Spacecraft telemetry data visualization method based on large language model

Abstract

The application discloses a kind of based on large language model's spacecraft telemetry data visualization method, comprising: deployment and configuration basic environment;Build basic knowledge base;According to the data in basic knowledge base, carry out data pre-training, and store data pre-training result in vector database;Response data query and display request, screening obtains the final query data matched with the data query and display request, and display;Based on final query data, feedback and correction are carried out.The spacecraft telemetry data visualization method based on large language model of the application solves the problems of low efficiency, heavy workload and high personnel occupation of existing spacecraft telemetry data visualization tools.

Description

Technical Field

[0001] This invention belongs to the field of spacecraft testing technology, and in particular relates to a method for visualizing spacecraft telemetry data based on a large language model. Background Technology

[0002] Spacecraft, as a typical example of complex systems engineering, have specific application scenarios. Whether it's pre-delivery integrated testing or on-orbit operation, telemetry data serves as a crucial basis for health assessment, and reliability is a vital dimension of spacecraft evaluation. As spacecraft functions become increasingly diverse, the complexity of telemetry data is gradually increasing, and the readability of complex, heterogeneous telemetry data is deteriorating. Therefore, the need for graphical representation of data characteristics, based on their physical meaning, is becoming more urgent. Furthermore, as typical time-series data, simply displaying the data linearly along the time axis cannot reflect all its characteristics. Appropriate statistical tools are needed to process the data to provide a more comprehensive assessment of its physical meaning.

[0003] In response to the two aforementioned data visualization needs, and considering the increasingly complex nature of telemetry data, the graphical display of telemetry data requires extensive data querying, curve plotting, statistical algorithm design, and statistical image design. Currently, this work is generally completed by staff designing numerous constraint query conditions and using specific drawing tools. However, spacecraft development has now progressed to a stage characterized by short development cycles, rapid progress, highly skilled personnel, and low costs. Such data visualization methods are inefficient and cannot meet the development requirements.

[0004] To date, there is no unified standard or mature tool for visualizing spacecraft telemetry data. Summary of the Invention

[0005] The technical problem solved by this invention is to overcome the shortcomings of the prior art and provide a spacecraft telemetry data visualization method based on a large language model, which aims to solve the problems of low efficiency, large workload and high personnel requirements of existing spacecraft telemetry data visualization tools.

[0006] To address the aforementioned technical problems, this invention discloses a spacecraft telemetry data visualization method based on a large language model, comprising: Deploy and configure the basic environment; Build a basic knowledge base; Based on the data in the basic knowledge base, data pre-training is carried out, and the data pre-training results are stored in the vector database; In response to data query and display requests, the system filters and obtains the final query data that matches the data query and display requests, and then displays it. Feedback and corrections will be made based on the final query data.

[0007] In the aforementioned spacecraft telemetry data visualization method based on large language models, the basic environment is deployed and configured, including: Deploy a general-purpose large language model; Deploy a vector database to provide vector storage after embedding the pre-trained results; Deploy a spacecraft telemetry timing data service database to store spacecraft telemetry timing data.

[0008] In the aforementioned spacecraft telemetry data visualization method based on a large language model, a basic knowledge base is constructed, including: cleaning and fragmenting the three types of data to obtain three types of fragmented corpus data; and constructing the basic knowledge base based on the three types of fragmented corpus data.

[0009] In the aforementioned spacecraft telemetry data visualization method based on a large language model, the three types of data are cleaned and segmented to obtain three types of segmented corpus data, including: The table creation statements of the engineering telemetry, data management telemetry, and ground telemetry data tables in the spacecraft telemetry time series data business database are divided into segments, and the remaining table creation statements are deleted to obtain the table creation file of the spacecraft telemetry time series data business database. The telemetry parameter definition document, technical document, and test summary document were cleaned and fragmented to obtain the spacecraft telemetry data document. Specifically, the telemetry parameter definition document defined the correspondence between data streams, telemetry names, codes, and numbers. Text content satisfying the syntax structure "data stream-telemetry code-telemetry number-physical meaning" was extracted as a single corpus entry and then fragmented. The technical document described the relevant subsystems of spacecraft telemetry, using periods as identifiers. The text data in the technical document was fragmented, and other charts were removed. The test summary document provided specific telemetry values ​​for spacecraft at certain times. Text satisfying the syntax structure "data stream-telemetry code or telemetry number-physical value" was extracted, fragmented, and other charts were removed. Using the log files of the existing spacecraft telemetry time series data service database query software, the query question content and the corresponding query SQL statement are extracted as a corpus. Irrelevant information in the log is deleted to obtain the spacecraft telemetry data "question-query" pair, that is, the specific telemetry query question content and its corresponding query SQL statement.

[0010] In the aforementioned spacecraft telemetry data visualization method based on a large language model, data pre-training is performed based on data from the basic knowledge base, and the pre-training results are stored in a vector database, including: Based on the table creation files of the spacecraft telemetry time-series data business database, the spacecraft telemetry data documents, and the spacecraft telemetry data "question-query" pairs in the basic knowledge base, data pre-training was carried out to obtain the pre-training results of three types of fragmented corpus data; Using the embedding functions provided by the vector database, three vector sets are generated to store the pre-training results of three types of segmented corpus data. The three vector sets are: a business database table creation vector set, used to store the pre-training results of the business database table creation files for spacecraft telemetry time series data; a telemetry data document vector set, used to store the pre-training results of spacecraft telemetry data documents; and a "question-query" pair vector set, used to store the pre-training results of the "question-query" pairs for spacecraft telemetry data.

[0011] In the aforementioned spacecraft telemetry data visualization method based on a large language model, in response to data query and display requests, the final query data matching the data query and display requests is filtered out and displayed, including: In response to data query and display requests, specific time-series data query requirements are proposed. The syntax structure is: "Time period + Subsystem name + Telemetry parameter name". The "Time period + Subsystem name + Telemetry parameter name" is sent as three vector sets to the vector database as retrieval information. The retrieval information recall data is obtained respectively, and the recall data and retrieval information are rearranged in a second time. Construct intermediate LLM prompts with the syntax structure of "initial prompt + user question + data recalled from basic knowledge base"; input the intermediate LLM prompts into a general large language model to obtain intermediate SQL query statements; input the intermediate SQL query statements into the spacecraft telemetry time series data business database for data query; and save the query results as intermediate query data. The final LLM prompt is constructed, with the syntax structure of: "intermediate LLM prompt + 'the following is the data retrieved using the intermediate SQL query statement' + intermediate SQL query statement + intermediate query data". The final LLM prompt is then fed into the general large language model to obtain the final SQL statement. The final SQL statement is then fed into the spacecraft telemetry time series data business database for final data query. The query results are saved as the final query data and displayed in a table sorted by timestamp. The data processing and plotting prompts are constructed using the "data statistics requirements + plotting requirements" framework. The syntax is "generate Python code for data processing and Plotly plotting code based on the following requirements + 'data statistics requirements + plotting requirements' + final query data". The data processing and plotting prompts are then fed into a general large language model to obtain and execute the code, resulting in the final graph, which is then displayed.

[0012] In the aforementioned spacecraft telemetry data visualization method based on a large language model, the data recalled from the basic knowledge base is adjusted according to the number of recalled data. Specifically, if the number of data entries recalled in the business database table creation vector set is greater than 0 and does not exceed the context window size, the prompt message "You can use the following database table creation statement as a reference to find tables that may be used for data querying + recalled data" is added. If the number of data entries recalled in the telemetry data document vector set is greater than 0 and does not exceed the context window size, the prompt message "You can use the following document as a reference to find entries in the data table + recalled data" is added. If the number of data entries recalled in the "question-query" vector set is greater than 0 and does not exceed the context window size, the prompt message "You can use the following related query questions and corresponding SQL statements as a reference + recalled data" is added.

[0013] In the aforementioned spacecraft telemetry data visualization method based on a large language model, feedback and corrections are performed based on the final query data, including: Based on the final query data, determine the correctness of the SQL statement and provide feedback; If the final queried data matches the data query and display request, the current "question-query" pair will be updated in the basic knowledge base; If the final queried data does not match the data query and display request, the problem will be analyzed and corrected step by step based on the information fed back by the general large language model, including: adjusting the general large language model, supplementing the data in the basic knowledge base, increasing relevant data recall, adjusting the initial prompt words, the grammatical structure of the intermediate LLM prompt words, and the grammatical structure of the final LLM prompt words.

[0014] The present invention has the following advantages: (1) This invention discloses a spacecraft telemetry data visualization method based on a large language model. The spacecraft telemetry data query can be described by Chinese natural language, thereby bypassing the SQL structured language, decoupling the structured query language, improving the readability of the query, improving the query efficiency, reducing the personnel occupation, and providing a basic condition for subsequent full-text Chinese natural language description summary reports.

[0015] (2) This invention discloses a spacecraft telemetry data visualization method based on a large language model. The spacecraft telemetry data visualization can be described using Chinese natural language, which reduces graphic design time, enriches design effects, introduces time series analysis of natural language description, and provides graphical display. It provides a foundation for subsequent full-text Chinese natural language description summary reports.

[0016] (3) This invention discloses a spacecraft telemetry data visualization method based on a large language model, which decouples the fine-tuning requirements of the general large model. By performing language fragmentation and data cleaning for the fixed grammatical structure unique to spacecraft telemetry technical documents, it can be directly used as excellent data input for data embedding and data retrieval, further reducing the cost and versatility of use. The data retrieval process introduces a data rearrangement process to further improve the accuracy.

[0017] (4) This invention discloses a spacecraft telemetry data visualization method based on a large language model, which has the ability to iterate and upgrade. As the feedback from use accumulates, the accuracy gradually improves. At the same time, it can remove erroneous prior information at any time, and there will be no knowledge pollution problem when fine-tuning the large model.

[0018] (5) This invention discloses a spacecraft telemetry data visualization method based on a large language model, which completes the adaptation of the large language model to specific applications in vertical fields, has general engineering significance, and provides an important technical route basis for subsequent expansion applications. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a spacecraft telemetry data visualization method based on a large language model, as described in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments disclosed in the present invention will be described in further detail below with reference to the accompanying drawings.

[0021] Reference Figure 1 In this embodiment, the spacecraft telemetry data visualization method based on a large language model includes: S1, deploy and configure the basic environment.

[0022] In this embodiment, the deployment and configuration of the basic environment mainly includes the following aspects: 1) Deploying a general-purpose large language model involves a data application development framework, including but not limited to: preparing hardware resources with the necessary computing power and preparing software resources to adapt to the operating system.

[0023] 2) Deploy a vector database to provide vector storage after embedding the pre-trained results, and provide corresponding embedding functions, related recall interface functions, reordering models, etc.

[0024] 3) Deploy a spacecraft telemetry timing data business database to store spacecraft telemetry timing data. This is generally an existing business database of the enterprise.

[0025] S2, Building a basic knowledge base.

[0026] In this embodiment, the construction of the basic knowledge base mainly includes cleaning and fragmenting three types of data to obtain three types of fragmented corpus data, and then constructing the basic knowledge base. Specifically: 21) The table creation file for the spacecraft telemetry time-series data service database contains the table definitions for engineering telemetry, data management telemetry, and ground telemetry data tables in the spacecraft telemetry time-series data service database. The table creation file is segmented according to the single-row creation statement of the data table, and the remaining table creation statements are deleted.

[0027] 22) Spacecraft telemetry data documents, including telemetry parameter definition documents, technical documents, and test summary documents. These documents were cleaned and fragmented to obtain the spacecraft telemetry data documents. Specifically, the telemetry parameter definition documents defined the correspondence between data streams, telemetry names, codes, and numbers. Text content satisfying the syntax structure "data stream-telemetry code-telemetry number-physical meaning" was extracted as a single corpus entry and fragmented. The technical documents described the relevant subsystems of spacecraft telemetry, using periods as identifiers. The text data in the technical documents was fragmented, and other charts were removed. The test summary documents provided specific telemetry values ​​for certain time points. Text content satisfying the syntax structure "data stream-telemetry code or telemetry number-physical value" was extracted, fragmented, and other charts were removed.

[0028] 23) For spacecraft telemetry data “problem-query” pairs, that is, specific telemetry query problem content and its corresponding query SQL statement, use the log file of the existing spacecraft telemetry time series data business database query software to extract the query problem content and the corresponding query SQL statement as a single corpus, and delete irrelevant information from the log.

[0029] S3 performs data pre-training based on data in the basic knowledge base and stores the data pre-training results in the vector database.

[0030] In this embodiment, data pre-training is performed based on three types of fragmented corpus data constructed from a basic knowledge base (spacecraft telemetry time-series data business database table creation files, spacecraft telemetry data documents, and spacecraft telemetry data "question-query" pairs). The pre-training results are obtained. Using the embedding functions provided by the vector database, three vector sets are generated: a business database table creation vector set, a telemetry data document vector set, and a "question-query" pair vector set. Specifically, the business database table creation vector set stores the pre-training results of the spacecraft telemetry time-series data business database table creation files; the telemetry data document vector set stores the pre-training results of the spacecraft telemetry data documents; and the "question-query" pair vector set stores the pre-training results of the spacecraft telemetry data "question-query" pairs.

[0031] S4. Respond to the data query and display request, filter out the final query data that matches the data query and display request, and display it.

[0032] In this embodiment, the data query and display process is as follows: 41) Respond to data query and display requests, and put forward specific time series data query requirements. Its syntax structure is: "Time period + Subsystem name + Telemetry parameter name". The system sends "Time period + Subsystem name + Telemetry parameter name" as the three vector sets of the retrieval information to the vector database, retrieves the retrieval information recall data respectively, and performs a second rearrangement of the recall data and retrieval information.

[0033] 42) Construct intermediate LLM prompts with the syntax "initial prompt + user question + data recalled from the basic knowledge base". Input the intermediate LLM prompts into a general large language model to obtain intermediate SQL query statements. Send these intermediate SQL query statements to the spacecraft telemetry time series data business database for data querying, and save the query results as intermediate query data. The "data recalled from the basic knowledge base" needs to be adjusted based on the amount of recalled data: If the number of recalled data entries in the business database table creation vector set is greater than 0 and does not exceed the context window size, add the prompt "You can use the following database table creation statement as a reference to find tables that may be used for data querying + recalled data"; if the number of recalled data entries in the telemetry data document vector set is greater than 0 and does not exceed the context window size, add the prompt "You can use the following document as a reference to find entries in the data table + recalled data"; if the "question-query" retrieves more than 0 data entries from the vector set and does not exceed the context window size, add the prompt "You can use the following related query questions and corresponding SQL statements as reference + recalled data".

[0034] 43) Construct the final LLM prompt, whose syntax structure is: "intermediate LLM prompt + 'The following is the data retrieved using the intermediate SQL query statement' + intermediate SQL query statement + intermediate query data"; send the final LLM prompt into the general large language model to obtain the final SQL statement, send the final SQL statement into the spacecraft telemetry time series data business database for final data query, save the query results as the final query data, and display them in a table sorted by timestamp index.

[0035] 44) Construct data processing and plotting prompts using "data statistics requirements + plotting requirements". The syntax structure is "generate Python code for data processing and Plotly plotting code based on the following requirements + 'data statistics requirements + plotting requirements' + final query data". Input the data processing and plotting prompts into the general large language model to obtain the code and execute the obtained code to obtain the final graph and display it.

[0036] S5 provides feedback and corrections based on the final query data.

[0037] In this embodiment, the correctness of the SQL statement can be determined and feedback can be provided based on the final query data. If the final query data matches the data query and display request, the current "question-query" pair is updated in the basic knowledge base to continuously improve the dimensionality and accuracy of the pre-trained knowledge base. If the final query data does not match the data query and display request, the problem is analyzed and corrected step-by-step based on the information fed back from the general large language model, including: adjusting the general large language model, supplementing data in the basic knowledge base, increasing relevant data recall, adjusting the initial prompt words, the grammatical structure of the intermediate LLM prompt words, and the grammatical structure of the final LLM prompt words.

[0038] Based on the above embodiments, the following section will take the visualization requirements of telemetry data curves of a subsystem of a manned spacecraft based on a large language model as an example to provide a detailed description of the solution described in this invention.

[0039] I. Deploy and configure the basic environment ① A general-purpose large language model, involving a data application development framework, including the preparation of hardware resources for necessary computing power and software resources for operating system adaptation. For example, using the Qwen-7B model as a general-purpose large language model, using a standard rack server, and configuring GPU parallel computing resources of no less than 80GB of video memory.

[0040] ② Vector databases are used to provide data for pre-trained training. For example, deploying and using ChromaDB as a vector database.

[0041] ③ The spacecraft telemetry timing data service database is mainly used to store spacecraft telemetry timing data. For example, Oracle, a database that uses the SQL standard language interface, can be used as the spacecraft telemetry timing data service database.

[0042] II. Construction of Basic Knowledge Base ① The database creation file for spacecraft telemetry timing data uses Oracle to store spacecraft telemetry timing data, and each CREATE statement in the Oracle table creation DDL statement is extracted as a single corpus.

[0043] ② For spacecraft telemetry data documents, use the DATP data stream telemetry parameter definition table, and assign "DATP-C001-vcd002-XXXX" to each corpus entry.

[0044] ③ For the spacecraft telemetry data "problem-query" pair, the query content of the log file and the corresponding SQL statement in the query software of the Oracle spacecraft telemetry time series data business database are used as a single corpus.

[0045] III. Data Pre-training Based on the three types of fragmented corpus data constructed from the aforementioned basic knowledge base, the default Euclidean distance embedding function provided by the vector database ChromeDB is used to embed them into the vector database, forming three vector datasets: business database table creation vector set, telemetry data document vector set, and "question-query" pair vector set.

[0046] IV. Data Query and Display ① Specify the time-series data query requirements. The query requirement is: "Query the telemetry time-series data of the receiver power telemetry parameters for the day following 9:50:00 on April 21, 2025, and plot the data time curve." ② Construct "intermediate LLM prompts," whose lexical structure is "initial prompts (using the generated relevant SQL query language as described below) + user question + data recalled from the basic knowledge base." The "initial prompts" provide more constraints to the large language model, resulting in more accurate feedback, such as: "If the provided context is sufficient, generate a valid SQL query without any explanation of the question; if the provided context is insufficient, explain why it cannot be generated; please use the most relevant table; if the question has been asked and answered before, please repeat the previously given answer exactly." Furthermore, use the query functions provided by ChromaDB to recall user questions from three vector sets, setting the top three relevant pieces of information for recall, and use the bge-reranker-large model to rearrange the recall information. Finally, the constructed "intermediate LLM hints" are as follows: "If the provided context is sufficient, please generate a valid SQL query without any explanation of the question; if the provided context is insufficient, please explain why it cannot be generated; please use the most relevant table; if the question has been asked and answered before, please repeat the previously given answer exactly; query the receiver power telemetry parameters for the telemetry time series data within one day after 9:50:00 on April 21, 2025; you can use the following document as a reference for data lookup entries in the data table: DATP-C001-vcd002-XXXX". Further, the final intermediate LLM hints are fed into a general large language model to generate an intermediate SQL statement, which is then fed into the Oracle spacecraft telemetry time series data business database to retrieve the intermediate query data.

[0047] ③ Construct the final LLM prompt, whose syntax is: "Intermediate LLM prompt content + 'The following is the data retrieved using intermediate SQL' + intermediate SQL statement + actual data retrieved by the intermediate SQL statement". The final LLM prompt is sent to Qwen-7B to construct the final SQL statement and then sent to Oracle for execution to obtain the final queried data.

[0048] ④ Use “data statistics requirements + plotting requirements” to construct data processing and plotting prompts, such as: “Generate Python code for data processing and plotly plotting code according to the following requirements + ‘Plot data time curve’ + data obtained in step ③”. Send the prompts to Qwen-7B to obtain Python code that supports plotly, and execute the code to obtain the time curve of the corresponding telemetry data.

[0049] V. Feedback and Corrections Based on the data query and plotting results in step four, determine the correctness of the data. In this example, the query and plotting are correct, and the current question and the corresponding generated SQL query statement are added to the basic knowledge base construction.

[0050] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solutions of the present invention by utilizing the methods and techniques disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the protection scope of the technical solutions of the present invention.

[0051] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A spacecraft telemetry data visualization method based on a large language model, characterized in that, include: Deploy and configure the basic environment; Build a basic knowledge base; Based on the data in the basic knowledge base, data pre-training is carried out, and the data pre-training results are stored in the vector database; In response to data query and display requests, the system filters and obtains the final query data that matches the data query and display requests, and then displays it. Feedback and corrections will be made based on the final query data.

2. The spacecraft telemetry data visualization method based on a large language model according to claim 1, characterized in that, Deploy and configure the basic environment, including: Deploy a general-purpose large language model; Deploy a vector database to provide vector storage after embedding the pre-trained results; Deploy a spacecraft telemetry timing data service database to store spacecraft telemetry timing data.

3. The spacecraft telemetry data visualization method based on a large language model according to claim 2, characterized in that, The basic knowledge base is constructed by: cleaning and fragmenting the three types of data to obtain three types of fragmented corpus data; and constructing the basic knowledge base based on the three types of fragmented corpus data.

4. The spacecraft telemetry data visualization method based on a large language model according to claim 3, characterized in that, The three types of data were cleaned and fragmented separately to obtain three types of fragmented corpus data, including: The table creation statements of the engineering telemetry, data management telemetry, and ground telemetry data tables in the spacecraft telemetry time series data business database are divided into segments, and the remaining table creation statements are deleted to obtain the table creation file of the spacecraft telemetry time series data business database. The telemetry parameter definition document, technical document, and test summary document were cleaned and fragmented to obtain the spacecraft telemetry data document. Specifically, the telemetry parameter definition document defined the correspondence between data streams, telemetry names, codes, and numbers; text content satisfying the syntax structure "data stream-telemetry code-telemetry number-physical meaning" was extracted as a single corpus entry and then fragmented. The technical document described the relevant subsystems of spacecraft telemetry, using periods as identifiers; the text data in the technical document was fragmented, and other charts were removed. The test summary document provided specific telemetry values ​​for some time points; text satisfying the syntax structure "data stream-telemetry code or telemetry number-physical value" was extracted, fragmented, and other charts were removed. Using the log files of the existing spacecraft telemetry time series data service database query software, the query question content and the corresponding query SQL statement are extracted as a corpus. Irrelevant information in the log is deleted to obtain the spacecraft telemetry data "question-query" pair, that is, the specific telemetry query question content and its corresponding query SQL statement.

5. The spacecraft telemetry data visualization method based on a large language model according to claim 4, characterized in that, Based on the data in the basic knowledge base, data pre-training is performed, and the results are stored in a vector database, including: Based on the spacecraft telemetry time-series data business database table building file, spacecraft telemetry data document, and spacecraft telemetry data "question-query" pair in the basic knowledge base, data pre-training was carried out to obtain the pre-training results of three types of fragmented corpus data; Using the embedding functions provided by the vector database, three vector sets are generated to store the pre-training results of three types of segmented corpus data. The three vector sets are: a business database table creation vector set, used to store the pre-training results of the business database table creation files for spacecraft telemetry time series data; a telemetry data document vector set, used to store the pre-training results of spacecraft telemetry data documents; and a "question-query" pair vector set, used to store the pre-training results of the "question-query" pairs for spacecraft telemetry data.

6. The spacecraft telemetry data visualization method based on a large language model according to claim 5, characterized in that, In response to data query and display requests, the system filters and obtains the final query data that matches the data query and display requests, and displays it, including: In response to data query and display requests, specific time-series data query requirements are proposed. The syntax structure is: "time period + subsystem name + telemetry parameter name". The "time period + subsystem name + telemetry parameter name" is sent as three vector sets to the vector database as retrieval information. The retrieval information recall data is obtained respectively, and the recall data and retrieval information are rearranged in a second time. Construct intermediate LLM prompts with the syntax structure of "initial prompt + user question + data recalled from basic knowledge base"; input the intermediate LLM prompts into a general large language model to obtain intermediate SQL query statements; input the intermediate SQL query statements into the spacecraft telemetry time series data business database for data query; and save the query results as intermediate query data. The final LLM prompt is constructed, with the syntax structure: "intermediate LLM prompt + 'The following is the data retrieved using the intermediate SQL query statement' + intermediate SQL query statement + intermediate query data". The final LLM prompt is then fed into the general large language model to obtain the final SQL statement. The final SQL statement is then fed into the spacecraft telemetry time series data business database for final data query. The query results are saved as the final query data and displayed in a table sorted by timestamp index. The data processing and plotting prompts are constructed using the "data statistics requirements + plotting requirements" framework. The syntax is "generate Python code for data processing and Plotly plotting code based on the following requirements + 'data statistics requirements + plotting requirements' + final query data". The data processing and plotting prompts are then fed into a general large language model to obtain and execute the code, resulting in the final graph, which is then displayed.

7. The spacecraft telemetry data visualization method based on a large language model according to claim 6, characterized in that, The data recalled from the basic knowledge base is adjusted based on the number of recalled data items. Specifically, if the number of recalled data items in the business database table creation vector set is greater than 0 and does not exceed the context window size, the prompt message "You can use the following database table creation statement as a reference to find tables that may be used for data querying + recalled data" is added. If the number of recalled data items in the telemetry data document vector set is greater than 0 and does not exceed the context window size, the prompt message "You can use the following document as a reference to find entries in the data table + recalled data" is added. If the number of recalled data items in the "Question-Query" vector set is greater than 0 and does not exceed the context window size, the prompt message "You can use the following related query questions and corresponding SQL statements as a reference + recalled data" is added.

8. The spacecraft telemetry data visualization method based on a large language model according to claim 7, characterized in that, Based on the final query data, feedback and corrections are made, including: Based on the final query data, determine the correctness of the SQL statement and provide feedback; If the final queried data matches the data query and display request, the current "question-query" pair will be updated in the basic knowledge base; If the final queried data does not match the data query and display request, the problem will be analyzed and corrected step by step based on the information fed back by the general large language model, including: adjusting the general large language model, supplementing the data in the basic knowledge base, increasing relevant data recall, adjusting the initial prompt words, the grammatical structure of the intermediate LLM prompt words, and the grammatical structure of the final LLM prompt words.

Images