Heating industry senior report generation method and system based on large language model

Abstract

The application relates to a heat supply industry advanced report generation method and system based on a large language model, and relates to the field of large model applications, which comprises the following steps: creating a report outline; setting a checking intention and an analysis intention under a report outline directory title; sequentially matching a first similar fragment in a first vector knowledge base based on the checking intention, and generating SQL and obtaining a data result by a large language model according to the checking intention, the first similar fragment and a prompt word; matching a second similar fragment in a second vector knowledge base based on the checking intention and a corresponding analysis intention, and obtaining an analysis conclusion by the large language model according to the checking intention, the corresponding analysis intention and the second similar fragment; taking the data result and the analysis conclusion as content under the current directory title; and generating an advanced report after sequentially integrating the content and polishing according to the report outline. The application realizes the whole process of advanced report generation, ensures that the generated advanced report is highly professional, high in quality and uniform in expression, and is beneficial to improving decision-making efficiency.

Description

Technical Field

[0001] This application relates to the field of large-scale model applications in the heating industry, and in particular to a method and system for generating advanced reports for the heating industry based on a large language model. Background Technology

[0002] In the daily operation and management of the heating industry, report preparation is one of the core tasks supporting business decision-making, compliance filing, and operational review. Currently, report preparation mainly involves technical personnel first querying various data tables during the operation of the heating system to obtain the data needed for report preparation, and then analyzing and organizing the data to prepare reports according to an appropriate outline and content.

[0003] However, the heating industry currently faces the following common problems in the process of compiling reports and documents: First, because heating system data is scattered across multiple scenarios such as energy consumption monitoring systems, equipment operation platforms, and user payment databases, manual aggregation and querying require exporting data across systems and manual calculation and analysis, making the data acquisition process cumbersome, fragmented, and inefficient, and prone to inaccurate conclusions due to data entry errors; Second, the report and document compilation process is cumbersome and highly repetitive. Documents such as the "Monthly Energy Consumption Statistical Analysis Report" and the "Quarterly Heating System Operation Summary" require a fixed structural framework and conventional analysis dimensions. When compiling them manually, templates are repeatedly applied and similar content is copied, which is time-consuming and labor-intensive; Third, due to the differences in the professional capabilities of different technical personnel, the process of writing analysis conclusions relies heavily on personal experience, and the expression of industry terminology is inconsistent, resulting in inconsistent document quality and difficulty in standardization, affecting subsequent decision-making and archiving value; Fourth, the report production cycle is long, and a complete report often takes several hours or even days to complete, causing management to be unable to obtain timely information on the operating conditions and resulting in delayed decision-making responses to unexpected problems. Summary of the Invention

[0004] To address the aforementioned technical challenges, this application utilizes a large language model to understand the report writing requirements of technical personnel, query relevant data, and perform professional data analysis. This enables the automatic completion of report document structure arrangement and content filling, supports one-click export of multiple document types, and ensures that the reports possess strong professionalism and standardization. This application provides a method and system for generating advanced reports for the heating industry based on a large language model.

[0005] Firstly, this application provides a method for generating advanced reports for the heating industry based on a large language model, employing the following technical solution:

[0006] A method for generating advanced reports for the heating industry based on a large language model includes the following steps:

[0007] Create a report outline, including a custom report outline or by selecting a preset outline template;

[0008] Under the headings of the report outline, define several groups of data lookup and analysis intentions;

[0009] The content under each directory title is generated sequentially according to the report outline table of contents and based on the grouped data lookup intent and analysis intent under the directory title.

[0010] The search enhancement generation technology matches the first similar segment in the pre-built first vector knowledge base based on the search intent under the current directory title, and generates SQL based on the search intent, the first similar segment and prompt words through a large language model. Then, the query results are obtained from the database through the SQL query.

[0011] By using retrieval enhancement generation technology, based on the data search intent and the corresponding analysis intent under the current directory title, a second similar segment is matched in a pre-built second vector knowledge base. Then, a large language model is used to intelligently gain insights into the data results based on the data search intent, the analysis intent, and the second similar segment to obtain analysis conclusions.

[0012] The data results and analysis conclusions will be used as content under the current directory heading based on the data lookup intent and analysis intent of that group.

[0013] The large language model integrates the titles and contents of the report outlines at all levels in the order of the report outline directory, and then polishes them to generate an advanced report.

[0014] By adopting the above technical solution, technicians can create multi-level report outlines by customizing the report outline or selecting preset outline templates. They can also independently set several sets of data query intentions and analysis intentions so that the big data model can accurately identify the technicians' needs for report content and generate customized report content. The big data model accurately executes data queries based on the data query intentions, avoiding manual data summarization and querying by technicians or data entry errors. Furthermore, the big data model accurately performs insight analysis based on the analysis intentions, which not only ensures that the generated report content closely matches user needs but also avoids time-consuming and labor-intensive manual writing. At the same time, it ensures that the generated advanced reports are highly professional, of high quality, and use consistent professional terminology. The report documents are standardized and regulated, thus realizing the entire process from data acquisition to analysis and insight to report generation. It also compresses the report generation process from hours or even days to minutes, helping technicians and leaders respond quickly to changes in work conditions and improve decision-making efficiency.

[0015] In a specific implementation, the content under the directory headings is generated sequentially according to the report outline directory order and based on the grouped data lookup intentions and analysis intentions under the directory headings. This content is previewable and allows users to modify the current data lookup intention and analysis intention multiple times based on the previewable content, and then regenerate the content.

[0016] By adopting the above technical solution, during the data query and insight analysis process based on grouped query intentions and analysis intentions, the query results and insight analysis conclusions are generated in real time as previewable content. At this time, technicians can directly judge whether the generated previewable content meets the actual needs, and modify and adjust the current query intentions and analysis intentions to regenerate new content until the generated content meets the needs of technicians. This realizes the dynamic display of data results and analysis conclusions, and supports technicians to implement intervention and adjustment, thereby ensuring that the generated report content provides effective support for the actual decision-making process.

[0017] In a specific implementation scheme, after the data results and the analysis conclusions are used as the content under the current directory title based on the data lookup intent and the analysis intent, the scheme further includes summarizing all the content under the directory titles of each level of report outline through a large language model, and generating the summary content under the current report outline directory title.

[0018] By adopting the above technical solution, summary content can be generated for all existing content under the outline headings of reports at all levels that contain multiple sets of data lookup and analysis intentions. This helps technical personnel draw conclusions under the current headings based on the summary content, thereby assisting in decision-making and improving the user experience for technical personnel.

[0019] In one specific implementation, setting several sets of data lookup intentions and analysis intentions under the report outline directory headings includes setting several sets of data lookup intentions and analysis intentions under the lowest level directory heading of the report outline.

[0020] In a specific feasible implementation, it also includes:

[0021] It supports selecting preset or saved outline templates and setting regular generation rules, as well as generating new advanced reports based on the latest data at the time, and automatically pushing them according to preset methods;

[0022] The periodic generation rules include generation cycles set by hour / day / week / month and generation time points.

[0023] By adopting the above technical solution, technicians can set timed generation rules for commonly used preset or saved outline templates. The system will automatically execute the advanced report generation process according to the set generation cycle and generation time, which greatly improves work efficiency, further realizes the standardization and normalization of advanced reports, facilitates subsequent management, and helps to make timely decisions based on advanced reports.

[0024] In a specific implementation scheme, the scheme further includes establishing a question library, which includes several data query questions. It supports adding pre-defined data query intentions to the question library and allows selecting data query questions from the question library as data query intentions when setting data query intentions.

[0025] By adopting the above technical solutions, a problem database is established, which facilitates direct selection when setting the intent of data lookup, thereby improving work efficiency.

[0026] In a specific implementation plan, the plan also includes establishing a template library, which includes several preset outline templates. It supports adding custom report outlines to the template library, and the outline templates are preset in several ways according to the application scenario.

[0027] By adopting the above technical solutions, establishing a template library makes it easier to select existing outline templates when creating advanced reports, which not only further improves efficiency but also helps ensure that the outline structure of advanced reports conforms to the specifications.

[0028] In one specific implementation scheme, constructing the first vector knowledge base includes vectorizing the database table structure description document;

[0029] The construction of the second vector knowledge base involves vectorizing unstructured knowledge documents composed of heating-related standards, heating industry literature, heating theory, system operation manuals, equipment maintenance methods, heating platform operation manuals, equipment technical manuals, operating procedures, troubleshooting guides, fault case libraries, and expert experience summaries.

[0030] By adopting the above technical solution, given the high level of specialization and complexity of data in the heating industry, a first vector knowledge base is constructed to facilitate the large language model's accurate understanding of the database based on the first vector knowledge base, and then to accurately locate the data results based on the first similar segment that matches the data lookup intention.

[0031] Because general-purpose large language models have weak professional knowledge capabilities in the heating field and lack a deep understanding of heating industry-specific terminology, business processes, and industry rules, a second vector knowledge base is constructed. By matching second similar segments according to the data lookup intent and analysis intent, the large language model can combine the professional knowledge of the second similar segments to intelligently gain insights into the data results. This enables the analysis results to have the professionalism of the heating industry, avoids problems such as misunderstanding bias or reasoning errors in the analysis process of the large language model, and also realizes the efficient utilization of heating professional knowledge.

[0032] In one specific implementation, the advanced report is an editable advanced report that supports selecting multiple export formats and downloading.

[0033] By adopting the above technical solution, after the advanced report is generated, an editable format is automatically created. Technical personnel can manually review the advanced report again and directly modify any discrepancies or other issues without regenerating a new advanced report, ensuring its accuracy. The advanced report supports one-click export to multiple formats commonly used in the heating industry, eliminating the need for third-party conversion tools and thus meeting the needs of different usage scenarios.

[0034] Secondly, this application provides an advanced report generation system for the heating industry based on a large language model, employing the following technical solution:

[0035] A high-level report generation system for the heating industry based on a large language model includes:

[0036] The outline creation module is used to create report outlines, including customizing report outlines or selecting preset outline templates; and setting several sets of data lookup intents and analysis intents under the report outline directory headings;

[0037] The content generation module is used to call the data analysis module sequentially to generate the content under the directory headings according to the report outline directory order and based on the grouped data lookup intentions and analysis intentions under the directory headings;

[0038] The data analysis module is used to match the first similar segment in a pre-built first vector knowledge base based on the search intent under the current directory title using retrieval enhancement generation technology, and to generate SQL based on the search intent, the first similar segment and prompt words using a large language model, and then obtain the query data results from the database through SQL query;

[0039] And by using retrieval enhancement generation technology, based on the search intent and the corresponding analysis intent under the current directory title, a second similar segment is matched in a pre-built second vector knowledge base, and a large language model is used to intelligently gain insights into the data results based on the search intent, the analysis intent and the second similar segment to obtain analysis conclusions;

[0040] The content generation module is also used to use the data results and the analysis conclusions as content under the current directory title based on the data lookup intent and the analysis intent of that group;

[0041] The report generation module is used to integrate the titles and contents of the report outline at all levels in the order of the report outline directory using a large language model, and then generate advanced reports after polishing.

[0042] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed, as described above, a method for generating advanced reports for the heating industry based on a large language model.

[0043] In summary, this application includes at least one of the following beneficial technical effects:

[0044] By creating a report outline and then setting several sets of data lookup and analysis intentions, the large language model can accurately identify user needs and generate customized report content. This realizes the entire process from data acquisition to analysis and insight to report generation, ensuring that the generated report content closely matches user needs. It ensures that the generated advanced reports are highly professional, of high quality, and use consistent professional terminology, helping users to quickly respond to changes in working conditions and improve decision-making efficiency.

[0045] It can set timed generation rules for the outline templates of commonly used advanced reports, and automatically execute the generation process of advanced reports according to the set generation cycle and generation time, which greatly improves efficiency, further realizes the standardization and normalization of advanced reports, facilitates subsequent management, and helps to make timely decisions based on advanced reports.

[0046] By constructing a first vector knowledge base and a second vector knowledge base, the large language model can accurately understand the database based on the first vector knowledge base and then accurately query the data results. It also enables the large language model to combine the professional knowledge of the second vector knowledge base to make intelligent insights into the data results, avoiding problems such as misunderstanding bias or reasoning errors in the analysis process.

[0047] By automatically generating editable advanced reports, it is possible to manually review the advanced reports again and make direct modifications when content discrepancies occur, without having to regenerate new advanced reports, thus ensuring the accuracy of the advanced reports. Attached Figure Description

[0048] Figure 1 This is a flowchart of an embodiment of the advanced report generation method for the heating industry based on a large language model in this application;

[0049] Figure 2 This is a flowchart illustrating how report content is generated based on data lookup intent and analysis intent in an embodiment of this application. Detailed Implementation

[0050] The following is in conjunction with the appendix Figure 1-2 This application will be described in further detail.

[0051] This application discloses a method and system for generating advanced reports for the heating industry based on a large language model.

[0052] Reference Figure 1 A method for generating advanced reports for the heating industry based on a large language model includes the following steps:

[0053] S100: Create a report outline, including a custom report outline or a preset outline template.

[0054] Specifically, when creating a report outline, you set the title of the advanced report and choose either a custom report outline or a preset outline template. A custom report outline includes adding several levels of report outline directories and setting several report outline directory titles under each level.

[0055] After selecting a preset outline template, choose a suitable outline template from the pre-built template library. The selected outline template will automatically display the outline headings at all levels. It also supports adding or deleting outline headings at all levels of the selected outline template, as well as modifying the outline headings at all levels.

[0056] The template library includes several pre-set outline templates, each tailored to a specific application scenario. It also supports adding custom report outlines to the library. Specifically, based on application scenarios, the library includes templates customized for heating companies, such as annual or monthly operation report templates for a particular heating company; it also includes frequently used reports in the heating industry, such as energy consumption statistical analysis reports, heating system operation summary reports, equipment anomaly and fault diagnosis analysis reports, etc.

[0057] S200: Set several sets of data lookup intents and analysis intents under the report outline directory headings.

[0058] Specifically, several sets of data lookup intentions and analysis intentions are set under the lowest-level directory headings of the report outline. Data lookup intentions include how to query data and which part of the data to query for generating the report content under the current directory heading. Analysis intentions include how to analyze the queried data for generating the report content under the current directory heading, and the desired analysis results. In this embodiment, the report outline directory is set up with three levels, and several sets of data lookup intentions and analysis intentions are set under each third-level report outline directory heading.

[0059] When setting a query intent, users can select an appropriate query question from a pre-built question library as the current query intent. Creating a question library involves setting several data query questions within it, and users can also add pre-defined query intents to the question library.

[0060] Furthermore, after selecting to use a preset outline template, the system will automatically call up the outline headings at all levels of the selected outline template, as well as several sets of lookup intents and analysis intents set under the lowest level heading. It also supports adding, deleting, and modifying the lookup intents and analysis intents of the current outline template.

[0061] S300: Generates the content under the current lowest-level directory heading according to the report outline table of contents and based on the grouped lookup and analysis intentions under the lowest-level directory heading. Specifically, this includes:

[0062] S301: Using retrieval enhancement generation technology, the first similar segment is matched in the pre-built first vector knowledge base based on the search intent under the current lowest level directory title. The SQL is generated by the large language model based on the search intent, the first similar segment and the prompt words. The query results are then obtained from the database through the SQL query.

[0063] Specifically, constructing the first vector knowledge base includes defining the structure of several tables in the database to be queried. Each table structure definition sets the relevant query scenarios, relevant query questions, and table structure definitions. The content of the table structure definition documents is segmented and vectorized to construct the first vector knowledge base for subsequent retrieval and use.

[0064] We use pre-trained language models such as BERT and Sentence-BERT or embedding models such as Embedding to perform semantic parsing on the query intent, extract key features of the query intent and contextual information, and generate a high-dimensional vector representation. Then, we use the cosine vector similarity calculation method in RAG technology to retrieve the first vector knowledge base of SQL, and use a hybrid retrieval strategy to perform efficient matching and quickly locate the first similar segment that is closest to the user's query question.

[0065] The query intent, the first similar segment, and the prompt word are sent to the large language model. The large language model generates SQL based on the query intent, the first similar segment, and the prompt word, and then retrieves the query results from the database through the SQL query.

[0066] S302: Using retrieval enhancement generation technology, based on the search intent and corresponding analysis intent under the current directory title, a second similar segment is matched in a pre-built second vector knowledge base. The data results are then intelligently analyzed and conclusions are drawn based on the search intent, analysis intent, and second similar segment using a large language model.

[0067] The construction of the second vector knowledge base involves collecting and organizing knowledge documents related to heating, including heating industry standards, heating theoretical knowledge, system operation manuals, equipment maintenance methods, heating platform operation manuals, equipment technical manuals, operating procedures, troubleshooting guides, fault case libraries, and expert experience summaries. The content of these knowledge documents is then segmented and vectorized to construct the second vector knowledge base for subsequent retrieval and use.

[0068] Specifically, pre-trained language models such as BERT and Sentence-BERT, or embedding models such as Embedding, are used to perform semantic parsing on the query intent and the corresponding analysis intent, extracting key features of the query intent, analysis intent and contextual information, and generating high-dimensional vector representations. Then, the cosine vector similarity calculation method in RAG technology is used to retrieve the second vector knowledge base, and a hybrid retrieval strategy is used for efficient matching to quickly locate the second similar segment that is closest to the user's query question.

[0069] By combining a large language model with second-similar segments, intelligent insights are gained from the data results to obtain analytical conclusions. Specifically, this includes using a large language model to combine knowledge referenced in second-similar segments to perform intelligent inference on the data results and generate professional analytical conclusions.

[0070] S303: Generate visualization charts based on the data results.

[0071] Specifically, the data visualization engine automatically selects and generates appropriate chart types based on data type, such as line charts, area charts, pie charts, donut charts, scatter plots, and heatmaps. Simultaneously, the charts automatically label key industry thresholds and outliers, improving data readability.

[0072] S304: Include data results, analysis conclusions, and visualizations as content under the current directory heading based on the group's data lookup and analysis intent.

[0073] By integrating data results, analysis conclusions, and visualization charts through a large language model, the content under the current directory title is generated based on the data search intent and analysis intent according to the logical framework of "data search intent - problem explanation - data presentation - analysis and interpretation - suggested solutions".

[0074] In addition, the content generated under the current directory title based on the group of data query intents and analysis intents is previewable, and users can modify the current data query intent and analysis intent multiple times based on the previewable content and regenerate the content to ensure that the report content meets the user's needs.

[0075] S400: Summarizes all content under the headings of the report outlines at all levels using a large language model, and generates summary content under the headings of the current report outline.

[0076] Specifically, the hierarchical structure of the report outline is identified through a large language model, the headings of each level are extracted, and the content under each level heading is analyzed layer by layer from the lowest level to the highest level. All the content generated under each level heading of the report outline is summarized, and summary content is generated under the current heading of the report outline, so that technical personnel can intuitively understand the overall situation of the content under the current heading of the report outline.

[0077] S500: Through a large language model, the titles and contents of the report outlines at all levels are integrated sequentially according to the order of the report outline directory, and then a high-level report is generated after polishing.

[0078] Specifically, the large language model integrates the report outline headings from the lowest to the highest level, gradually piecing together the headings and the content generated under each heading. Then, the pieced content is intelligently polished, and a rich text editor is integrated to generate an advanced report that can be edited a second time, so that technicians can manually review it and directly modify any discrepancies or other issues.

[0079] Advanced reports support the selection and download of various export formats commonly used in the heating industry, such as PPT, Word, and PDF.

[0080] In addition, it supports selecting preset or saved outline templates and setting regular generation rules, as well as generating new advanced reports based on the latest data at the time, and automatically pushing them out in a preset manner.

[0081] Specifically, select a preset or saved outline template and set regular generation rules. Regular generation rules include the generation cycle set by hour / day / week / month and the generation time point set. Set the push method. The system will automatically generate scheduled tasks according to the set regular generation rules and push method, and trigger the scheduled tasks at the set cycle and time point to generate new advanced reports based on the latest data in the database. Then, the new advanced reports will be automatically pushed through the set push method.

[0082] The implementation principle of the advanced report generation method for the heating industry based on a large language model in this application embodiment is as follows: Technicians create report outlines by customizing them or selecting preset outline templates, and independently set several sets of data query intentions and analysis intentions so that the large language model can accurately identify the technicians' needs for report content and generate customized report content. The large language model accurately executes data queries based on the data query intentions, avoiding manual data summarization and querying by technicians or data entry errors. Furthermore, the large language model accurately performs insight analysis based on the analysis intentions, which not only ensures that the generated report content closely matches user needs, but also avoids time-consuming and labor-intensive manual writing. At the same time, it ensures that the generated advanced reports are highly professional, of high quality, and use consistent professional terminology. The report documents are standardized and normalized, thus realizing the entire process from data acquisition to analysis and insight to report generation, and compressing the report generation process from hours or even days to minutes, helping technicians and leaders to quickly respond to changes in operating conditions and improve decision-making efficiency.

[0083] It should be understood that, although Figure 1 and Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows; unless explicitly stated otherwise, there is no strict order requirement for the execution of these steps, and they can be executed in other orders; and Figure 1 and Figure 2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0084] This application also discloses an advanced report generation system for the heating industry based on a large language model.

[0085] A high-level report generation system for the heating industry based on a large language model includes:

[0086] The outline creation module is used to create report outlines, including customizing report outlines or selecting preset outline templates; and setting several sets of data lookup intents and analysis intents under the report outline directory headings.

[0087] Specifically, when creating a report outline, you set the title of the advanced report and choose either a custom report outline or a preset outline template. A custom report outline includes adding several levels of report outline directories and setting several report outline directory titles under each level.

[0088] After selecting a preset outline template, choose a suitable outline template from the pre-built template library. The selected outline template will automatically display the outline headings at all levels. It also supports adding or deleting outline headings at all levels of the selected outline template, as well as modifying the outline headings at all levels.

[0089] The template library includes several pre-set outline templates, each tailored to a specific application scenario. It also supports adding custom report outlines to the library. Specifically, based on application scenarios, the library includes templates customized for heating companies, such as annual or monthly operation report templates for a particular heating company; it also includes frequently used reports in the heating industry, such as energy consumption statistical analysis reports, heating system operation summary reports, equipment anomaly and fault diagnosis analysis reports, etc.

[0090] Specifically, several sets of data lookup intentions and analysis intentions are set under the lowest-level directory headings of the report outline. Data lookup intentions include how to query data and which part of the data to query for generating the report content under the current directory heading. Analysis intentions include how to analyze the queried data for generating the report content under the current directory heading, and the desired analysis results. In this embodiment, the report outline directory is set up with three levels, and several sets of data lookup intentions and analysis intentions are set under each third-level report outline directory heading.

[0091] When setting a query intent, users can select an appropriate query question from a pre-built question library as the current query intent. Creating a question library involves setting several data query questions within it, and users can also add pre-defined query intents to the question library.

[0092] Furthermore, after selecting to use a preset outline template, the system will automatically call up the outline headings at all levels of the selected outline template, as well as several sets of lookup intents and analysis intents set under the lowest level heading. It also supports adding, deleting, and modifying the lookup intents and analysis intents of the current outline template.

[0093] The content generation module is used to generate content under the current lowest-level directory title in sequence according to the report outline directory order and based on the grouped data lookup intentions and analysis intentions under the lowest-level directory title.

[0094] Specifically, the content generation module is also used to match the first similar segment in the pre-built first vector knowledge base based on the search intent under the current directory title through retrieval enhancement generation technology, and generate SQL based on the search intent, the first similar segment and prompt words through a large language model, and then obtain the query data results from the database through SQL query.

[0095] Furthermore, constructing the first vector knowledge base includes defining the structure of several tables in the database to be queried. Each table structure definition specifies the relevant query scenarios, related query questions, and table structure definitions. The content of the table structure definition documents is segmented and vectorized to construct the first vector knowledge base for subsequent retrieval and use.

[0096] We use pre-trained language models such as BERT and Sentence-BERT or embedding models such as Embedding to perform semantic parsing on the query intent, extract key features of the query intent and contextual information, and generate a high-dimensional vector representation. Then, we use the cosine vector similarity calculation method in RAG technology to retrieve the first vector knowledge base of SQL, and use a hybrid retrieval strategy to perform efficient matching and quickly locate the first similar segment that is closest to the user's query question.

[0097] The query intent, the first similar segment, and the prompt word are sent to the large language model. The large language model generates SQL based on the query intent, the first similar segment, and the prompt word, and then retrieves the query results from the database through the SQL query.

[0098] Specifically, the content generation module uses retrieval-enhanced generation technology to match second similar segments in a pre-built second vector knowledge base based on the search intent and corresponding analysis intent under the current directory title. It also uses a large language model to intelligently analyze the data results based on the search intent, analysis intent, and second similar segments to obtain analytical conclusions.

[0099] The construction of the second vector knowledge base involves collecting and organizing knowledge documents related to heating, including heating industry standards, heating theoretical knowledge, system operation manuals, equipment maintenance methods, heating platform operation manuals, equipment technical manuals, operating procedures, troubleshooting guides, fault case libraries, and expert experience summaries. The content of these knowledge documents is then segmented and vectorized to construct the second vector knowledge base for subsequent retrieval and use.

[0100] Furthermore, pre-trained language models such as BERT and Sentence-BERT, or embedding models such as Embedding, are used to perform semantic parsing on the query intent and the corresponding analysis intent, extracting key features of the query intent, analysis intent, and contextual information, and generating high-dimensional vector representations. Then, the cosine vector similarity calculation method in RAG technology is used to retrieve the second vector knowledge base, and a hybrid retrieval strategy is used for efficient matching to quickly locate the second similar segment that is closest to the user's query question.

[0101] By combining a large language model with second-similar segments, intelligent insights are gained from the data results to obtain analytical conclusions. Specifically, this includes using a large language model to combine knowledge referenced in second-similar segments to perform intelligent inference on the data results and generate professional analytical conclusions.

[0102] Specifically, the content generation module is also used to generate visual charts based on the data results.

[0103] Furthermore, the data visualization engine automatically selects and generates appropriate chart types based on data type, such as line charts, area charts, pie charts, donut charts, scatter plots, and heatmaps. Simultaneously, the charts automatically label key industry thresholds and outliers, improving data readability.

[0104] The content generation module is also used to convert data results, analysis conclusions, and visualization charts into content under the current directory title based on the group's data lookup and analysis intent.

[0105] Specifically, the data results, analysis conclusions, and visualization charts are integrated through a large language model, and the content under the current directory title is generated based on the group of data search intentions and analysis intentions according to the logical framework of "data search intention - problem explanation - data presentation - analysis and interpretation - suggested solutions".

[0106] In addition, the content generated under the current directory title based on the group of data query intents and analysis intents is previewable, and users can modify the current data query intent and analysis intent multiple times based on the previewable content and regenerate the content to ensure that the report content meets the user's needs.

[0107] The content generation module is also used to summarize all content under the headings of the report outlines at all levels using a large language model, and to generate summary content under the headings of the current report outline.

[0108] Specifically, the hierarchical structure of the report outline is identified through a large language model, the headings of each level are extracted, and the content under each level heading is analyzed layer by layer from the lowest level to the highest level. All the content generated under each level heading of the report outline is summarized, and summary content is generated under the current heading of the report outline, so that technical personnel can intuitively understand the overall situation of the content under the current heading of the report outline.

[0109] The report generation module is used to integrate the titles and contents of the report outline at all levels in the order of the report outline directory using a large language model, and then generate advanced reports after polishing.

[0110] Specifically, the large language model integrates the report outline headings from the lowest to the highest level, gradually piecing together the headings and the content generated under each heading. Then, the pieced content is intelligently polished, and a rich text editor is integrated to generate an advanced report that can be edited a second time, so that technicians can manually review it and directly modify any discrepancies or other issues.

[0111] Advanced reports support the selection and download of various export formats commonly used in the heating industry, such as PPT, Word, and PDF.

[0112] In addition, it supports selecting preset or saved outline templates and setting regular generation rules, as well as generating new advanced reports based on the latest data at the time, and automatically pushing them out in a preset manner.

[0113] Specifically, select a preset or saved outline template and set regular generation rules. Regular generation rules include the generation cycle set by hour / day / week / month and the generation time point set. Set the push method. The system will automatically generate scheduled tasks according to the set regular generation rules and push method, and trigger the scheduled tasks at the set cycle and time point to generate new advanced reports based on the latest data in the database. Then, the new advanced reports will be automatically pushed through the set push method.

[0114] This application also discloses a computer-readable storage medium.

[0115] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed, such as the aforementioned advanced report generation method for the heating industry based on a large language model. The computer-readable storage medium includes, for example, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for generating advanced reports for the heating industry based on a large language model, characterized in that, Includes the following steps: Create a report outline, including a custom report outline or by selecting a preset outline template; Under the table of contents of the report outline, set several sets of data query intents and analysis intents. The data query intents include how to query data and which part of the data to query when generating the report content under the current table of contents. The analysis intents include how to analyze the queried data when generating the report content under the current table of contents, and the desired analysis results. The content under each directory title is generated sequentially according to the report outline directory order and based on the grouped data lookup intent and analysis intent under the directory title. The search enhancement generation technology matches the first similar segment in the pre-built first vector knowledge base based on the search intent under the current directory title, and generates SQL based on the search intent, the first similar segment and prompt words through a large language model. Then, the query results are obtained from the database through the SQL query. By using retrieval enhancement generation technology, based on the data search intent and the corresponding analysis intent under the current directory title, a second similar segment is matched in a pre-built second vector knowledge base. Then, a large language model is used to intelligently gain insights into the data results based on the data search intent, the analysis intent, and the second similar segment to obtain analysis conclusions. The data results and analysis conclusions will be used as content under the current directory heading based on the data lookup intent and analysis intent of that group. The large language model integrates the titles and contents of the report outlines at all levels in the order of the report outline directory, and then polishes them to generate an advanced report.

2. The method for generating advanced reports for the heating industry based on a large language model according to claim 1, characterized in that, The system generates previewable content under each directory heading according to the report outline directory order and based on the grouped data lookup intentions and analysis intentions under the directory headings. It also supports users to modify the current data lookup intention and analysis intention multiple times based on the previewable content and regenerate the content.

3. The method for generating advanced reports for the heating industry based on a large language model according to claim 1, characterized in that, After taking the data results and analysis conclusions as the content under the current directory title based on the data lookup intent and the analysis intent, the method further includes summarizing all the content under the directory titles of each level of report outline through a large language model, and generating the summary content under the current report outline directory title.

4. The method for generating advanced reports for the heating industry based on a large language model according to claim 1, characterized in that, The method of setting several sets of data lookup and analysis intentions under the table of contents of the report outline includes setting several sets of data lookup and analysis intentions under the lowest level table of contents of the report outline.

5. The method for generating advanced reports for the heating industry based on a large language model according to claim 1, characterized in that, Also includes: It supports selecting preset or saved outline templates and setting regular generation rules, as well as generating new advanced reports based on the latest data at the time, and automatically pushing them according to preset methods; The periodic generation rules include generation cycles set by hour / day / week / month and generation time points.

6. The method for generating advanced reports for the heating industry based on a large language model according to claim 1, characterized in that, It also includes the establishment of a question library, which includes several data query questions. It supports adding pre-defined data query intentions to the question library and allows users to select data query questions from the question library as data query intentions when setting them.

7. The method for generating advanced reports for the heating industry based on a large language model according to claim 1, characterized in that, It also includes the creation of a template library, which includes several preset outline templates and supports adding custom report outlines to the template library. The outline templates are preset according to the application scenario.

8. The method for generating advanced reports for the heating industry based on a large language model according to claim 1, characterized in that, Constructing the first vector knowledge base includes vectorizing the database table structure description document; The construction of the second vector knowledge base involves vectorizing unstructured knowledge documents composed of heating-related standards, heating industry literature, heating theory, system operation manuals, equipment maintenance methods, heating platform operation manuals, equipment technical manuals, operating procedures, troubleshooting guides, fault case libraries, and expert experience summaries.

9. The method for generating advanced reports for the heating industry based on a large language model according to claim 1, characterized in that, The advanced report is an editable report that supports multiple export formats and downloads.

10. A high-level report generation system for the heating industry based on a large language model, characterized in that, include: The outline creation module is used to create report outlines, including custom report outlines or selecting preset outline templates; And set several sets of data query intents and analysis intents under the report outline directory title. The data query intents include how to query data and which part of the data to query when generating report content under the current directory title. The analysis intents include how to analyze the queried data when generating report content under the current directory title and the desired analysis results. The content generation module is used to call the data analysis module sequentially to generate the content under the directory headings according to the report outline directory order and based on the grouped data lookup intentions and analysis intentions under the directory headings; The data analysis module is used to match the first similar segment in a pre-built first vector knowledge base based on the search intent under the current directory title using retrieval enhancement generation technology, and to generate SQL based on the search intent, the first similar segment and prompt words using a large language model, and then obtain the query data results from the database through SQL query; And by using retrieval enhancement generation technology, based on the search intent and the corresponding analysis intent under the current directory title, a second similar segment is matched in a pre-built second vector knowledge base, and a large language model is used to intelligently gain insights into the data results based on the search intent, the analysis intent and the second similar segment to obtain analysis conclusions; The content generation module is also used to use the data results and the analysis conclusions as content under the current directory title based on the data lookup intent and the analysis intent of that group; The report generation module is used to integrate the titles and contents of the report outline at all levels in the order of the report outline directory using a large language model, and then generate advanced reports after polishing.

Images