Building pig disease early warning and prevention and control intelligent recommendation method based on LLM-Agent

Abstract

The present application relates to a building pig farm epidemic disease early warning and prevention and control intelligent recommendation method based on LLM-Agent, belongs to the technical field of intelligent breeding and pig farm epidemic disease risk management, through combining abnormal detection algorithm and epidemic disease prevention and control field knowledge resources, the pig farm epidemic disease risk is graded early warning and the corresponding risk cause analysis result is generated, further aiming at different risk grades to generate various prevention and control and management strategies, and from the prevention and control effect, the execution difficulty and the economic cost and so on multiple dimensions are comprehensively evaluated and sorted, the output optimization prevention and control strategy scheme, simultaneously, through introducing user feedback, prevention and control execution result and system cooperative linkage mechanism, the prevention and control strategy is dynamically adjusted and continuously optimized, forms the closed loop control system of '' prevention and control recommendation-execution-feedback-optimization ''. The present application can realize the prospective early warning and scientific decision of building pig farm epidemic disease risk, improves the intelligentization, the explainability and the economic feasibility of epidemic disease prevention and control, has good application value and popularization prospect.

Description

Technical Field

[0001] This invention relates to the field of smart farming and pig farm disease risk management technology, and in particular to an intelligent recommendation method for early warning and prevention of diseases in multi-story pig farming based on LLM-Agent. Background Technology

[0002] In modern pig farming, timely early warning and scientific prevention and control of disease risks are key technical aspects for ensuring pig herd health, reducing the probability of disease outbreaks, and improving farming efficiency. Continuous monitoring and comprehensive analysis of pig farm operations can identify potential disease risks in advance, clarify their causes, and formulate targeted prevention and management strategies, thereby effectively reducing production losses caused by disease spread. However, in actual production, the assessment of pig farm disease risks and the formulation of prevention and control measures still heavily rely on the experience of veterinarians or managers, lacking a systematic analysis of changes in the pig farm environment, pig herd behavior characteristics, and historical events. This leads to delayed disease early warnings and insufficiently targeted prevention and control measures.

[0003] With the continuous improvement of the scale and intensification of the pig farming industry, especially the promotion and application of high-density farming models such as multi-story (building) pig farms, the operating environment and disease prevention and control situation of pig farms are becoming increasingly complex. The closed, multi-level, and large-scale characteristics of multi-story pig farms result in significant multivariate coupling and time-series evolution characteristics in environmental indicators such as temperature and humidity, ammonia concentration, and carbon dioxide concentration, as well as health indicators such as pig activity, feed intake, abnormal behavior, and disease records. The cumulative changes of these indicators over time are often an important factor in the gradual increase of disease risk. However, existing management methods are difficult to achieve real-time fusion and analysis of the above-mentioned multi-source heterogeneous data, making it difficult to capture potential disease risks in a timely manner. Prevention and control decisions are mostly at the post-event response level, which cannot meet the disease prevention and control needs in high-density farming scenarios.

[0004] In the practice of disease prevention and control in pig farms, the occurrence of diseases is usually not caused by a single abnormal indicator, but rather by the combined effect of multiple factors such as environmental deterioration, improper management, stress factors, and changes in the health status of the pig herd. Taking multi-story pig farms as an example, insufficient ventilation due to abnormal operation of the ventilation system, long-term excessive ammonia concentration, improper adjustment of stocking density, or disordered timing of key management operations can all lead to a decline in the immunity of the pig herd within a certain period, thereby increasing the probability of disease occurrence. Therefore, the assessment of disease risk not only needs to clarify "whether there is a risk," but also needs to accurately determine the risk level and its causes, providing support for the formulation of differentiated and graded prevention and control strategies, and achieving precise and scientific prevention and control.

[0005] Existing disease monitoring and early warning methods primarily rely on threshold alerts or single-indicator analysis, making it difficult to accurately characterize the intrinsic relationships and time-series impact mechanisms among multiple variables. Even when some systems can identify abnormal states, their outputs are mostly simple alarm messages, lacking systematic summarization and scientific explanation of the causes of the anomalies, thus failing to provide managers with clear and definite decision-making basis. Furthermore, for the same risk state, different prevention and control measures vary significantly in terms of implementation difficulty, economic cost, and control effectiveness. Current technologies struggle to comprehensively evaluate and weigh multiple prevention and control options, severely hindering the implementation of precise prevention and control strategies.

[0006] Furthermore, the professional knowledge involved in pig farm disease prevention and control comes from a wide range of sources, including environmental control experience, veterinary clinical knowledge, historical disease cases, and industry management standards. This information is often stored in a scattered manner and lacks a unified organization, sorting, and intelligent reasoning mechanism, making it difficult to efficiently access and utilize in actual prevention and control scenarios. At the same time, existing systems generally lack good human-computer interaction capabilities, failing to support managers in asking follow-up questions, adjusting parameters, and providing feedback based on the current risk status. This results in prevention and control strategies being difficult to dynamically optimize according to the actual situation of the pig farm, leading to insufficient adaptability.

[0007] Therefore, there is an urgent need for an intelligent disease risk early warning and control decision-making method and system for multi-story pig farms. Its core requirements are: to integrate multi-source time-series data from pig farms, combine anomaly detection results with professional knowledge resources, and achieve graded early warning and accurate analysis of the causes of disease risks; to generate a variety of optional prevention and control and management strategies based on this, and to comprehensively evaluate the implementation difficulty, economic cost and control effect of each strategy; and to support user interaction and feedback optimization to continuously improve the prevention and control strategies, ultimately enhancing the scientific, forward-looking and intelligent level of disease prevention and control in pig farms, and adapting to the high-density and refined breeding management needs of multi-story pig farms. Summary of the Invention

[0008] This invention overcomes the shortcomings of existing technologies and provides an intelligent recommendation method for early warning and prevention of diseases in multi-story pig farming based on LLM-Agent.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0010] The first aspect of this invention provides an intelligent recommendation method for early warning and prevention of diseases in multi-story pig farming based on LLM-Agent, comprising the following steps:

[0011] The environmental parameters, pig behavior data, health status data, and historical management event information of each floor of the multi-story pig farm are structured and preprocessed.

[0012] Based on preprocessed data from multi-story pig farms, a contextual representation of disease risk scenarios is constructed to reflect the current environmental status of the pig farm, the health level of the pig herd, and historical abnormal events, thus characterizing the disease risk features at different time points and on different floors.

[0013] The context of the epidemic risk scenario is input into the LLM-Agent, which has the ability to reason, retrieve and call tools. The anomaly detection algorithm is invoked and a comprehensive analysis is carried out in combination with the knowledge-enhanced retrieval mechanism. Finally, the epidemic risk level judgment result and the corresponding risk cause analysis are generated.

[0014] Based on the analysis results of different risk levels and causes, a variety of candidate prevention and control strategies are generated, and the strategies are comprehensively evaluated and ranked from three dimensions: prevention and control effectiveness, implementation difficulty and economic cost, and the optimal prevention and control strategy is output.

[0015] By combining user feedback, strategy execution results, and system collaborative control mechanisms, the prevention and control strategies are dynamically adjusted and continuously optimized, ultimately outputting a complete implementation plan for epidemic prevention and control in multi-story pig farms.

[0016] Furthermore, in the intelligent recommendation method for disease early warning and prevention in multi-story pig farms based on LLM-Agent, the environmental parameters, pig behavior data, health status data, and historical management event information of each floor of the multi-story pig farm are structured and preprocessed, specifically as follows:

[0017] The raw data from the pig farm's IoT devices, management system, and manual records are deduplicated, formatted, and time-aligned. Interpolation and neighbor-value filling methods are used to handle missing data.

[0018] Normalize or standardize the numerical characteristics of environmental parameters and behavioral indicators to eliminate the influence between different units, and unstructured data is formatted to form a unified representation that can be processed by large models.

[0019] Abnormal data is identified through statistical analysis, threshold rules, and anomaly detection algorithms, and features reflecting the operational status of multi-story pig farms are constructed.

[0020] Furthermore, in the intelligent recommendation method for disease early warning and prevention in multi-story pig farming based on LLM-Agent, a contextual representation of the epidemic risk scenario is constructed based on preprocessed multi-story pig farm data. This representation reflects the current environmental status of the pig farm, the health level of the pig herd, and historical abnormal events, characterizing the disease risk features at different time points and on different floors. Specifically:

[0021] The temperature and humidity, ammonia concentration, carbon dioxide concentration, pig activity, detection data, and disease record characteristics of each floor are coded to form multidimensional risk characteristics.

[0022] By integrating historical anomaly detection results, key management events, and environmental mutation information with real-time features, a unified representation of epidemic risk status is constructed;

[0023] The risk status is represented and organized as a structured contextual hint, describing the characteristics of disease risk changes in a pig farm over a specific time period.

[0024] Furthermore, in the intelligent recommendation method for early warning and prevention of swine diseases based on LLM-Agent, the context of the epidemic risk scenario is input into the LLM-Agent, which has the capabilities of reasoning, retrieval, and tool invocation. Anomaly detection algorithms are invoked, and a comprehensive analysis is conducted in conjunction with a knowledge-enhanced retrieval mechanism. Finally, an epidemic risk level judgment result and corresponding risk cause analysis are generated, specifically:

[0025] LLM-Agent automatically generates analysis intent based on the current risk scenario and constructs trend data by calling anomaly detection algorithms to obtain key anomaly indicators.

[0026] Retrieve information related to the current abnormal pattern from the disease prevention and control rule base, veterinary clinical knowledge base, expert rule system and disease vaccine knowledge graph, and return it in the form of structured summary;

[0027] LLM-Agent integrates anomaly detection results, search results, and pig farm disease risks through contextual fusion reasoning to classify them as low-risk, medium-risk, or high-risk. It outputs the corresponding risk causes and implementation details for prevention and control management, along with corresponding recommendations and existing reference materials.

[0028] Furthermore, in the intelligent recommendation method for early warning and prevention of swine diseases based on LLM-Agent, multiple candidate prevention and management strategies are generated based on different risk levels and causal analysis results. These strategies are then comprehensively evaluated and ranked from three dimensions: prevention effectiveness, implementation difficulty, and economic cost, outputting the optimal prevention strategy. Specifically:

[0029] A systematic evaluation of the disease control effectiveness of each candidate prevention and control strategy will be conducted. The evaluation indicators include the actual suppression efficiency of the disease spread after the implementation of the strategy, the speed of mitigation of the risks related to disease transmission, and the continuous stability of the prevention and control effect during the implementation period. At the same time, the actual effectiveness of the prevention and control strategy in achieving the core objectives of disease control and improving the overall level of prevention and control will be measured by combining the disease transmission patterns and scenario characteristics.

[0030] A comprehensive assessment of the feasibility of implementing candidate prevention and control strategies is conducted. This includes analyzing the scale of personnel input and professional capabilities required for strategy implementation, the complexity and technical threshold of the operational process, and the extent of adjustments to the existing management system and production and operation processes. Additionally, the assessment considers the hardware conditions and scenario adaptability of the actual application scenario to evaluate the operability, ease of execution, and suitability of implementing the strategy under given on-site conditions.

[0031] A comprehensive assessment of the economic costs of the entire process of prevention and control strategies is conducted, taking into account the direct input costs of material procurement, allocation, and consumption during the implementation of the strategies, the scope and recovery period of the impact of the implementation of the strategies on normal production and operation activities, the potential loss of revenue and opportunity costs caused by the implementation of prevention and control strategies, and the sustainability of cost input in combination with the length of the prevention and control period. From the perspective of economic feasibility, the degree of impact of different prevention and control strategies on overall operating costs and the rationality of cost input are analyzed.

[0032] Based on preset multi-dimensional weight parameters or user-defined personalized preference strategies, the three evaluation results of prevention and control effectiveness, implementation feasibility, and economic cost are comprehensively weighed and quantitatively scored. Based on the scoring results, all candidate prevention and control strategies are prioritized and selected, and the prevention and control strategies with the best comprehensive performance or those that meet the constraints of specific scenarios are selected to form multiple prevention and control plans.

[0033] Furthermore, in the intelligent recommendation method for disease early warning and prevention in multi-story pig farms based on LLM-Agent, the prevention and control strategies are dynamically adjusted and continuously optimized by combining user feedback, strategy execution results, and system collaborative control mechanisms. The final output is a complete implementation plan for disease prevention and control in multi-story pig farms, specifically:

[0034] The immunization recommendations generated by LLM-Agent are presented through a human-computer interaction interface, allowing users to confirm, adjust, or reject the immunization plan, and convert the confirmed immunization plan into standardized execution instructions.

[0035] The execution instructions are linked with the automatic immunization equipment or pig farm management system through the platform interface to realize the actual deployment of the immunization plan, including generating the vaccination list, grouping the immunized subjects and allocating vaccine batches, and collecting execution status and result data in real time during the immunization process;

[0036] Collect users' manual adjustment opinions, preference information, and immunization execution result data, and uniformly input them into LLM-Agent as feedback to update the immunization strategy generation logic and decision context;

[0037] Based on continuously accumulated execution feedback and user interaction information, LLM-Agent adaptively optimizes subsequent immunization programs, forming personalized vaccine immunization implementation plans for different pig farm sizes, feeding models, and health conditions, and constructing a closed-loop control system of immunization recommendation—execution—feedback—optimization.

[0038] Furthermore, the intelligent recommendation method for early warning and prevention of swine diseases in multi-story buildings based on LLM-Agent also includes:

[0039] Each floor is divided into multiple zones, environmental data for each zone is collected, an environmental distribution cloud map for that floor is generated using a spatial interpolation algorithm, and a three-dimensional environmental field model is established by combining the floor height and updated in real time.

[0040] Define a comprehensive stress index, obtain environmental parameter data from the three-dimensional environmental field model, and perform nonlinear correction using weighted summation or fuzzy logic methods;

[0041] Based on the environmental parameter data, a comprehensive stress index is calculated using a comprehensive stress index calculation method. A correlation model between the stress index and the physiological indicators of pigs is established based on deep learning. Through training with historical data, the immune risk level corresponding to different stress index intervals is determined.

[0042] The immune risk level of each floor is estimated based on the comprehensive stress index, and the immunization plan is automatically adjusted according to the immune risk level.

[0043] Furthermore, the intelligent recommendation method for early warning and prevention of swine diseases in multi-story buildings based on LLM-Agent also includes:

[0044] A digital profile of each pig's immunity is constructed based on historical immunization records. For piglets, a maternal antibody decay curve is fitted for each individual based on maternal immunization information, piglet age, and regular antibody test data. The time point when antibody protection disappears is predicted based on the maternal antibody decay curve.

[0045] By combining feeding behavior, vocal characteristics, and body temperature data, an immune system activity level is constructed based on the feeding behavior, vocal characteristics, and body temperature data.

[0046] Collect ammonia concentration, dust concentration, and respiratory symptom sounds for each pig to estimate the mucosal barrier damage index;

[0047] By fusing the time point of antibody protection loss, immune system activity, and mucosal barrier damage index through an attention mechanism, a comprehensive immune profile vector of the pig herd is generated.

[0048] Based on the profile, the best vaccine in the vaccine database is matched, and the immunization regimen is adjusted based on the best vaccine in the vaccine database.

[0049] This invention addresses the shortcomings of the prior art and has the following beneficial effects:

[0050] This invention introduces an LLM-Agent with perception, tool invocation, retrieval reasoning, and feedback learning capabilities. It integrates multi-source environmental and production data from multi-story pig farms, anomaly detection results, disease prevention and control knowledge resources, and expert experience rules to achieve tiered early warning, causal induction, prevention and control strategy generation, and dynamic optimization of pig farm disease prevention and control. This improves the foresight, scientific rigor, and intelligence of pig farm disease prevention and control. This invention proposes an intelligent recommendation framework for disease early warning and prevention and control in multi-story pig farming based on an LLM-Agent, achieving intelligent processing from anomaly identification to prevention and control strategy recommendation. By integrating anomaly detection algorithms, professional knowledge, and user feedback, it enables interpretable analysis of disease risks and dynamic optimization of multiple prevention and control strategies. By introducing multi-objective evaluation and user feedback mechanisms, it balances disease prevention and control and biosecurity in multi-story pig farms with economic costs, achieving intelligent, personalized, and closed-loop management of disease prevention and control solutions for multi-story pig farms. Attached Figure Description

[0051] 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.

[0052] Figure 1 The overall flowchart of the intelligent recommendation method for disease early warning and prevention and control in multi-story pig farms based on LLM-Agent is shown;

[0053] Figure 2 The flowchart of the multi-source data real-time perception and epidemic scenario modeling module is shown;

[0054] Figure 3 The flowchart shows the invocation of tools such as anomaly detection, knowledge-enhanced retrieval, and risk causation reasoning generation modules.

[0055] Figure 4 The flowchart of the multi-objective modeling and prevention and control management strategy evaluation and ranking module is shown;

[0056] Figure 5 The flowcharts for the user interaction feedback, prevention and control strategies, and closed-loop optimization modules are shown.

[0057] Figure 6 The overall structure diagram of the intelligent recommendation system for disease early warning and prevention in multi-story pig farms based on LLM-Agent is shown. Detailed Implementation

[0058] 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.

[0059] 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.

[0060] like Figure 1 As shown, the first aspect of this invention provides an intelligent recommendation method for early warning and prevention of diseases in multi-story pig farming based on LLM-Agent, comprising the following steps:

[0061] The environmental parameters, pig behavior data, health status data, and historical management events of each floor of the multi-story pig farm are structured and processed, and the epidemic prevention and control related data are preprocessed.

[0062] Based on preprocessed data from multi-story pig farms, a contextual representation of the epidemic risk scenario is constructed to reflect the current environmental status of the pig farm, the health level of the pig herd, and historical abnormal events, in order to characterize the disease risk features at different time points and on different floors.

[0063] The context of the epidemic risk scenario is input into the LLM-Agent, which has the ability to reason, retrieve and call tools. The agent calls existing tools such as anomaly detection algorithms and combines them with knowledge-enhanced retrieval mechanisms to carry out comprehensive analysis, and finally generates the epidemic risk level judgment result and the corresponding risk cause analysis.

[0064] Based on the analysis results of different risk levels and causes, a variety of candidate prevention and control strategies are generated, and the strategies are comprehensively evaluated and ranked from three dimensions: prevention and control effectiveness, implementation difficulty and economic cost, and the optimal prevention and control strategy is output.

[0065] By combining user feedback, strategy execution results, and system collaborative control mechanisms, the prevention and control strategies are dynamically adjusted and continuously optimized, ultimately outputting a complete implementation plan for epidemic prevention and control in multi-story pig farms.

[0066] It should be noted that this invention introduces an LLM-Agent with perception, tool invocation, retrieval reasoning, and feedback learning capabilities. This LLM-Agent integrates multi-source environmental and production data from multi-story pig farms, anomaly detection results, disease prevention and control knowledge resources, and expert experience rules to achieve tiered early warning, causal induction, prevention and control strategy generation, and dynamic optimization of pig farm disease prevention and control. This improves the foresight, scientific rigor, and intelligence of pig farm disease prevention and control. This invention proposes an intelligent recommendation framework for disease early warning and prevention and control in multi-story pig farming based on an LLM-Agent, achieving intelligent processing from anomaly identification to prevention and control strategy recommendation. By integrating anomaly detection algorithms, professional knowledge, and user feedback, it enables interpretable analysis of disease risks and dynamic optimization of multiple prevention and control strategies. By introducing multi-objective evaluation and user feedback mechanisms, it balances disease prevention and control and biosecurity in multi-story pig farms with economic costs, achieving intelligent, personalized, and closed-loop management of disease prevention and control solutions for multi-story pig farms.

[0067] In this solution, information such as environmental parameters of each floor of the multi-story pig farm, pig behavior data, health status data, and historical management events (such as feeding events and disease prevention and control events) are structured and processed. The disease prevention and control related data are preprocessed, specifically as follows:

[0068] First, the raw data from the pig farm's IoT devices, management system, and manual records are deduplicated, formatted, and time-aligned to ensure consistency in time scale and data structure across multiple data sources.

[0069] Secondly, interpolation and neighbor-value filling methods are used to process missing data to avoid the impact of missing information caused by sensor failure or incomplete manual recording on the accuracy of epidemic situation modeling.

[0070] Then, numerical features such as environmental parameters and behavioral indicators are normalized or standardized to eliminate the influence between different units. Furthermore, unstructured data is formatted to form a unified representation that can be processed by large models.

[0071] Finally, abnormal data were identified through statistical analysis, threshold rules, and anomaly detection algorithms, and features reflecting the operational status of multi-story pig farms were constructed to provide a reliable data foundation for subsequent epidemic risk inference.

[0072] It should be noted that during the data acquisition and preprocessing phase, the system continuously receives data from the pig farm's IoT devices and management system, including but not limited to temperature and humidity on each floor, ammonia concentration, carbon dioxide concentration, pig herd behavior indicators, disease records, and vaccination logs. For the raw data, firstly, deduplication and format standardization are performed to ensure data consistency and comparability. Then, missing data is repaired using interpolation or historical adjacent value filling strategies. Based on this, numerical features are normalized or standardized to eliminate differences in different units of measurement. Finally, statistical analysis and threshold rules are used to identify and remove outlier data, improving overall data quality and system robustness. After the above processing, the data can effectively support subsequent scenario modeling and intelligent reasoning. Furthermore, by introducing an LLM-Agent with sensing, retrieval, reasoning, and feedback capabilities, and integrating multi-source data from multi-story pig farms, disease epidemiological information, and immunization knowledge resources, intelligent evaluation, dynamic optimization, and interpretable recommendations of vaccine immunization programs can be achieved, thereby improving immunization effectiveness, reducing immunization risks, and optimizing breeding costs.

[0073] Figure 2 The flowchart of the multi-source data real-time perception and epidemic scenario modeling module is shown.

[0074] In this embodiment, semantic parsing and feature fusion are performed on multi-source heterogeneous data from pig farms to construct a scenario context for epidemic early warning and risk assessment. Specifically:

[0075] First, data from different sources are aligned according to floor, pig herd batch, and time dimension, and features are extracted from environmental, behavioral, and health indicators to form a structured feature set.

[0076] Secondly, based on the preset scenario modeling rules, statistical feature analysis methods and historical disease patterns, a comprehensive judgment is made on the current operating status of the pig farm, and the health information of the pig herd is summarized.

[0077] Finally, the above analysis results are encoded into a unified epidemic risk context representation to describe the overall epidemic prevention and control decision-making background of the pig farm at a specific point in time, and to serve as input conditions for subsequent tool calls, knowledge retrieval and reasoning generation.

[0078] It should be noted that the scenario context is not a simple data splicing, but rather a semantic hierarchical modeling approach that integrates and expresses abnormal environmental features, trends in pig herd behavior, evolution of health status, and historical event information, thereby improving the accuracy and interpretability of epidemic risk warning and causal analysis.

[0079] In this solution, based on preprocessed operational data from multi-story pig farms, a disease risk scenario context representation reflecting the current environmental status of the pig farm, the health status of the pig herd, and historical abnormal events is constructed, specifically as follows:

[0080] First, the characteristics of temperature and humidity, ammonia concentration, carbon dioxide concentration, pig activity, test data and disease records of each floor are coded to form multidimensional risk characteristics.

[0081] Secondly, historical anomaly detection results, key management events, and environmental mutation information are integrated with real-time features to construct a unified representation of epidemic risk status;

[0082] Finally, the risk status is organized into structured contextual hints to describe the characteristics of disease risk changes in pig farms over a specific time period, providing relevant support for subsequent intelligent reasoning.

[0083] In this solution, the context of the epidemic risk scenario is input into the LLM-Agent, combined with knowledge base retrieval and the invocation of tools such as anomaly detection, to generate risk assessment and causal analysis results, specifically as follows:

[0084] First, LLM-Agent automatically generates analysis intent based on the current risk scenario, and constructs trend data by calling tools such as anomaly detection algorithms to obtain key anomaly indicators.

[0085] Secondly, information related to the current abnormal pattern is retrieved from the disease prevention and control rule base, veterinary clinical knowledge base, expert rule system and disease vaccine knowledge graph, such as keyword search, vector search, etc., and returned in the form of structured summary.

[0086] Finally, LLM-Agent will use contextual fusion reasoning to classify pig farm disease risks into low, medium, or high risk levels by combining anomaly detection results, search results, and contextual fusion reasoning. It will then output the corresponding risk causes and implementation details for prevention and control management, along with corresponding recommendations and relevant existing data.

[0087] Figure 3 The flowchart shows the invocation of tools such as anomaly detection, knowledge enhancement, and causal reasoning generation modules.

[0088] In this embodiment, the system, based on the constructed epidemic risk scenario context, utilizes the tool invocation and retrieval enhancement capabilities of LLM-Agent for risk warning and causal analysis. Specifically: First, LLM-Agent automatically generates analysis tasks based on the current epidemic risk scenario and invokes the variable fusion time-series pigsty environment anomaly detection method and related algorithm tools in the system to identify anomalies in key environmental and behavioral indicators. Second, based on the relevant disease intent, it initiates a request to the integrated knowledge retrieval module. The knowledge sources include the veterinary clinical knowledge base, expert experience rule base, historical epidemic case base, and prevention and control management standard documents. Then, the retrieved knowledge information is structured and organized, and joint reasoning is performed with the relevant data returned by the tool to summarize the key indicator anomalies, triggering events, and potential transmission paths that lead to the current disease risk. Finally, the system outputs the disease risk level determination result and the corresponding causal analysis conclusion, providing a decision-making basis for the subsequent generation of prevention and control management strategies.

[0089] It should be noted that by introducing a tool-calling mechanism and knowledge-enhanced reasoning, the risk of knowledge bias in the field of professional veterinary medicine using large language models can be effectively reduced, and the professionalism and reliability of disease early warning results and causal analysis can be improved.

[0090] Figure 4 The flowchart of the multi-objective modeling and prevention and control management strategy evaluation and ranking module is shown.

[0091] In this embodiment, firstly, the control capability of candidate strategies against the risk of disease transmission is evaluated from the perspective of prevention and control effectiveness, including the magnitude of risk level reduction, the recovery speed of key abnormal indicators, and the expected trend of incidence rate changes. Secondly, the implementation risk of the strategies is evaluated from the perspective of execution difficulty and biosafety, including operational complexity, the degree of interference with existing production processes, and potential secondary risks. Thirdly, the implementation cost of the prevention and control strategies is evaluated from the perspective of economic cost, including management adjustment costs, labor input, and potential production losses. Finally, based on preset weight parameters or user preferences, the above evaluation results are comprehensively quantified and scored, the candidate prevention and control management strategies are ranked, and the recommended scheme with the best overall performance or that meets specific constraints is output.

[0092] This plan generates multiple prevention and management strategies for different risk levels, and comprehensively evaluates them from three dimensions: prevention and control effectiveness, implementation difficulty, and economic cost. Based on a multi-objective trade-off mechanism, the prevention and control strategies are ranked and selected, and the optimal prevention and control plan is output, specifically:

[0093] First, a systematic evaluation of the disease control effectiveness of each candidate prevention and control strategy is conducted. The evaluation indicators include the actual suppression efficiency of the disease spread after the implementation of the strategy, the speed of mitigation of the risks related to disease transmission, and the continuous stability of the prevention and control effect during the implementation period. At the same time, the actual effectiveness of the prevention and control strategy in achieving the core objectives of disease prevention and control and improving the overall level of prevention and control is measured by combining the disease transmission patterns and scenario characteristics.

[0094] Secondly, a comprehensive assessment of the feasibility of implementing candidate prevention and control strategies is conducted. The focus is on analyzing the scale of personnel input and professional capabilities required for strategy implementation, the complexity and technical threshold of the operation and execution process, and the extent of adjustment to the existing management system and production and operation processes. At the same time, factors such as the hardware conditions and scenario adaptability of the actual application scenario are considered to evaluate the operability, ease of execution, and adaptability of the strategy under the given on-site conditions.

[0095] Then, a comprehensive assessment of the economic costs of the entire prevention and control strategy process is conducted. This assessment takes into account the direct input costs of material procurement, allocation, and consumption during the strategy implementation process, the scope and recovery period of the impact of the strategy implementation on normal production and operation activities, the potential loss of revenue and opportunity costs caused by the implementation of the prevention and control strategy, and the sustainability of cost input in combination with the length of the prevention and control period. From the perspective of economic feasibility, the assessment analyzes the degree of impact of different prevention and control strategies on overall operating costs and the rationality of cost input.

[0096] Finally, based on the preset multi-dimensional weight parameters or the user-defined personalized preference strategy, the three evaluation results of prevention and control effect, implementation feasibility and economic cost are comprehensively weighed and quantitatively scored. Based on the scoring results, all candidate prevention and control strategies are prioritized and selected. The prevention and control strategy with the best comprehensive performance or that meets the constraints of specific scenarios is selected to form multiple prevention and control plans. The core advantages, applicable scenarios and potential limitations and precautions in the implementation process of each plan are clarified.

[0097] This solution dynamically adjusts and continuously optimizes prevention and control strategies by combining user feedback, implementation results, and system collaboration mechanisms, ultimately outputting a complete prevention and control implementation plan for pig farms, as follows:

[0098] First, the prevention and control recommendations generated by LLM-Agent are presented to pig farm managers through a human-computer interaction interface, showing the core content of the prevention and control plan, risk explanations and strategy priorities. Users can confirm, adjust or reject the prevention and control strategy, and the confirmed prevention and control strategy is converted into standardized execution instructions.

[0099] Secondly, the execution instructions are linked with the pig farm management system through the platform interface to realize the actual deployment of the prevention and control plan, including task allocation, clear execution standards and acceptance indicators, and real-time collection of execution status, environmental changes and result data during the prevention and control execution process.

[0100] Next, user feedback, usage preferences, prevention and control implementation results, and environmental change data are collected and uniformly fed into the LLM-Agent to update the prevention and control strategy generation logic and decision context.

[0101] Finally, based on continuously accumulated execution feedback, environmental data, and user interaction information, LLM-Agent adaptively optimizes subsequent prevention and control strategies, forming personalized prevention and control implementation plans that can adapt to different pig farm sizes, feeding models, and environmental conditions, thereby constructing a closed-loop control system of prevention and control recommendation—execution—feedback—optimization.

[0102] Furthermore, the intelligent recommendation method for early warning and prevention of swine diseases in multi-story buildings based on LLM-Agent also includes:

[0103] Each floor is divided into multiple zones, environmental data for each zone is collected, an environmental distribution cloud map for that floor is generated using a spatial interpolation algorithm, and a three-dimensional environmental field model is established by combining the floor height and updated in real time.

[0104] Define a comprehensive stress index, obtain environmental parameter data from a three-dimensional environmental field model, and perform nonlinear correction using weighted summation or fuzzy logic methods;

[0105] The comprehensive stress index is calculated based on environmental parameter data. A correlation model between the stress index and pig physiological indicators is established based on deep learning. The immune risk level corresponding to different stress index intervals is determined by training with historical data.

[0106] For example, stress index data and pig physiological index data can be input into deep learning networks (including convolutional neural networks, recurrent neural networks, etc.) for training, thereby using the correlation model between stress index and pig physiological index to determine the immune risk level corresponding to different stress index intervals.

[0107] The immune risk level of each floor is estimated based on the comprehensive stress index, and the immunization plan is automatically adjusted according to the immune risk level.

[0108] It should be noted that the formula for calculating the Comprehensive Stress Index (SI) is as follows:

[0109]

[0110] Where T is temperature, H is humidity, W is wind speed, G is the pig's age / weight factor, and A is ammonia concentration. Weighted summation or fuzzy logic methods are employed. For example, referencing the temperature-humidity index (THI) from pig heat stress studies, and combining it with wind speed correction, a modified temperature-humidity-wind index suitable for pigs is constructed: THI = (1.8T + 32) - (0.55 - 0.0055H)(1.8T - 26) - 0.2W, which is then nonlinearly corrected based on ammonia concentration and age.

[0111] Secondly, a correlation model was established between stress index and physiological indicators of pigs (such as cortisol levels and heat shock protein expression). This model was trained using historical data to determine the corresponding immune risk levels (low, medium, and high) for different stress index ranges. Finally, the system automatically adjusted the immunization plan based on the current comprehensive stress index of each floor: if the stress index of a floor was in the low-risk range (e.g., SI < 30), immunization was performed according to the original plan. If it was in the medium-risk range (30 ≤ SI < 60), immunization was recommended during the optimal environmental time of the day (e.g., early morning or evening), and an alert was issued. If it was in the high-risk range (SI ≥ 60), immunization was automatically postponed, and the immunization task for that floor was suspended until the stress index decreased, at which point it was reactivated.

[0112] Furthermore, the intelligent recommendation method for early warning and prevention of swine diseases in multi-story buildings based on LLM-Agent also includes:

[0113] A digital profile of each pig's immunity is constructed based on historical immunization records. For piglets, a maternal antibody decay curve is fitted for each individual based on maternal immunization information, piglet age, and regular antibody testing data. The time point when antibody protection disappears, such as the antibody window period, is predicted based on the maternal antibody decay curve.

[0114] By combining feeding behavior, vocal characteristics, and body temperature data, the activity of the immune system is constructed based on these data. For example, when the feeding behavior, vocal characteristics, and body temperature data are normal, it indicates high activity; when the feeding behavior, vocal characteristics, and body temperature data are abnormal, it indicates low activity.

[0115] The ammonia concentration, dust concentration, and respiratory symptom sounds of each pig were collected to estimate the mucosal barrier damage index. The spatiotemporal features of the environment and sound were extracted by a convolutional neural network (CNN) to output the barrier index.

[0116] By fusing the time point of antibody protection loss, immune system activity, and mucosal barrier damage index through an attention mechanism, a comprehensive immune profile vector of the pig herd is generated.

[0117] Based on the profile, the best vaccine in the vaccine database is matched, and the immunization regimen is adjusted based on the best vaccine in the vaccine database.

[0118] It should be noted that this invention can achieve quantitative assessment of the immune status of pig herds, providing a data foundation for precision immunization. Furthermore, the fusion of multi-source data improves the accuracy and real-time performance of immunity assessment. Finally, the profiling can assist veterinarians in developing personalized immunization methods and improve the accuracy of recommendations.

[0119] In addition, this method also includes:

[0120] Obtain the architectural structural dimensions of each floor and railing in the building, and construct a three-dimensional geometric model of the building based on the architectural structural dimensions of each floor and railing.

[0121] Each pig in each pen is treated as an intelligent agent, and attributes such as age, weight, breed, antibody level, and health status are added to the intelligent agent to form a pig herd agent model. Real-time temperature data and humidity data are integrated to form an environmental model.

[0122] A SIR model was constructed, and contact transmission and airborne transmission processes were added to the SIR model to establish a vaccine-antibody kinetic model to describe the process of antibody production, maintenance, and decay after vaccination. The model parameters were dynamically adjusted according to the vaccine type, pig age, and immunization dose.

[0123] For example, the vaccine-antibody kinetic model uses dA / dt = k1 * D * S - k2 * A, where A is the antibody concentration, D is the dose, S is the immune system state, and k1 and k2 are constants.

[0124] By integrating the geometric 3D model, environmental model, vaccine-antibody kinetic model, and SIR model, a digital twin model is constructed. The digital twin model is then used for virtual simulation to model the antibody growth curve, changes in the number of infected cases, and changes in production indicators (feed ratio, daily weight gain) within a preset time period from the immunization time.

[0125] One or more immunization protocols are proposed, and the economic benefits of each protocol are scored. The protocol with the highest score is recommended, and the actual antibody detection data and morbidity data are used to calibrate the digital twin model.

[0126] The method utilizes an economic benefit evaluation function for scoring, such as Economic Benefit = Weight Gain Gain - Vaccine Cost - Mortality Loss. This approach enables virtual validation of immunization programs, avoiding blind trial and error. Furthermore, comparisons of multiple programs provide a scientific basis for decision-making, optimizing economic benefits. The digital twin model continuously learns, constantly improving prediction accuracy.

[0127] In addition, this method also includes:

[0128] Construct a reinforcement learning model, pre-setting a state space, action space, and reward function. The state space includes the average antibody level, infection rate, health status distribution, environmental stress index, and vaccine inventory of each pig population.

[0129] In the action space, standard immunization is performed on all healthy individuals to form group actions, while delayed immunization, isolation treatment, and booster immunization are performed on specific individuals to form individual actions.

[0130] The reward function is designed so that when it is a positive reward, the antibody qualification rate after immunization increases, the incidence rate of the population decreases, and the feed conversion ratio is optimized; when it is a negative reward, individual immunization leads to stress-induced death and the spread of infection in the population.

[0131] The reward function is: Reward = w1 * (group production efficiency) - w2 * (individual loss) - w3 * (vaccine cost), where w1, w2, and w3 are weighting coefficients, and the sum of w1, w2, and w3 is 1.

[0132] The PPO algorithm is used to train the reinforcement learning model. Given a state, the optimal action is selected to maximize the long-term cumulative reward. On the day of execution, the system outputs suggestions based on the current state using the trained model.

[0133] It should be noted that if the model outputs a herd immunity action, all healthy pigs will be immunized, and the immunization program will be adjusted. If the model identifies individual pigs at the disease threshold and outputs an individual action of delayed immunization or "isolation treatment," the pig will be automatically removed from the immunization list, and an alert will be sent to the veterinarian, suggesting isolation observation or treatment to further optimize the immunization program. Furthermore, the interaction between herd immunity and individual immunity is considered. For example, if an individual is a source of pathogen transmission, timely isolation may reduce group infection, thus allowing herd immunity to be implemented more smoothly. This solution achieves an intelligent balance between herd immunity and precise individual management, ensuring herd protection while avoiding individual losses. Moreover, the reinforcement learning model can adapt to complex dynamic environments, making optimal trade-offs and further optimizing the accuracy of the recommended immunization program.

[0134] Figure 5 The flowchart of the user interaction feedback and closed-loop optimization control module is shown.

[0135] In this embodiment, firstly, the user can confirm, adjust, or ask questions about the recommended strategy, and can also ask follow-up questions about the causes of risk. The system continuously performs interactive reasoning based on the context. Secondly, the system converts the prevention and control strategy selected by the user into standardized execution instructions and links with the pig farm management system to complete the strategy deployment and execution. Finally, the execution results and user feedback are fed back to the LLM-Agent to update the risk assessment and strategy recommendation logic, thereby achieving continuous optimization and closed-loop control.

[0136] Figure 6 The diagram illustrates the structural block of an intelligent recommendation system for disease early warning and prevention in multi-story pig farms based on LLM-Agent. The system includes a data preprocessing module, a disease scenario modeling module, a knowledge-enhanced reasoning module, a multi-objective evaluation module, and a user interaction feedback and closed-loop optimization control module.

[0137] The data preprocessing module is used to acquire and process data related to the pig farm environment, behavior, and health.

[0138] The disease scenario modeling module is used to construct a disease risk scenario context that reflects the operational status of the pig farm;

[0139] The knowledge-enhanced reasoning module is used to complete disease risk early warning, cause analysis, and strategy generation.

[0140] The multi-objective evaluation module is used to comprehensively evaluate and rank prevention and control management strategies;

[0141] The user interaction feedback and closed-loop optimization control module is used to support user decision-making participation and follow-up questioning, as well as to complete the deployment, execution and feedback of prevention and control strategies.

[0142] 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.

[0143] 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.

[0144] 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.

[0145] 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.

[0146] 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.

[0147] 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 method for intelligent recommendation of disease early warning and prevention in multi-story pig farming based on LLM-Agent, characterized in that, Includes the following steps: The environmental parameters, pig behavior data, health status data, and historical management event information of each floor of the multi-story pig farm are structured and preprocessed. Based on preprocessed data from multi-story pig farms, a contextual representation of disease risk scenarios is constructed to reflect the current environmental status of the pig farm, the health level of the pig herd, and historical abnormal events, thus characterizing the disease risk features at different time points and on different floors. The context of the epidemic risk scenario is input into the LLM-Agent, which has the ability to reason, retrieve and call tools. The anomaly detection algorithm is invoked and a comprehensive analysis is carried out in combination with the knowledge-enhanced retrieval mechanism. Finally, the epidemic risk level judgment result and the corresponding risk cause analysis are generated. Based on the analysis results of different risk levels and causes, a variety of candidate prevention and control strategies are generated, and a comprehensive evaluation and ranking are conducted from three dimensions: prevention and control effectiveness, implementation difficulty and economic cost, and a prevention and control strategy plan is output. By combining user feedback, strategy execution results, and system collaborative control mechanisms, the prevention and control strategies are dynamically adjusted and continuously optimized, ultimately outputting a complete implementation plan for epidemic prevention and control in multi-story pig farms. The context of the epidemic risk scenario is input into an LLM-Agent with reasoning, retrieval, and tool invocation capabilities. Anomaly detection algorithms are invoked, and a comprehensive analysis is conducted using a knowledge-enhanced retrieval mechanism. Ultimately, an epidemic risk level assessment result and corresponding risk causal analysis are generated, specifically: LLM-Agent automatically generates analysis intent based on the current risk scenario and constructs trend data by calling anomaly detection algorithms to obtain key anomaly indicators. Retrieve information related to the current abnormal pattern from the disease prevention and control rule base, veterinary clinical knowledge base, expert rule system and disease vaccine knowledge graph, and return it in the form of structured summary; LLM-Agent integrates anomaly detection results, search results, and pig farm disease risk through contextual fusion reasoning to classify them as low-risk, medium-risk, or high-risk. It outputs the corresponding risk causes and implementation details for prevention and control management, along with corresponding recommendations and relevant existing data. By combining user feedback, strategy execution results, and system collaborative control mechanisms, the prevention and control strategies are dynamically adjusted and continuously optimized, ultimately outputting a complete implementation plan for epidemic prevention and control in multi-story pig farms, specifically: The immunization recommendations generated by LLM-Agent are presented through a human-computer interaction interface, allowing users to confirm, adjust, or reject the immunization plan. The confirmed immunization plan is then converted into standardized execution instructions. The execution instructions are linked with the automatic immunization equipment or pig farm management system through the platform interface to realize the actual deployment of the immunization plan, including generating the vaccination list, grouping the immunized subjects and allocating vaccine batches, and collecting execution status and result data in real time during the immunization process; Collect users' manual adjustment opinions, preference information, and immunization execution result data, and uniformly input them as feedback into LLM-Agent to update the immunization strategy generation logic and decision context; Based on continuously accumulated execution feedback and user interaction information, LLM-Agent adaptively optimizes subsequent immunization programs, forming personalized vaccine immunization implementation plans for different pig farm sizes, feeding models, and health conditions, and constructing a closed-loop control system of immunization recommendation—execution—feedback—optimization.

2. The intelligent recommendation method for early warning and prevention of swine diseases based on LLM-Agent according to claim 1, characterized in that, The environmental parameters, pig behavior data, health status data, and historical management event information of each floor of the multi-story pig farm were structured and preprocessed, specifically as follows: The raw data from the pig farm's IoT devices, management system, and manual records are deduplicated, formatted, and time-aligned. Interpolation and neighbor-value filling methods are used to handle missing data. Normalize or standardize the numerical characteristics of environmental parameters and behavioral indicators to eliminate the influence between different units, and unstructured data is formatted to form a unified representation that can be processed by large models. Abnormal data is identified through statistical analysis, threshold rules, and anomaly detection algorithms, and features reflecting the operational status of multi-story pig farms are constructed.

3. The intelligent recommendation method for early warning and prevention of swine diseases based on LLM-Agent according to claim 1, characterized in that, Based on preprocessed data from multi-story pig farms, a contextual representation of disease risk scenarios is constructed to reflect the current environmental status of the pig farm, the health level of the pig herd, and historical abnormal events. This representation characterizes the disease risk features at different time points and on different floors. The temperature and humidity, ammonia concentration, carbon dioxide concentration, pig activity, detection data, and disease record characteristics of each floor are coded to form multidimensional risk characteristics. By integrating historical anomaly detection results, key management events, and environmental mutation information with real-time features, a unified representation of epidemic risk status is constructed; The risk status is represented and organized as a structured contextual hint, describing the characteristics of disease risk changes in a pig farm over a specific time period.

4. The intelligent recommendation method for early warning and prevention of swine diseases based on LLM-Agent according to claim 1, characterized in that, Based on the analysis results of different risk levels and causes, multiple candidate prevention and control strategies are generated. These strategies are then comprehensively evaluated and ranked from three dimensions: prevention and control effectiveness, implementation difficulty, and economic cost. The resulting prevention and control strategy plan is as follows: A systematic evaluation of the disease control effectiveness of each candidate prevention and control strategy will be conducted. The evaluation indicators include the actual suppression efficiency of the disease spread after the implementation of the strategy, the speed of mitigation of the risks related to disease transmission, and the continuous stability of the prevention and control effect during the implementation period. At the same time, the actual effectiveness of the prevention and control strategy in achieving the core objectives of disease control and improving the overall level of prevention and control will be measured by combining the disease transmission patterns and scenario characteristics. A comprehensive assessment of the feasibility of implementing candidate prevention and control strategies is conducted. This includes analyzing the scale of personnel input and professional capabilities required for strategy implementation, the complexity and technical threshold of the operational process, and the extent of adjustments to the existing management system and production and operation processes. Additionally, the assessment considers the hardware conditions and scenario adaptability of the actual application scenario to evaluate the operability, ease of execution, and suitability of implementing the strategy under given on-site conditions. A comprehensive assessment of the economic costs of the entire process of prevention and control strategies is conducted, taking into account the direct input costs of material procurement, allocation, and consumption during the implementation of the strategies, the scope and recovery period of the impact of the implementation of the strategies on normal production and operation activities, the potential loss of revenue and opportunity costs caused by the implementation of prevention and control strategies, and the sustainability of cost input in combination with the length of the prevention and control period. From the perspective of economic feasibility, the degree of impact of different prevention and control strategies on overall operating costs and the rationality of cost input are analyzed. Based on preset multi-dimensional weight parameters or user-defined personalized preference strategies, the three evaluation results of prevention and control effectiveness, implementation feasibility, and economic cost are comprehensively weighed and quantitatively scored. Based on the scoring results, all candidate prevention and control strategies are prioritized and selected, and the prevention and control strategies with the best comprehensive performance or those that meet the constraints of specific scenarios are selected to form multiple prevention and control plans.

5. The intelligent recommendation method for early warning and prevention of swine diseases based on LLM-Agent according to claim 1, characterized in that, Also includes: Each floor is divided into multiple zones, environmental data for each zone is collected, an environmental distribution cloud map for that floor is generated using a spatial interpolation algorithm, and a three-dimensional environmental field model is established by combining the floor height and updated in real time. Define a comprehensive stress index, obtain environmental parameter data from the three-dimensional environmental field model, and perform nonlinear correction using weighted summation or fuzzy logic methods; Based on the environmental parameter data, a comprehensive stress index is calculated using a comprehensive stress index calculation method. A correlation model between the stress index and the physiological indicators of pigs is established based on deep learning. Through training with historical data, the immune risk level corresponding to different stress index intervals is determined. The immune risk level of each floor is estimated based on the comprehensive stress index, and the immunization plan is automatically adjusted according to the immune risk level.

6. The intelligent recommendation method for early warning and prevention of swine diseases based on LLM-Agent according to claim 2, characterized in that, Also includes: A digital profile of each pig's immunity is constructed based on historical immunization records. For piglets, a maternal antibody decay curve is fitted for each individual based on maternal immunization information, piglet age, and regular antibody test data. The time point when antibody protection disappears is predicted based on the maternal antibody decay curve. By combining feeding behavior, vocal characteristics, and body temperature data, an immune system activity level is constructed based on the feeding behavior, vocal characteristics, and body temperature data. Collect ammonia concentration, dust concentration, and respiratory symptom sounds for each pig to estimate the mucosal barrier damage index; By fusing the time point of antibody protection loss, immune system activity, and mucosal barrier damage index through an attention mechanism, a comprehensive immune profile vector of the pig herd is generated. Based on the profile, the best vaccine in the vaccine database is matched, and the immunization regimen is adjusted based on the best vaccine in the vaccine database.

Images