Automatic simulation confrontation method based on large model and RAG technology

Abstract

The invention relates to an automatic simulation confrontation method based on a large model and an RAG technology, and belongs to the technical field of software development, the method is based on Ollan, Qwen2 and RAG (AnythingLLM), according to historical red and blue army simulation confrontation mass data, an internal diversified simulation confrontation AI knowledge base and small sample fine tuning, and the large model and the RAG are subjected to large-scale simulation confrontation. According to the method, a red and blue army simulation deployment scheme and simulation confrontation can be imported and automatically generated according to user-level requirements, conversion from a general large model to a private large model in the field is completed, the completed private large model can quickly recommend the red and blue army simulation confrontation deployment scheme according to manual instructions, and the accuracy evaluation of the deduction process is improved.

Description

Technical Field

[0001] The present invention relates to an automatic simulation confrontation method based on a large model and RAG technology, belonging to the technical field of software development, through the debugging of Qwen2 general large model and RAG (AnythingLLM) technology and small samples. Background Art

[0002] With the discovery of modern big data and the continuous changes in its form, traditional big data training and combat simulation methods can no longer meet the needs of the new situation. As an important means to improve the level of military training, the design and application of the simulation system of the Red Army and Blue Army confrontation is particularly important. However, the existing Red Army and Blue Army simulation system often lacks intelligence and automation in the deployment and confrontation of combat personnel and weapons and equipment, resulting in unsatisfactory training results. Therefore, it is necessary to develop a system architecture design method that can automatically deploy combat personnel and weapons and equipment according to simulation requirements, so as to improve the simulation effect and promote the application of big data and big models in training. Summary of the invention

[0003] In order to solve the above problems, the present invention discloses an automatic simulation confrontation method based on a large model and RAG technology, and its specific technical solution is as follows: An automatic simulation confrontation method based on a large model and RAG technology includes the following steps: Step S1, collecting massive data of historical Red Army and Blue Army simulated confrontation, including combat assumptions, force deployment, model resources, weapon status, combat technical index data, weapon equipment performance data, combat standard data and combat performance evaluation; Step S2: According to the installation manual of Ollama, build and install Ollama to run the large language model tool locally; Step S3, download the Qwen2 large model, and use the Ollama run Qwen2 command to import Qwen2 into the Ollama platform. Ollama provides an API that can create, run, and manage large language model instances; Step S4, install and deploy the open source RAG AnythingLLM service according to the installation manual of AnythingLLM; Step S5, calling the AnythingLLM API interface through the developed front-end interface, setting up the connection to the Ollama platform, and accessing the Qwen2 large model; Step S6, through the upload button of the AnythingLLM front-end visualization interface, save the internal diversified simulation adversarial knowledge documents into the vector database and parse the document data; Step S7, the user manually inputs the requirements, and the Qwen2 large model understands the requirements input by the user; Step S8, according to the set workflow, calling the platform interface of the system design modeling platform SEMX, and according to the document data of step S1 parsed in step S6, perfecting and enriching the creation of relevant strategic deployment models; Step S9, the client interface presents the system model results; Step S10: According to the set workflow, after the model is created, the result is fed back to the workflow. The workflow notifies the squad-level intelligent equipment system policy deduction system to call the system construction model API interface and generate a simulation deduction scenario file according to the set scenario. Step S11, the big model deploys the troops deployment, firepower configuration and tactical coordination of the red and blue sides according to the identifiers defined in the scenario file and the deployment plan data parsed in step S6; Step S12, based on the recommendation solution automatically generated by the large model, a visual interface is provided to allow the user to confirm, fine-tune with a small sample, or cancel the simulated confrontation; Step S13, the assumption file confirmed by the user calls the simulation deduction interface to conduct a simulated confrontation, and feeds back the results to the visualization interface to support user comments and confirm the score of this confrontation, providing data support for the accuracy and rationality of the subsequent optimization of the large model recommendation model strategy.

[0004] Furthermore, the specific implementation steps of step S6 are as follows: Step 6.1, document reception and parsing: AnythingLLM receives the files uploaded by users and uses the built-in parser to extract the content through text extraction, format conversion, and content structuring steps; Step 6.2, document storage: the extracted data is stored in the LanceDB vector database; Step 6.3, Document Management: AnythingLLM provides users with the function of creating workspaces. The contents of workspaces will not interfere with or pollute each other, ensuring that the context of each workspace is clear; Step 6.4, provide services: query the vectorized document content through the query API interface provided by AnythingLLM.

[0005] Furthermore, the file formats uploaded by the user in step 6.1 include Word documents, PDF files, txt documents and picture files.

[0006] Furthermore, the Qwen2 large model is a Qwen2 32B large model, which is downloaded from the Mota community and is an open source model in the Ali Tongyi Qianwen series.

[0007] Furthermore, the Ollama is an open source large language model service tool that simplifies the local operation of the large language model AnythingLLM.

[0008] Furthermore, the AnythingLLM is an open source enterprise-level chat solution based on retrieval enhancement to generate RAG.

[0009] The working process of the present invention is: Data collection and preprocessing: Collect massive data of historical Red Army and Blue Army simulated confrontations, including combat scenarios, force deployment, model resources, weapon status and combat technical indicator data, weapon equipment performance data, combat standard data, combat performance evaluation and other data. At the same time, these massive data are governed and stored in the database in a certain format and structure to facilitate subsequent analysis and retrieval.

[0010] Knowledge base construction: Upload a variety of internal simulation confrontation knowledge documents, including tactical manuals, combat guides, and case analyses to RAG to form an AI knowledge base.

[0011] Integration of Qwen2 model and RAG technology: The Qwen2 large model integrates RAG technology to realize the interaction between the Qwen2 model and the external knowledge base.

[0012] Small sample fine-tuning strategy: Provides an interactive interface for users to fine-tune the Qwen2 model to make it more suitable for scenario requirements.

[0013] Automatic recommendation implementation: Provides a user-friendly interface, supports user demand input or import, calls the SEMX (system design modeling platform) platform API interface, and automatically generates the red and blue army simulation deployment plan.

[0014] System optimization: Based on each user's score, the Qwen2 model is continuously optimized to improve the accuracy and practicality of the recommendation plan.

[0015] The beneficial effects of the present invention are: The present invention is based on Ollama, Qwen2 and RAG (AnythingLLM), according to the massive data of historical Red Army and Blue Army simulation confrontation, the internal diversified simulation confrontation AI knowledge base and small sample fine-tuning, so as to realize the automatic generation of Red Army simulation deployment plan and simulation confrontation according to user-level demand import.

[0016] The present invention uses the Qwen2 large model and RAG (AnythingLLM) technology and small sample debugging to improve the conversion from the general large model to the private large model in the field. The improved private large model can quickly recommend the red and blue army simulation confrontation deployment plan according to manual instructions and improve the accuracy of the deduction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0018] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0019] The components involved in the present invention and their corresponding functions are shown in Table 1 below.

[0020] Table 1

[0021] Combined with Figure 1 It can be seen that the automatic simulation confrontation method based on the large model and RAG technology includes the following steps: Step S1, collecting massive data of historical Red Army and Blue Army simulated confrontation, including combat assumptions, force deployment, model resources, weapon status, combat technical index data, weapon equipment performance data, combat standard data and combat performance evaluation data; Step S2: According to the installation manual of Ollama, build and install Ollama to run the large language model tool locally; Step S3, download the Qwen2 large model, and use the Ollama run Qwen2 command to import Qwen2 into the Ollama platform. Ollama provides an API that can create, run, and manage large language model instances; Step S4, install and deploy the open source RAG AnythingLLM service according to the installation manual of AnythingLLM; Step S5, calling the AnythingLLM API interface through the developed front-end interface, setting up the connection to the Ollama platform, and accessing the Qwen2 large model; Step S6, through the upload button of the AnythingLLM front-end visualization interface, upload the internal diverse simulation confrontation knowledge documents, including tactical manuals, combat guides, and case analysis to RAG to form an AI knowledge base, and parse the document data, specifically: Step 6.1, document reception and parsing: AnythingLLM receives the files uploaded by users and uses the built-in parser to extract the content through text extraction, format conversion, and content structuring steps; Step 6.2, document storage: the extracted data is stored in the LanceDB vector database; Step 6.3, Document Management: AnythingLLM provides users with the function of creating workspaces. The contents of workspaces will not interfere with or pollute each other, ensuring that the context of each workspace is clear; Step 6.4, provide services: query the vectorized document content through the query API interface provided by AnythingLLM; Step S7, the user manually inputs the requirements, and the Qwen2 large model parses the semantic content of the requirements; Step S8, according to the set workflow, call the platform interface of the system design modeling platform SEMX, conduct system capability research and analysis on the server side, and improve and enrich the creation of relevant strategic deployment models according to the document data of step S1 parsed in step S6; Step S9, constructing a system model on the client according to the system capability research and analysis, and presenting the system model results on the client interface; Step S10, the squadron-level intelligent equipment system policy deduction system sets a simulation deduction scenario file according to the requirements of system construction; Step S11, the big model automatically sets the force deployment, firepower configuration, tactical coordination, etc. of the red and blue sides through the scenario file according to the identifiers defined in the scenario file and the deployment plan data analyzed in step S6; Step S12, based on the recommendation solution automatically generated by the large model, a visual interface is provided to allow the user to confirm, fine-tune with a small sample, or cancel the simulated confrontation; Step S13, the assumption file confirmed by the user calls the simulation deduction interface to conduct a simulated confrontation, and feeds back the results to the visualization interface to support user comments and confirm the score of this confrontation, providing data support for the accuracy and rationality of the subsequent optimization of the large model recommendation model strategy.

[0022] The specific process of optimizing the accuracy and rationality of the large model recommendation model strategy based on user reviews is as follows: Use user reviews as additional training data for further training and optimization of the model; By associating user reviews with the model’s output, a feedback loop is formed in which the model adjusts its parameters or strategies based on user reviews to improve the accuracy of future reasoning and the model’s generalization ability; Based on the user's review history, the model learns the user's personal preferences and needs, thereby providing more personalized results in future reasoning.

[0023] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0024] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for automatic simulation confrontation based on a large model and RAG technology, characterized in that: The following steps are involved: Step S1, collecting massive data of historical Red Army and Blue Army simulated confrontation, including combat assumptions, force deployment, model resources, weapon status, combat technical index data, weapon equipment performance data, combat standard data and combat performance evaluation; Step S2: According to the installation manual of Ollama, build and install Ollama to run the large language model tool locally; Step S3, download the Qwen2 large model, and use the Ollama run Qwen2 command to import Qwen2 into the Ollama platform. Ollama provides an API that can create, run, and manage large language model instances; Step S4, install and deploy the open source RAG AnythingLLM service according to the installation manual of AnythingLLM; Step S5, calling the AnythingLLM API interface through the developed front-end interface, setting up the connection to the Ollama platform, and accessing the Qwen2 large model; Step S6, through the upload button of the AnythingLLM front-end visualization interface, save the internal diversified simulation adversarial knowledge documents into the vector database and parse the document data; Step S7, the user manually inputs the requirements, and the Qwen2 large model understands the requirements input by the user; Step S8, according to the set workflow, calling the platform interface of the system design modeling platform SEMX, and according to the document data of step S1 parsed in step S6, perfecting and enriching the creation of relevant strategic deployment models; Step S9, the client interface presents the system model results; Step S10: According to the set workflow, after the model is created, the result is fed back to the workflow. The workflow notifies the squad-level intelligent equipment system policy deduction system to call the system construction model API interface and generate a simulation deduction scenario file according to the set scenario. Step S11, the big model deploys the troops deployment, firepower configuration and tactical coordination of the red and blue sides according to the identifiers defined in the scenario file and the deployment plan data parsed in step S6; Step S12, based on the recommendation solution automatically generated by the large model, a visual interface is provided to allow the user to confirm, fine-tune with a small sample, or cancel the simulated confrontation; Step S13, the assumption file confirmed by the user calls the simulation deduction interface to conduct a simulated confrontation, and feeds back the results to the visualization interface to support user comments and confirm the score of this confrontation, providing data support for the accuracy and rationality of the subsequent optimization of the large model recommendation model strategy.

2. The automatic simulation confrontation method based on large model and RAG technology according to claim 1 is characterized in that: The specific implementation steps of step S6 are as follows: Step 6.1, document reception and parsing: AnythingLLM receives the files uploaded by users and uses the built-in parser to extract the content through text extraction, format conversion, and content structuring steps; Step 6.2, document storage: the extracted data is stored in the LanceDB vector database; Step 6.3, Document Management: AnythingLLM provides users with the function of creating workspaces. The contents of workspaces will not interfere with or pollute each other, ensuring that the context of each workspace is clear; Step 6.4, provide services: query the vectorized document content through the query API interface provided by AnythingLLM.

3. The automatic simulation confrontation method based on large model and RAG technology according to claim 2 is characterized in that: The file formats uploaded by the user in step 6.1 include Word documents, PDF files, txt documents and picture files.

4. The automatic simulation confrontation method based on large model and RAG technology according to claim 1 is characterized in that: The Qwen2 large model is a Qwen2 32B large model, downloaded from the Mota community, and is an open source model in the Ali Tongyi Qianwen series.

5. The automatic simulation confrontation method based on large model and RAG technology according to claim 1 is characterized in that: The Ollama is an open source large language model service tool that simplifies running the large language model AnythingLLM locally.

6. The automatic simulation confrontation method based on large model and RAG technology according to claim 1 is characterized in that: The AnythingLLM is an open source enterprise-level chat solution based on retrieval enhancement and RAG generation.

Images