Automatic report generation system and method based on combination of large language model and digital twinning

Abstract

The invention discloses an automatic report generation system and method based on a large language model in combination with digital twinning, and belongs to the technical field of digital twinning. In order to quickly summarize and screen a large amount of information and data generated by a digital twinborn model by using a large language model technology, the invention comprises a report system service layer which comprises a digital twinborn model connection module, a user information database, a system scheduling module, an information sending module and an inter-level connection module, the report system service layer processes user requirements, communicates with the natural language model layer, the data processing layer, the early warning information processing layer and the file service layer, outputs a report and pushes the report to a digital twin system; the natural language model layer comprises a large language model, a role selection module, a language model reasoning module and a language model role module, and the natural language model layer uses prompt engineering to construct roles, generate data analysis abstracts and reports and send the data analysis abstracts and reports back to the report system service layer. According to the method, text summarization can be directly carried out.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital twins, and specifically relates to a report automatic generation system and method based on a large language model combined with digital twins. Background Art

[0002] Digital twin is an industrial monitoring and management technology that utilizes industrial Internet of Things, computer simulation, and three-dimensional visualization. It is widely applied in multiple fields such as manufacturing, energy and power, social governance, and transportation. In production scenarios, the technology of digital twin has been proven to be an effective application. However, for managers, the method of finding the required information from complex digital twin models is not efficient enough.

[0003] With the progress of computer processing power, large language models have made it possible to have automated, intelligent, and text generation programs with deep semantic understanding capabilities. Such models can be used in digital twins to quickly and instantaneously generate meaningful and accurate reports. However, large language models have not been well applied in digital twin systems that already have a large amount of available data and descriptive information. To address the above problems, it becomes very important to find a method to utilize large language model technology to quickly summarize and screen a large amount of information and data generated by digital twin models. Summary of the Invention

[0004] The problem to be solved by the present invention is to utilize large language model technology to quickly summarize and screen a large amount of information and data generated by digital twin models, and propose a report automatic generation system and method based on a large language model combined with digital twins.

[0005] To achieve the above object, the present invention is realized through the following technical solutions:

[0006] A report automatic generation system based on a large language model combined with digital twins, comprising a report system service layer, a natural language model layer, a data processing layer, an early warning information processing layer, and a file service layer;

[0007] The report system service layer includes a digital twin model connection module, a user information database, a system scheduling module, an information sending module, and an inter-layer connection module. The report system service layer processes user requirements, communicates with the natural language model layer, the data processing layer, the early warning information processing layer, and the file service layer, and outputs and pushes reports to the digital twin system;

[0008] The natural language model layer includes a large language model, a role selection module, a language model reasoning module, and a language model role module. The natural language model layer constructs roles using prompt engineering, generates data analysis summaries and reports, and sends them back to the report system service layer;

[0009] The data processing layer processes the collected data or generates charts based on data statistics and processing algorithms, and sends them back to the report system service layer;

[0010] The early warning information processing layer filters and summarizes information based on the early warning information screening algorithm, organizes it into early warning text information, and sends it back to the report system service layer;

[0011] The file service layer connects to the file system in the application scenario of the digital twin system, finds the corresponding file according to the final requirement description vector received from the report system service layer, and returns the file to the report system service layer.

[0012] Furthermore, the language model inference module includes a large inference language model and a streamlined dialogue request module; the language model role module includes a role prompt word module, a role description module, and a role identifier module.

[0013] Furthermore, the report system service layer stores and identifies user information through the user information database, communicates with the natural language model layer, data processing layer, early warning information processing layer, and file service layer based on the requirement description vector using the digital twin model connection module and the inter-layer connection module, and sends the communication results to the user device through the information sending module;

[0014] The requirement description vector D is expressed as follows:

[0015] D = [D EQ , D EV , D W , D M , D F , D T , T]

[0016] Among them, D EQ represents the device requirement vector, D EV represents the environmental requirement vector, D W represents the worker requirement vector, D M represents the material requirement vector, D F represents the file requirement vector, D T represents the time period requirement vector, and T represents the time when the request is initiated.

[0017] Furthermore, a trigger is set in the system scheduling module.

[0018] Further, the language model role module in the natural language model layer constructs a language model role using prompt engineering, then passes it to the language model inference module for model input, and utilizes the role prompt word module to invoke the language inference ability in the large-scale inference language model to generate model input. The model output is input into the refined dialogue request module to request a refined dialogue, and the final output obtained is used as part of the final report or integrated into the digital twin system. The expression is as follows:

[0019]

[0020] Among them, is the completed text content, T roleprompt is the large language model according to the prompt words of the role, T input is the input prompt word, and LLM is the large language model text completion function.

[0021] Further, the language model role module uses code generation to generate a role for internal information processing of the system. The code generation role is a language model role containing code generation prompt words, which generates code through an interface call and delivers it to the digital twin system for execution by the digital twin system;

[0022] The role selection module consists of a structured output language model constructed by prompt engineering and the corresponding control program. Multiple selection roles in the role selection module correspond one-to-one with the model roles in the language model role module. All selection roles determine whether they need their corresponding model roles. If so, they structurally output the data list and prompt words required by the model role, read the structured output, and return the language model code list and the corresponding data list and prompt word list.

[0023] Further, the data processing layer establishes connections with different databases in the digital twin system. Based on the final requirement description vector received in the report system service layer, it uses the requirement description vector to generate Structured Query Language (SQL) statements, executes the statements to query data and obtain descriptive statistical data, and sends them back to the report system service layer.

[0024] Further, the early warning information processing layer establishes connections with different early warning data interfaces in the digital twin system. Based on the final requirement description vector received in the report system service layer, it invokes the application programming interfaces provided by the interfaces to complete the query of early warning information and the generation or acquisition of descriptive statistical data, and sends them back to the report system service layer.

[0025] A method for automatically generating a report based on a large language model combined with a digital twin is realized relying on the described system for automatically generating a report based on a large language model combined with a digital twin, and includes the following steps:

[0026] S1. The user triggers the trigger in the system scheduling module, and the system scheduling module issues a report generation instruction;

[0027] S2. The report system service layer queries the user information database to obtain user information. The report system service layer aggregates the user information, user requests, and current time to the digital twin model connection module for digital twin model response, generates a requirement description vector, and sends it to the data processing layer, warning information processing layer, and file service layer;

[0028] S3. The report system service layer receives the response files from the data processing layer, warning information processing layer, and file service layer;

[0029] S4. The report system service layer combines the received response files from the data processing layer, warning information processing layer, and file service layer with the user requests to obtain a response request, and sends it to the natural language model layer;

[0030] S5. The role selection module in the natural language model layer receives the response request, generates a list of language model codes in the language model role module, and a corresponding data list and prompt word list;

[0031] S6. The natural language model layer generates a response combination based on the list of language model codes, data list, and prompt word list obtained in step S5, and sends it to the report system service layer;

[0032] S7. The report system service layer combines the obtained response combination with the response files from the data processing layer, warning information processing layer, and file service layer into a manuscript;

[0033] S8. The information sending module in the report system service layer sends the manuscript to the user, completing the automated report generation based on the large language model combined with digital twin.

[0034] Advantages of the present invention:

[0035] The automated report generation system based on the large language model combined with digital twin described in the present invention can be directly used in different application scenarios without any modification. When using the large language model deployed in the cloud, it can perform a large number of real-time report generation and sending operations, greatly improving the practical value and popularity of the current digital twin model. It can quickly and timely contact the staff in each position in a customized manner using text, meet the data requirements, reduce the usage cost of digital twin and data visualization, and can directly summarize the text by setting roles using prompt engineering. Description of the Drawings

[0036] Figure 1 It is a schematic structural diagram of the automated report generation system based on the large language model combined with digital twin described in the present invention;

[0037] Figure 2 This is a schematic structural diagram of the service layer of the invention's reporting system;

[0038] Figure 3 This is a schematic structural diagram of the natural language model layer of the invention;

[0039] Figure 4 This is a schematic structural diagram of the language model inference module of the invention;

[0040] Figure 5 This is a schematic structural diagram of the language model role module of the invention;

[0041] Figure 6 This is a schematic diagram of the processing flow of the data processing module. Specific Embodiments

[0042] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only a part of the embodiments of the present invention, rather than all of the specific embodiments. The components of the specific embodiments of the present invention usually described and shown in the accompanying drawings here can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0043] Therefore, the detailed description of the specific embodiments of the present invention provided in the accompanying drawings below is not intended to limit the scope of the claimed present invention, but only represents the selected specific embodiments of the present invention. All other specific embodiments obtained by those skilled in the art based on the specific embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0044] To further understand the content, features and effects of the present invention, the following specific embodiments are exemplified and accompanied by Figure 1 - Attachment Figure 6 The details are as follows:

[0045] Embodiment 1:

[0046] A report automatic generation system based on a large language model combined with digital twins, including a report system service layer 10, a natural language model layer 20, a data processing layer 30, an early warning information processing layer 40, and a file service layer 50;

[0047] The reporting system service layer 10 includes a digital twin model connection module 110, a user information database 120, a system scheduling module 130, an information sending module 140, and an inter-layer connection module 150. The reporting system service layer 10 processes user requirements, communicates with the natural language model layer 20, the data processing layer 30, the early warning information processing layer 40, and the file service layer 50, and outputs and pushes the report to the digital twin system;

[0048] The natural language model layer 20 includes a large language model 210, a role selection module 220, a language model inference module 230, and a language model role module 240. The natural language model layer 20 constructs roles using prompt engineering, generates data analysis summaries and reports, and sends them back to the reporting system service layer 10;

[0049] The data processing layer 30 processes the collected data or generates charts based on data statistics and processing algorithms, and sends them back to the reporting system service layer 10;

[0050] The early warning information processing layer 40 screens and summarizes information based on the early warning information screening algorithm, organizes it into early warning text information, and sends it back to the reporting system service layer 10;

[0051] The file service layer 50 connects to the file system in the application scenario of the digital twin system, finds the corresponding file according to the final requirement description vector received from the reporting system service layer 10, and returns the file to the reporting system service layer 10.

[0052] Furthermore, the language model inference module 230 includes a large inference language model 231 and a streamlined dialogue request module 232; the language model role module 240 includes a role prompt word module 241, a role description module 242, and a role identifier module 243.

[0053] Furthermore, the reporting system service layer 10 stores and identifies user information through the user information database 120, communicates with the natural language model layer 20, the data processing layer 30, the early warning information processing layer 40, and the file service layer 50 based on the requirement description vector using the digital twin model connection module 110 and the inter-layer connection module 150, and sends the communication result to the user device through the information sending module 140;

[0054] The requirement description vector D is expressed as follows:

[0055] D = [D EQ , D EV , D W , D M , D F , D T , T]

[0056] Among them, D EQ represents the equipment demand vector, D EV represents the environmental demand vector, D W represents the worker demand vector, D M represents the material demand vector, D F represents the document demand vector, D T represents the time period demand vector, and T represents the time when the request is initiated.

[0057] Furthermore, a trigger is set in the system scheduling module 130.

[0058] Furthermore, the language model role module 240 of the natural language model layer 20 constructs a language model role using prompt engineering, then passes it to the language model inference module 230 for model input, and uses the role prompt word module 241 to invoke the language inference ability in the large-scale inference language model 231 to generate model input. The model output is input into the refined dialogue request module 232 to request a refined dialogue, and the final output is used as part of the final report or connected to the digital twin system. The expression is:

[0059]

[0060] Among them, is the completed text content, T roleprompt is the large language model according to the prompt words of the role, T input is the input prompt word, and LLM is the large language model text completion function.

[0061] Furthermore, the language model role module 240 uses code generation roles to process information within the system. The code generation role is a language model role containing code generation prompt words, which generates code through an interface call and hands it over to the digital twin system for execution by the digital twin system;

[0062] The role selection module 220 consists of a language model with a structured output constructed by prompt engineering and the corresponding control program. The multiple selection roles in the role selection module 220 correspond one-to-one with the model roles in the language model role module 240. All selection roles determine whether they need their corresponding model roles. If so, they structurally output the data list and prompt words required by the model role, read the structured output, and return the language model code list and the data list and prompt word list corresponding to it one by one.

[0063] Further, the data processing layer 30 establishes connections with different databases in the digital twin system. Based on the final requirement description vector received in the report system service layer 10, it generates Structured Query Language statements using the requirement description vector, executes the statements to query data and obtain descriptive statistical data, and sends them back to the report system service layer 10.

[0064] Further, the warning information processing layer 40 establishes connections with different warning data interfaces in the digital twin system. Based on the final requirement description vector received in the report system service layer 10, it calls the application programming interfaces provided by the interfaces to complete the query of warning information and the generation or acquisition of descriptive statistical data, and sends them back to the report system service layer 10.

[0065] Embodiment 2:

[0066] A method for automatically generating reports based on a large language model combined with digital twins, implemented relying on the report automatic generation system based on a large language model combined with digital twins described in Embodiment 1, includes the following steps:

[0067] S1. The user triggers the trigger in the system scheduling module 130, and the system scheduling module 130 issues a report generation instruction;

[0068] S2. The report system service layer 10 queries the user information database 120 to obtain user information. The report system service layer 10 aggregates the user information, user requests, and current time to the digital twin model connection module 110 for digital twin model response, generates a requirement description vector, and sends it to the data processing layer 30, the warning information processing layer 40, and the file service layer 50;

[0069] S3. The report system service layer 10 accepts the response files from the data processing layer 30, the warning information processing layer 40, and the file service layer 50;

[0070] S4. The report system service layer 10 combines the response files from the data processing layer 30, the warning information processing layer 40, and the file service layer 50 received with the user requests to obtain a response request, and sends it to the natural language model layer 20;

[0071] S5. The role selection module 220 in the natural language model layer 20 accepts the response request, generates a list of language model codes in the language model role module 240, and a corresponding data list and prompt word list;

[0072] S6. The natural language model layer 20 generates a response combination based on the list of language model codes, data list, and prompt word list obtained in step S5, and sends it to the report system service layer 10;

[0073] S7. The report system service layer 10 combines the obtained response combination with the response files from the data processing layer 30, the early warning information processing layer 40, and the file service layer 50 into a document.

[0074] S8. The information sending module 140 in the report system service layer 10 sends the document to the user, completing the automated report generation based on the combination of the large language model and the digital twin.

[0075] It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0076] Although the present application has been described above with reference to specific embodiments, various improvements can be made thereto and components thereof can be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the various features in the specific embodiments disclosed in the present application can be combined with each other in any way, and the exhaustive description of the situations of these combinations is omitted in this specification only for the sake of saving space and resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. An automatic report generation system based on a large language model combined with digital twins, characterized in that: It includes a reporting system service layer (10), a natural language model layer (20), a data processing layer (30), an early warning information processing layer (40), and a file service layer (50); The reporting system service layer (10) includes a digital twin model connection module (110), a user information database (120), a system scheduling module (130), an information sending module (140), and an inter-level connection module (150). The reporting system service layer (10) processes user needs, communicates with the natural language model layer (20), the data processing layer (30), the warning information processing layer (40), and the file service layer (50), and outputs reports and pushes them to the digital twin system. The natural language model layer (20) includes a large language model (210), a role selection module (220), a language model reasoning module (230), and a language model role module (240). The natural language model layer (20) uses prompt engineering to construct roles, generates data analysis summaries and reports, and sends them back to the reporting system service layer (10); The data processing layer (30) processes the collected data or generates charts based on data statistics and processing algorithms, and sends reports back to the system service layer (10); The warning information processing layer (40) screens and summarizes the information based on the warning information screening algorithm, organizes it into warning text information, and sends it back to the reporting system service layer (10); The file service layer (50) is connected to the file system in the application scenario of the digital twin system, finds the corresponding file according to the final demand description vector received from the reporting system service layer (10), and returns the file to the reporting system service layer (10).

2. According to claim 1, the automatic report generation system based on a large language model combined with digital twins is characterized in that: The language model inference module (230) includes a large inference language model (231) and a simplified dialogue request module (232); the language model role module (240) includes a role prompt word module (241), a role description module (242), and a role identifier module (243).

3. According to claim 2, the automatic report generation system based on a large language model combined with digital twins is characterized in that: The reporting system service layer (10) stores and identifies user information through a user information database (120), communicates with a natural language model layer (20), a data processing layer (30), an early warning information processing layer (40), and a file service layer (50) based on a demand description vector using a digital twin model connection module (110) and an inter-level connection module (150), and sends the communication result to a user device through an information sending module (140); The demand description vector D is expressed as follows: D=[D EQ ,D EV ,D W ,D M ,D F ,D T ,T] Among them, D EQ represents the equipment demand vector, D EV represents the environmental demand vector, D W represents the worker demand vector, D M represents the material demand vector, D F represents the file demand vector, D T represents the time period demand vector, and T represents the time when the request is initiated.

4. According to claim 3, the automatic report generation system based on a large language model combined with digital twins is characterized in that: The system scheduling module (130) is provided with a trigger.

5. According to claim 4, the automatic report generation system based on a large language model combined with digital twins is characterized in that: The language model role module (240) of the natural language model layer (20) uses the prompt engineering to construct the language model role, which is then passed to the language model reasoning module (230) for model input, and the language reasoning capability in the large-scale reasoning language model (231) is retrieved using the role prompt word module (241) to generate the model input, and the model output is input into the simplified dialogue request module (232) to request the simplified dialogue, and the final output obtained is used as part of the final report or connected to the digital twin system, and the expression is: in, is the completed text content, T roleprompt For large language models, T input For input prompt words, LLM is the large language model text completion function.

6. The automatic report generation system based on a large language model combined with digital twins according to claim 5 is characterized in that: The language model role module (240) uses a code generation role to perform information processing within the system. The code generation role is a language model role that includes a code generation prompt word. The code is generated through an interface call and handed over to the digital twin system, which is executed by the digital twin system. The role selection module (220) is composed of a structured output language model constructed by the prompting project and a corresponding control program. The multiple selected roles in the role selection module (220) correspond one-to-one to the model roles in the language model role module (240). All selected roles determine whether they need a model role corresponding to them. If necessary, the data list and prompt words required by the model role are structured output, the structured output is read, and a language model code list and a data list and prompt word list corresponding to them are returned.

7. The automatic report generation system based on a large language model combined with digital twins according to claim 6 is characterized in that: The data processing layer (30) establishes connections with different databases in the digital twin system, generates structured query language statements based on the final demand description vector received in the reporting system service layer (10), executes the statements to query data and obtain descriptive statistical data, and sends them back to the reporting system service layer (10).

8. The automatic report generation system based on a large language model combined with digital twins according to claim 7 is characterized in that: The warning information processing layer (40) establishes a connection with different warning data interfaces in the digital twin system, and based on the final demand description vector received in the reporting system service layer (10), calls the application programming interface provided by the interface to complete the warning information query and the generation or acquisition of descriptive statistical data, and sends it back to the reporting system service layer (10).

9. A method for automatically generating reports based on a large language model combined with digital twins, which is implemented by relying on a system for automatically generating reports based on a large language model combined with digital twins as described in one of claims 1 to 8, characterized in that: The steps include: S1. The user triggers the trigger in the system scheduling module (130), and the system scheduling module (130) issues a report generation instruction; S2. The reporting system service layer (10) queries the user information database (120) to obtain user information, and the reporting system service layer (10) aggregates the user information, user request, and current time to the digital twin model connection module (110) for digital twin model response, generates a demand description vector and sends it to the data processing layer (30), the warning information processing layer (40), and the file service layer (50); S3. The reporting system service layer (10) receives the response file from the data processing layer (30), the warning information processing layer (40), and the file service layer (50); S4. The reporting system service layer (10) combines the response files received from the data processing layer (30), the warning information processing layer (40), and the file service layer (50) with the user request to obtain a response request, and sends it to the natural language model layer (20); S5. The role selection module (220) in the natural language model layer (20) receives the response request and generates a language model code list in the language model role module (240), and a data list and a prompt word list corresponding thereto; S6. The natural language model layer (20) generates a response combination based on the language model code list, data list and prompt word list obtained in step S5, and sends it to the reporting system service layer (10); S7. The reporting system service layer (10) combines the obtained response combination with the response files from the data processing layer (30), the warning information processing layer (40), and the file service layer (50) into a document; S8. The information sending module (140) in the reporting system service layer (10) sends the document to the user, completing the automatic generation of reports based on the large language model combined with digital twins.

Images