Animal epidemic disease prevention and control system based on intelligent AI

Abstract

The invention relates to the technical field of disease prevention and control information, and particularly discloses an animal epidemic disease prevention and control system based on intelligent AI, which comprises a data acquisition module, a data transmission module, a central database module, an individual anomaly detection module, a group propagation prediction module, an intervention strategy optimization module and a decision execution and interaction module. The system collects animal behavior data through the intelligent necklace, adopts double-algorithm fusion analysis to realize accurate identification of individual abnormity, overcomes the problem that epidemic diseases are difficult to find, prevent and control in time in a field grazing environment, constructs a dynamic contact network to predict an epidemic situation propagation path, and applies multi-target reinforcement learning to generate an optimal prevention and control strategy. The intelligent management from individual monitoring to group prevention and control is realized, the accuracy of epidemic disease identification and the timeliness of early warning are remarkably improved, and meanwhile, the prevention and control cost and the influence on the ecological environment are reduced through resource optimization configuration.

Description

Technical Field

[0001] This invention relates to the field of disease prevention and control information technology, specifically to an animal disease prevention and control system based on intelligent AI. Background Technology

[0002] Animal diseases pose a significant biosecurity threat to the development of animal husbandry. They are characterized by their sudden onset, rapid spread, and difficulty in prevention and control. In grazing animal populations, outbreaks and spreads of diseases not only lead to mass mortality and direct economic losses, but also affect animal productivity, reduce the quality of livestock products, and may even trigger regional outbreaks, threatening public health and safety. Furthermore, the occurrence of animal diseases significantly increases breeding costs, including investments in veterinary drugs, vaccines, and manpower for prevention and control, as well as the costs of treating and harmlessly disposing of sick animals, posing a severe challenge to the sustainable development of animal husbandry.

[0003] Currently, existing animal disease prevention and control technologies mainly rely on manual patrols and routine monitoring methods, which suffer from problems such as untimely monitoring, low accuracy of early warning, and weak targeting of prevention and control measures. Traditional methods are difficult to detect early outbreaks in a timely manner, cannot effectively predict the spread trend, and lack a comprehensive assessment of the economic benefits and ecological impact of prevention and control measures. Although some intelligent monitoring technologies have been applied, most systems have limited functions and have failed to form a complete technical system from data collection and intelligent analysis to decision-making and execution. Especially in the complex and ever-changing field grazing environment, existing technologies are difficult to meet the needs of precise and intelligent disease prevention and control. Therefore, there is an urgent need for an intelligent prevention and control system that can achieve early warning, accurate prediction, and optimized decision-making. Summary of the Invention

[0004] (a) Technical problems to be solved This invention provides an animal disease prevention and control system based on intelligent AI, which solves the problems mentioned in the background art above.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: an animal disease prevention and control system based on intelligent AI, comprising... The data acquisition module is used to collect animal movement data, location data, and physiological data; The data transmission module is used to receive and transmit the motion data, position data, and physiological data; The central database module is used to store and manage all data of the system, including raw sensing data, intermediate analysis results, and strategy records; The individual anomaly detection module is used to perform parallel dual-channel analysis based on the received motion data, location data, and physiological data to generate a list of suspected diseased animals that includes animal identifiers and anomaly confidence levels. The group transmission prediction module is used to construct a dynamic contact network based on the list of suspected diseased animals, predict the epidemic transmission path through a spatiotemporal graph neural network, and generate an epidemic transmission risk map. The intervention strategy optimization module is used to generate the optimal intervention strategy based on the epidemic transmission risk map and combined with economic data, ecological data and resource data through multi-objective reinforcement learning. This module includes a strategy simulation and verification module for simulating and verifying the effect of the generated strategy. The decision execution and interaction module is used to execute the optimal intervention strategy and realize human-computer interaction.

[0006] Furthermore, the data acquisition module specifically includes: Data is collected through a smart collar worn around the animal's neck. The smart collar integrates a nine-axis inertial measurement unit, a high-precision GPS positioning module, and a digital body temperature sensor. The nine-axis inertial measurement unit continuously collects the animal's three-dimensional motion acceleration, angular velocity, and direction data at a sampling frequency of no less than 100Hz. The high-precision GPS positioning module records the animal's latitude and longitude coordinates once per second and generates movement trajectory data. The digital body temperature sensor collects the animal's body surface temperature data every five minutes and calculates the core body temperature using an environmental temperature compensation algorithm. All sensor data undergoes timestamp alignment, data packaging, and preliminary compression processing in a microcontroller unit inside the collar.

[0007] Furthermore, the data transmission module specifically includes: The edge computing gateway deployed in the grazing area continuously listens for data transmission requests from the smart collar. The gateway device uses multi-protocol communication and automatically selects LoRaWAN, 4G LTE, or 5G communication links based on network conditions. After receiving data, the gateway's built-in preprocessing algorithm detects and removes outliers from the raw data and uses a sliding window method for data smoothing. When the network connection is unstable, the gateway activates a local caching mechanism to temporarily store the data in a large-capacity storage device and records the data reception timestamp. After the network recovers, the gateway performs breakpoint resumption in chronological order to ensure the integrity and timeliness of data transmission. All processed data is forwarded to the central database module for storage.

[0008] Furthermore, the central database module specifically includes: Four specialized data storage areas are established: raw data area, model parameter area, strategy record area, and knowledge base area. The raw data area receives and categorizes various sensor data from the data transmission module, and establishes a spatiotemporal index to support fast querying. The model parameter area stores the algorithm models and their parameters for the individual anomaly detection, group propagation prediction, and intervention strategy optimization modules. The strategy record area records all generated intervention strategies and their execution effect feedback data in chronological order. The knowledge base area accumulates historical anomaly cases, disease transmission patterns, and expert experience knowledge. The database module provides a unified data access interface for the individual anomaly detection module, group propagation prediction module, intervention strategy optimization module, and decision execution and interaction module, supporting concurrent read / write and data synchronization.

[0009] Furthermore, the individual anomaly detection module specifically includes: The system reads historical behavioral data for each animal over the past 30 days from the central database module, including movement trajectory, activity intensity, and physiological indicators. Then, it uses a sliding window method to segment the historical data, with a window size of 24 hours and a sliding step of 1 hour. Next, it extracts temporal and frequency domain features from the data within each time window. Temporal features include mean, variance, and extreme values; frequency domain features include dominant frequency and power spectral density. Based on these features, an autoencoder network is used for unsupervised training. The encoder maps the input features to a low-dimensional latent space, and the decoder reconstructs the original features. The network parameters are optimized to minimize the reconstruction error, thereby establishing a baseline model of normal behavior for each animal. In the real-time detection phase, the first analysis channel uses a deep temporal convolutional network containing causal convolutional layers and an attention mechanism for real-time analysis. The current behavioral sequence deviates from the baseline, outputting an anomaly score based on time-series reconstruction. The second analysis channel uses an improved isolated forest algorithm, constructing multiple isolation trees by randomly selecting features and segmentation values, calculating the path length required for each sample point to be isolated, and outputting an anomaly index based on feature isolation. The intelligent fusion module performs fusion calculations on the dual-channel results based on Dempster evidence theory. First, the two anomaly scores are converted into basic probability assignments, then evidence is synthesized using Dempster's combination rule, and finally, the comprehensive anomaly confidence is calculated based on the synthesis result. When the confidence exceeds a set threshold, the animal is added to the list of suspected diseased animals, and the anomaly type, occurrence time, duration, and geographical location information are recorded. All intermediate results, final judgment results, and model update parameters in the detection process are transmitted back to the central database module for storage in real time.

[0010] Furthermore, the group propagation prediction module specifically includes: The system retrieves a list of suspected diseased animals from the individual anomaly detection module, using them as the initial source of the epidemic. An initial infection probability is assigned to each source, proportional to the anomaly confidence level. Then, the system reads the movement trajectory data of all animals in the group over the past 24 hours from the central database module. The trajectory data undergoes preprocessing, including trajectory compression using the Douglas-Peucker algorithm, noise filtering using Kalman filtering, and handling missing values ​​using linear interpolation. Next, a dynamic contact network is constructed, where network nodes represent individual animals. Node attributes include species, age, and health status. Edge weights are dynamically calculated based on trajectory co-occurrence probability and movement pattern similarity. Specifically, the geographic space is divided into 100m × 100m grid cells. The probability of any two animals appearing in the same grid cell during the same time period is calculated by statistically analyzing the co-occurrence frequency of their location coordinates within a set time window. Then, similarity in movement direction, movement speed, and activity range is calculated. Direction similarity is calculated using cosine similarity to determine the movement direction vector, and speed similarity is calculated... Similarity is represented by the reciprocal of the relative velocity difference. Activity range similarity is measured by calculating the intersection-union ratio of the polygon areas of the two animals' activity areas. Finally, the spatial co-occurrence probability is weighted and fused with the three movement similarity indices. The weight coefficients are obtained by training a logistic regression model based on historical contact infection data. Specifically, the weights are: spatial co-occurrence probability 0.4, directional similarity 0.2, velocity similarity 0.2, and activity range similarity 0.2. The weighted sum is used to obtain the final contact probability value. Spatiotemporal graph convolutional neural network is used for propagation prediction. The spatial convolutional layer uses Chebyshev polynomial approximation graph convolution operation to capture local propagation patterns in the contact network. The temporal convolutional layer uses dilated causal convolution to capture the temporal pattern of epidemic propagation. The network outputs the infection probability of each node in the next 72 hours. Based on the prediction results, a visualized epidemic propagation risk heat map is generated. The kernel density estimation method is used to transform the discrete infection probability into a continuous risk distribution, identifying high-risk areas and main propagation paths. All intermediate results, final prediction results, and model parameters in the prediction process are updated to the central database module in real time.

[0011] Furthermore, the intervention strategy optimization module specifically includes: The system obtains an epidemic transmission risk map from the group transmission prediction module, analyzes the distribution of high-risk areas, transmission speed, and impact range, and then reads economic value data from the central database module, including market prices, breeding costs, and potential losses for different types of animals. It also reads information on ecologically sensitive areas, including the geographical boundaries and sensitivity levels of water source protection areas, biodiversity hotspots, and ecologically fragile areas, and the status of resource reserves, including the quantity of available vaccines, reserves of disinfection supplies, personnel allocation, and equipment status. Finally, it constructs a multi-objective reinforcement learning environment, which includes three core components: a state space, an action space, and a reward function. The state space is defined as containing epidemic risk characteristics, resource constraints, economic impact assessment, and ecological sensitivity assessment. The multidimensional feature representation defines the action space as a combination of various executable interventions, including the dosage and timing of vaccination, the delineation and duration of isolation areas, the scope and intensity of movement restrictions, and the frequency and scope of environmental disinfection. The reward function is designed as a weighted sum of three objectives: epidemic control effectiveness, economic cost, and ecological impact. Specifically, the epidemic control reward is based on the reduction of the number of infected animals and the decrease in the rate of transmission; the economic cost penalty is based on material consumption and output loss; and the ecological impact penalty is based on the intensity and duration of operations in sensitive areas. A proximal policy optimization algorithm is used to search for the Pareto optimal policy in this environment. An action distribution is generated through a policy network, and a value network evaluates the state value. The `clip` function constrains the policy update magnitude to ensure training stability. The policy simulation verification module predicts the effectiveness of the proposed solutions. This module constructs a simulation environment based on an agent model. Each animal is modeled as an agent with independent attributes, including species type, age and weight, health status, immune level, movement pattern, and social behavior. The simulation system imports the current epidemic transmission risk map, animal population distribution data, and environmental geographic information from the central database module as initial conditions. By integrating the SEIR infectious disease dynamics model and a rule-based behavioral decision-making model, it simulates the state evolution of each agent under different intervention strategies. The specific simulation process includes: first, setting the vaccination coverage rate and isolation area range according to the policy parameters. The simulation is conducted under various intervention conditions, with a time step of 1 hour. Within each time step, the contact probability between agents, the risk of virus transmission, and the immune response mechanism are calculated, and the health status of each agent is dynamically updated. The simulation system runs multiple policy schemes in parallel and conducts repeated experiments using the Monte Carlo method. Key performance indicators for each scheme are statistically analyzed, including indicators of epidemic control effectiveness (number of new infections, transmission interruption time), economic cost indicators (material consumption, human cost), and ecological impact indicators (disturbance degree in sensitive areas). Finally, based on a multi-objective decision analysis method, the comprehensive performance of each scheme in the three dimensions is compared, and a policy effectiveness evaluation report is generated. The final policy schemes and simulation results are stored in the central database module.

[0012] Furthermore, the decision execution and interaction module specifically includes: The optimal intervention strategy is obtained from the intervention strategy optimization module. The specific action instructions, execution sequence, and resource allocation within the strategy are analyzed and displayed multidimensionally on a visualization platform. A geographic information system (GIS) is used to display the action area, a time-series chart to show the execution plan, and a dashboard to display resource usage. The platform provides a strategy details viewing function, including the action basis, expected effects, and risk assessment. It also offers effect simulation and extrapolation capabilities, supports parameter modification for what-if analysis, and provides manual review and confirmation functionality. Multi-person countersigning and approval processes are supported. After approval, the system automatically generates execution instructions and distributes them to the corresponding execution terminals based on the action type, including drone systems, intelligent isolation equipment, mobile inspection terminals, and on-site personnel. Upon receiving the vaccination instruction, the drone system first plans the optimal flight path, considering terrain obstacles, weather conditions, and airspace restrictions. The system automatically completes vaccine loading, flight navigation, and precise delivery tasks, transmitting operational videos and status data in real time. Intelligent isolation equipment automatically deploys electronic fences, configures monitoring cameras and sensor networks based on isolation area planning, dynamically adjusts the isolation range, and records entry / exit logs and alarm information. Mobile inspection terminals receive inspection tasks, navigate to designated locations, collect on-site samples, record observation results, and upload inspection reports in real time. During execution, the system monitors the status of each terminal in real time, including equipment operating status, task execution progress, and abnormal situations. It collects feedback data on execution effects, including animal reactions, environmental changes, and operational problems. All execution records and feedback data are synchronized to the central database module in real time. Based on the execution effect data, a closed-loop optimization mechanism is established to compare actual results with expected goals, analyze the causes of deviations, optimize strategy parameters and operating procedures, and continuously improve system performance.

[0013] Furthermore, the system also includes an online learning and optimization mechanism, the workflow of which includes: New monitoring data and execution effect data are regularly extracted from the central database module. The new data is used to incrementally train the individual anomaly detection model to adapt to changes in animal behavior patterns. The parameters of the group transmission prediction model are adjusted according to the actual transmission situation to improve prediction accuracy. Based on the execution effect of historical strategies, the reward function weight of the intervention strategy generation algorithm is optimized. All model update and parameter adjustment records are stored in the central database module. A model version management mechanism is established to support model rollback and performance comparison analysis.

[0014] Furthermore, the system also includes a distributed computing architecture, the workflow of which includes: Lightweight anomaly detection algorithms are deployed on edge computing gateways to enable real-time local early warning. Complex model training and large-scale simulation computing tasks are deployed in the cloud. Data and model synchronization between the edge and the cloud are achieved through a central database module. Computing resources are dynamically allocated according to task load to ensure system response timeliness. Data backup and disaster recovery mechanisms are established to ensure continuous and reliable system operation.

[0015] (III) Beneficial Effects This invention provides an animal disease prevention and control system based on intelligent AI. It has the following beneficial effects: (I) This AI-based animal disease prevention and control system adopts a dual-analysis channel parallel analysis and intelligent fusion architecture. Through mutual verification of temporal pattern recognition and feature space anomaly detection, it significantly improves the processing accuracy and result accuracy of raw data. On the basis of disease prevention and control, it effectively solves the pain point of insufficient reliability of single algorithms in complex field environments. It can keenly capture real early signals of disease and effectively filter environmental interference factors, providing a high-quality data foundation for subsequent prevention and control decisions.

[0016] (II) This AI-based animal disease prevention and control system, by constructing a dynamic animal contact network and deeply integrating spatiotemporal graph neural network algorithms, achieves quantitative simulation and visualization of the epidemic transmission path. It breaks through the limitations of traditional epidemiological models, can accurately capture changes in the contact patterns of animal groups under natural conditions, and identify high-risk areas and key transmission paths in advance. This provides a scientific basis for implementing precise interception and key deployment, and realizes the transformation from passive response to active defense.

[0017] (III) This AI-based animal disease prevention and control system utilizes a multi-objective reinforcement learning framework to simultaneously optimize three major objectives during strategy generation: epidemic control effectiveness, economic cost-effectiveness, and ecological environmental protection. By constructing a decision-making model with multi-dimensional constraints and a strategy simulation verification mechanism, the system can quickly generate Pareto optimal solutions under specific resource conditions, ensuring that prevention and control measures are both scientific and effective and economically feasible, while minimizing disturbance to the ecological environment and maximizing the comprehensive benefits of prevention and control. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall structure of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] First embodiment: as follows Figure 1 As shown, the present invention provides a technical solution: an animal disease prevention and control system based on intelligent AI, comprising... The data acquisition module is used to collect animal movement data, location data, and physiological data; The data transmission module is used to receive and transmit motion data, location data, and physiological data. The central database module is used to store and manage all data of the system, including raw sensing data, intermediate analysis results, and strategy records; The individual anomaly detection module is used to perform parallel dual-channel analysis based on the received motion data, location data, and physiological data to generate a list of suspected diseased animals that includes animal identifiers and anomaly confidence levels. The group transmission prediction module is used to construct a dynamic contact network based on a list of suspected infected animals, predict the epidemic transmission path through a spatiotemporal graph neural network, and generate an epidemic transmission risk map. The intervention strategy optimization module is used to generate the optimal intervention strategy based on the epidemic transmission risk map and combined with economic data, ecological data and resource data through multi-objective reinforcement learning. This module includes a strategy simulation and verification module, which is used to simulate and verify the effect of the generated strategy. The decision execution and interaction module is used to execute the optimal intervention strategy and realize human-computer interaction.

[0021] Furthermore, the data acquisition module specifically includes: Data is collected through a smart collar worn around the animal's neck. The smart collar integrates a nine-axis inertial measurement unit, a high-precision GPS positioning module, and a digital body temperature sensor. The nine-axis inertial measurement unit continuously collects the animal's three-dimensional motion acceleration, angular velocity, and direction data at a sampling frequency of no less than 100Hz. The high-precision GPS positioning module records the animal's latitude and longitude coordinates once per second and generates movement trajectory data. The digital body temperature sensor collects the animal's body surface temperature data every five minutes and calculates the core body temperature through an environmental temperature compensation algorithm. All sensor data undergoes timestamp alignment, data packaging, and preliminary compression processing in the microcontroller unit inside the collar.

[0022] The data transmission module specifically includes: The edge computing gateway deployed in the grazing area continuously listens for data transmission requests from the smart collar. The gateway device uses multi-protocol communication and automatically selects LoRaWAN, 4G LTE, or 5G communication links based on network conditions. After receiving data, the gateway's built-in preprocessing algorithm detects and removes outliers from the raw data and uses a sliding window method for data smoothing. When the network connection is unstable, the gateway activates a local caching mechanism to temporarily store the data in a large-capacity storage device and records the data reception timestamp. After the network recovers, the gateway performs breakpoint resumption in chronological order to ensure the integrity and timeliness of data transmission. All processed data is forwarded to the central database module for storage.

[0023] The central database module specifically includes: Four specialized data storage areas are established: raw data area, model parameter area, strategy record area, and knowledge base area. The raw data area receives and categorizes various sensor data from the data transmission module, and establishes a spatiotemporal index to support fast querying. The model parameter area stores the algorithm models and their parameters for the individual anomaly detection, group propagation prediction, and intervention strategy optimization modules. The strategy record area records all generated intervention strategies and their execution effect feedback data in chronological order. The knowledge base area accumulates historical anomaly cases, disease transmission patterns, and expert experience knowledge. The database module provides a unified data access interface for the individual anomaly detection module, group propagation prediction module, intervention strategy optimization module, and decision execution and interaction module, supporting concurrent read and write operations and data synchronization.

[0024] Second embodiment: as follows Figure 1 As shown, the individual anomaly detection module specifically includes: First, historical behavioral data for each animal over the past 30 days is retrieved from the central database module, including movement trajectory, activity intensity, and physiological indicators. Then, a sliding window method is used to segment the historical data, with a window size of 24 hours and a sliding step of 1 hour. Next, temporal and frequency domain features are extracted from the data within each time window. Temporal features include mean, variance, and extreme values; frequency domain features include dominant frequency and power spectral density. Based on these features, an autoencoder network is used for unsupervised training. The encoder maps the input features to a low-dimensional latent space, and the decoder reconstructs the original features. The network parameters are optimized to minimize the reconstruction error, thereby establishing a baseline model of normal behavior for each animal. In the real-time detection phase, the first analysis channel uses a deep temporal convolutional network containing causal convolutional layers and an attention mechanism to analyze the deviation of the current behavioral sequence from the baseline in real time, outputting anomaly analysis based on temporal reconstruction. The second analysis channel employs an improved isolated forest algorithm, constructing multiple isolation trees by randomly selecting features and segmentation values. It calculates the path length required for each sample point to be isolated and outputs an anomaly index based on feature isolation. The intelligent fusion module performs fusion calculations on the dual-channel results based on Dempster evidence theory. First, it converts the two anomaly scores into basic probability assignments, then uses Dempster's combination rules to synthesize evidence, and finally calculates the comprehensive anomaly confidence based on the synthesis results. When the confidence exceeds a set threshold, the animal is added to the list of suspected diseased animals, and the anomaly type, occurrence time, duration, and geographical location information are recorded. All intermediate results, final judgment results, and model update parameters in the detection process are transmitted back to the central database module in real time for storage. This dual-channel fusion architecture, through complementary advantages, can effectively distinguish between environmental interference and real disease symptoms, significantly improving the accuracy of early disease identification and greatly reducing the false alarm rate.

[0025] Third embodiment: as follows Figure 1 As shown, the group propagation prediction module specifically includes: First, a list of suspected diseased animals is obtained from the individual anomaly detection module, serving as the initial source of the epidemic. An initial infection probability is assigned to each source, proportional to the anomaly confidence level. Then, the movement trajectory data of all animals in the group over the past 24 hours is read from the central database module. The trajectory data is preprocessed, including trajectory compression using the Douglas-Peucker algorithm, noise filtering using Kalman filtering, and handling missing values ​​using linear interpolation. Next, a dynamic contact network is constructed, where network nodes represent individual animals. Node attributes include species, age, and health status. Edge weights are dynamically calculated based on trajectory co-occurrence probability and movement pattern similarity. Specifically, the geographic space is divided into 100m × 100m grid cells. The probability of any two animals appearing in the same grid cell during the same time period is calculated by statistically analyzing the co-occurrence frequency of their location coordinates within a set time window. Then, similarity in movement direction, movement speed, and activity range is calculated. Direction similarity is calculated using cosine similarity to determine the movement direction vector, speed similarity is represented by the reciprocal of the relative speed difference, and activity range similarity is calculated by... The intersection-union ratio (IUGR) of the polygon areas of two animal activity regions is used as a metric. Finally, the spatial co-occurrence probability is weighted and fused with three movement similarity indices. The weight coefficients are obtained through a logistic regression model trained on historical contact infection data: spatial co-occurrence probability weight 0.4, orientation similarity weight 0.2, velocity similarity weight 0.2, and activity range similarity weight 0.2. The weighted sum is used to obtain the final contact probability value. A spatiotemporal graph convolutional neural network is used for propagation prediction. The spatial convolutional layer uses Chebyshev multinomial approximation graph convolution operation to capture local propagation patterns in the contact network. The temporal convolution... The layer uses dilated causal convolution to capture the temporal patterns of epidemic transmission. The network outputs the infection probability of each node in the next 72 hours. Based on the prediction results, a visualized epidemic transmission risk heat map is generated. The kernel density estimation method is used to transform the discrete infection probability into a continuous risk distribution, identifying high-risk areas and main transmission paths. All intermediate results, final prediction results and model parameters in the prediction process are updated to the central database module in real time. This dynamic contact network model can accurately quantify the effective contact probability between animals, significantly improve the accuracy of epidemic transmission prediction, and achieve early warning of high-risk areas.

[0026] Fourth embodiment: as Figure 1 As shown, the intervention strategy optimization module specifically includes: First, an epidemic transmission risk map is obtained from the group transmission prediction module, analyzing the distribution of high-risk areas, transmission speed, and impact range. Then, economic value data, including market prices, breeding costs, and potential losses for different animal species, is retrieved from the central database module. Information on ecologically sensitive areas, including the geographical boundaries and sensitivity levels of water source protection areas, biodiversity hotspots, and ecologically fragile areas, is also retrieved. Resource inventory status, including the quantity of available vaccines, reserves of disinfection supplies, personnel allocation, and equipment status, is also retrieved. Next, a multi-objective reinforcement learning environment is constructed, comprising three core components: a state space, an action space, and a reward function. The state space is defined as a multi-dimensional feature table containing epidemic risk characteristics, resource constraints, economic impact assessment, and ecological sensitivity assessment. The action space is defined as a combination of various interventions that the system can execute, including the dosage and timing of vaccination, the delineation and duration of isolation areas, the scope and intensity of movement restrictions, and the frequency and scope of environmental disinfection. The reward function is designed as a weighted sum of three objectives: epidemic control effectiveness, economic cost, and ecological impact. Specifically, the epidemic control reward is based on the reduction of the number of infected animals and the decrease in the rate of transmission; the economic cost penalty is based on material consumption and output loss; and the ecological impact penalty is based on the intensity and duration of operations in sensitive areas. A proximal policy optimization algorithm is used to search for the Pareto optimal policy in this environment. An action distribution is generated through a policy network, a value network evaluates the state value, and a clip function is used to constrain the policy update magnitude. To ensure training stability, the effectiveness of the intervention plan is predicted through a strategy simulation verification module. This module constructs a simulation environment based on an agent model, where each animal is modeled as an agent with independent attributes, including species type, age and weight, health status, immune level, movement pattern, and social behavior. The simulation system imports the current epidemic transmission risk map, animal population distribution data, and environmental geographic information from the central database module as initial conditions. By integrating the SEIR infectious disease dynamics model and a rule-based behavioral decision-making model, it simulates the state evolution of each agent under different intervention strategies. The specific simulation process includes: first, setting intervention conditions such as vaccination coverage and isolation area range according to strategy parameters; then, advancing the simulation in 1-hour time steps. Within each time step, the system calculates the contact probability between agents, the risk of virus transmission, and the immune response mechanism, and dynamically updates the health status of each agent. The simulation system runs multiple policy schemes in parallel, conducting repeated experiments using the Monte Carlo method to statistically analyze key performance indicators for each scheme, including epidemic control effectiveness indicators (number of new infections, transmission interruption time), economic cost indicators (material consumption, labor costs), and ecological impact indicators (disturbance degree in sensitive areas). Finally, based on a multi-objective decision analysis method, the system compares the comprehensive performance of each scheme across the three dimensions, generating a policy effectiveness evaluation report. The final policy schemes and simulation results are stored in the central database module. This multi-objective optimization and simulation verification mechanism can quickly generate optimal policies, significantly improving decision-making efficiency.This effectively reduces the economic costs of prevention and control while minimizing disturbance to the ecological environment.

[0027] Fifth embodiment: as follows Figure 1 As shown, the decision execution and interaction module specifically includes: First, the optimal intervention strategy is obtained from the intervention strategy optimization module. The specific action instructions, execution sequence, and resource allocation within the strategy are analyzed and displayed in a multi-dimensional visualization platform. A geographic information system (GIS) is used to display the action area, a time-series chart to show the execution plan, and a dashboard to display resource usage. The platform provides a strategy details viewing function, including the action basis, expected effects, and risk assessment. It also offers effect simulation and extrapolation capabilities, supports parameter modification for what-if analysis, and provides manual review and confirmation functionality, supporting multi-person countersigning and approval processes. After approval, the system automatically generates execution instructions and distributes them to the corresponding execution terminals based on the action type, including drone systems, intelligent isolation equipment, mobile inspection terminals, and on-site personnel. Upon receiving the vaccination instruction, the drone system first plans the optimal flight path, considering terrain obstacles, weather conditions, and airspace restrictions. Then, it automatically completes vaccine loading, flight navigation, and precise delivery tasks, transmitting data back in real time. Based on video and status data, intelligent isolation equipment automatically deploys electronic fences, configures surveillance cameras and sensor networks according to the isolation area plan, dynamically adjusts the isolation range, records entry and exit logs and alarm information, and mobile inspection terminals receive inspection tasks, navigate to designated locations, collect on-site samples, record observation results, and upload inspection reports in real time. During execution, the status of each terminal is monitored in real time, including equipment operating status, task execution progress and abnormal situations. Execution effect feedback data is collected, including animal reactions, environmental changes and operational problems. All execution records and feedback data are synchronized to the central database module in real time. Based on the execution effect data, a closed-loop optimization mechanism is established to compare the actual effect with the expected goal, analyze the reasons for deviations, optimize strategy parameters and operation processes, and continuously improve system performance. This automated execution and closed-loop optimization system can significantly improve strategy execution efficiency, reduce human resource input, and continuously improve the overall prevention and control performance of the system through continuous learning.

[0028] The system also includes an online learning and optimization mechanism, the workflow of which includes: New monitoring data and execution effect data are regularly extracted from the central database module. The new data is used to incrementally train the individual anomaly detection model to adapt to changes in animal behavior patterns. The parameters of the group transmission prediction model are adjusted according to the actual transmission situation to improve prediction accuracy. Based on the execution effect of historical strategies, the reward function weight of the intervention strategy generation algorithm is optimized. All model update and parameter adjustment records are stored in the central database module. A model version management mechanism is established to support model rollback and performance comparison analysis.

[0029] The system also includes a distributed computing architecture, whose workflow includes: Lightweight anomaly detection algorithms are deployed on edge computing gateways to enable real-time local early warning. Complex model training and large-scale simulation computing tasks are deployed in the cloud. Data and model synchronization between the edge and the cloud are achieved through a central database module. Computing resources are dynamically allocated according to task load to ensure system response timeliness. Data backup and disaster recovery mechanisms are established to ensure continuous and reliable system operation.

[0030] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0031] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An animal disease prevention and control system based on intelligent AI, characterized in that: include The data acquisition module is used to collect animal movement data, location data, and physiological data; The data transmission module is used to receive and transmit the motion data, position data, and physiological data; The central database module is used to store and manage all data of the system, including raw sensing data, intermediate analysis results, and strategy records; The individual anomaly detection module is used to perform parallel dual-channel analysis based on the received motion data, location data, and physiological data to generate a list of suspected diseased animals that includes animal identifiers and anomaly confidence levels. The group transmission prediction module is used to construct a dynamic contact network based on the list of suspected diseased animals, predict the epidemic transmission path through a spatiotemporal graph neural network, and generate an epidemic transmission risk map. The intervention strategy optimization module is used to generate the optimal intervention strategy based on the epidemic transmission risk map and combined with economic data, ecological data and resource data through multi-objective reinforcement learning. This module includes a strategy simulation and verification module for simulating and verifying the effect of the generated strategy. The decision execution and interaction module is used to execute the optimal intervention strategy and realize human-computer interaction.

2. The animal disease prevention and control system based on intelligent AI according to claim 1, characterized in that: The data acquisition module specifically includes: Data is collected through a smart collar worn around the animal's neck. The smart collar integrates a nine-axis inertial measurement unit, a high-precision GPS positioning module, and a digital body temperature sensor. The nine-axis inertial measurement unit continuously collects the animal's three-dimensional motion acceleration, angular velocity, and direction data at a sampling frequency of no less than 100Hz. The high-precision GPS positioning module records the animal's latitude and longitude coordinates once per second and generates movement trajectory data. The digital body temperature sensor collects the animal's body surface temperature data every five minutes and calculates the core body temperature using an environmental temperature compensation algorithm. All sensor data undergoes timestamp alignment, data packaging, and preliminary compression processing in a microcontroller unit inside the collar.

3. The animal disease prevention and control system based on intelligent AI according to claim 2, characterized in that: The data transmission module specifically includes: The edge computing gateway deployed in the grazing area continuously listens for data transmission requests from the smart collar. The gateway device uses multi-protocol communication and automatically selects LoRaWAN, 4G LTE, or 5G communication links based on network conditions. After receiving data, the gateway's built-in preprocessing algorithm detects and removes outliers from the raw data and uses a sliding window method for data smoothing. When the network connection is unstable, the gateway activates a local caching mechanism to temporarily store the data in a large-capacity storage device and records the data reception timestamp. After the network recovers, the gateway performs breakpoint resumption in chronological order to ensure the integrity and timeliness of data transmission. All processed data is forwarded to the central database module for storage.

4. The animal disease prevention and control system based on intelligent AI according to claim 3, characterized in that: The central database module specifically includes: Four specialized data storage areas are established: raw data area, model parameter area, strategy record area, and knowledge base area. The raw data area receives and categorizes various sensor data from the data transmission module, and establishes a spatiotemporal index to support fast querying. The model parameter area stores the algorithm models and their parameters for the individual anomaly detection, group propagation prediction, and intervention strategy optimization modules. The strategy record area records all generated intervention strategies and their execution effect feedback data in chronological order. The knowledge base area accumulates historical anomaly cases, disease transmission patterns, and expert experience knowledge. The database module provides a unified data access interface for the individual anomaly detection module, group propagation prediction module, intervention strategy optimization module, and decision execution and interaction module, supporting concurrent read / write and data synchronization.

5. The animal disease prevention and control system based on intelligent AI according to claim 4, characterized in that: The individual anomaly detection module specifically includes: The system reads historical behavioral data of each animal from the central database module to establish a baseline model of normal behavior for each animal. Then, the first analysis channel uses a deep temporal convolutional network to analyze the deviation of the current behavioral sequence from the baseline in real time and outputs an anomaly score based on temporal reconstruction. The second analysis channel uses an improved isolated forest algorithm to extract multidimensional statistical features for spatial anomaly detection and outputs an anomaly index based on feature isolation. Finally, the intelligent fusion module performs fusion calculation on the dual-channel results based on DS evidence theory to generate a comprehensive anomaly confidence score. When the confidence score exceeds a set threshold, the animal is added to the list of suspected diseased animals, and the anomaly type and time information are recorded. All detection results and model update parameters are sent back to the central database module for storage in real time.

6. The animal disease prevention and control system based on intelligent AI according to claim 5, characterized in that: The group propagation prediction module specifically includes: The system obtains a list of suspected infected animals from the individual anomaly detection module to determine the initial source of the epidemic. Then, it reads the movement trajectory data of all animals in the group from the central database module to construct a dynamic contact network. In this network, network nodes represent individual animals, and edge weights are dynamically calculated based on trajectory co-occurrence probability and movement pattern similarity. A spatiotemporal graph convolutional neural network is used for propagation prediction. Spatial convolutional layers capture local propagation patterns in the contact network, while temporal convolutional layers learn the temporal patterns of epidemic propagation to predict the infection risk distribution in the next 72 hours. A visualized epidemic propagation risk map is generated, identifying high-risk areas and propagation paths. The prediction results and model parameters are updated to the central database module in real time.

7. The animal disease prevention and control system based on intelligent AI according to claim 6, characterized in that: The intervention strategy optimization module specifically includes: The system obtains an epidemic transmission risk map from the group transmission prediction module to assess the current epidemic situation. Then, it reads economic value data, ecologically sensitive area information, and resource inventory status from the central database module to construct a multi-objective reinforcement learning environment. This environment includes three core components: state space, action space, and reward function. The state space is defined as a multi-dimensional feature representation that includes epidemic risk and resource constraints. The action space is defined as a combination of various intervention measures that the system can execute, including operations such as vaccination, regional isolation, and movement restrictions. The reward function is designed as a weighted sum of three objectives that simultaneously optimize the epidemic control effect, economic cost, and ecological impact. The proximal policy optimization algorithm searches for the Pareto optimal policy in this environment to generate detailed intervention plans. The policy simulation verification module predicts the effect of the plan, supports manual adjustment and optimization, and finally stores the policy and simulation results in the central database module.

8. The animal disease prevention and control system based on intelligent AI according to claim 7, characterized in that: The decision execution and interaction module specifically includes: The optimal intervention strategy is obtained from the intervention strategy optimization module and displayed in multiple dimensions on the visualization platform. The platform provides functions such as viewing strategy details, effect simulation, and manual review and confirmation. After approval, the system automatically generates execution instructions and distributes them to the execution terminals of the drone system and intelligent isolation equipment. The drone system receives the vaccination instructions, plans the optimal route, and executes the precise delivery task. The intelligent isolation equipment dynamically adjusts the range of the electronic fence according to the isolation area plan. During the execution process, the status of the execution terminals is monitored in real time, and the execution effect feedback data is collected. All execution records and feedback data are synchronized to the central database module in real time. Based on the execution effect data, a closed-loop optimization mechanism is established to continuously improve system performance.

9. The animal disease prevention and control system based on intelligent AI according to claim 8, characterized in that: The system also includes an online learning and optimization mechanism, the workflow of which includes: New monitoring data and execution effect data are regularly extracted from the central database module. The new data is used to incrementally train the individual anomaly detection model to adapt to changes in animal behavior patterns. The parameters of the group transmission prediction model are adjusted according to the actual transmission situation to improve prediction accuracy. Based on the execution effect of historical strategies, the reward function weight of the intervention strategy generation algorithm is optimized. All model update and parameter adjustment records are stored in the central database module. A model version management mechanism is established to support model rollback and performance comparison analysis.

10. The animal disease prevention and control system based on intelligent AI according to claim 9, characterized in that: The system also includes a distributed computing architecture, the workflow of which includes: Lightweight anomaly detection algorithms are deployed on edge computing gateways to enable real-time local early warning. Complex model training and large-scale simulation computing tasks are deployed in the cloud. Data and model synchronization between the edge and the cloud are achieved through a central database module. Computing resources are dynamically allocated according to task load to ensure system response timeliness. Data backup and disaster recovery mechanisms are established to ensure continuous and reliable system operation.

Images