Configuration data binding method and system based on large model

Abstract

The invention discloses a configuration data binding method and system based on a large model, and the method comprises the steps: binding corresponding target data sources according to different system application scenes, so as to obtain execution result data meeting the data format required by each component; the execution result data is rendered into corresponding components and displayed on a visual interface, the system comprises a plurality of components, and the target data source comprises an API data source and an SQL data source. According to the method, the large model is introduced into the Internet of Things platform system to automatically generate the data conversion script and the SQL statement, so that a user can easily perform configuration data source binding without mastering a programming language or an SQL writing skill. Through the design, the operation difficulty of the user is remarkably reduced, and the overall experience of the user is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic control of Internet of Things platforms, and in particular relates to a configuration data binding method and system based on a large model. Background Art

[0002] One of the core businesses of the Internet of Things (IoT) platform is to effectively manage the large and diverse group of connected devices. In order to achieve efficient management and configuration of devices and systems, the IoT platform usually adopts configuration strategies to achieve visual management, real-time monitoring and operational control of devices. The configuration process is generally performed by equipment operation and maintenance or engineering and technical personnel.

[0003] However, traditional configuration solutions often require collaboration between engineers and software developers, mainly because traditional configuration relies on specific programming skills and SQL knowledge for data binding. For example, when using an API as a data source, if the data format returned by the API does not match the format required for configuration, you need to write a JavaScript script to convert the data format. Similarly, when using an SQL data source, you also need to write an SQL query statement to retrieve the required data. This configuration method that relies on programming and database queries cannot meet the needs of engineers to flexibly adjust the configuration according to actual business needs. Summary of the invention

[0004] (I) Purpose of the invention

[0005] The purpose of the present invention is to provide a configuration data binding method and system based on a large model. With the help of a large language model, the method enables engineers to independently perform configuration design and management without writing any program code or SQL statements to meet the configuration requirements of the Internet of Things platform.

[0006] (II) Technical solution

[0007] To solve the above problems, a first aspect of the present invention provides a configuration data binding method based on a large model, the method comprising:

[0008] Bind the corresponding target data source according to different system application scenarios to generate corresponding prompt words, and input the prompt words into the corresponding target big model to obtain the execution result data that meets the data format required by each component; wherein the system includes several components, and the target data source includes an API data source and an SQL data source;

[0009] The execution result data is rendered into the corresponding component and displayed on the visual interface.

[0010] Furthermore, the binding of the API data source includes:

[0011] Configure interface request information, where the interface is an interface corresponding to the target API data source bound to the first component;

[0012] After verifying the request information, the interface return parameters are obtained;

[0013] Establishing a corresponding mapping relationship according to the interface return parameter and the first component parameter;

[0014] Generate a first prompt word according to the interface return parameter, the first component parameter and the corresponding mapping relationship;

[0015] Input the first prompt word into the first target macro model to generate a corresponding target JavaScript script;

[0016] The target JavaScript script is executed to convert the interface return parameters into the parameter format required by the first component, and the interface return parameters after the conversion are the first execution result data.

[0017] Furthermore, establishing a corresponding mapping relationship according to the interface return parameter and the first component parameter includes:

[0018] If the data format of the interface return parameter is consistent with that of the first component parameter, a field mapping relationship is automatically established;

[0019] If the data formats of the interface return parameter and the first component parameter are inconsistent, the interface return parameter is matched with the first component parameter and then a field mapping relationship is established.

[0020] Furthermore, the first target large model is generated based on the following steps:

[0021] Acquire a first data set and divide it into a first training set and a first validation set, the first data set comprising: the interface return parameter, the first component parameter and a JavaScript script of system configuration;

[0022] The first training set is used to train the general large language model, and the first validation set is used to evaluate the model during the training process, and the optimal first target large model is determined according to the evaluation results.

[0023] Furthermore, the binding of the SQL data source includes:

[0024] Create a SQL data source;

[0025] Select the corresponding database table in the SQL data source according to the business requirements, and click Generate Target SQL Statement according to the business requirements and the selected database table;

[0026] Execute the target SQL statement to generate an SQL execution result, convert the SQL execution result into a parameter format required by the second component, and the SQL execution result after the conversion is the second execution result data.

[0027] Furthermore, before executing the target SQL statement, if the target SQL statement needs to be configured with query parameters, the SQL query parameters are configured according to the target SQL statement before execution.

[0028] Further, the step of generating a target SQL statement according to the business requirements and the selected database table includes:

[0029] Enter user query text in the selected database table;

[0030] Generate an SQL request according to the user query text and the selected database table;

[0031] Based on the SQL request, a basic prompt word template preset in the SQL data source, a schema corresponding to the selected database table, and a history list of related SQL generation are sequentially acquired to generate a second prompt word;

[0032] The second prompt word is input into the second target large model to generate the target SQL statement required by the second component, and then it is determined whether the target SQL statement meets expectations, and the data that meets expectations is saved in the history list of SQL generation for calling.

[0033] Furthermore, the expected data includes user query text, embedding information corresponding to the query text, target SQL statements and related product information.

[0034] Furthermore, the second target large model is generated based on the following steps:

[0035] Collect open source data of text2sql in the field of Internet of Things and general fields to obtain the second data set;

[0036] Cleaning and converting the second data set to obtain a valid data set;

[0037] The effective data set is divided into a second training set and a second validation set, the second training set is used to train the general large language model, and the second validation set is used to evaluate the model during the training process, and the optimal second target large model is determined according to the evaluation result.

[0038] Another aspect of the present invention provides a configuration data binding system based on a large model, the system comprising:

[0039] The target data source binding module is used to bind the corresponding target data source according to different system application scenarios to generate corresponding prompt words, and input the prompt words into the corresponding target big model to obtain the execution result data that meets the data format required by each component; wherein, the system includes several components, and the target data source includes API data source and SQL data source;

[0040] The rendering and display module is used to render the execution result data into the corresponding component and display it on the visual interface.

[0041] (III) Beneficial effects

[0042] The above technical scheme of the present invention has the following beneficial technical effects: the present invention provides a configuration data binding method and system based on a large model. With the help of a large language model, the method enables engineers to independently perform configuration design and management without writing any program code or SQL statements. The process is: bind the corresponding target data source according to the different application scenarios of the system to generate corresponding prompt words, input the prompt words into the corresponding target large model, the target data source includes an API data source and an SQL data source, and finally generate execution result data that meets the data format required by each component; then render the execution result data into the corresponding component and display it on the visual interface. With the help of a large language model, the present invention allows users to bind the data format that is inconsistent with the data format required by the configuration component without any programming, greatly reducing the use threshold and greatly improving the data binding efficiency. At the same time, through the above method, even business personnel who do not have the ability to write SQL can easily generate SQL query statements that meet business needs through visual tools, greatly reducing the use threshold of configuration and improving the efficiency and convenience of data processing. The present invention introduces an innovative solution of automatically generating data conversion scripts and SQL statements from a large model into the Internet of Things platform system, so that users can easily configure data source binding without having to master programming languages ​​or SQL writing skills. This design significantly reduces the user's operating difficulty, greatly improves the user's enthusiasm and convenience in using the platform, lowers the threshold for configuration use, and improves the user's overall experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of a configuration data binding method based on a large model of the present invention;

[0044] Figure 2 is a flowchart of API data source binding according to a specific embodiment of the present invention;

[0045] Figure 3 This is an example diagram of establishing a field mapping relationship in API data source binding in a specific embodiment of the present invention;

[0046] Figure 4 Partial flow chart of SQL data source binding according to a specific embodiment of the present invention;

[0047] Figure 5 The invention discloses a flow chart of a configuration data binding system based on a large model. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concept of the present invention.

[0049] like Figure 1 As shown, one aspect of the present invention provides a configuration data binding method based on a large model, the method comprising:

[0050] S1, bind the corresponding target data source according to different system application scenarios to generate corresponding prompt words, input the prompt words into the corresponding target big model to obtain the execution result data that meets the data format required by each component; wherein the system is, for example, an IoT platform configuration system, which includes several components, such as maps, alarm records, buttons, digital display panels, statistical charts and other components, and the target data source includes API data source and SQL data source. One component corresponds to only one data source (API / SQL);

[0051] S2, rendering the execution result data into corresponding components and displaying them on a visual interface.

[0052] Furthermore, the binding of the API data source is as follows: Figure 2 As shown, including:

[0053] 1) configuring interface request information (interface information), where the interface is an interface corresponding to the target API data source bound to the first component;

[0054] 2) After the request information is verified (interface test module), the interface return parameter (interface return parameter) is obtained. In this step: the correctness of the request is verified by supporting two methods: the http status code or the return parameter includes / does not include the specified content;

[0055] 3) Establishing a corresponding mapping relationship (obtaining the data after interface mapping) according to the interface return parameter and the first component parameter; in this step, if the data format of the interface return parameter and the first component parameter is consistent, a field mapping relationship is automatically established to obtain the data after interface mapping, such as Figure 3The example shown is as follows, Figure 3 The left and right columns are the field lists of the interface return parameters and the field lists of the first component parameters, respectively. For example, if the first component requires a parameter with the device name (deviceName), and the interface return parameter (interface return parameter) also has a deviceName field, a field mapping relationship will be automatically established;

[0056] If the data formats of the interface return parameters and the first component parameters are inconsistent, the interface return parameters are matched with the first component parameters, and then a field mapping relationship is established. The specific operation is: drag the corresponding field in the field list of the interface return parameters and move it to the blank column corresponding to the first component parameters for matching, and then establish a field mapping relationship. For example, the parameter data returned by the configured API interface for obtaining device location information and the parameter data format required by the map component are shown in the first and second columns of Table 1, respectively, and the corresponding display in the user interface is as follows Figure 3 As shown in the figure, the location parameter of the map component does not automatically match the corresponding interface return parameter, so it is blank. You need to manually drag the corresponding longitude field and latitude field from the left to the two blank columns corresponding to the location to complete the data mapping between the interface return parameter field and the corresponding configuration parameters of the chart. Finally, you can get the data example after interface mapping, as shown in the third column of Table 1.

[0057] Table 1 Mapping relationship example description

[0058]

[0059]

[0060] 4) Generate a first prompt word (generate prompt) according to the interface return parameter, the first component parameter and the corresponding mapping relationship; in this step, taking Table 1 as an example, first, the interface response parameters, component required parameters (component parameters) and interface mapped data in Table 1 are transmitted to the server in the first step, and then the server loads the basic prompt word template from the configuration, and then fills the interface response parameters, component required parameters and mapped data transmitted by the front end in the first step into the basic prompt word template to generate prompt words, as shown in Table 2:

[0061] Table 2 Basic prompt word template example description

[0062]

[0063] 5) Input the first prompt word into the first target macro model to generate a corresponding target JavaScript script; in this step: the generated target JavaScript script will be saved to the database and returned to the front end to display the generated JavaScript script, as shown in Table 3:

[0064] Table 3 Example of generating target JavaScript script

[0065]

[0066] 6) Execute the target JavaScript script and convert the interface return parameters into the parameter format required by the first component (corresponding to Figure 2 In the data content module), the interface return parameter after the conversion format is the first execution result data. In this step: while the front end displays JavaScript, it will automatically execute the target JavaScript script and convert the interface return parameter into the parameter format required by the first component.

[0067] Finally, the first execution result data is rendered into the corresponding component and displayed on the visual interface (corresponding to Figure 2 Component rendering module). Through the above API data source binding method, the present invention uses a large language model, and the user does not need to perform any programming to bind the API data format that is inconsistent with the data format required by the configuration component, which greatly reduces the usage threshold and can greatly improve the data binding efficiency.

[0068] Furthermore, the first target large model is generated based on the following steps:

[0069] 1) obtaining a first data set and dividing it into a first training set and a first validation set, wherein the first data set includes: the interface return parameters, the first component parameters, and a JavaScript script of system configuration;

[0070] 2) Use the first training set to train the general large language model, and use the first validation set to evaluate the model during the training process, and determine the optimal first target large model based on the evaluation results. In this step: when the validation results indicate that the answer accuracy of the general large language large model reaches the target threshold of 99%, save the current model parameters to obtain the optimal first target large model.

[0071] Furthermore, the binding of the SQL data source includes:

[0072] 1) Create a SQL data source, including configuring the SQL data source name, database address, database and account password;

[0073] 2) Select the corresponding database table in the SQL data source according to the business needs. According to the business needs and the selected database table, click to generate the target SQL statement. The key link of the SQL data source is the generation of SQL statements. The present invention generates SQL statements with the help of a large language model. Users select the database table to be queried according to business needs. Compared with matching tables according to task descriptions, this can narrow the query scope and improve the accuracy of subsequent SQL generation. Figure 4 , detailing the step:

[0074] a. Enter the user query text (query requirements) in the selected database table;

[0075] b. Generate an SQL request according to the user query text and the selected database table;

[0076] c. based on the SQL request, sequentially obtain the basic prompt word template preset in the SQL data source, the schema corresponding to the selected database table, and the history list of related SQL generation to generate a second prompt word;

[0077] d. Input the second prompt word into the second target large model to generate the target SQL statement required by the second component, then determine whether the target SQL statement meets expectations, and save the data that meets expectations into the history list of SQL generation, which is used to provide relevant examples (history list of relevant SQL generation) when generating SQL in step c, thereby improving the accuracy of SQL generation; the data that meets expectations includes user query text, embedding information corresponding to the query text, target SQL statement and related product information.

[0078] In addition, before executing the target SQL statement, if the target SQL statement needs to configure query parameters, configure the SQL query parameters according to the target SQL statement before executing. For example: if the SQL statement is dynamic SQL, query parameters need to be configured. For example, perhaps the requirement is to query a range of alarm records, and the generated SQL will be similar to: select * from iot_alter_log where create_time>=${startTime}and create_time<=${endTime}, where query parameters need to be configured, and the parameter name corresponds to the dynamic parameter ${param} in the query SQL.

[0079] 3) Execute the target SQL statement to generate an SQL execution result. Similar to the above API, convert the SQL execution result into the parameter format required by the second component. The SQL execution result after the conversion is the second execution result data.

[0080] Finally, the second execution result data is rendered into the corresponding component and displayed on the visual interface.

[0081] To facilitate understanding of the above SQL generation process, a specific example is provided: querying the alarm records of a specified product on the day. The detailed steps are as follows:

[0082] 1) The user selects the table in the data query range. In this example, the iot_alter_log table needs to be selected.

[0083] 2) The user enters the query requirement. The query requirement in this example is: Query the alarm records of the specified product on the day

[0084] 3) SQL generation, this step is divided into two steps:

[0085] The first step is to generate prompt words (second prompt words). Load the basic prompt words preset in the SQL generation module, user requirements, the schema corresponding to the selected table iot_alter_log, and the most relevant SQL example found. The generated prompt word examples are shown in Table 4:

[0086] Step 2: Input the generated prompt into the large language model to guide the generated SQL statement as follows:

[0087] select*from iot_alert_log ial where ial.product_id=${productId}andial.create_time>=${startTime}and ial.create_time<=${endTime}

[0088] Table 4 Examples of generated second prompt words

[0089]

[0090] Through the above method, even business personnel who do not have the ability to write SQL can easily generate SQL query statements that meet business needs through visual tools, which greatly reduces the threshold for configuration use and improves the efficiency and convenience of data processing.

[0091] Furthermore, the second target large model is generated based on the following steps:

[0092] 1) Collect open source text2sql data in the field of Internet of Things and general fields to obtain the second data set. In the SQL data source binding solution, a large language model that can generate SQL based on text is required. The large language model is based on the general large language model, and the text2sql data set in the field of Internet of Things is constructed as professional domain knowledge. The data set in the field of general database is combined to fine-tune the model to obtain a special model applied to the configuration of Internet of Things;

[0093] 2) Cleaning and converting the second data set (converting it into a required format) to obtain a valid data set;

[0094] 3) The effective data set is divided into a second training set and a second validation set, and the general large language model is trained using the second training set to update the model parameters; and the model in the training process is evaluated using the second validation set, and the optimal second target large model is determined according to the evaluation results. When the validation results indicate that the SQL accuracy of the general large model reaches the target threshold of 99%, the current model parameters are saved to obtain the optimal second target large model. The training set accounts for 70%-80% of the data set, and the validation set accounts for 20%-30% of the data set.

[0095] Another aspect of the present invention provides a configuration data binding system based on a large model, such as Figure 5 As shown, the system comprises:

[0096] The target data source binding module is used to bind the corresponding target data source according to different system application scenarios to generate corresponding prompt words, and input the prompt words into the corresponding target big model to obtain the execution result data that meets the data format required by each component; wherein, the system includes several components, and the target data source includes API data source and SQL data source;

[0097] The rendering and display module is used to render the execution result data into the corresponding component and display it on the visual interface.

[0098] Furthermore, the binding target data source module includes an API data source binding submodule, which includes:

[0099] A configuration interface unit, used to configure interface request information, wherein the interface is an interface corresponding to a target API data source bound to the first component;

[0100] The interface testing unit is used to obtain the interface return parameters after verifying the request information;

[0101] Establishing a mapping unit, used to establish a corresponding mapping relationship according to the interface return parameter and the first component parameter;

[0102] A first prompt word generating unit, configured to generate a first prompt word according to the interface return parameter, the first component parameter and the corresponding mapping relationship;

[0103] A first input model unit, used for inputting the first prompt word into a first target large model to generate a corresponding target JavaScript script;

[0104] The first conversion format unit is used to execute the target JavaScript script and convert the interface return parameters into the parameter format required by the first component. The interface return parameters after the conversion are the first execution result data.

[0105] Furthermore, the mapping establishment unit is specifically used for:

[0106] If the data format of the interface return parameter is consistent with that of the first component parameter, a field mapping relationship is automatically established;

[0107] If the data formats of the interface return parameter and the first component parameter are inconsistent, drag the corresponding field in the field list of the interface return parameter and move it to the blank column corresponding to the first component parameter for matching, and then establish a field mapping relationship.

[0108] Furthermore, the binding target data source module includes a SQL data source binding submodule, which includes:

[0109] Create a unit to create a SQL data source;

[0110] Generate target statement unit, used to select the corresponding database table in the SQL data source according to business requirements, and click to generate target SQL statement according to the business requirements and the selected database table;

[0111] The second format conversion unit is used to execute the target SQL statement, generate an SQL execution result, and convert the SQL execution result into a parameter format required by the second component. The SQL execution result after the conversion is the second execution result data.

[0112] Furthermore, the SQL data source binding submodule also includes a query parameter configuration unit, which is used to configure the SQL query parameters according to the target SQL statement before executing the target SQL statement if the target SQL statement needs to configure query parameters.

[0113] Furthermore, the target sentence generation unit includes:

[0114] An input text subunit is used to input user query text into a selected database table;

[0115] A SQL request generating subunit, used for generating a SQL request according to the user query text and the selected database table;

[0116] A second prompt word generating subunit is used to sequentially obtain a basic prompt word template in the SQL data source, a schema corresponding to the selected database table, and a history list of related SQL generation based on the SQL request to generate a second prompt word;

[0117] The second input sub-unit is used to input the second prompt word into the second target large model, generate the target SQL statement required by the second component, then determine whether the target SQL statement meets expectations, and save the data that meets expectations into the history list of SQL generation.

[0118] Furthermore, the rendering and display module includes a first rendering and display module and a second rendering and display module, the first rendering and display module is used to render the first execution result data into a corresponding component and display it on a visual interface, and the second rendering and display module is used to render the second execution result data into a corresponding component and display it on a visual interface.

[0119] A person skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes the following process of the above-mentioned embodiment of each method. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

Claims

1. A configuration data binding method based on a large model, characterized in that: The method comprises: Bind the corresponding target data source according to different system application scenarios to generate corresponding prompt words, and input the prompt words into the corresponding target big model to obtain the execution result data that meets the data format required by each component; wherein the system includes several components, and the target data source includes an API data source and an SQL data source; The execution result data is rendered into the corresponding component and displayed on the visual interface.

2. The configuration data binding method based on a large model according to claim 1 is characterized in that: The binding of the API data source includes: Configure interface request information, where the interface is an interface corresponding to the target API data source bound to the first component; After verifying the request information, the interface return parameters are obtained; Establishing a corresponding mapping relationship according to the interface return parameter and the first component parameter; Generate a first prompt word according to the interface return parameter, the first component parameter and the corresponding mapping relationship; Input the first prompt word into the first target macro model to generate a corresponding target JavaScript script; The target JavaScript script is executed to convert the interface return parameters into the parameter format required by the first component, and the interface return parameters after the conversion are the first execution result data.

3. The configuration data binding method based on a large model according to claim 2 is characterized in that: The establishing a corresponding mapping relationship according to the interface return parameter and the first component parameter comprises: If the data format of the interface return parameter is consistent with that of the first component parameter, a field mapping relationship is automatically established; If the data formats of the interface return parameter and the first component parameter are inconsistent, the interface return parameter is matched with the first component parameter and then a field mapping relationship is established.

4. The configuration data binding method based on a large model according to claim 2 is characterized in that: The first target large model is generated based on the following steps: Acquire a first data set and divide it into a first training set and a first validation set, the first data set comprising: the interface return parameter, the first component parameter and a JavaScript script of system configuration; The first training set is used to train the general large language model, and the first validation set is used to evaluate the model during the training process, and the optimal first target large model is determined according to the evaluation results.

5. The configuration data binding method based on a large model according to claim 1 is characterized in that: The binding of the SQL data source includes: Create a SQL data source; Select the corresponding database table in the SQL data source according to the business requirements, and click Generate Target SQL Statement according to the business requirements and the selected database table; Execute the target SQL statement to generate an SQL execution result, convert the SQL execution result into a parameter format required by the second component, and the SQL execution result after the conversion is the second execution result data.

6. The configuration data binding method based on a large model according to claim 5 is characterized in that: Before executing the target SQL statement, if the target SQL statement needs to be configured with query parameters, the SQL query parameters are configured according to the target SQL statement before execution.

7. The configuration data binding method based on a large model according to claim 5 is characterized in that: According to the business requirements and the selected database table, clicking to generate a target SQL statement includes: Enter user query text in the selected database table; Generate an SQL request according to the user query text and the selected database table; Based on the SQL request, a basic prompt word template preset in the SQL data source, a schema corresponding to the selected database table, and a history list of related SQL generation are sequentially acquired to generate a second prompt word; The second prompt word is input into the second target large model to generate the target SQL statement required by the second component, and then it is determined whether the target SQL statement meets expectations, and the data that meets expectations is saved in the history list of SQL generation for calling.

8. The configuration data binding method based on a large model according to claim 7 is characterized in that: The expected data includes user query text, embedding information corresponding to the query text, target SQL statements and related product information.

9. The configuration data binding method based on a large model according to claim 7, characterized in that: The second target large model is generated based on the following steps: Collect open source data of text2sql in the field of Internet of Things and general fields to obtain the second data set; Cleaning and converting the second data set to obtain a valid data set; The effective data set is divided into a second training set and a second validation set, the second training set is used to train the general large language model, and the second validation set is used to evaluate the model during the training process, and the optimal second target large model is determined according to the evaluation result.

10. A configuration data binding system based on a large model, characterized in that: The system comprises: The target data source binding module is used to bind the corresponding target data source according to different system application scenarios to generate corresponding prompt words, and input the prompt words into the corresponding target big model to obtain the execution result data that meets the data format required by each component; wherein, the system includes several components, and the target data source includes API data source and SQL data source; The rendering and display module is used to render the execution result data into the corresponding component and display it on the visual interface.

Images