Livestock breeding management system and method based on LLM and multi-source data fusion, and medium

Abstract

This invention relates to a livestock farming management system, method, and medium based on LLM and multi-source data fusion, belonging to the field of livestock farming management technology. It aims to solve problems such as scarce expert resources, tool misuse, and low solution reuse rates. The system comprises three main modules: a virtual expert consultant module, integrating LLM and RAG technologies to provide real-time responses to lean problem consultations; an intelligent guidance generation module, combining NLP and a visualization engine to automatically generate operation guides and graphical documents for tools such as Kanban boards and 5 Whys; and a cross-scenario recommendation module, based on a vectorized case library and semantic matching, to intelligently push historically successful solutions. This invention transforms expert experience into iterative and reusable digital assets, constructing a closed loop of "demand identification—intelligent response—continuous optimization," significantly reducing reliance on manual labor, improving the accuracy and response efficiency of tool applications, and promoting the transformation of lean management from experience-driven to data-driven. It is applicable to industries such as manufacturing, supply chain, and services, and has strong applicability and scalability.

Description

Technical Field

[0001] This invention relates to the field of livestock breeding management technology, and in particular to a livestock breeding management system, method and medium based on LLM and multi-source data fusion. Background Technology

[0002] Lean management, as a business management method centered on eliminating waste and continuous improvement, has been widely applied in the manufacturing industry. However, it faces three major bottlenecks in practice: Scarcity of expert resources: The number of lean experts within the company is limited, making it difficult to cover all business units or independent operating units, resulting in untimely and insufficient guidance, and loss of control over project quality.

[0003] Non-standard application of tools: Frontline staff do not have a proper understanding of tools such as Kanban, 5 Whys, and value stream mapping, often resulting in misuse, formalization, and inability to implement them, which affects the improvement effect.

[0004] Experience is difficult to reuse: successful improvement cases are scattered across different departments, lacking structured accumulation and intelligent matching mechanisms, leading to "reinventing the wheel" and low efficiency.

[0005] Traditional solutions, such as offline training, on-site expert support, and standardized manuals, suffer from drawbacks such as high costs, slow response, poor flexibility, and lagging knowledge updates, and cannot meet the needs of modern enterprises for rapid iteration and data-driven operations.

[0006] Especially in the livestock industry, lean management faces even more severe challenges: the live animal production process is greatly affected by biological characteristics and is highly volatile, making it difficult to directly apply traditional Kanban boards; experience in key aspects such as disease prevention and control, batch scheduling, and feed loss is highly dependent on "veteran workers," and the training cycle for new employees is long; and best practices across farms and regions (such as a certain farm's "improvement plan to reduce the mortality rate of chicks in the first 7 days") are difficult to effectively accumulate and reuse.

[0007] There is an urgent need for an intelligent auxiliary system that does not rely on on-site experts and can be adapted to the characteristics of biological production. This system should: solidify expert knowledge, such as disease prevention and control SOPs, batch scheduling experience, and feeding optimization strategies, into readily available assets; provide dynamically adjustable lean tool variants (such as "flexible Kanban" and "biological beat analysis") to address the volatility of live animal production; and break down data silos across multiple farms within the group, enabling "innovation in one location, reuse in multiple locations." Experience should be made explicit, tools should be scenario-based, and solutions should be transferable. Summary of the Invention

[0008] This invention overcomes the shortcomings of the prior art and provides a livestock breeding management system, method and medium based on LLM and multi-source data fusion.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a livestock farming management system based on LLM and multi-source data fusion, comprising: The natural language interaction unit is used to receive lean management-related questions or requirements from users and to perform intent recognition. The virtual expert advisor unit couples a large language model with retrieval enhancement and generation technology, dynamically retrieves a structured knowledge base based on intent, and generates diagnostic conclusions and natural language responses for lean problems; The intelligent guidance generation unit, based on natural language processing and visualization engine, responds to tool usage needs, generates charts based on data, and automatically generates standardized operation guidelines and corresponding graphical documents for at least one lean tool in Kanban, 5 Whys analysis, and value stream mapping. The cross-scenario recommendation unit, based on a vectorized historical case library and semantic matching mechanism, recommends the most suitable cross-departmental and cross-business line improvement solutions according to the semantic features of the current problem. The feedback optimization unit is used to collect user evaluation and solution implementation effect data, and to update the knowledge base, case library and model parameters accordingly.

[0010] Furthermore, in the livestock breeding management system based on LLM and multi-source data fusion, the structured knowledge base stores lean management theories, tool standards, typical problem answers, and industry best practices; the knowledge base is constructed using a vector database, supports dynamic retrieval based on semantic similarity, and RAG technology injects the retrieved Top-K knowledge fragments as context into the LLM before generating a response.

[0011] Furthermore, in the livestock farming management system based on LLM and multi-source data fusion, the intelligent guidance generation unit includes: The instruction parsing subunit transforms natural language requirements into structured operation steps; The graphics generation subunit calls the visualization component to automatically generate graphical representations of the corresponding lean tools based on the data; The document synthesis subunit integrates step-by-step instructions, diagrams, and precautions to output a downloadable standardized document.

[0012] Furthermore, in the livestock farming management system based on LLM and multi-source data fusion, based on a vectorized historical case library and semantic matching mechanism, and according to the semantic characteristics of the current problem, the system recommends the most suitable cross-departmental and cross-business line improvement solutions, specifically including: Historical improvement cases are encoded using text embedding to construct a high-dimensional vector case library; The semantic similarity between the current problem and historical cases is calculated using an approximate nearest neighbor algorithm. Based on the matching score, case success rate, and time decay factor, the system outputs a list of recommended solutions and the reasons for the recommendations.

[0013] Furthermore, in a livestock farming management system based on LLM and multi-source data fusion, the RAG retrieval process includes: The user's question is vectorized, the Top-K similar knowledge fragments are retrieved from the vector knowledge base, and the fragments are concatenated with the question to form a Prompt, which is then input into the LLM to generate the final response.

[0014] Furthermore, in the livestock farming management system based on LLM and multi-source data fusion, the generated graphical documents include: 5Whys fishbone diagram, value stream map, and Kanban status signal diagram.

[0015] The second aspect of this invention provides a livestock farming management method based on LLM and multi-source data fusion, comprising the following steps: Receive lean management-related questions or requirements from users and perform intent recognition; By coupling a large language model with retrieval enhancement generation technology, a structured knowledge base is dynamically retrieved based on intent to generate diagnostic conclusions and natural language responses for lean problems; Based on natural language processing and visualization engines, it responds to tool usage needs, generates charts based on data, and automatically generates standardized operation guidelines and corresponding graphical documents for at least one lean tool, such as Kanban boards, 5 Whys analysis, and value stream mapping. Based on a vectorized historical case library and semantic matching mechanism, and according to the semantic characteristics of the current problem, the most suitable cross-departmental and cross-business line improvement solutions are recommended. Collect user feedback and implementation effectiveness data, and update the knowledge base, case library, and model parameters accordingly.

[0016] A third aspect of the present invention provides a computer-readable storage medium, including a livestock breeding management method program based on LLM and multi-source data fusion, wherein when the livestock breeding management method program based on LLM and multi-source data fusion is executed by a processor, it implements the steps of the livestock breeding management method based on LLM and multi-source data fusion.

[0017] This invention addresses the shortcomings of the prior art and has the following beneficial effects: This invention transforms expert experience into iterative and reusable digital assets, constructing a closed loop of "demand identification - intelligent response - continuous optimization," significantly reducing reliance on manual labor, improving the accuracy and efficiency of tool application, and driving the transformation of lean management from experience-driven to data-driven. It is applicable to industries such as manufacturing, supply chain, and services, and has strong feasibility and scalability. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the overall methodology of a livestock farming management system based on LLM and multi-source data fusion is presented. Figure 2 The overall system architecture diagram of the livestock breeding management system based on LLM and multi-source data fusion is shown. Detailed Implementation

[0020] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0022] like Figure 1 and Figure 2 As shown, the first aspect of the present invention provides a livestock farming management system based on LLM and multi-source data fusion, comprising: The natural language interaction unit is used to receive lean management-related questions or requirements from users and to perform intent recognition. The virtual expert advisor unit couples a large language model with retrieval enhancement and generation technology, dynamically retrieves a structured knowledge base based on intent, and generates diagnostic conclusions and natural language responses for lean problems; The intelligent guidance generation unit, based on natural language processing and visualization engine, responds to tool usage needs, generates charts based on data, and automatically generates standardized operation guidelines and corresponding graphical documents for at least one lean tool in Kanban, 5 Whys analysis, and value stream mapping. The cross-scenario recommendation unit, based on a vectorized historical case library and semantic matching mechanism, recommends the most suitable cross-departmental and cross-business line improvement solutions according to the semantic features of the current problem. The feedback optimization unit is used to collect user evaluation and solution implementation effect data, and to update the knowledge base, case library and model parameters accordingly.

[0023] It should be noted that this invention constructs a lean management intelligent agent architecture of "four-unit collaboration and one closed-loop drive", the core of which includes: Natural Language Interaction Unit: Serves as a unified user entry point, enabling intelligent distribution of user requests; Virtual Expert Advisor Unit: Integrating LLM and RAG, providing precise diagnosis and knowledge Q&A; Intelligent guidance generation unit: Based on NLP and visualization engine, it generates scenario-based tool guidance; Cross-scenario recommendation unit: Based on a vectorized case library, it achieves highly adaptable solution recommendations; A supporting feedback optimization unit drives the continuous evolution of the system.

[0024] The system operates in a closed loop of "demand identification → intelligent response → continuous optimization". It does not require mandatory connection to external systems and can be deployed independently on the enterprise intranet or cloud, supporting access from multiple terminals such as web pages and apps.

[0025] It should be noted that the intelligent guidance generation unit is used to respond to tool usage needs. It has pre-built lean tool templates suitable for different industry scenarios, including general manufacturing templates and special template variants for livestock farming. This unit performs natural language parsing of user needs, extracts key parameters, matches the corresponding templates, and calls visualization components to generate standardized guidance documents, including operation steps and graphical representations (such as 5Whys fishbone diagrams, value stream maps, and Kanban layout diagrams), and supports exporting to PDF, Word, or HTML formats.

[0026] The natural language interaction unit is used to receive user input in the form of text or voice questions or needs, and to perform intent recognition, classifying the input into knowledge consultation, tool usage, or solution seeking.

[0027] The virtual expert consultant unit couples a large language model with retrieval enhancement generation technology. When it receives a knowledge consultation request, it retrieves relevant knowledge fragments from the structured knowledge base based on the identified intent, combines the fragments with the original question to form a prompt message, and inputs it into the large language model to generate a diagnostic conclusion or answer in natural language form.

[0028] Furthermore, in the livestock breeding management system based on LLM and multi-source data fusion, the structured knowledge base stores lean management theories, tool standards, typical problem answers, and industry best practices; the knowledge base is constructed using a vector database, supports dynamic retrieval based on semantic similarity, and RAG technology injects the retrieved Top-K knowledge fragments as context into the LLM before generating a response.

[0029] Standardized guidance documents, and support exporting to PDF, Word, or HTML formats.

[0030] The cross-scenario recommendation unit is used to respond to the needs of solution seeking. It has built a vectorized database of historical improvement cases. After semantically encoding the current problem, the unit performs similarity retrieval in the case library, and performs weighted sorting based on metadata such as the success rate and time freshness of the cases, and outputs several recommended solutions and their key implementation information.

[0031] The feedback optimization unit is used to collect user evaluation information on the system response results and the effect data of subsequent solution implementation. This data is then used to update knowledge base entries, adjust the ranking weight of the case library, and optimize the prompt words or search parameters of the large language model, enabling the system to have continuous iteration capabilities.

[0032] Furthermore, in the livestock farming management system based on LLM and multi-source data fusion, the intelligent guidance generation unit includes: The instruction parsing subunit transforms natural language requirements into structured operation steps; The graphics generation subunit calls the visualization component to automatically generate graphical representations of the corresponding lean tools based on the data; The document synthesis subunit integrates step-by-step instructions, diagrams, and precautions to output a downloadable standardized document.

[0033] Furthermore, in the livestock farming management system based on LLM and multi-source data fusion, based on a vectorized historical case library and semantic matching mechanism, and according to the semantic characteristics of the current problem, the system recommends the most suitable cross-departmental and cross-business line improvement solutions, specifically including: Historical improvement cases are encoded using text embedding to construct a high-dimensional vector case library; The semantic similarity between the current problem and historical cases is calculated using an approximate nearest neighbor algorithm. Based on the matching score, case success rate, and time decay factor, the system outputs a list of recommended solutions and the reasons for the recommendations.

[0034] Furthermore, in a livestock farming management system based on LLM and multi-source data fusion, the RAG retrieval process includes: The user's question is vectorized, the Top-K similar knowledge fragments are retrieved from the vector knowledge base, and the fragments are concatenated with the question to form a Prompt, which is then input into the LLM to generate the final response.

[0035] Furthermore, in the livestock farming management system based on LLM and multi-source data fusion, the generated graphical documents include: 5Whys fishbone diagram, value stream map, and Kanban status signal diagram.

[0036] In summary, the present invention converts expert experience into iterable and reusable digital assets, constructs a closed loop of "demand identification - intelligent response - continuous optimization", significantly reduces manual dependence, improves the accuracy of tool application and response efficiency, promotes the transformation of lean management from experience-driven to data-driven, is applicable to manufacturing, supply chain, service and other industries, and has strong implementability and scalability.

[0037] Existing breeding management systems have single functions, and all management links (feeding, environmental regulation, epidemic prevention and control, reproduction management, etc.) are independent of each other, lacking global collaborative decision-making capability. Managers need to switch operations among multiple subsystems, and it is difficult to obtain an overall optimization scheme. The present system further includes: Deploy a perception agent, an environmental regulation agent, a nutrition and feeding agent, an epidemic prevention and control agent, a reproduction management agent, and a marketing and supply chain agent, where each agent exchanges information through a unified communication protocol, and adopts a publish-subscribe mode to realize loose coupling collaboration; When the perception agent detects an abnormal event, it broadcasts the event information to relevant agents to trigger a collaborative decision-making process, and a master large language model is configured to receive the breeding management instruction input by the user for semantic parsing and task decomposition, generate a subtask tree, and distribute each subtask to the corresponding professional agent; When generating a decision scheme, each agent comprehensively considers the target dimensions of production benefit, animal welfare, economic cost and biosafety, and carries out multi-objective comprehensive evaluation and sorting; Record the execution result of each agent's decision and user feedback, and continuously optimize the decision strategy of each agent through a feedback learning mechanism.

[0038] It should be noted that the perception agent is responsible for interfacing with various IoT sensors, cameras, and wearable devices, continuously collecting multi-modal data such as environmental parameters, individual physiological data, and behavior videos, and performing data cleaning and standardization processing; the environmental regulation agent focuses on the analysis and regulation decision-making of the breeding environment (temperature, humidity, ammonia concentration, carbon dioxide concentration, illumination, etc.); the nutrition and feeding agent formulates a precise feeding scheme based on factors such as the individual growth stage, body weight, and health status; the epidemic prevention and control agent performs epidemic risk identification and prevention and control strategy generation based on anomaly detection and an epidemic knowledge base; the reproduction management agent is responsible for the whole reproduction process decision-making such as estrus monitoring, mating timing optimization, and pregnancy management. The marketing and supply chain agent is responsible for growth prediction, optimal marketing timing judgment and supply chain optimization.

[0039] It should be noted that the present method realizes autonomous and intelligent decision-making for the whole breeding process, greatly reduces the dependence on manual experience; moreover, the multi-agents work in division and collaboration, performing their respective duties and cooperating in a linked manner, taking into account both professional depth and global optimization; finally, the closed-loop feedback mechanism enables the system to evolve continuously and become more intelligent with use.

[0040] Current aquaculture environment control primarily relies on reactive control based on real-time monitoring data, lacking the ability to predict future environmental changes. This results in delayed control and high energy consumption. Furthermore, adjustments to control strategies lack visual verification methods, making it difficult to assess the expected effects of strategy adjustments. This system also includes: Based on the building information model (BIM) and computational fluid dynamics (CFD) model of the aquaculture shed, a digital twin 3D model of the aquaculture environment is constructed. The model includes geometric and physical information such as the internal spatial structure, ventilation layout, and equipment location. By deploying a distributed sensor array in the breeding shed, including temperature, humidity, wind speed, ammonia, and carbon dioxide sensors, environmental data is continuously collected and mapped to a digital twin model in real time, achieving synchronization between physical and digital spaces. Historical environmental data sequences are input into a large language model, combined with meteorological forecast data, and LLM generates a text predicting environmental change trends for the next few hours, including temperature fluctuation ranges and extreme weather impact assessments. The prediction results and candidate control strategies generated by LLM are input into a digital twin platform for simulation. The system simulates the execution effects of different control strategies in a virtual environment and outputs visualized simulation results of environmental changes; the strategies include ventilation adjustment and the activation of cooling measures. Based on simulation results, LLM selects the best control scheme from multiple candidate strategies and generates executable control commands. The selected strategy is first virtually verified on the digital twin platform. After confirming that there are no negative effects, it is then sent to the actual control system for execution. It should be noted that this invention utilizes digital twin simulation and LLM to reduce the cost of strategy trial and error, avoid the negative impact of improper regulation on farmed animals, and provide decision explanations, enhancing the interpretability of the system and the trust of managers. It can significantly reduce energy consumption in the farming environment and achieve a balance between energy conservation and farming efficiency.

[0041] The diagnosis of livestock and poultry diseases heavily relies on the expertise and experience of veterinary specialists, while grassroots farms often lack such personnel. Existing disease diagnosis systems are mostly rule-matching based, lacking the ability to deeply reason about complex symptom combinations, and their prevention and control strategies lack personalized consideration. This system also includes: Based on veterinary textbooks, research papers, clinical cases, and epidemic prevention guidelines, a knowledge graph of livestock and poultry diseases is constructed. The entities in the knowledge graph include disease name, pathogen, symptoms, transmission route, susceptible animals, incubation period, diagnostic methods, treatment drugs, and preventive measures. Entity relationships include disease-symptom associations, disease-pathogen associations, symptom-symptom co-occurrences, disease-treatment drug associations, and disease-prevention measure associations. Data collected includes symptom descriptions from feeders, abnormal behavior / appearance features from computer vision analysis, environmental monitoring data, and production data. Among these, environmental monitoring data includes abnormal temperature and humidity and air quality; production data includes decreased feed intake and changes in milk / egg production. The collected symptom information is input into a knowledge graph for reasoning. The system uses a graph retrieval algorithm to match a set of candidate diseases highly correlated with the input symptoms within the knowledge graph, and then sorts the candidate disease sets and their related knowledge graph subgraphs according to the strength of the correlation before inputting them into the large language model. LLM performs in-depth reasoning analysis to assess the matching degree and likelihood probability of each candidate disease, analyze whether the combination of symptoms conforms to the typical or atypical manifestations of a certain disease, identify key symptoms or test items that need further confirmation, and generate differential diagnosis suggestions. After a confirmed or highly suspected case, LLM (Laboratory Management) generates a personalized prevention and control strategy based on the specific conditions of the farm, the current season and climate, the farm's biosecurity level and existing disease prevention measures, and economic costs. The effectiveness of the prevention and control strategy is continuously monitored. If no improvement is seen within the expected time or new symptoms appear, LLM re-examines the diagnosis and adjusts the prevention and control strategy.

[0042] It should be noted that the LLM method in this paper provides flexible deep reasoning capabilities to deal with complex and atypical cases, and the prevention and control strategies are generated in a personalized manner, which can effectively alleviate the problem of insufficient veterinary resources in grassroots farms.

[0043] The second aspect of this invention provides a livestock farming management method based on LLM and multi-source data fusion, comprising the following steps: Receive lean management-related questions or requirements from users and perform intent recognition; By coupling a large language model with retrieval enhancement generation technology, a structured knowledge base is dynamically retrieved based on intent to generate diagnostic conclusions and natural language responses for lean problems; Based on natural language processing and visualization engines, it responds to tool usage needs, generates charts based on data, and automatically generates standardized operation guidelines and corresponding graphical documents for at least one lean tool, such as Kanban boards, 5 Whys analysis, and value stream mapping. Based on a vectorized historical case library and semantic matching mechanism, and according to the semantic characteristics of the current problem, the most suitable cross-departmental and cross-business line improvement solutions are recommended. Collect user feedback and implementation effectiveness data, and update the knowledge base, case library, and model parameters accordingly.

[0044] It should be noted that the method is based on the above system and includes the following steps: receiving user input and performing intent recognition; generating a response by calling the virtual expert advisor unit, intelligent guidance generation unit, or cross-scenario recommendation unit according to the intent type; outputting results to the user in text, graphic, or document form; collecting user feedback and implementation effects; and updating the knowledge base, case library, and model parameters using the feedback data to complete a closed-loop optimization.

[0045] A third aspect of the present invention provides a computer-readable storage medium, including a livestock breeding management method program based on LLM and multi-source data fusion, wherein when the livestock breeding management method program based on LLM and multi-source data fusion is executed by a processor, it implements the steps of the livestock breeding management method based on LLM and multi-source data fusion.

[0046] It should be noted that the computer-readable storage medium stores a computer program, which, when executed by a processor, implements all the steps of the above method.

[0047] The system can be deployed on servers, cloud platforms, or local terminals, and supports access via web pages, mobile applications, or enterprise instant messaging tools. It is suitable for lean improvement assistance scenarios in multiple industries such as animal husbandry, discrete manufacturing, process industries, and modern services.

[0048] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0049] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0050] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0051] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0052] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0053] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A livestock farming management system based on LLM and multi-source data fusion, characterized in that, include: The natural language interaction unit is used to receive lean management-related questions or requirements from users and to perform intent recognition. The virtual expert advisor unit couples a large language model with retrieval enhancement and generation technology, dynamically retrieves a structured knowledge base based on intent, and generates diagnostic conclusions and natural language responses for lean problems; The intelligent guidance generation unit, based on natural language processing and visualization engine, responds to tool usage needs, generates charts based on data, and automatically generates standardized operation guidelines and corresponding graphical documents for at least one lean tool in Kanban, 5 Whys analysis, and value stream mapping. The cross-scenario recommendation unit, based on a vectorized historical case library and semantic matching mechanism, recommends the most suitable cross-departmental and cross-business line improvement solutions according to the semantic features of the current problem. The feedback optimization unit is used to collect user evaluation and solution implementation effect data, and to update the knowledge base, case library and model parameters accordingly.

2. The livestock farming management system based on LLM and multi-source data fusion according to claim 1, characterized in that, The structured knowledge base stores lean management theories, tool standards, answers to typical questions, and industry best practices. The knowledge base is built using a vector database and supports dynamic retrieval based on semantic similarity. Before generating a response, RAG technology injects the retrieved Top-K knowledge fragments as context into the LLM.

3. The livestock farming management system based on LLM and multi-source data fusion according to claim 1, characterized in that, The intelligent guidance generation unit includes: The instruction parsing subunit transforms natural language requirements into structured operation steps; The graphics generation subunit calls the visualization component to automatically generate graphical representations of the corresponding lean tools based on the data; The document synthesis subunit integrates step-by-step instructions, diagrams, and precautions to output a downloadable standardized document.

4. The livestock farming management system based on LLM and multi-source data fusion according to claim 1, characterized in that, Based on a vectorized historical case library and semantic matching mechanism, and according to the semantic characteristics of the current problem, the most suitable cross-departmental and cross-business line improvement solutions are recommended, including: Historical improvement cases are encoded using text embedding to construct a high-dimensional vector case library; The semantic similarity between the current problem and historical cases is calculated using an approximate nearest neighbor algorithm. Based on the matching score, case success rate, and time decay factor, the system outputs a list of recommended solutions and the reasons for the recommendations.

5. The livestock farming management system based on LLM and multi-source data fusion according to claim 1, characterized in that, The RAG retrieval process includes: The user's question is vectorized, and Top-K similar knowledge fragments are retrieved from the vector knowledge base. The fragments are then concatenated with the question to form a Prompt, which is then input into the LLM to generate the final response.

6. The livestock farming management system based on LLM and multi-source data fusion according to claim 1, characterized in that, The generated graphical documents include: 5 Whys fishbone diagram, value stream map, and Kanban status signal diagram.

7. A livestock farming management method based on LLM and multi-source data fusion, characterized in that, Includes the following steps: Receive lean management-related questions or requirements from users and perform intent recognition; By coupling a large language model with retrieval enhancement generation technology, a structured knowledge base is dynamically retrieved based on intent to generate diagnostic conclusions and natural language responses for lean problems; Based on natural language processing and visualization engines, it responds to tool usage needs, generates charts based on data, and automatically generates standardized operation guidelines and corresponding graphical documents for at least one lean tool, such as Kanban boards, 5 Whys analysis, and value stream mapping. Based on a vectorized historical case library and semantic matching mechanism, and according to the semantic characteristics of the current problem, the most suitable cross-departmental and cross-business line improvement solutions are recommended. Collect user feedback and implementation effectiveness data, and update the knowledge base, case library, and model parameters accordingly.

8. A computer-readable storage medium, characterized in that, The method includes a livestock farming management program based on LLM and multi-source data fusion. When the livestock farming management program based on LLM and multi-source data fusion is executed by a processor, it implements the steps of the livestock farming management method based on LLM and multi-source data fusion as described in claim 7.

Images