge base can be understood as a database that stores relevant and historical data of line loss business; the standardized field definition can be understood as the specific meaning of the standardized field corresponding to the line loss business type; the historical alias set can be understood as other naming information of the line loss business type; and the business rule fragment can be understood as the business rule information corresponding to the line loss business type.

[0070] Optionally, server 102 uses the line loss business type in the data description summary to retrieve the business type definition table in the power line loss domain knowledge base to obtain the core standard field ID corresponding to the line loss business type. Then, it uses the core standard field ID to retrieve the standard field main table in the power line loss domain knowledge base to obtain the corresponding core standard field. It then retrieves the historical mapping relationship table in the power line loss domain knowledge base that stores the mapping relationship between non-standard line loss expressions and standard line loss expressions according to the data description summary to obtain the mapping relationship corresponding to the line loss related data. It then maps the core standard field according to the mapping relationship. If the mapping is successful, it extracts the corresponding standardized field definition, historical alias set and business rule fragment from the standard field main table according to the core standard field.

[0071] Step S203: Generate a line loss related data extraction instruction based on the standardized field definitions, historical alias set, and business rule fragments.

[0072] Among them, the line loss related data extraction instruction can be understood as an instruction used to constrain and guide the second-level multimodal large model.

[0073] For example, server 102 obtains a pre-configured machine-readable structured instruction template, fills the standardized field definitions, historical alias sets and business rule fragments into the corresponding positions of the machine-readable structured instruction template, and obtains line loss related data extraction instructions.

[0074] Step S204: Input the line loss related data extraction instruction and line loss related data into the pre-constructed second-level multimodal large model, and output the power line loss analysis dataset through the second-level multimodal large model.

[0075] The second-level multimodal large model can be understood as a reinforcement learning model that can be bidirectionally input, where one input guides the model to process data from the other input.

[0076] Optionally, server 102 extracts line loss-related data and inputs it into a pre-built second-level multimodal large model. First, it performs semantic understanding on the line loss-related data to obtain semantically understood line loss-related data. Then, it matches the semantically understood line loss-related data with the standardized fields in the line loss-related data extraction instruction. After a successful match, it extracts the corresponding standardized data from the semantically understood line loss table in JSON format. Finally, it combines the standardized data to obtain the power line loss analysis dataset.

[0077] In the aforementioned method for constructing power line loss analysis data, line loss-related data exported from various power systems in the distribution network are acquired and input into a pre-constructed first-level multimodal large-scale model. Based on the structural parsing prompts in the first-level multimodal large-scale model, the line loss-related data is parsed to obtain metadata. Based on this metadata, summary synthesis prompts corresponding to the line loss-related data are generated. Data description summaries corresponding to the line loss-related data are obtained based on these summary synthesis prompts. A knowledge base in the power line loss domain is retrieved based on the data description summaries to obtain standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss-related data. Then, based on these standardized field definitions, historical alias sets, and business rule fragments, a line loss-related data extraction instruction is generated. This instruction and the line loss-related data are input into a pre-constructed second-level multimodal large-scale model, which outputs the power line loss analysis dataset. By introducing the first and second-level multimodal large-scale models to deeply analyze the relevant data and clarify its actual meaning, the accuracy of the final constructed dataset is improved through this progressive approach.

[0078] In one embodiment, obtaining a data description summary corresponding to line loss-related data based on summary synthesis prompts includes: re-inputting the summary synthesis prompts into the first-level multimodal large model to obtain the original data description summary corresponding to the line loss-related data; and filling the original data description summary into a pre-configured classification and discrimination prompt template to obtain the data description summary corresponding to the line loss-related data.

[0079] The original data description summary can be understood as line loss related data generated according to the standardized sentence pattern of "column name (type): sample value"; the classification and discrimination prompt word template can be understood as a blank template containing several parts, including role definition, context injection, task and format constraints, with corresponding data filling requirements for each part.

[0080] For example, server 102 re-inputs the summary synthesis prompt words into the first-level multimodal large model. The summary synthesis prompt words instruct the first-level multimodal large model to generate the corresponding original data description summary according to the normalized sentence structure. Then, the original data description summary is filled into the pre-configured classification and discrimination prompt word template to obtain a data description summary corresponding to the line loss related data with a regular format.

[0081] Based on the above implementation method, the powerful zero-shot understanding and instruction following capabilities of the multimodal large model are utilized to achieve generalized parsing of heterogeneous tables. There is no need to write specific parsing code for each file format. By dynamically injecting knowledge of line loss domain as context and combining it with strict output format control, a data foundation is laid for subsequent retrieval of relevant information.

[0082] In one embodiment, a line loss-related data extraction instruction is generated based on standardized field definitions, a set of historical aliases, and business rule fragments. This includes: obtaining a pre-configured machine-readable structured instruction template; and filling the standardized field definitions, the set of historical aliases, and the business rule fragments into the machine-readable structured instruction template to obtain the line loss-related data extraction instruction.

[0083] Optionally, server 102 obtains a machine-readable structured instruction template pre-designed for line loss scenarios. The template has reserved placeholders to wait for relevant information to be filled in. According to the correspondence between standardized field definitions, historical alias sets, business rule fragments and machine-readable structured instruction templates, the standardized field definitions, historical alias sets and business rule fragments are filled into the relevant positions of the machine-readable structured instruction template to obtain line loss related data extraction instructions.

[0084] According to the aforementioned implementation method, by filling the recalled knowledge points into the relevant positions, a complete and executable instruction for adding line loss scenario context is generated, providing precise constraints and guidance for the second-level multimodal large model.

[0085] In an exemplary embodiment, the line loss related data extraction instruction and the line loss related data are input into a pre-constructed second-level multimodal large model. The second-level multimodal large model outputs a power line loss analysis dataset, including: performing semantic understanding on the line loss related data to obtain semantically understood line loss related data; matching the semantically understood line loss related data with the standardized fields in the line loss related data extraction instruction, and extracting the corresponding standardized data from the semantically understood line loss related data according to JSON format; and combining the standardized data to obtain the power line loss analysis dataset.

[0086] For example, server 102 inputs the line loss related data extraction instruction and line loss related data into a pre-constructed second-level multimodal large model. First, it performs semantic understanding on the line loss related data to obtain semantically understood line loss related data. Then, it matches the semantically understood line loss related data with the standardized fields in the line loss related data extraction instruction, and extracts the corresponding standardized data from the semantically understood line loss related data according to JSON format. Finally, it combines the standardized data to obtain the power line loss analysis dataset.

[0087] Based on the above implementation method, the task of multimodal large model is transformed from the open "understanding the table" to the closed "extraction under specified rules" by bidirectional input, which greatly improves the determinism of the task and the reliability of the results. Secondly, the multimodal large model strictly follows the format requirements defined in the "output specification" part of the instruction to perform data reading and encapsulation, ensuring the structural consistency of the output data.

[0088] In one embodiment, the power line loss knowledge base includes a business type definition table, a standard field master table, and a historical mapping relationship table; the historical mapping relationship table is used to store the mapping relationship from non-standard line loss descriptions to standard line loss descriptions; the data description summary includes the line loss business type corresponding to the line loss related data;

[0089] The system retrieves the knowledge base for power line loss based on data description summaries, obtaining corresponding standardized field definitions, historical alias sets, and business rule fragments. This includes: retrieving the business type definition table using the line loss business type to obtain the core standard field ID corresponding to the line loss business type; retrieving the standard field main table based on the core standard field ID to obtain the core standard field corresponding to the core standard field ID; retrieving the historical mapping relationship table according to the data description summary to obtain the mapping relationship corresponding to the line loss-related data; and obtaining the standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss-related data based on the mapping relationship and the core standard field.

[0090] The business type definition table is used to store the abstracted and standardized line loss business table types. The table structure includes type ID (integer), type name (string, such as "Daily load line loss data"), core line loss standard field ID set (integer array), line loss scenario description (string, such as "Adapted to 10kV line daily line loss calculation scenario"), and timing requirements (string, such as "The data statistics period is natural days, and the power supply and sales volume must be synchronized in time"). The system should clearly define the set of core standard fields and business constraints necessary for line loss calculation. The main table of standard fields defines unique line loss-related standard field information for the entire system. The table structure includes field ID (integer), standard field name (string, such as "input active power," "line resistance," "transformer no-load loss," etc.), line loss business meaning (string, such as "total active power input to the line, a fundamental indicator for line loss calculation"), unit of measurement (string, such as "kWh," "Ω," "kW"), field type (string, such as numeric, text, date), and line loss calculation association rules (string, such as "the difference between the output active power and the line loss amount"). The historical mapping relationship table stores verified mapping relationships from "non-standard line loss descriptions" to "line loss standard fields." For example, when the system (or manual assistance) confirms that the "ZXYGZ" column in a table should be mapped to the standard field active_power_in (input active power), this relationship will be persistently saved as a record, containing the original description, target standard field ID, mapping confidence, and source context.

[0091] Optionally, server 102 retrieves the business type definition table based on the line loss business type corresponding to the line loss related data in the data description summary to obtain the core standard field ID corresponding to the line loss business type. Then, it retrieves the standard field main table according to the core standard field ID to obtain the core standard field corresponding to the core standard field ID. At the same time, it retrieves the historical mapping relationship table based on the data description summary to obtain the mapping relationship corresponding to the line loss related data. Finally, it uses the mapping relationship to map the core standard field. If the mapping is successful, it extracts the standardized field definition, historical alias set, and business rule fragment corresponding to the line loss related data from the standard field main table based on the core standard field.

[0092] According to the aforementioned implementation method, by performing vectorized table storage on the line loss data, and sequentially retrieving the corresponding information according to the relevant data, it is ensured that the relevant data can be directly retrieved and called, thereby improving the recall efficiency and recall success rate of knowledge points.

[0093] In one embodiment, the process of retrieving a historical mapping table based on a data description summary to obtain the mapping relationship corresponding to the line loss-related data includes: constructing a corresponding semantic retrieval query vector based on the data description summary; retrieving a pre-constructed vector database based on the semantic retrieval query vector to obtain a non-standard line loss expression corresponding to the semantic retrieval query vector; the vector database is used to store the non-standard line loss expressions in the historical mapping table; and determining the corresponding mapping relationship from the historical mapping table based on the non-standard line loss expressions.

[0094] For example, server 102 constructs a corresponding semantic retrieval query vector based on the data description summary, checks a pre-built vector database containing non-standard line loss expressions from a historical mapping table according to the semantic retrieval query vector, obtains the non-standard line loss expression corresponding to the semantic retrieval vector, and finally determines the corresponding mapping relationship from the historical mapping table based on the non-standard line loss expression.

[0095] Based on the above implementation method, vector queries are executed by constructing corresponding semantic retrieval query vectors, thereby quickly retrieving the corresponding data and realizing accurate retrieval and reuse of relevant data.

[0096] In one embodiment, based on the mapping relationship and core standard fields, the standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss related data are obtained, including: mapping the core standard fields according to the mapping relationship; and if the mapping is successful, extracting the standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss related data from the standard field master table according to the core standard fields.

[0097] Optionally, server 102 maps the core standard fields according to the mapping relationship. If the mapping is successful, it extracts the standardized field definitions, historical alias sets and business rule fragments corresponding to the line loss related data from the standard field master table based on the core standard fields.

[0098] According to the aforementioned implementation method, by performing mapping according to the determined mapping relationship and extracting relevant knowledge points when the mapping is successful, the real-time reliability of the data is guaranteed and the accuracy of the constructed power line loss analysis dataset is improved.

[0099] In one exemplary embodiment, a method and system for constructing a power line loss analysis dataset based on a large model are provided, such as... Figure 3 and Figure 4 As shown, by constructing a collaborative processing architecture containing hierarchical intelligent modules, end-to-end automated governance of multi-source heterogeneous electricity table data is achieved. This method aims to solve the core problems of high reliance on manual labor, rigid rules, and lack of semantic understanding in traditional data preparation methods.

[0100] Figure 3 Description: This application's system consists of four core functional modules: a table type discriminator (module 1), a power line loss domain knowledge base (module 2), a knowledge retrieval and prompt word builder based on RAG (Retrieval-Augmented Generation) (module 3), and a line loss data extractor (module 4). The collaborative relationship and data flow between the modules are as follows: raw table data is input to the table type discriminator; the output of the discriminator triggers the RAG-based knowledge retrieval and prompt word builder to initiate a query to the power line loss domain knowledge base; the knowledge base returns line loss domain knowledge, and the builder generates standardized extraction instructions for line loss scenarios; finally, the line loss data extractor receives the raw data and the instructions, and completes line loss data extraction, semantic mapping, association verification, timing verification, and standardized output. The arrows in the diagram clearly indicate the information flow of "table summary," "query," "retrieved knowledge," "extraction instructions," and "standardized data."

[0101] Figure 4 Description: The process of this application begins with receiving raw line loss-related data from multiple business systems. First, a table type discriminator intelligently classifies the input tables and outputs their line loss business type. Second, based on this table type, the system dynamically acquires relevant knowledge by searching the domain knowledge base and constructs precise line loss data extraction instructions. Finally, the line loss data extractor parses the raw tables according to the instructions, processing them iteratively until all data has undergone standardization transformation and output, forming the final line loss analysis dataset.

[0102] (1) Table type discriminator (first-level multimodal large model):

[0103] The core innovation of this module lies in proposing a "table data extraction and classification method based on multimodal large model and prompt word engineering", which optimizes the traditional technical path that relies on hard-coded rules or training a dedicated classification model.

[0104] Core Technical Approach 1 (Table Parsing and Summary Generation Based on Prompt Terms): To achieve unified processing of tables of any format, this module designs a progressive prompt term engineering scheme. The system first directly inputs the binary or text content of the original table file (such as Excel or CSV) into the first-level multimodal large model, along with a structure parsing prompt term. This prompt term explicitly instructs the model to act as a "data extraction assistant," identifying and extracting the header row, partial data samples, and data types (such as text, numeric, and date) from the original content. Subsequently, the system inputs these extracted discrete metadata items into the same multimodal large model through a summary synthesis prompt term, instructing it to generate a well-structured data description summary according to the standardized sentence format "column name (type): sample value". This method leverages the powerful zero-shot understanding and instruction-following capabilities of the multimodal large model, achieving universal parsing of heterogeneous tables without needing to write specific parsing code for each file format.

[0105] Core Technology Method Two (Interactive Table Type Discrimination Based on Context Injection): After obtaining the data description summary, the system performs classification discrimination. Its key technology lies in designing a classification discrimination prompt word template. This template contains several key parts: 1) Role Definition: The model is role-based as a "data classification expert in the power sector"; 2) Context Injection: The complete "list of known business type names" obtained from the domain knowledge base is dynamically injected as the context for discrimination options; 3) Task and Format Constraints: The instruction model strictly follows the provided summary, selecting the most matching item from the injected type list, and the output must only be the type name string. Example of the classification discrimination prompt word template: "As a data classification expert in the power sector, please select the most matching item from the list of known business type names based on the following data description summary, outputting only the type name string and the data description summary, without additional explanation. Data Description Summary Template: [Enter the desired data description summary template here]; List of Known Business Type Names: [Enter the line loss business type names and brief scenario descriptions stored in the power line loss domain knowledge base here]." By dynamically injecting domain knowledge of line loss as context and combining it with strict output format control, this method guides a multimodal large model to complete a reliable and controlled sample classification inference, replacing the traditional feature engineering-based classifier training process.

[0106] (2) Knowledge base for power line loss:

[0107] The core innovation of this module lies in the construction of a "knowledge base for the power line loss domain based on expert knowledge vectorization management". Its design aims to solve the problem in the construction of traditional line loss datasets where expert experience exists in the form of unstructured documents or tacit knowledge, which cannot be directly and efficiently called upon and reused by the system.

[0108] Core technical means 1 (vectorized storage of expert knowledge):

[0109] To address the challenge of directly applying knowledge in the field of line loss to automated processing, this solution designs a dedicated structured storage system for line loss analysis data. Its core is a set of core data tables defined and populated by line loss experts within the relational database PostgreSQL (PostgreStructured Query Language), encoding the core knowledge of line loss.

[0110] Historical Mapping Relationship Table: This table stores verified mapping relationships from "non-standard line loss descriptions" to "standard line loss fields". For example, when the system (or with manual assistance) confirms that the "ZXYGZ" column in a table should be mapped to the standard field active_power_in, this relationship will be persistently saved as a record, containing the original description, the target standard field ID, the mapping confidence level, and the source context.

[0111] Business Type Definition Table: This table stores the abstracted and standardized line loss business table types. The table structure includes a type ID (integer), a type name (string, such as "Daily Line Load Loss Data"), a set of core line loss standard field IDs (an array of integers), a line loss scenario description (string, such as "Suitable for 10kV line daily line loss calculation scenario"), and timing requirements (string, such as "Data statistics period is natural days, and power supply and sales volume must be synchronized in timing"). It clearly defines the set of core standard fields necessary for line loss calculation and the business constraints that should be included.

[0112] Standard Fields Main Table: Defines unique standard field information related to line loss across the entire system. The table structure includes field ID (integer), standard field name (string, such as "input active power", "line resistance", "transformer no-load loss", etc.), line loss business meaning (string, such as "total active power input to the line, which is the basic indicator for line loss calculation"), unit of measurement (string, such as "kWh", "Ω", "kW"), field type (string, such as numeric, text, date), and line loss calculation association rule (string, such as "the difference between the output active power and the line loss power").

[0113] Core Technical Meaning Two (Index Construction for Line Loss Semantic Retrieval):

[0114] To address the challenge of quickly and accurately retrieving and reusing historical experience in the field of massive line loss data, this solution builds a parallel semantic indexing layer on top of structured storage. Specifically, the system transforms the "non-standard line loss related data descriptions" (such as "total positive active power") and their related contextual text stored in the historical mapping table into high-dimensional vectors through an embedding model and stores them in a vector database.

[0115] The innovation of this mechanism lies in its ability to enable the knowledge base to support not only precise queries based on line loss field names, but also fuzzy searches based on line loss semantics. When subsequent processing encounters new, unseen non-standard column names, module 3 can quickly recall historically semantically similar expressions and their corresponding standard field mapping schemes and association rules from the vector index by calculating semantic similarity. This essentially endows the system with the ability to "associate" and "learn from" historical experience, significantly reducing the reliance on immediate manual intervention when processing new line loss data.

[0116] (3) RAG-based knowledge retrieval and instruction generator:

[0117] The core technological breakthrough of this module lies in the realization of a "dynamic knowledge retrieval and instruction synthesis mechanism for data governance," which deeply integrates retrieval enhancement generation technology with line loss business needs and is specifically used for the mapping and extraction guidance of line loss data.

[0118] Technical Feature 1 (Multi-dimensional Query Vector Generation): This module receives the "line loss business type name" from Module 1. Unlike simple keyword queries, this module concatenates the line loss type name with the header information and time-series statistical period of the original line loss table to form a multi-dimensional query text rich in line loss context (e.g., "Daily load data table for power lines, including columns: line number, data time, total positive active power"). Subsequently, the query text is transformed into a query vector using the same embedding model used in the construction of the power line loss domain knowledge base.

[0119] Technical Feature Two (Automated Editing of Context-Enhanced Instructions): A machine-readable structured instruction template specifically designed for line loss scenarios was developed. This template includes placeholders for filling in "Line Loss Target Field List," "Line Loss Mapping Reference," "Calculation Association Rules," "Time Series Verification Requirements," and "Output Rules." When the module retrieves relevant knowledge points from the vector database, the system automatically fills in "Line Loss Standard Field Names," "Line Loss Business Meaning," "Unit of Measurement," "Field Type," "Line Loss Calculation Association Rules," "Historical Alias ​​Set," and "Time Series Verification Requirements" into the corresponding positions in the template based on the correspondence between the retrieved results and the line loss instruction template. This synthesizes a complete and executable line loss scenario context-enhanced instruction, providing precise constraints and guidance for the second-level multimodal large model.

[0120] (4) Line loss data extractor (second-level multimodal large model):

[0121] The core technological innovation of this module lies in the design of a "multimodal large model semantic alignment and data extraction method under the knowledge base experience and context constraints of the line loss domain, which is specifically designed for the semantic characteristics and temporal requirements of line loss data to ensure the accuracy, relevance and temporal consistency of line loss data transformation".

[0122] Technical Feature 1 (Task Setting with Dual Input Constraints): This module is configured to simultaneously receive two types of input: A) the complete or segmented content of the original table (text format, with each segment set to 200 rows of data); B) structured extraction instructions (JSON format) retrieved from Module 3. This input method transforms the task of multimodal large-scale models from the open-ended "understanding the table" to the closed-ended "performing extraction under specified rules," significantly improving the determinism of the task and the reliability of the results.

[0123] Technical Feature Two (Semantic Matching and Structured Output Based on Line Loss Knowledge Reference): The second-level multimodal large model operates under the strict constraints of the structured instructions. Its core is to perform a semantic alignment task: using the "historical alias set" provided under the "line loss target field" item in the instructions as a key reference, combined with the "line loss business meaning" and "calculation association rules," it performs semantic similarity comparison and inference with the original line loss column names in input A (using the cosine similarity calculation method, with a similarity threshold set to 0.85), thereby establishing an accurate mapping from non-standard column names to standard fields. Subsequently, the model strictly follows the format requirements defined in the "output specification" section of the instructions, performing data reading and encapsulation to ensure the structural consistency of the output data.

[0124] The four modules described in this application constitute a complete system flow for constructing a line loss dataset. Module 1 (Discriminator) solves the problem of automated perception and triggering of line loss scenarios; Module 2 (Knowledge Base) provides a structured and semantic knowledge foundation for line loss, covering core content such as line loss calculation principles, business rules, and historical experience; Module 3 (Knowledge Retrieval and Instruction Generator) uses the output of Module 1 to dynamically retrieve line loss-specific knowledge from Module 2 and generate "line loss data processing operation instructions," realizing the flexibility and knowledge-based nature of the processing logic for line loss scenarios; Module 4 (Extractor), under the precise guidance of these "operation instructions," completes the final semantic alignment, association verification, temporal verification, and data conversion. The collaborative work of this architecture enables the efficient, accurate, and low-maintenance-cost automated construction of standardized line loss analysis datasets from multi-source line loss data. Its technical effects are reflected in the breakthrough in semantic understanding of non-standard data in the field of power line loss and the strong adaptability to the variable line loss data environment, ultimately outputting a high-quality dataset that meets the needs of accurate line loss calculation and anomaly diagnosis.

[0125] Compared with the prior art, this application has the following technical advantages:

[0126] 1. By introducing an intelligent workflow architecture based on a two-layer multimodal large model and a knowledge base for line loss, the entire process of "line loss scenario classification - line loss knowledge retrieval - line loss data extraction" is automatically applied to the construction of power line loss datasets for the first time. It specifically focuses on the core requirements of the line loss field for data semantic consistency, correlation integrity, and temporal synchronization, which significantly improves processing efficiency and realizes the transformation from a type of highly dependent on the human experience of line loss experts to a type dominated by machine intelligence, making large-scale, routine line loss analysis (such as daily line loss and full coverage analysis of transformer area line loss) possible.

[0127] 2. By designing a dynamic knowledge retrieval and instruction generation mechanism for the field of line loss based on retrieval enhancement generation technology, the system can adaptively retrieve and organize line loss-specific knowledge from the structured knowledge base according to the specific scenario of the input line loss data, and automatically synthesize processing instructions. This fundamentally overcomes the defects of traditional script methods based on hard-coded rules, such as fragility and high maintenance costs, and greatly enhances the system's adaptability and robustness to the variable line loss data environment.

[0128] 3. By leveraging the powerful semantic understanding capabilities of multimodal large models and receiving precise instructions rich in domain context from the line loss knowledge base, intelligent semantic recognition and accurate mapping of complex non-standard data in the power line loss industry (such as "positive active power total", "ZXYGZ", etc.) are achieved. At the same time, line loss data association verification and time sequence verification are completed. This solves the fundamental limitations of traditional dictionary- or rule-based methods, which can only perform mechanical literal matching, cannot understand the business connotation of line loss, and ignore time sequence synchronization. As a result, a high-quality line loss analysis dataset with more accurate association, clearer business logic, and meeting the needs of line loss analysis is constructed.

[0129] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0130] Based on the same inventive concept, this application also provides a device for constructing a power line loss analysis dataset based on a large model to implement the aforementioned method for constructing a power line loss analysis dataset based on a large model. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the device for constructing a power line loss analysis dataset based on a large model provided below can be found in the limitations of the method for constructing a power line loss analysis dataset based on a large model described above, and will not be repeated here.

[0131] In one exemplary embodiment, such as Figure 5 As shown, a device for constructing a power line loss analysis dataset based on a large model is provided, including: an acquisition module 501, a retrieval module 502, a generation module 503, and a construction module 504, wherein:

[0132] The acquisition module 501 is used to acquire line loss related data exported from each power system in the distribution network, input the line loss related data into the pre-constructed first-level multimodal large model, parse the line loss related data based on the structural parsing prompt words in the first-level multimodal large model, obtain the metadata of the line loss related data, and generate the summary synthesis prompt words corresponding to the line loss related data based on each metadata, and obtain the data description summary corresponding to the line loss related data based on the summary synthesis prompt words.

[0133] The retrieval module 502 is used to retrieve the knowledge base in the field of power line loss based on the data description summary, and obtain the standardized field definitions, historical alias sets and business rule fragments corresponding to the line loss related data.

[0134] The generation module 503 is used to generate instructions for extracting line loss-related data based on standardized field definitions, historical alias sets, and business rule fragments.

[0135] Module 504 is used to input the line loss related data extraction instructions and line loss related data into the pre-built second-level multimodal large model, and output the power line loss analysis dataset through the second-level multimodal large model.

[0136] In one embodiment, the acquisition module 501 is further configured to re-input the summary synthesis prompt words into the first-level multimodal large model to obtain the original data description summary corresponding to the line loss related data; and to fill the original data description summary into the pre-configured classification and discrimination prompt word template to obtain the data description summary corresponding to the line loss related data.

[0137] In one embodiment, the generation module 503 is further configured to obtain a pre-configured machine-readable structured instruction template; and to fill the standardized field definitions, historical alias sets and business rule fragments into the machine-readable structured instruction template to obtain line loss related data extraction instructions.

[0138] In an exemplary embodiment, the construction module 504 is further configured to perform semantic understanding on the line loss related data to obtain semantically understood line loss related data; match the semantically understood line loss related data with the standardized fields in the line loss related data extraction instruction, and extract the corresponding standardized data from the semantically understood line loss related data in JSON format; combine the standardized data to obtain a power line loss analysis dataset.

[0139] In one embodiment, the power line loss domain knowledge base includes a business type definition table, a standard field master table, and a historical mapping relationship table. The historical mapping relationship table stores the mapping relationship from non-standard line loss descriptions to standard line loss descriptions. The data description summary includes the line loss business type corresponding to the line loss related data. The retrieval module 502 is further used to retrieve the business type definition table using the line loss business type to obtain the core standard field ID corresponding to the line loss business type; retrieve the standard field master table based on the core standard field ID to obtain the core standard field corresponding to the core standard field ID; retrieve the historical mapping relationship table according to the data description summary to obtain the mapping relationship corresponding to the line loss related data; and obtain the standardized field definition, historical alias set, and business rule fragment corresponding to the line loss related data based on the mapping relationship and the core standard field.

[0140] In one embodiment, the retrieval module 502 is further configured to construct a corresponding semantic retrieval query vector based on the data description summary; retrieve a pre-constructed vector database according to the semantic retrieval query vector to obtain a non-standard line loss representation corresponding to the semantic retrieval query vector; the vector database is used to store the non-standard line loss representations in the historical mapping relationship table; and determine the corresponding mapping relationship from the historical mapping relationship table based on the non-standard line loss representations.

[0141] In an exemplary embodiment, the retrieval module 502 is further configured to map the core standard fields according to the mapping relationship; if the mapping is successful, it extracts the standardized field definitions, historical alias sets and business rule fragments corresponding to the line loss related data from the standard field master table according to the core standard fields.

[0142] Each module in the aforementioned power line loss analysis dataset construction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0143] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media. The database stores line loss-related data, data description summaries, standardized field definitions, historical alias sets, business rule fragments, line loss-related data extraction instructions, and power line loss analysis datasets. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for constructing a power line loss analysis dataset based on a large model.

[0144] Those skilled in the art will understand that Figure 6The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0145] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method for constructing a power line loss analysis dataset based on a large model as described in the above embodiments.

[0146] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a power line loss analysis dataset based on a large model as described in the above embodiment.

[0147] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the method for constructing a power line loss analysis dataset based on a large model as described in the above embodiments.

[0148] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0149] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0150] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0151] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for constructing a power line loss analysis dataset based on a large model, characterized in that, The method includes: The system acquires line loss-related data exported from various power systems in the distribution network and inputs this data into a pre-constructed first-level multimodal large model. Based on the structural parsing prompts in the first-level multimodal large model, it parses the line loss-related data to obtain metadata. Based on this metadata, it generates summary synthesis prompts corresponding to the line loss-related data. These prompts are then re-inputted into the first-level multimodal large model to obtain the original data description summary corresponding to the line loss-related data. The original data description summary is then filled into a pre-configured classification and discrimination prompt template to obtain the data description summary corresponding to the line loss-related data. The data description summary includes the line loss service type corresponding to the line loss-related data. The core standard field ID corresponding to the line loss business type is obtained by retrieving the business type definition table in the power line loss domain knowledge base using the line loss business type. Based on the core standard field ID, retrieve the main table of standard fields in the power line loss knowledge base to obtain the core standard field corresponding to the core standard field ID; A corresponding semantic retrieval query vector is constructed based on the data description summary. A pre-constructed vector database is retrieved according to the semantic retrieval query vector to obtain the non-standard line loss description corresponding to the semantic retrieval query vector. Based on the non-standard line loss description, the corresponding mapping relationship is determined from the historical mapping relationship table in the power line loss domain knowledge base. The historical mapping relationship table is used to store the mapping relationship from non-standard line loss description to standard line loss description, and the vector database is used to store the non-standard line loss description in the historical mapping relationship table. Based on the mapping relationship and the core standard fields, the standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss related data are obtained; Based on the standardized field definitions, historical alias set, and business rule fragments, generate instructions for extracting line loss-related data; The line loss related data extraction command and the line loss related data are input into a pre-constructed second-level multimodal large model, and the power line loss analysis dataset is output through the second-level multimodal large model.

2. The method according to claim 1, characterized in that, The step of generating line loss-related data extraction instructions based on the standardized field definitions, historical alias set, and business rule fragments includes: Obtain a pre-configured machine-readable structured instruction template; The standardized field definitions, historical alias sets, and business rule fragments are filled into the machine-readable structured instruction template to obtain the line loss related data extraction instruction.

3. The method according to claim 2, characterized in that, The step of inputting the line loss-related data extraction instruction and the line loss-related data into a pre-constructed second-level multimodal large model, and outputting the power line loss analysis dataset through the second-level multimodal large model, includes: The line loss related data is semantically understood to obtain semantically understood line loss related data; The semantically understood line loss-related data is matched with the standardized fields in the line loss-related data extraction instruction, and the corresponding standardized data is extracted from the semantically understood line loss-related data in JSON format. By combining the standardized data, the power line loss analysis dataset is obtained.

4. The method according to claim 1, characterized in that, Based on the mapping relationship and the core standard fields, the standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss related data are obtained, including: The core standard fields are mapped according to the mapping relationship; If the mapping is successful, the standardized field definitions, historical alias sets, and business rule fragments corresponding to the line loss-related data are extracted from the standard field master table based on the core standard fields.

5. A device for constructing a power line loss analysis dataset based on a large model, characterized in that, The device includes: The acquisition module is used to acquire line loss-related data exported from various power systems in the distribution network, input the line loss-related data into a pre-constructed first-level multimodal large model, parse the line loss-related data based on the structural parsing prompts in the first-level multimodal large model to obtain the metadata of the line loss-related data, and generate summary synthesis prompts corresponding to the line loss-related data according to the metadata. The summary synthesis prompts are then re-inputted into the first-level multimodal large model to obtain the original data description summary corresponding to the line loss-related data. The original data description summary is then filled into a pre-configured classification and discrimination prompt template to obtain the data description summary corresponding to the line loss-related data. The data description summary includes the line loss service type corresponding to the line loss-related data. The retrieval module is used to retrieve the business type definition table in the power line loss domain knowledge base using the line loss business type to obtain the core standard field ID corresponding to the line loss business type; retrieve the standard field main table in the power line loss domain knowledge base according to the core standard field ID to obtain the core standard field corresponding to the core standard field ID; construct the corresponding semantic retrieval query vector according to the data description summary; retrieve the pre-constructed vector database according to the semantic retrieval query vector to obtain the non-standard line loss description corresponding to the semantic retrieval query vector; determine the corresponding mapping relationship from the historical mapping relationship table in the power line loss domain knowledge base based on the non-standard line loss description; the historical mapping relationship table is used to store the mapping relationship from non-standard line loss description to standard line loss description; the vector database is used to store the non-standard line loss description in the historical mapping relationship table; and obtain the standardized field definition, historical alias set, and business rule fragment corresponding to the line loss related data based on the mapping relationship and the core standard field. The generation module is used to generate line loss related data extraction instructions based on the standardized field definitions, historical alias set, and business rule fragments. The construction module is used to input the line loss related data extraction instruction and the line loss related data into a pre-constructed second-level multimodal large model, and output the power line loss analysis dataset through the second-level multimodal large model.

6. The apparatus according to claim 5, characterized in that, The device further includes: The generation module is also used to obtain a pre-configured machine-readable structured instruction template; and to fill the standardized field definitions, historical alias sets and business rule fragments into the machine-readable structured instruction template to obtain the line loss related data extraction instruction.

7. The apparatus according to claim 5, characterized in that, The device further includes: The construction module is further configured to perform semantic understanding on the line loss related data to obtain semantically understood line loss related data; match the semantically understood line loss related data with the standardized fields in the line loss related data extraction instruction, and extract the corresponding standardized data from the semantically understood line loss related data in JSON format; and combine the standardized data to obtain the power line loss analysis dataset.

8. The apparatus according to claim 5, characterized in that, The device further includes: The retrieval module is further configured to map the core standard fields according to the mapping relationship; if the mapping is successful, it extracts the standardized field definitions, historical alias sets and business rule fragments corresponding to the line loss related data from the standard field master table according to the core standard fields.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images