Knowledge enhancement method and system for data insight of heating industry based on large language model

Abstract

The application relates to a knowledge enhancement method for heat supply industry data insight based on a large language model, relates to the field of application of a heat supply industry large model, and comprises constructing a database, a SQL example question knowledge base, a database table structure knowledge base and a heat supply professional document knowledge base; first associated information is matched in the SQL example question knowledge base and the database table structure knowledge base based on a user query question, and second associated information is matched in the heat supply professional document knowledge base; enhanced prompt words are obtained by fusing the user query question, the first associated information and preset prompt words, SQL is generated through the large language model, and data results are obtained in the database through SQL query; intelligent insight is performed on the data results in combination with the second associated information to obtain analysis conclusions; and answers are generated by integrating the data results and the analysis conclusions. The application enables the large language model technology to deeply understand heat supply system professional knowledge and deeply analyze heat supply system data, and realizes accurate data query, data analysis and decision support.

Description

Technical Field

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

[0002] Currently, heating system data is stored in multiple source structured databases, exhibiting diverse data types, complex business coupling, and high real-time operational requirements. Due to the high level of specialization and complexity of heating industry data, and the weak professional knowledge capabilities of existing general-purpose language models in the heating field, lacking a deep understanding of industry-specific terminology, business processes, and rules, these models struggle to understand the user's true intent during data queries, failing to accurately locate query results. Furthermore, they are prone to misunderstandings or inference errors during data analysis. Summary of the Invention

[0003] In order to enable large language model technology to deeply understand the professional knowledge of heating systems, deeply analyze heating system data, quickly mobilize system functions through colloquial expressions, and realize accurate querying of heating system data, as well as data analysis and decision support, this application provides a knowledge enhancement method and system for heating industry data insight based on large language model.

[0004] Firstly, this application provides a knowledge enhancement method for data insights in the heating industry based on a large language model, employing the following technical solution:

[0005] A knowledge enhancement method for heating industry data insights based on a large language model includes the following steps:

[0006] Access real-time sensing data and business operation data and build a database;

[0007] Build a knowledge base for SQL example questions, a knowledge base for database table structures, and a knowledge base for heating industry documents;

[0008] Receive user queries;

[0009] Based on the user query question, retrieve the first matching related information from the SQL example question knowledge base and the database table structure knowledge base;

[0010] The enhanced prompt words are obtained by fusing the user query question, the first associated information, and the preset prompt words;

[0011] SQL is generated based on the enhanced prompt words using a large language model, and then the query results are obtained from the database using the SQL query.

[0012] Based on the user's query, retrieve matching second related information from the heating professional document knowledge base;

[0013] By combining the large language model with the second association information, intelligent insights are gained into the data results to obtain analytical conclusions.

[0014] The results of the data and the conclusions of the analysis are combined to generate an answer.

[0015] By adopting the above technical solution, real-time sensing data and business operation data of the heating system are stored in a database. A knowledge base for SQL example questions and a knowledge base for database table structure are constructed for data retrieval. Based on the user's query, the system searches and matches the SQL example question knowledge base and the database table structure knowledge base to obtain the first related information. Enhanced prompts are obtained by fusing the user's query with the first related information and preset prompt words, thus supplementing the completeness of the user's query intent. SQL is generated based on the enhanced prompts using a large language model, ensuring that the SQL generated by the large language model accurately reflects the user's query intent. Finally, the query results are retrieved from the database through SQL queries, effectively improving the accuracy of the query results.

[0016] Due to the high level of specialization and complexity of data in the heating industry, by constructing an SQL example question knowledge base and a database table structure knowledge base, the first association information matched in the SQL example question knowledge base and the database table structure knowledge base can play a linking role between the user's query question and the corresponding data result in the database. This facilitates the large language model to accurately understand the database based on the knowledge in the SQL example question knowledge base and the database table structure knowledge base, thereby accurately locating the data result corresponding to the user's query question.

[0017] 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 heating professional document knowledge base is constructed. This knowledge base is then matched with secondary related information based on user queries. This allows the large language model to intelligently analyze data results by incorporating the professional knowledge within the secondary related information. The secondary related information includes professional knowledge related to the user's query, summarizing relevant information from the heating professional document knowledge base. This ensures the analysis results possess the professionalism of the heating industry, avoiding misunderstandings or reasoning errors that can occur with the large language model during analysis, and achieving efficient utilization of heating professional knowledge.

[0018] Based on the user's query in natural language, the above process integrates data results and analysis conclusions to generate an answer and provide feedback to the user. This upgrades the system from traditional intelligent querying to in-depth data insights, breaks down data barriers between multiple business platforms, and enables non-technical personnel to obtain the required data results and analysis conclusions through natural language queries. The clear data explanation and presentation process helps users better understand the analysis conclusions, enabling them to efficiently complete data analysis and decision-making.

[0019] In one specific implementation scheme, the construction of the SQL example question knowledge base includes:

[0020] Several SQL examples and a group of similar questions for each SQL example are set up according to query type, query subject, query content, organizational structure, time range, data type, data granularity, and analysis type.

[0021] By adopting the above technical solution, several SQL examples and similar question groups corresponding to each SQL example are set. The first association information can be matched in the similar question group according to the user's query question. This allows the large model to accurately generate the corresponding SQL based on the questions in the similar question group and the corresponding SQL examples contained in the first association information, thereby effectively improving the accuracy of the data results obtained from the SQL query.

[0022] In a specific implementation scheme, the similar question group corresponding to each SQL example includes several questions from different people's perspectives, several questions formed by different levels of specialization, and several questions after synonym replacement.

[0023] By adopting the above technical solution, the similar question group corresponding to each SQL example can cover the questions queried by users with different roles, users with different levels of expertise, and user questions that have been replaced with synonyms, thereby improving the recall accuracy and ensuring that the most relevant key information can be retrieved.

[0024] In one specific implementation scheme, the construction of the database table structure knowledge base includes:

[0025] The database table structure is defined according to the query content;

[0026] For each table structure description, define the relevant query scenarios, relevant query questions, and table structure definitions.

[0027] By adopting the above technical solution, the corresponding first association information is matched in the relevant query scenarios, related query questions, and data table structure definitions set in several table structure descriptions based on the user's query question. This allows for the accurate location of the corresponding data table in the database involved in the user's query question based on the table structure description content in the first association information, and clarifies the fields and business attributes in the data table. As a result, the large model can accurately generate SQL based on the enhanced prompts containing the first association information, and accurately locate the corresponding data table based on the SQL query for querying data results, further improving the accuracy of the query data results.

[0028] In a specific implementation plan, the plan also includes pre-setting or allowing users to define several keywords for different query content, and supporting users to select multiple keywords to construct the user query question.

[0029] By adopting the above technical solution, several keywords can be preset for different query content or several keywords can be defined by the user. When entering a user query question, the user can select multiple keywords to concatenate, which not only facilitates the quick input of the user query question, but also ensures that the user query question is within the query range of the database.

[0030] In a specific implementation scheme, the step of pre-setting or user-defined keywords for different query contents includes:

[0031] Several query categories can be preset or user-defined based on different query content;

[0032] Each query category can be pre-defined or user-defined with several keyword categories;

[0033] Each keyword category can be pre-set or user-defined with several keywords.

[0034] By adopting the above technical solution, several query categories are first preset according to different query content, or several query categories are defined by the user. Then, several keyword categories are preset or defined by the user under each query category. Finally, several keywords are preset or defined by the user under each keyword category. When entering a query question, the user first selects the corresponding query category according to the query content, and then selects the corresponding keywords for different keyword categories under the selected query category. This allows the keywords under multiple keyword categories to be combined to form the user's query question, effectively improving the accuracy of the user's query question when selecting keywords to combine into an input query question.

[0035] In one specific implementation plan, the construction of the heating industry document knowledge base includes:

[0036] Unstructured knowledge documents comprised 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.

[0037] By adopting the above technical solution, unstructured knowledge documents of the heating industry are integrated to build a heating professional document knowledge base. This enables the large language model to refer to professional heating industry knowledge and generate analysis conclusions during the process of intelligently insighting data results, thereby realizing the effective utilization of unstructured professional knowledge of the heating industry.

[0038] In a specific feasible implementation, it also includes:

[0039] Construct corresponding SQL example question vector knowledge bases, database table structure vector knowledge bases, and heating professional document vector knowledge bases based on the SQL example question knowledge base, database table structure vector knowledge base, and heating professional document vector knowledge base, respectively.

[0040] The step of retrieving matching first related information from the SQL example question knowledge base and the database table structure knowledge base based on the user query question includes:

[0041] The user query question is vectorized, and a first similar fragment is matched in the SQL example question vector knowledge base and the database table structure vector knowledge base using RAG technology. The first similar fragment is the first associated information.

[0042] The step of retrieving matching second related information from the heating industry document knowledge base based on the user query includes:

[0043] The user query question is vectorized, and a second similar fragment is matched in the heating professional document vector knowledge base using RAG technology. The second similar fragment is the second associated information.

[0044] In one specific implementation scheme, the method further includes generating a visualization chart based on the data results, rendering a table based on the data results to generate an exportable table file, and integrating the data results, the visualization chart, the table file, and the analysis conclusions to generate an answer.

[0045] Secondly, this application provides a knowledge enhancement system for data insights in the heating industry based on a large language model, employing the following technical solution:

[0046] A knowledge enhancement system for data insights in the heating industry based on a large language model, characterized by comprising:

[0047] The database management module is used to access multimodal real-time sensing data and business operation data and build a database;

[0048] The knowledge base management module is used to build a knowledge base for SQL example questions, a knowledge base for database table structures, and a knowledge base for heating-related professional documents.

[0049] The user interaction module is used to receive user queries.

[0050] The retrieval enhancement module is used to retrieve matching first related information from the SQL example question knowledge base and the database table structure knowledge base based on the user query question;

[0051] The enhanced prompt words are obtained by fusing the user query question, the first associated information, and the preset prompt words; and

[0052] Based on the user's query, retrieve matching second related information from the heating professional document knowledge base;

[0053] The data query module is used to generate SQL based on the enhanced prompt words using a large language model, and then use the SQL to query the database to obtain the query results.

[0054] The intelligent insight module is used to perform intelligent insight analysis on the data results by combining the large language model with the second correlation information to obtain analytical conclusions;

[0055] The answer generation module is used to integrate the data results and the analysis conclusions to generate an answer.

[0056] Thirdly, this application provides a computer-readable storage medium, which adopts the following technical solution:

[0057] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed, such as the aforementioned knowledge enhancement method for data insights in the heating industry based on a large language model.

[0058] In summary, this application includes at least one of the following beneficial technical effects:

[0059] This application utilizes the existing information platform and data of the heating company, taking data-driven approaches as its starting point. Through the understanding and generation capabilities of natural language using a large language model, it calls upon the core database of the entire network to upgrade from traditional intelligent queries to in-depth data insights. This breaks through the data barriers of multiple business platforms, enabling non-technical personnel to obtain the required data and analysis results through natural language queries. Furthermore, clear data explanations and visualizations help users better understand the analysis results, empowering users to efficiently complete data analysis and decision-making.

[0060] By building a knowledge base of SQL example questions and a knowledge base of database table structure, the large language model can accurately generate SQL statements that meet user needs, and the SQL statements conform to the table structure logic of the heating industry database, ensuring the accuracy of data query results.

[0061] By building a knowledge base of heating professional documents, large language models can conduct in-depth analysis of data results based on a deep understanding of the professional knowledge of heating systems, avoiding comprehension biases or reasoning errors in large models and ensuring the accuracy and professionalism of analysis results.

[0062] By using preset keywords and supporting user-defined keywords, users can input their queries by selecting multiple keywords, which not only improves interaction efficiency but also ensures that the user's query falls within the scope of the database. Attached Figure Description

[0063] Figure 1 This is a schematic diagram illustrating the steps of the knowledge enhancement method for data insights in the heating industry based on a large language model, as described in this application embodiment.

[0064] Figure 2 This is a flowchart illustrating the knowledge enhancement method for data insights in the heating industry based on a large language model, as described in this application. Detailed Implementation

[0065] The following is in conjunction with the appendix Figure 1-2 This application will be described in further detail.

[0066] This application discloses a knowledge enhancement method for data insights in the heating industry based on a large language model.

[0067] Reference Figure 1 and Figure 2 A knowledge enhancement method for data insights in the heating industry based on large language models includes the following steps:

[0068] S100: Integrates real-time sensing data and business operation data and builds a database, specifically including:

[0069] Real-time sensing data includes time-series data (temperature, pressure, flow rate, heat, etc.) from multiple sensors at various levels, including heat sources, pipelines, heat exchange stations, and user terminals. Operational data includes structured data such as user information, equipment ledgers, work order management, energy consumption metering, and billing records. The acquisition of real-time sensing data and operational data forms raw data. This raw data is then cleaned to eliminate noise and errors, and standardized and normalized before being stored in a database.

[0070] S200: Construct a knowledge base for SQL example questions, a knowledge base for database table structures, and a knowledge base for heating industry-related documents, specifically including:

[0071] S201: Construct a database table structure knowledge base, including:

[0072] First, define the structure of several tables in the database according to the query content.

[0073] Specifically, the query content includes basic information of the generating unit, basic information of the heat exchange station, basic information of the organization, basic information of the heat source, real-time data of the generating unit, historical data of the generating unit, weather forecast data, summary data of the generating unit, basic information of the building valve, real-time data of the building valve, historical data of the building valve, building valve strategy information, basic information of the household valve, real-time data of the household valve, historical data of the household valve, household valve strategy information, real-time data of the room temperature, and historical data of the room temperature. The structure descriptions for the following tables are set according to the query content: basic information of the generating unit, basic information of the heat exchange station, basic information of the organization, basic information of the heat source, real-time data of the generating unit, historical data of the generating unit, weather forecast data, summary data of the generating unit, basic information of the building valve, real-time data of the building valve, historical data of the building valve, strategy information of the building valve, basic information of the household valve, real-time data of the household valve, historical data of the household valve, strategy information of the household valve, real-time data of the household valve, and historical data of the household valve.

[0074] Then, for each table structure description, set the relevant query scenarios, relevant query questions, and table structure definitions.

[0075] Specifically, the structure description of the unit basic information table includes relevant query scenarios such as the heat exchange station to which the unit belongs, heating type, whether it is an energy-saving building, and the organization to which it belongs; relevant query questions include querying the number or list of units under a certain organization, querying the list or number of units that meet the query conditions, and querying the heat exchange station, heating type, whether it is an energy-saving building, and the organization to which a certain unit belongs; the table structure definition includes an explanation of the structure of the unit basic information table in the order of table name, field name, data type, and field description.

[0076] The description of the basic information table structure for heat exchange stations includes relevant query scenarios such as basic information of heat exchange stations, heating area, and affiliated organization; relevant query questions include querying the correspondence between heat exchange stations and generating units, querying the heating area or affiliated organization of heat exchange stations; the table structure definition includes an explanation of the structure of the basic information table for heat exchange stations in the order of table name, field name, data type, and field description.

[0077] The organizational structure description includes relevant query scenarios such as the hierarchical relationship of the organization; relevant query questions include querying basic information such as the heating companies under the heating group and the heating centers under the heating companies; the table structure definition includes an explanation of the organizational structure in the order of table name, field name, data type, and field description.

[0078] The description of the basic information table structure for heat sources includes relevant query scenarios such as heat source name, location, number of boilers, activation status, energy type, and the organization mentioned; relevant query questions include the number or list of heat sources under a certain organization, and querying the name, location, number of boilers, activation status, energy type, and organization mentioned for heat sources that meet the query conditions; the table structure definition includes an explanation of the basic information table structure for heat sources in the order of table name, field name, data type, and field description.

[0079] The description of the unit's real-time data table structure includes relevant query scenarios for real-time parameters and indicators of the unit's heating process, including real-time parameters such as valve opening degree of the primary and secondary networks, inlet water temperature, return water temperature, flow rate, and heat output, as well as indicators such as terminal temperature difference and heat exchange efficiency. Relevant query questions include querying a specific real-time parameter or indicator, several real-time parameters or indicators, or all real-time parameters or indicators of a specific unit. The table structure definition includes a description of the unit's real-time data table structure in the order of table name, field name, data type, and field description.

[0080] The description of the unit's historical data table structure includes relevant query scenarios such as parameters and indicators of the unit's heating process over historical time periods or time intervals, including parameters such as valve opening degree of the primary and secondary networks, inlet water temperature, return water temperature, flow rate, heat output, and indicators such as terminal temperature difference and heat exchange efficiency; relevant query questions include querying a specific parameter or indicator, several parameters or indicators, or all parameters or indicators of a specific unit at a historical time or time interval; the table structure definition includes a description of the unit's historical data table structure in the order of table name, field name, data type, and field description.

[0081] The description of the weather forecast data table structure includes query scenarios for weather forecast data for a future time or a period of time, and related query questions such as querying the weather changes for the next seven days or querying the weather changes for tomorrow; the table structure definition includes an explanation of the weather forecast data table structure in the order of table name, field name, data type, and field description.

[0082] The structure description of the unit summary data table includes relevant query scenarios such as summary statistics of unit heating process parameters and indicators, including parameters such as valve opening, inlet water temperature, return water temperature, flow rate, and heat output of primary and secondary networks, as well as indicators such as terminal temperature difference and heat exchange efficiency. Related query questions include querying summary statistics or rankings of a specific parameter or indicator, several parameters or indicators, or all parameters or indicators of a specified unit on a specified date in the current heating season or the previous heating season. It also includes querying the same-year comparison of a specific parameter or indicator, several parameters or indicators, or all parameters or indicators of a specified unit. The table structure definition includes an explanation of the unit historical data table structure in the order of table name, field name, data type, and field description.

[0083] The structure description of the valve basic information table includes relevant query scenarios such as valve manufacturer, location, equipment number, and affiliated organization; relevant query questions include querying the manufacturer, location, equipment number, and affiliated organization of a specific valve, or querying the number or list of valves under a specific organization, or querying the number or list of valves that meet the query conditions; the table structure definition includes an explanation of the valve basic information table structure in the order of table name, field name, data type, and field description.

[0084] The description of the valve real-time data table structure includes relevant query scenarios such as valve opening degree and online status; relevant query questions include querying the opening degree and online status of a specific valve; the table structure definition includes an explanation of the valve real-time data table structure in the order of table name, field name, data type, and field description.

[0085] The description of the valve historical data table structure includes relevant query scenarios such as valve opening degree and online status at historical time or historical time period; relevant query questions include querying the opening degree and online status of a specific valve at historical time or historical time period; the table structure definition includes an explanation of the valve historical data table structure in the order of table name, field name, data type, and field description.

[0086] The description of the valve strategy information table structure includes relevant query scenarios such as valve control strategy details, strategy type, or a specific strategy; relevant query questions include querying the control strategy details, strategy type, or a specific strategy for a valve in the current or historical time period, or querying the number or list of valves that meet a specified control strategy or control type; the table structure definition includes a description of the valve strategy information table structure in the order of table name, field name, data type, and field description.

[0087] The description of the basic information table for customer valves includes relevant query scenarios such as customer valve manufacturer, location, equipment number, and affiliated organization; relevant query questions include querying the manufacturer, location, equipment number, and affiliated organization of a specific customer valve, or querying the number or list of customer valves under a specific organization, or querying the number or list of customer valves that meet the query conditions; the table structure definition includes a description of the basic information table structure for customer valves in the order of table name, field name, data type, and field description.

[0088] The description of the real-time data table structure for household valves includes relevant query scenarios such as valve opening degree and online status; relevant query questions include querying the opening degree and online status of a specific household valve; the table structure definition includes an explanation of the real-time data table structure for household valves in the order of table name, field name, data type, and field description.

[0089] The description of the valve historical data table structure includes relevant query scenarios such as the valve opening degree and online status during historical time or historical time period; relevant query questions include querying the opening degree and online status of a specific valve during historical time or historical time period; the table structure definition includes an explanation of the valve historical data table structure in the order of table name, field name, data type, and field description.

[0090] The description of the valve strategy information table structure includes relevant query scenarios such as valve control strategy details, strategy type, or a specific strategy; relevant query questions include querying the control strategy details, strategy type, or a specific strategy for a valve at the current or historical time, or querying the number or list of valves that meet a specified control strategy or control type; the table structure definition includes a description of the valve strategy information table structure in the order of table name, field name, data type, and field description.

[0091] The description of the real-time room temperature data table structure includes relevant query scenarios such as the real-time indoor temperature of an organization, a household, or a unit, or the number or list of households within a specified real-time room temperature range; relevant query questions include querying the real-time indoor temperature of an organization, a household, or a unit, or querying the number or list of households within a specified real-time room temperature range; the table structure definition includes a description of the real-time room temperature data table structure in the order of table name, field name, data type, and field description.

[0092] The description of the historical room temperature data table structure includes relevant query scenarios such as the indoor temperature of a certain organization, household, or unit over a historical period, the average indoor temperature of a certain organization, household, or unit over a historical period, or the number or list of households within a specified room temperature range during a historical period. Related query questions include querying the indoor temperature of a certain organization, household, or unit over a historical period, querying the average indoor temperature of a certain organization, household, or unit over a historical period, or querying the number or list of households within a specified room temperature range during a historical period. The table structure definition includes a description of the historical room temperature data table structure in the order of table name, field name, data type, and field description.

[0093] S202: Construct a knowledge base of sample SQL questions, including:

[0094] Several SQL examples and a group of similar questions for each SQL example are set up according to query type, query subject, query content, organizational structure, time range, data type, data granularity, and analysis type.

[0095] Specifically, query types include querying specific data, querying data details, and querying lists; query subjects include querying a specific unit, a specific heat exchange station, a specific building valve, a specific household valve, and a specific household; query content includes querying basic information, querying historical data, querying real-time data, querying energy consumption data, and querying weather data; organizational structures include querying a specific group, a specific company, a specific center, and a heat exchange station managed by a specific manager; time ranges include querying real-time data, data from the past week, data from the current heating season, and data from the previous heating season; data types include querying all parameters, querying partial parameters, querying average opening degree, querying real-time opening degree, querying average room temperature, and querying real-time room temperature; data granularity includes querying historical data statistically analyzed by day and querying historical data statistically analyzed by hour; and analysis types include analysis percentage, analysis ranked from high to low, analysis ranked from low to high, analysis details, number of analyses, and analysis interval distribution.

[0096] Each SQL example corresponds to a group of similar questions, including several questions from different perspectives, several questions based on different levels of expertise, and several questions that have been replaced with synonyms.

[0097] Several questions from different perspectives include those from the perspectives of housekeepers, operators, and residents; questions arising from different levels of specialization include colloquial versions and technical document-style versions of the same query; and several questions resulting from synonym replacement include new questions formed after multiple synonym replacements of the query's wording.

[0098] S203: Construct a knowledge base for heating industry documents, including:

[0099] We collect and organize 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 into unstructured knowledge documents. These unstructured heating industry knowledge documents are then stored in a heating professional document knowledge base.

[0100] S204: Construct corresponding SQL example question knowledge bases, database table structure vector knowledge bases, and heating professional document vector knowledge bases based on the SQL example question knowledge base, database table structure knowledge base, and heating professional document vector knowledge base, respectively.

[0101] Specifically, the content in the SQL example question knowledge base is segmented and vectorized, and a vector knowledge base for SQL example questions is constructed; the content in the database table structure knowledge base is segmented and vectorized, and a vector knowledge base for database table structure is constructed; the knowledge documents in the heating professional document knowledge base are batch parsed, entities are identified, and relations are extracted to construct a knowledge graph; the knowledge documents in the heating professional document knowledge base are segmented and vectorized, and the entity relations in the knowledge graph are vectorized to construct a vector knowledge base for heating professional documents, so as to facilitate subsequent retrieval and use.

[0102] S300: Receives user inquiries, specifically including:

[0103] Users can input their queries via text or voice, or by selecting multiple keywords and combining them into a query.

[0104] The user selects multiple keywords to form a user query question, which includes pre-setting or user-defined keywords for different query content, and then selecting multiple keywords to form the user query question.

[0105] Specifically, several query categories are first preset or user-defined based on different query content. Query categories include basic information, real-time data, historical data, weather data, and room temperature data.

[0106] Secondly, several keyword categories are preset or user-defined under each query category. Keyword categories include query type, query subject, organization, time range, data type, data granularity, and analysis type.

[0107] Then, several keywords are preset or user-defined under each keyword category. Keywords under query type include specific data, details, list, etc.; keywords under query subject include a specific unit, a specific heat exchange station, a specific building valve, a specific household valve, a specific household, etc.; keywords under organization include a specific group, a specific company, a specific center, a specific housekeeper, etc.; keywords under time range include real-time data, the past week, the current heating season, the previous heating season, etc.; keywords under data type include full parameters, partial parameters, average opening degree, real-time opening degree, average room temperature, real-time room temperature, etc.; keywords under data granularity include statistics by day, statistics by hour, etc.; keywords under analysis type include analysis percentage, ranking from high to low, ranking from low to high, details, quantity, interval distribution, etc.

[0108] Finally, the user selects multiple keywords and combines them to form their query question.

[0109] S400: Based on the user query question, retrieve the first matching related information in the SQL example question knowledge base and the database table structure knowledge base, including vectorizing the user query question and matching the first similar fragment in the SQL example question vector knowledge base and the database table structure vector knowledge base using RAG technology. The first similar fragment is the first related information.

[0110] Specifically, pre-trained language models such as BERT and Sentence-BERT, or embedding models such as Embedding, are used to perform semantic parsing of user queries, extract key features of core intent and contextual information, and generate high-dimensional vector representations. Then, the cosine similarity calculation method in RAG technology is used to retrieve the SQL example question vector knowledge base and the database table structure vector knowledge base. Through a hybrid retrieval strategy, efficient matching is performed to quickly locate the first similar segment that is closest to the user query.

[0111] S500: Enhanced suggestions are generated by merging the user's query, primary related information, and preset suggestions. Specifically, the Prompt process merges the user's query, primary related information, and preset suggestions to generate enhanced suggestions that better match the user's actual query requirements.

[0112] S600: Generates SQL based on enhanced prompt words using a large language model, and then retrieves the query results from the database using the SQL query.

[0113] S700: Based on the user's query question, retrieve matching second related information in the heating professional document knowledge base. Specifically, this includes vectorizing the user's query question and matching second similar segments in the heating professional document vector knowledge base using RAG technology. The second similar segments are the second related information.

[0114] Specifically, pre-trained language models such as BERT and Sentence-BERT, or embedding models such as Embedding, are used to perform semantic parsing of user queries, extract key features of core intent and contextual information, and generate high-dimensional vector representations. Then, the cosine vector similarity calculation method in RAG technology is used to retrieve the heating professional document vector knowledge base, and a hybrid retrieval strategy is used for efficient matching to quickly locate the second most similar segment that is closest to the user query.

[0115] S800: Intelligent insights are gained from data results by combining a large language model with second-related information to obtain analytical conclusions. Specifically, this includes intelligent insights that use a large language model combined with knowledge fragments referenced in the second-related information to perform associative reasoning on data results, generating professional analytical conclusions.

[0116] S900: Generates visualization charts based on data results, and renders tables based on data results to generate exportable table files, specifically including:

[0117] The data results are rendered into exportable table files. A data visualization engine automatically selects the appropriate chart type based on the data type and generates visual charts. For trend analysis (such as monthly heating changes), line charts or area charts are generated; for percentage analysis (such as the heating percentage of different heat source types), pie charts or donut charts are used; for correlation analysis (such as the relationship between outdoor temperature and heating load), scatter plots or heat maps are output. Simultaneously, the charts automatically label key industry thresholds (such as the optimal operating temperature range for the heating system) and outliers (such as sudden increases in energy consumption during a certain period), and support interactive operations (such as zooming in on the time range and switching data dimensions) to improve data readability.

[0118] S010: Integrate data results, visualizations, tables, and analysis conclusions to generate an answer, specifically including:

[0119] The large language model integrates data results, visualization charts, exportable table files, and analysis conclusions, and generates natural language answers according to the logical framework of "user intent analysis - problem explanation - data presentation - analysis interpretation - suggested solutions", enabling accurate querying of dynamic data, intelligent insights, interpretable processes, and visual presentation in the heating industry.

[0120] 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.

[0121] This application discloses a knowledge enhancement system for data insights in the heating industry based on a large language model.

[0122] Reference Figure 1 and Figure 2 The knowledge enhancement system for data insights in the heating industry based on a large language model includes a database management module, a knowledge base management module, a user interaction module, a retrieval enhancement module, a data query module, an intelligent insight module, a data processing module, and an answer generation module.

[0123] The database management module is used to access multimodal real-time sensing data and business operation data and build a database.

[0124] Real-time sensing data includes time-series data (temperature, pressure, flow rate, heat, etc.) from multiple sensors at various levels, including heat sources, pipelines, heat exchange stations, and user terminals. Operational data includes structured data such as user information, equipment ledgers, work order management, energy consumption metering, and billing records. The acquisition of real-time sensing data and operational data forms raw data. This raw data is then cleaned to eliminate noise and errors, and standardized and normalized before being stored in a database.

[0125] The knowledge base management module is used to build a knowledge base for SQL example questions, a knowledge base for database table structures, and a knowledge base for heating-related professional documents.

[0126] Construct a database table structure knowledge base, including:

[0127] First, define the structure of several tables in the database according to the query content.

[0128] Specifically, the query content includes basic information of the generating unit, basic information of the heat exchange station, basic information of the organization, basic information of the heat source, real-time data of the generating unit, historical data of the generating unit, weather forecast data, summary data of the generating unit, basic information of the building valve, real-time data of the building valve, historical data of the building valve, building valve strategy information, basic information of the household valve, real-time data of the household valve, historical data of the household valve, household valve strategy information, real-time data of the room temperature, and historical data of the room temperature. The structure descriptions for the following tables are set according to the query content: basic information of the generating unit, basic information of the heat exchange station, basic information of the organization, basic information of the heat source, real-time data of the generating unit, historical data of the generating unit, weather forecast data, summary data of the generating unit, basic information of the building valve, real-time data of the building valve, historical data of the building valve, strategy information of the building valve, basic information of the household valve, real-time data of the household valve, historical data of the household valve, strategy information of the household valve, real-time data of the household valve, and historical data of the household valve.

[0129] Then, for each table structure description, set the relevant query scenarios, relevant query questions, and table structure definitions.

[0130] Specifically, the structure description of the unit basic information table includes relevant query scenarios such as the heat exchange station to which the unit belongs, heating type, whether it is an energy-saving building, and the organization to which it belongs; relevant query questions include querying the number or list of units under a certain organization, querying the list or number of units that meet the query conditions, and querying the heat exchange station, heating type, whether it is an energy-saving building, and the organization to which a certain unit belongs; the table structure definition includes an explanation of the structure of the unit basic information table in the order of table name, field name, data type, and field description.

[0131] The description of the basic information table structure for heat exchange stations includes relevant query scenarios such as basic information of heat exchange stations, heating area, and affiliated organization; relevant query questions include querying the correspondence between heat exchange stations and generating units, querying the heating area or affiliated organization of heat exchange stations; the table structure definition includes an explanation of the structure of the basic information table for heat exchange stations in the order of table name, field name, data type, and field description.

[0132] The organizational structure description includes relevant query scenarios such as the hierarchical relationship of the organization; relevant query questions include querying basic information such as the heating companies under the heating group and the heating centers under the heating companies; the table structure definition includes an explanation of the organizational structure in the order of table name, field name, data type, and field description.

[0133] The description of the basic information table structure for heat sources includes relevant query scenarios such as heat source name, location, number of boilers, activation status, energy type, and the organization mentioned; relevant query questions include the number or list of heat sources under a certain organization, and querying the name, location, number of boilers, activation status, energy type, and organization mentioned for heat sources that meet the query conditions; the table structure definition includes an explanation of the basic information table structure for heat sources in the order of table name, field name, data type, and field description.

[0134] The description of the unit's real-time data table structure includes relevant query scenarios for real-time parameters and indicators of the unit's heating process, including real-time parameters such as valve opening degree of the primary and secondary networks, inlet water temperature, return water temperature, flow rate, and heat output, as well as indicators such as terminal temperature difference and heat exchange efficiency. Relevant query questions include querying a specific real-time parameter or indicator, several real-time parameters or indicators, or all real-time parameters or indicators of a specific unit. The table structure definition includes a description of the unit's real-time data table structure in the order of table name, field name, data type, and field description.

[0135] The description of the unit's historical data table structure includes relevant query scenarios such as parameters and indicators of the unit's heating process over historical time periods or time intervals, including parameters such as valve opening degree of the primary and secondary networks, inlet water temperature, return water temperature, flow rate, heat output, and indicators such as terminal temperature difference and heat exchange efficiency; relevant query questions include querying a specific parameter or indicator, several parameters or indicators, or all parameters or indicators of a specific unit at a historical time or time interval; the table structure definition includes a description of the unit's historical data table structure in the order of table name, field name, data type, and field description.

[0136] The description of the weather forecast data table structure includes query scenarios for weather forecast data for a future time or a period of time, and related query questions such as querying the weather changes for the next seven days or querying the weather changes for tomorrow; the table structure definition includes an explanation of the weather forecast data table structure in the order of table name, field name, data type, and field description.

[0137] The structure description of the unit summary data table includes relevant query scenarios such as summary statistics of unit heating process parameters and indicators, including parameters such as valve opening, inlet water temperature, return water temperature, flow rate, and heat output of primary and secondary networks, as well as indicators such as terminal temperature difference and heat exchange efficiency. Related query questions include querying summary statistics or rankings of a specific parameter or indicator, several parameters or indicators, or all parameters or indicators of a specified unit on a specified date in the current heating season or the previous heating season. It also includes querying the same-year comparison of a specific parameter or indicator, several parameters or indicators, or all parameters or indicators of a specified unit. The table structure definition includes an explanation of the unit historical data table structure in the order of table name, field name, data type, and field description.

[0138] The structure description of the valve basic information table includes relevant query scenarios such as valve manufacturer, location, equipment number, and affiliated organization; relevant query questions include querying the manufacturer, location, equipment number, and affiliated organization of a specific valve, or querying the number or list of valves under a specific organization, or querying the number or list of valves that meet the query conditions; the table structure definition includes an explanation of the valve basic information table structure in the order of table name, field name, data type, and field description.

[0139] The description of the valve real-time data table structure includes relevant query scenarios such as valve opening degree and online status; relevant query questions include querying the opening degree and online status of a specific valve; the table structure definition includes an explanation of the valve real-time data table structure in the order of table name, field name, data type, and field description.

[0140] The description of the valve historical data table structure includes relevant query scenarios such as valve opening degree and online status at historical time or historical time period; relevant query questions include querying the opening degree and online status of a specific valve at historical time or historical time period; the table structure definition includes an explanation of the valve historical data table structure in the order of table name, field name, data type, and field description.

[0141] The description of the valve strategy information table structure includes relevant query scenarios such as valve control strategy details, strategy type, or a specific strategy; relevant query questions include querying the control strategy details, strategy type, or a specific strategy for a valve in the current or historical time period, or querying the number or list of valves that meet a specified control strategy or control type; the table structure definition includes a description of the valve strategy information table structure in the order of table name, field name, data type, and field description.

[0142] The description of the basic information table for customer valves includes relevant query scenarios such as customer valve manufacturer, location, equipment number, and affiliated organization; relevant query questions include querying the manufacturer, location, equipment number, and affiliated organization of a specific customer valve, or querying the number or list of customer valves under a specific organization, or querying the number or list of customer valves that meet the query conditions; the table structure definition includes a description of the basic information table structure for customer valves in the order of table name, field name, data type, and field description.

[0143] The description of the real-time data table structure for household valves includes relevant query scenarios such as valve opening degree and online status; relevant query questions include querying the opening degree and online status of a specific household valve; the table structure definition includes an explanation of the real-time data table structure for household valves in the order of table name, field name, data type, and field description.

[0144] The description of the valve historical data table structure includes relevant query scenarios such as the valve opening degree and online status during historical time or historical time period; relevant query questions include querying the opening degree and online status of a specific valve during historical time or historical time period; the table structure definition includes an explanation of the valve historical data table structure in the order of table name, field name, data type, and field description.

[0145] The description of the valve strategy information table structure includes relevant query scenarios such as valve control strategy details, strategy type, or a specific strategy; relevant query questions include querying the control strategy details, strategy type, or a specific strategy for a valve at the current or historical time, or querying the number or list of valves that meet a specified control strategy or control type; the table structure definition includes a description of the valve strategy information table structure in the order of table name, field name, data type, and field description.

[0146] The description of the real-time room temperature data table structure includes relevant query scenarios such as the real-time indoor temperature of an organization, a household, or a unit, or the number or list of households within a specified real-time room temperature range; relevant query questions include querying the real-time indoor temperature of an organization, a household, or a unit, or querying the number or list of households within a specified real-time room temperature range; the table structure definition includes a description of the real-time room temperature data table structure in the order of table name, field name, data type, and field description.

[0147] The description of the historical room temperature data table structure includes relevant query scenarios such as the indoor temperature of a certain organization, household, or unit over a historical period, the average indoor temperature of a certain organization, household, or unit over a historical period, or the number or list of households within a specified room temperature range during a historical period. Related query questions include querying the indoor temperature of a certain organization, household, or unit over a historical period, querying the average indoor temperature of a certain organization, household, or unit over a historical period, or querying the number or list of households within a specified room temperature range during a historical period. The table structure definition includes a description of the historical room temperature data table structure in the order of table name, field name, data type, and field description.

[0148] Build a knowledge base of SQL example questions, including:

[0149] Several SQL examples and a group of similar questions for each SQL example are set up according to query type, query subject, query content, organizational structure, time range, data type, data granularity, and analysis type.

[0150] Specifically, query types include querying specific data, querying data details, and querying lists; query subjects include querying a specific unit, a specific heat exchange station, a specific building valve, a specific household valve, and a specific household; query content includes querying basic information, querying historical data, querying real-time data, querying energy consumption data, and querying weather data; organizational structures include querying a specific group, a specific company, a specific center, and a heat exchange station managed by a specific manager; time ranges include querying real-time data, data from the past week, data from the current heating season, and data from the previous heating season; data types include querying all parameters, querying partial parameters, querying average opening degree, querying real-time opening degree, querying average room temperature, and querying real-time room temperature; data granularity includes querying historical data statistically analyzed by day and querying historical data statistically analyzed by hour; and analysis types include analysis percentage, analysis ranked from high to low, analysis ranked from low to high, analysis details, number of analyses, and analysis interval distribution.

[0151] Each SQL example corresponds to a group of similar questions, including several questions from different perspectives, several questions based on different levels of expertise, and several questions that have been replaced with synonyms.

[0152] Several questions from different perspectives include those from the perspectives of housekeepers, operators, and residents; questions arising from different levels of specialization include colloquial versions and technical document-style versions of the same query; and several questions resulting from synonym replacement include new questions formed after multiple synonym replacements of the query's wording.

[0153] A heating industry knowledge base will be constructed, including the collection and organization of heating-related standards, industry literature, theoretical knowledge, system operation manuals, equipment maintenance methods, platform operation manuals, equipment technical manuals, operating procedures, troubleshooting guides, case studies, and expert experience summaries, forming unstructured knowledge documents. These unstructured heating industry knowledge documents will be stored in the heating industry knowledge base.

[0154] Based on the SQL example question knowledge base, the database table structure knowledge base, and the heating professional document knowledge base, respectively construct the corresponding SQL example question knowledge base, database table structure vector knowledge base, and heating professional document vector knowledge base.

[0155] Specifically, the content in the SQL example question knowledge base is segmented and vectorized, and a vector knowledge base for SQL example questions is constructed; the content in the database table structure knowledge base is segmented and vectorized, and a vector knowledge base for database table structure is constructed; the knowledge documents in the heating professional document knowledge base are batch parsed, entities are identified, and relations are extracted to construct a knowledge graph; the knowledge documents in the heating professional document knowledge base are segmented and vectorized, and the entity relations in the knowledge graph are vectorized to construct a vector knowledge base for heating professional documents, so as to facilitate subsequent retrieval and use.

[0156] The user interaction module is used to receive user queries.

[0157] Specifically, users can input their query questions via text or voice input, or by selecting multiple keywords and combining them into a query question.

[0158] The user selects multiple keywords to form a user query question, which includes pre-setting or user-defined keywords for different query content, and then selecting multiple keywords to form the user query question.

[0159] Furthermore, several query categories are preset or user-defined based on different query content. Query categories include basic information, real-time data, historical data, weather data, and room temperature data.

[0160] Secondly, several keyword categories are preset or user-defined under each query category. Keyword categories include query type, query subject, organization, time range, data type, data granularity, and analysis type.

[0161] Then, several keywords are preset or user-defined under each keyword category. Keywords under query type include specific data, details, list, etc.; keywords under query subject include a specific unit, a specific heat exchange station, a specific building valve, a specific household valve, a specific household, etc.; keywords under organization include a specific group, a specific company, a specific center, a specific housekeeper, etc.; keywords under time range include real-time data, the past week, the current heating season, the previous heating season, etc.; keywords under data type include full parameters, partial parameters, average opening degree, real-time opening degree, average room temperature, real-time room temperature, etc.; keywords under data granularity include statistics by day, statistics by hour, etc.; keywords under analysis type include analysis percentage, ranking from high to low, ranking from low to high, details, quantity, interval distribution, etc.

[0162] Finally, the user selects multiple keywords and combines them to form their query question.

[0163] The retrieval enhancement module is used to retrieve matching first-related information from the SQL example question knowledge base and the database table structure knowledge base based on the user's query question; to fuse the user's query question, the first-related information, and preset prompt words to obtain enhanced prompt words; and to retrieve matching second-related information from the heating professional document knowledge base based on the user's query question.

[0164] Specifically, the user query question is vectorized, and the first similar fragment is matched in the SQL example question vector knowledge base and the database table structure vector knowledge base using RAG technology. The first similar fragment is the first related information.

[0165] Furthermore, pre-trained language models such as BERT and Sentence-BERT, or embedding models such as Embedding, are used to perform semantic parsing of user queries, extract key features of core intent and contextual information, and generate high-dimensional vector representations. Then, the cosine similarity calculation method in RAG technology is used to retrieve the SQL example question vector knowledge base and the database table structure vector knowledge base. Through a hybrid retrieval strategy, efficient matching is performed to quickly locate the first similar segment that is closest to the user query.

[0166] The Prompt project integrates the user's query, primary related information, and preset prompts to generate enhanced prompts, ensuring that the generated enhanced prompts match the user's actual query needs.

[0167] Specifically, the user's query is vectorized, and a second similar segment is matched in the heating professional document vector knowledge base using RAG technology. The second similar segment is the second related information.

[0168] Furthermore, pre-trained language models such as BERT and Sentence-BERT, or embedding models such as Embedding, are used to perform semantic parsing of user queries, extract key features of core intent and contextual information, and generate high-dimensional vector representations. Then, the cosine vector similarity calculation method in RAG technology is used to retrieve the heating professional document vector knowledge base, and a hybrid retrieval strategy is used for efficient matching to quickly locate the second most similar segment that is closest to the user query.

[0169] The data query module is used to generate SQL based on enhanced prompt words using a large language model, and then retrieve the query results from the database using the SQL query.

[0170] The Intelligent Insight module is used to intelligently analyze data results by combining a large language model with secondary related information to arrive at analytical conclusions. Specifically, it uses a large language model combined with knowledge fragments referenced in the secondary related information to perform intelligent inference on the data results, generating professional analytical conclusions.

[0171] The data processing module is used to render tables based on the data results, generate exportable table files, and generate visual charts based on the data results.

[0172] Specifically, the data visualization engine automatically selects the appropriate chart type based on the data type and generates visual charts. For trend analysis (such as monthly heating changes), line charts or area charts are generated; for percentage analysis (such as the heating percentage of different heat source types), pie charts or donut charts are used; and for correlation analysis (such as the relationship between outdoor temperature and heating load), scatter plots or heat maps are output. At the same time, the charts automatically label key industry thresholds (such as the optimal operating temperature range of the heating system) and outliers (such as sudden increases in energy consumption during a certain period), and support interactive operations (such as zooming in on the time range and switching data dimensions) to improve data readability.

[0173] The answer generation module is used to integrate data results, visualization charts, tables, and analysis conclusions to generate answers.

[0174] Specifically, the large language model integrates data results, visualization charts, exportable table files, and analysis conclusions, and generates natural language answers according to the logical framework of "user intent analysis - problem explanation - data presentation - analysis interpretation - suggested solutions", enabling accurate querying of dynamic data in the heating industry, intelligent insights, interpretable processes, and visual presentation.

[0175] This application also discloses a computer device.

[0176] Specifically, the computer device includes a memory and a processor, with the memory storing computer programs that can be loaded by the processor and executed using the aforementioned knowledge enhancement method for data insights into the heating industry based on a large language model.

[0177] This application also discloses a computer-readable storage medium.

[0178] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed, such as the knowledge enhancement method for data insights in the heating industry based on a large language model described above. 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.

[0179] 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 knowledge enhancement method for data insights in the heating industry based on a large language model, characterized in that, Includes the following steps: Access real-time sensing data and business operation data and build a database; Build a knowledge base for SQL example questions, a knowledge base for database table structures, and a knowledge base for heating industry documents; Receive user queries; Based on the user query question, retrieve the first matching related information from the SQL example question knowledge base and the database table structure knowledge base; The enhanced prompt words are obtained by fusing the user query question, the first associated information, and the preset prompt words; SQL is generated based on the enhanced prompt words using a large language model, and then the query results are obtained from the database using the SQL query. Based on the user's query, retrieve matching second related information from the heating professional document knowledge base; By combining the large language model with the second association information, intelligent insights are gained into the data results to obtain analytical conclusions. The answer is generated by integrating the data results and the analysis conclusions. The construction of the SQL example question knowledge base includes: Several SQL examples and a group of similar questions for each SQL example are set up according to query type, query subject, query content, organizational structure, time range, data type, data granularity, and analysis type. The similar question groups corresponding to each SQL example include several questions from different people's perspectives, several questions formed by different levels of specialization, and several questions after synonym replacement; The construction of the database table structure knowledge base includes: The database table structure is defined according to the query content; For each table structure description, define the relevant query scenarios, relevant query questions, and table structure definitions; The construction of the heating industry professional document knowledge base includes: Unstructured knowledge documents comprised 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.

2. The knowledge enhancement method for data insights in the heating industry based on a large language model as described in claim 1, characterized in that, It also includes pre-setting or allowing users to define several keywords for different query content, and supports users to select multiple keywords to form the user query question.

3. The knowledge enhancement method for data insights in the heating industry based on a large language model as described in claim 2, characterized in that, The method of pre-setting or allowing users to define several keywords for different query content includes: Several query categories can be preset or user-defined based on different query content; Each query category can be pre-defined or user-defined with several keyword categories; Each keyword category can be pre-set or user-defined with several keywords.

4. The knowledge enhancement method for data insights in the heating industry based on a large language model as described in claim 1, characterized in that, Also includes: Construct corresponding SQL example question vector knowledge bases, database table structure vector knowledge bases, and heating professional document vector knowledge bases based on the SQL example question knowledge base, database table structure vector knowledge base, and heating professional document vector knowledge base, respectively. The step of retrieving matching first related information from the SQL example question knowledge base and the database table structure knowledge base based on the user query question includes: The user query question is vectorized, and a first similar fragment is matched in the SQL example question vector knowledge base and the database table structure vector knowledge base using RAG technology. The first similar fragment is the first associated information. The step of retrieving matching second related information from the heating industry document knowledge base based on the user query includes: The user query question is vectorized, and a second similar fragment is matched in the heating professional document vector knowledge base using RAG technology. The second similar fragment is the second associated information.

5. The knowledge enhancement method for data insights in the heating industry based on a large language model as described in claim 1, characterized in that, It also includes generating visualization charts based on the data results, rendering tables based on the data results to generate exportable table files, and integrating the data results, the visualization charts, the table files, and the analysis conclusions to generate an answer.

6. A knowledge enhancement system for data insights in the heating industry based on a large language model, characterized in that, A knowledge enhancement method for implementing data insights in the heating industry based on a large language model as described in any one of claims 1-5, comprising: The database management module is used to access multimodal real-time sensing data and business operation data and build a database; The knowledge base management module is used to build a knowledge base for SQL example questions, a knowledge base for database table structures, and a knowledge base for heating-related professional documents. The user interaction module is used to receive user queries. The retrieval enhancement module is used to retrieve matching first related information from the SQL example question knowledge base and the database table structure knowledge base based on the user query question; The enhanced prompt words are obtained by fusing the user query question, the first associated information, and the preset prompt words; and Based on the user's query, retrieve matching second related information from the heating professional document knowledge base; The data query module is used to generate SQL based on the enhanced prompt words using a large language model, and then use the SQL to query the database to obtain the query results. The intelligent insight module is used to perform intelligent insight analysis on the data results by combining the large language model with the second correlation information to obtain analytical conclusions; The answer generation module is used to integrate the data results and the analysis conclusions to generate an answer.

Images