Livestock health monitoring and disease early warning system and method based on environmental data fusion

Abstract

The invention relates to the technical field of livestock health management systems, in particular to a livestock health monitoring and disease early warning system and method based on environmental data fusion, and the system comprises a data collection module which transmits environmental data and livestock physiological data to a data analysis module; the data analysis module is in communication connection with the data acquisition module and is used for receiving the environment data and the livestock physiological data sent by the data acquisition module; based on the environmental data and the livestock physiological data, constructing livestock health features; analyzing livestock health features by using the improved graph neural network model; the early warning module is in communication connection with the data analysis module and is used for receiving an analysis result of the data analysis module; based on the analysis result, a disease risk early warning report of the livestock is generated; and the application end displays a disease risk early warning report to the user, so that the health condition of the livestock can be evaluated more comprehensively, and the monitoring accuracy and the early warning timeliness are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of livestock health management systems, and more specifically, to a livestock health monitoring and disease early warning system and method based on environmental data fusion. Background Art

[0002] With the rapid development of modern animal husbandry, livestock health monitoring and disease early warning systems are playing an increasingly important role in improving breeding efficiency and ensuring the quality of livestock products. Traditional livestock health monitoring mainly relies on manual observation and regular inspections, which is not only time-consuming and labor-intensive, but also difficult to detect potential health problems in a timely manner. In recent years, with the development of Internet of Things technology and artificial intelligence, automated livestock health monitoring systems have gradually become a research hotspot.

[0003] At present, there are some livestock health monitoring systems based on sensor technology on the market. These systems usually collect physiological data, such as body temperature and heart rate, through sensor devices worn on livestock, and transmit these data to a central processing unit for analysis. However, this approach has several obvious limitations. First, wearable devices may cause discomfort to livestock, affecting their normal behavior and physiological state. Second, this method mainly focuses on the physiological indicators of individual livestock, and it is difficult to fully consider the impact of environmental factors on the health of the entire group.

[0004] Another common approach is health monitoring systems based on image recognition. These systems capture images of livestock behavior through cameras installed on farms, and then use computer vision algorithms to analyze livestock behavior patterns. Although this approach avoids direct interference with livestock, it mainly relies on external manifestations and has difficulty accurately assessing the internal physiological state of livestock. In addition, this approach often faces challenges in computing efficiency and real-time performance when processing massive video data from large-scale farms.

[0005] Some studies have also attempted to combine environmental monitoring with livestock health monitoring. These systems use environmental sensors to collect data such as temperature, humidity, and gas concentration in farms, and conduct comprehensive analysis based on livestock physiological indicators. However, existing methods often simply analyze environmental data and physiological data in parallel, lacking in-depth exploration of the complex relationship between the two. This approach makes it difficult to fully utilize the inherent connection between environmental factors and livestock health, thus affecting the accuracy of the system's early warning.

[0006] In addition, existing livestock health monitoring systems also have some deficiencies in data analysis. Most systems use traditional statistical methods or simple machine learning algorithms for data processing, which are difficult to effectively handle high-dimensional, multi-source heterogeneous data. In particular, when dealing with livestock health data with temporal characteristics and spatial correlation, existing methods often have difficulty capturing complex patterns and long-term dependencies in the data.

[0007] Finally, the performance of existing systems in disease early warning is also unsatisfactory. Many systems use fixed thresholds to trigger early warnings, which lacks flexibility and is difficult to adapt to the characteristics of different breeding environments and different breeds of livestock. At the same time, the interpretability of early warning results is also a common problem. Many systems are unable to provide clear explanations and specific intervention recommendations for early warning results, which greatly limits the practicality of the system in actual applications.

[0008] In view of the above-mentioned deficiencies of existing technologies, there is an urgent need for a livestock health monitoring and disease early warning system that can fully integrate environmental data and physiological data, has strong data analysis capabilities, and can provide accurate and interpretable early warning results. Summary of the invention

[0009] The livestock health monitoring and disease early warning system and method based on environmental data fusion of the present invention is proposed to address the above technical problems. The system achieves comprehensive and accurate monitoring of livestock health status and timely early warning through innovative data fusion methods, improved graph neural network models and advanced early warning mechanisms.

[0010] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0011] Livestock health monitoring and disease early warning system based on environmental data fusion, including:

[0012] Data acquisition module for:

[0013] Collect environmental data of livestock breeding space and livestock physiological data;

[0014] transmitting the environmental data and the livestock physiological data to a data analysis module;

[0015] A data analysis module is connected to the data acquisition module for:

[0016] Receiving environmental data and livestock physiological data sent by the data acquisition module;

[0017] constructing a livestock health signature based on the environmental data and the livestock physiological data;

[0018] Analyzing the livestock health characteristics using an improved graph neural network model;

[0019] The early warning module is connected to the data analysis module for:

[0020] Receiving the analysis result of the data analysis module;

[0021] Based on the analysis results, a livestock disease risk warning report is generated;

[0022] The application end is connected to the early warning module for:

[0023] Receiving the disease risk warning report generated by the warning module;

[0024] The disease risk warning report is displayed to the user.

[0025] Preferably, the data acquisition module comprises:

[0026] Temperature and humidity sensors are used to collect temperature and humidity data of livestock breeding spaces;

[0027] CO2 sensor, used to collect CO2 concentration data in livestock breeding spaces;

[0028] Cameras to collect data on livestock’s body temperature, shape, posture, and behavior;

[0029] A data converter is connected to the temperature and humidity sensor, the CO2 sensor and the camera, and is used to convert the collected data into a processable format.

[0030] Preferably, the data analysis module comprises:

[0031] A data cleaning unit, used for cleaning the environmental data and the livestock physiological data;

[0032] A feature construction unit, used to construct livestock health features based on the cleaned data;

[0033] A graph neural network analysis unit is used to analyze the health characteristics of the livestock using an improved graph neural network model.

[0034] Preferably, the graph neural network analysis unit comprises:

[0035] A graph structure building module, used to build a graph structure based on the livestock health characteristics;

[0036] A multi-layer graph self-attention network module, used for extracting features from the graph structure;

[0037] A global-local graph convolutional neural network module is used to combine global information and local features for analysis.

[0038] Preferably, the early warning module comprises:

[0039] A conditional random field model is used to perform sequence labeling on the analysis results of the data analysis module;

[0040] A Bayesian inference unit, used for inferring disease risk based on the output results of the conditional random field model;

[0041] The threshold adjustment unit is used to dynamically adjust the disease risk warning threshold based on historical data.

[0042] Preferably, the application end includes:

[0043] Visual display module, used to display disease risk warning results in the form of charts or text;

[0044] Event analysis module, used to track and analyze warning events;

[0045] The early warning management module is used to set system alarm thresholds and manage user permissions.

[0046] Preferably, data is transmitted between the data analysis module and the early warning module via an encrypted communication protocol.

[0047] Preferably, the system further comprises a data storage module for storing the environmental data, the livestock physiological data, the livestock health characteristics and the disease risk warning report.

[0048] Preferably, the improved graph neural network model includes adding additional edges based on the original graph, and the additional edges are used to connect each node and its most adjacent node.

[0049] The livestock health monitoring and disease early warning method based on environmental data fusion using the system comprises the following steps:

[0050] Collect environmental data of livestock breeding space and livestock physiological data;

[0051] Cleaning the environmental data and the livestock physiological data;

[0052] construct livestock health signatures based on the cleaned data;

[0053] Analyzing the livestock health characteristics using an improved graph neural network model;

[0054] Based on the analysis results, sequence labeling is performed using a conditional random field model;

[0055] Generate early warning reports on livestock disease risks using a Bayesian inference framework;

[0056] The disease risk warning report is displayed to the user.

[0057] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0058] First of all, the system of the present invention not only solves the above technical problems, but also brings a series of significant technical effects. First, through multi-source data fusion technology, the system realizes the deep integration of environmental data and livestock physiological data, which not only improves the utilization efficiency of data, but also reveals the complex relationship between environmental factors and livestock health status. This fusion method enables the system to more comprehensively evaluate the health status of livestock, greatly improving the accuracy of monitoring and the timeliness of early warning.

[0059] Secondly, the improved graph neural network model adopted by the present invention significantly enhances the data analysis capability of the system. By adding additional edges to the original graph structure, the model can better capture the local similarities and global dependencies between data. This innovation not only improves the model's ability to recognize complex patterns, but also enhances the system's ability to predict long-term health trends. In particular, when processing massive data from large-scale farms, the model exhibits excellent computational efficiency and scalability.

[0060] Furthermore, the early warning mechanism of the present invention achieves a more accurate and reliable disease risk assessment by combining conditional random fields and Bayesian inference framework. This method not only takes into account the temporal dependence of health status, but also quantifies the uncertainty of early warning results, providing a more comprehensive and reliable decision-making basis for breeding managers. At the same time, the system's dynamic threshold adjustment mechanism enables the early warning standard to be adaptively adjusted according to actual conditions, effectively improving the sensitivity and specificity of the early warning.

[0061] In addition, the system of the present invention also has significant advantages in practicality and scalability. The modular design of the system enables each functional unit to be independently upgraded and optimized while maintaining good overall synergy. This design not only improves the maintenance efficiency of the system, but also provides convenience for future functional expansion. In particular, the data storage module of the system adopts a distributed architecture and data hierarchical storage strategy, which not only ensures the efficiency of big data processing, but also takes into account the economy of long-term data storage.

[0062] Finally, the system of the present invention greatly improves the interpretability and practicality of the early warning results through an intuitive visual interface and detailed early warning reports. The system can not only issue early warning signals in a timely manner, but also provide specific risk factor analysis and intervention suggestions, which is of great significance for guiding breeding management practices and improving disease prevention effects.

[0063] In general, the livestock health monitoring and disease early warning system and method based on environmental data fusion of the present invention effectively solves the problems existing in the prior art through innovative technical solutions, and significantly improves the accuracy, comprehensiveness and practicality of livestock health monitoring. The system can not only help farms improve production efficiency and reduce disease risks, but also provide strong technical support for the sustainable development of animal husbandry. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a logic block diagram of the overall structure of the system of the present invention.

[0065] Figure 2 It is a logic block diagram of the data acquisition module of the present invention.

[0066] Figure 3 It is a logic block diagram of the data analysis module of the present invention.

[0067] Figure 4 It is a logical block diagram of the graph neural network analysis unit of the present invention.

[0068] Figure 5 It is a logic block diagram of the data cleaning unit of the present invention.

[0069] Figure 6 A logical block diagram of a feature building unit of the present invention.

[0070] Figure 7 It is a logic block diagram of the early warning module of the present invention. DETAILED DESCRIPTION

[0071] like Figure 1-7 As shown, the present invention provides a livestock health monitoring and disease early warning system and method based on environmental data fusion. The system includes a data acquisition module 1, a data analysis module 2, an early warning module 3 and an application terminal 4.

[0072] The data acquisition module 1 is used to collect environmental data and livestock physiological data of the livestock breeding space, and transmit these data to the data analysis module 2. In a preferred embodiment of the present invention, the environmental data may include parameters such as temperature, humidity and CO2 concentration. The livestock physiological data may include indicators such as body surface temperature, heart rate and respiratory rate. The data acquisition module 1 collects these data in real time through a variety of sensors and devices to provide a basis for subsequent analysis.

[0073] The data analysis module 2 is in communication with the data acquisition module 1 and is used to receive environmental data and livestock physiological data from the data acquisition module 1. Based on these data, the data analysis module 2 constructs livestock health characteristics and analyzes these characteristics using an improved graph neural network model. In the present invention, the construction of livestock health characteristics takes into account the combined effects of environmental factors and physiological indicators, making the analysis results more comprehensive and accurate.

[0074] The improved graph neural network model is an important innovation of the present invention. The model improves the connectivity between data and the efficiency of information transmission by adding additional edges to the original graph. Specifically, the mathematical expression of the model can be described by the following formula:

[0075]

[0076] Among them, H (l) represents the node feature matrix of the lth layer, To add the adjacency matrix after the additional edge, is the corresponding degree matrix, W (l) is the learnable weight matrix and σ is the activation function.

[0077] The early warning module 3 is connected to the data analysis module 2 for receiving the analysis results of the data analysis module 2. Based on these results, the early warning module 3 generates a disease risk early warning report for livestock. In one embodiment of the present invention, the early warning module 3 uses a conditional random field (CRF) model for sequence labeling, and then uses a Bayesian inference framework to generate a final risk early warning report. This method combines the advantages of the discriminant model and the generative model, and improves the accuracy and interpretability of the early warning.

[0078] The application terminal 4 is connected to the early warning module 3 for receiving the disease risk early warning reports generated by the early warning module 3 and displaying these reports to the user. The application terminal 4 of the present invention is not only a simple data display platform, but also provides an interactive analysis tool, so that the user can deeply understand the reasons for the early warning results and possible countermeasures.

[0079] In a specific embodiment of the present invention, the data acquisition module 1 includes a temperature and humidity sensor 11, a CO2 sensor 12, a camera 13 and a data converter 14. The temperature and humidity sensor 11 is used to collect temperature and humidity data of the livestock breeding space. Preferably, the sampling frequency of the temperature and humidity sensor 11 can be set to once every 5 minutes to balance data accuracy and system load. The CO2 sensor 12 is used to collect CO2 concentration data of the livestock breeding space. The CO2 sensor 12 used in the present invention has high sensitivity and can detect CO2 concentration changes in the range of 50ppm to 5000ppm, covering all concentration levels from normal to dangerous.

[0080] The camera 13 is used to collect the body surface temperature, body shape, posture and behavior data of the livestock. The camera 13 used in the present invention is equipped with an infrared thermal imaging function, which can accurately measure the body surface temperature of the livestock with a resolution of 0.1°C. At the same time, the camera 13 can also capture the behavior patterns of the livestock, such as feeding frequency, movement status, etc., which are crucial for early disease warning.

[0081] The data converter 14 is connected to the temperature and humidity sensor 11, the CO2 sensor 12 and the camera 13, and is used to convert the collected data into a processable format. In the present invention, the data converter 14 adopts a 16-bit ADC (analog-to-digital converter) to ensure high accuracy of data conversion. At the same time, the data converter 14 also has a data compression function, which can reduce the burden of data transmission and storage while ensuring data quality.

[0082] The data analysis module 2 of the present invention includes a data cleaning unit 21, a feature construction unit 22 and a graph neural network analysis unit 23. The data cleaning unit 21 first cleans the received environmental data and livestock physiological data. This process includes steps such as removing outliers, filling missing values ​​and data standardization. In the present invention, the determination of outliers adopts an improved Z-score method, namely:

[0083]

[0084] Where X is the original data, median(X) is the median, and MAD is the median absolute deviation. When the Z value is greater than 3.5, the data point is considered an outlier. This method is more robust than the traditional Z-score method and is not easily affected by extreme values.

[0085] The feature construction unit 22 constructs livestock health features based on the cleaned data. The feature construction process of the present invention not only considers the statistical characteristics of a single indicator (such as mean, variance, etc.), but also introduces the correlation and time series characteristics between multiple indicators. For example, we define an "environmental stress index" (EPI), and its calculation formula is:

[0086]

[0087] Among them, T, H, and C represent the current temperature, humidity, and CO2 concentration respectively. opt , H opt , C opt is the corresponding optimal value, T max , T min etc. are the maximum and minimum values ​​allowed, and w1, w2, and w3 are weight coefficients. This index comprehensively reflects the impact of environmental factors on livestock health.

[0088] The graph neural network analysis unit 23 uses the improved graph neural network model to analyze the constructed livestock health characteristics. The graph neural network model of the present invention introduces an attention mechanism and a jump connection on the basis of the traditional GCN, wherein the calculation formula of the attention mechanism is:

[0089]

[0090] Among them, h i and h j Represent the features of node i and node j respectively, W is the weight matrix, a is the attention vector, || represents the concatenation operation, Represents the set of neighbors of node i. This design enables the model to adaptively focus on important features and connections, improving the accuracy and efficiency of analysis.

[0091] From the above description, it can be seen that the system of the present invention has unique innovations in data collection, analysis and early warning, and can comprehensively and accurately monitor the health status of livestock and timely warn of potential disease risks. This method based on environmental data fusion not only improves the accuracy of early warning, but also provides valuable decision support for breeding management, and has important practical application value.

[0092] In the system of the present invention, the graph neural network analysis unit 23 includes a graph structure construction module 231, a multi-layer graph self-attention network module 232, and a global-local graph convolutional neural network module 233. This modular design enables the system to process complex livestock health data more flexibly.

[0093] The graph structure construction module 231 is used to construct a graph structure based on livestock health features. In a preferred embodiment of the present invention, the construction of the graph structure not only considers the correlation between features, but also introduces time series information. Specifically, each node represents a set of health features at a time point, and the edge represents the association between different time points or different features. This graph structure can effectively capture the dynamic changes in the health status of livestock.

[0094] Preferably, the graph structure building module 231 uses an adaptive threshold method to determine the connection of edges. Its mathematical expression is:

[0095]

[0096] Among them, A i j represents the element of the adjacency matrix, x i and x j denote the feature vectors of node i and node j respectively, sin(·,·) is the similarity function, θ i and θ jis an adaptive threshold. This method can automatically adjust the connection threshold according to the local data distribution, thereby improving the expressiveness of the graph structure. The multi-layer graph self-attention network module 232 is used to extract features from the graph structure. The present invention adopts a multi-head attention mechanism, where each attention head independently learns different feature representations and then splices these representations together. Its mathematical expression is:

[0097]

[0098] in, represents the feature representation of node i in layer l, K is the number of attention heads, The attention weight calculated for the kth attention head, W k is the corresponding weight matrix, and σ is the activation function. In the embodiment of the present invention, the number of attention heads K is set to 8, which is the result of a trade-off between model complexity and expressiveness.

[0099] The global-local graph convolutional neural network module 233 is used to combine global information and local features for analysis. The module first uses a global average pooling operation to obtain a global representation of the graph, then concatenates it with the local features of each node, and finally performs information fusion through a graph convolution operation. Its mathematical expression is:

[0100]

[0101] Among them, d i and d j denote the degrees of node i and node j respectively, is the global representation of the lth layer, and || represents the concatenation operation. This design enables the model to consider both local and global information at the same time, improving the ability to understand the health status of livestock.

[0102] The early warning module 3 of the present invention includes a conditional random field model 31, a Bayesian inference unit 32 and a threshold adjustment unit 33. This combined design makes full use of the advantages of sequence modeling and probabilistic reasoning, and can generate a more reliable disease risk early warning report. The conditional random field model 31 is used to perform sequence annotation on the analysis results of the data analysis module 2. In a preferred embodiment of the present invention, a linear chain conditional random field (linear-chain CRF) is used to model the temporal dependency of livestock health status. Its potential function is defined as:

[0103]

[0104] Among them, y t represents the label at time t, X is the observation sequence, and f k is the characteristic function, λ k is the corresponding weight. In the present invention, the feature function includes not only the transfer feature (such as ), also includes state characteristics (such as ). This design can make full use of the information of environmental and physiological indicators and improve the accuracy of labeling.

[0105] The Bayesian inference unit 32 is used to infer the disease risk based on the output results of the conditional random field model 31. The present invention adopts a variational inference method to approximate the posterior distribution, and its objective function is:

[0106] ELBO(q)=E q [logp(X,Y)]-E q [logq(Y)]

[0107] Where q(Y) is the approximate posterior distribution and p(X,Y) is the joint distribution. By maximizing ELBO, the system can achieve a good balance between computational efficiency and inference accuracy. The threshold adjustment unit 33 is used to dynamically adjust the disease risk warning threshold according to historical data. The present invention adopts an adaptive threshold algorithm, and its mathematical expression is:

[0108] θ t =βθ t-1 +(1-β)percentile(R t-w:t ,p)

[0109] Among them, θ t is the threshold at time t, β is the smoothing coefficient, R t -w:t is the risk score sequence of the past w days, and p is the percentile. In one embodiment of the present invention, β is set to 0.9, w is set to 30 days, and p is set to 95%. This dynamic adjustment mechanism enables the system to adapt to changes in different seasons and different breeding environments, and improves the accuracy and practicality of early warning.

[0110] The application end 4 of the present invention includes a visualization display module 41, an event analysis module 42 and an early warning management module 43. This modular design not only provides intuitive data display, but also supports in-depth event analysis and flexible system management.

[0111] The visualization display module 41 is used to display the disease risk warning results in the form of charts or text. In a preferred embodiment of the present invention, the module adopts a responsive design and can adapt to different display devices. At the same time, the module also supports interactive visualization of multidimensional data, allowing users to understand the health status of livestock from different angles. For example, users can intuitively see the distribution of environmental stress index (EPI) in different areas through heat maps, and can also observe the changing trend of health indicators of specific livestock through time series diagrams.

[0112] The event analysis module 42 is used to track and analyze warning events. This module implements a graph-based event propagation analysis algorithm, the core idea of ​​which is to construct an event propagation graph and then use graph mining technology to identify key nodes and propagation paths. Specifically, each warning event is regarded as a node in the graph, and the edges between nodes represent the association between events (such as temporal sequence, spatial proximity, etc.). By analyzing the topological structure of this graph, the system can identify the source and main propagation path of the disease outbreak, providing an important basis for the formulation of prevention and control measures.

[0113] The early warning management module 43 is used to set the system alarm threshold and manage user permissions. In one embodiment of the present invention, the module uses a role-based access control (RBAC) model to manage user permissions. The system predefines several roles (such as administrator, veterinarian, breeder, etc.), and each role has different operating permissions. This design not only ensures the security of the system, but also provides a flexible permission management mechanism.

[0114] Preferably, the early warning management module 43 also supports a multi-level early warning mechanism. The system divides early warnings into three levels according to the risk level: reminder, warning and emergency. Each level corresponds to a different processing flow and notification method. For example, for an early warning at the "reminder" level, the system may only display a message on the management interface; while for an early warning at the "emergency" level, the system will immediately send SMS and email notifications to relevant personnel and activate the preset emergency plan.

[0115] Through the above detailed description, it can be seen that the system of the present invention has innovative designs in data analysis, risk warning and application management, etc. These designs fully consider the special needs of livestock health monitoring, can effectively improve the accuracy of monitoring and the timeliness of warning, and are of great significance to improving the management level and economic benefits of the breeding industry.

[0116] In the system of the present invention, data analysis module 2 and early warning module 3 are transmitted through an encrypted communication protocol. This design not only ensures the security of data transmission, but also improves the overall reliability of the system. In a preferred embodiment of the present invention, an encryption scheme based on elliptic curve cryptography (ECC) is adopted. Compared with traditional RSA encryption, ECC can provide the same level of security at a shorter key length, which is more suitable for resource-constrained IoT environments.

[0117] Specifically, the present invention uses the ECDH (Elliptic Curve Diffie-Hellman) key exchange protocol to establish the session key, and then uses the AES (Advanced Encryption Standard) algorithm to perform actual data encryption. The mathematical basis of ECDH is the discrete logarithm problem on the elliptic curve, and its security can be expressed by the following equation:

[0118] Q=dG

[0119] Among them, Q is the public key, d is the private key, and G is the base point on the elliptic curve. Since it is difficult to solve d on the elliptic curve when Q and G are known, this scheme has high security.

[0120] Preferably, the system of the present invention also implements a forward security mechanism. Each communication session uses a newly generated temporary key pair, and even if the key of a certain communication is cracked, it will not affect the security of other communications. This mechanism greatly improves the long-term security of the system and is particularly suitable for livestock health monitoring systems that need to run for a long time.

[0121] The system of the present invention further comprises a data storage module 5 for storing environmental data, livestock physiological data, livestock health characteristics and disease risk warning reports. In one embodiment of the present invention, the data storage module 5 adopts a distributed storage architecture, combining the advantages of relational databases and non-relational databases.

[0122] For structured data (such as environmental parameters, physiological indicators, etc.), the system uses a relational database (such as PostgreSQL) for storage. The table structure of the database has been carefully designed, taking into full consideration the temporal and relevance of the data. This design allows the system to efficiently query and analyze environmental data for a specific time period or a specific area.

[0123] For unstructured or semi-structured data (such as image data, behavior logs, etc.), the system uses document databases (such as MongoDB) for storage. This storage method provides greater flexibility and can easily store and retrieve complex nested data structures. For example, a livestock behavior record may be as follows:

[0124] This storage method enables the system to flexibly record and analyze complex behavioral data, providing a rich information foundation for in-depth health analysis.

[0125] Preferably, the data storage module 5 of the present invention also implements a data tiered storage strategy. Hot data (recently generated data) is stored on a high-speed storage device, such as a solid-state drive (SSD), to ensure fast read and write performance. Cold data (historical data) is stored on a large-capacity mechanical hard drive to reduce storage costs. The system automatically migrates data based on the access frequency and importance of the data to optimize the use of storage resources.

[0126] The improved graph neural network model of the present invention includes adding additional edges to the original graph, and these additional edges are used to connect each node with its most adjacent node. This design significantly improves the ability of the graph model to capture local structural information, and is particularly suitable for processing complex data relationships in livestock health monitoring.

[0127] In a preferred embodiment of the present invention, the construction process of the additional edge is as follows:

[0128] 1. For each node in the graph, calculate its similarity with all other nodes. The similarity can be calculated based on the Euclidean distance, cosine similarity, etc. of the node features.

[0129] 2. Select K most similar nodes for each node (K is an adjustable parameter, and in one embodiment of the present invention, K is set to 5).

[0130] 3. Based on the original graph, add edges connecting each node with its K most similar nodes.

[0131] The construction of this additional edge can be expressed mathematically as follows:

[0132] A′=A+TopK(S,K)

[0133] Among them, A′ is the adjacency matrix after adding additional edges, A is the original adjacency matrix, S is the similarity matrix between nodes, and the TopK operation selects the K largest elements in each row and sets the other elements to zero.

[0134] Preferably, the present invention also introduces the concept of edge weight, that is, the weight of the additional edge is proportional to the similarity between nodes. This can be achieved by modifying the above formula to:

[0135] A′=A+α·TopK(S,K)

[0136] Wherein, α is an adjustable parameter used to control the relative importance of the original edge and the additional edge. In one embodiment of the present invention, α is set to 0.5, which is a good value found in experiments to balance model performance and overfitting risk.

[0137] This improved graph structure enables the model to better capture local similarities in the data, especially when dealing with livestock health data, which often contains important health status information. For example, livestock under similar environmental conditions may show similar health characteristics, and by increasing the connections between these livestock nodes, the model can more effectively utilize this similarity information.

[0138] The present invention also provides a livestock health monitoring and disease early warning method based on environmental data fusion, the method comprising the following steps:

[0139] First, collect environmental data of the livestock breeding space and livestock physiological data. In a preferred embodiment of the present invention, the environmental data is collected once every 5 minutes, including temperature (accuracy ±0.1°C), relative humidity (accuracy ±2%RH) and CO2 concentration (accuracy ±50ppm). The collection of livestock physiological data has different frequencies according to different indicators. For example, body temperature may be measured once an hour, while behavioral data may be recorded continuously.

[0140] Next, the collected environmental data and livestock physiological data are cleaned. The data cleaning process includes removing outliers, filling missing values ​​and data standardization. In the present invention, the improved Z-score method is used to determine the outliers, and the threshold is set to 3.5. For missing values, different filling strategies are used according to the type of data. For example, for environmental data, linear interpolation may be used; for discrete behavioral data, the previous value filling method may be used.

[0141] Then, livestock health features are constructed based on the cleaned data. The feature construction process of the present invention not only considers the statistical characteristics of a single indicator, but also introduces the correlation and time series characteristics between multiple indicators. For example, the temperature and humidity coupling index (Temperature-Humidity Index, THI) can be calculated, and its calculation formula is:

[0142] THI=(1.8×T+32)-(0.55-0.0055×RH)×(1.8×T-26)

[0143] Where T is temperature (℃) and RH is relative humidity (96). THI is an important indicator for assessing heat stress levels and has an important impact on livestock health. Next, the improved graph neural network model was used to analyze livestock health characteristics.

[0144] Based on the analysis results, sequence labeling is performed using a conditional random field model. In the present invention, the potential function of the conditional random field is defined as:

[0145]

[0146] Among them, yi and j is the label of the adjacent time, x is the observation sequence, f k is the characteristic function, λ k is the corresponding weight. The design of the characteristic function fully considers the temporal characteristics of livestock health status and the influence of environmental factors.

[0147] Finally, the Bayesian inference framework is used to generate a warning report on the disease risk of livestock. The present invention uses a variational inference method to approximate the posterior distribution, and its objective function (ELBO) is:

[0148]

[0149] Among them, q(y) is the approximate posterior distribution, p(y|x) is the true posterior distribution, p(y) is the prior distribution, and KL represents the KL divergence. By maximizing ELBO, the system is able to achieve a good balance between computational efficiency and inference accuracy.

[0150] The generated disease risk warning report contains not only the risk level, but also the possible causes of the risk and recommended interventions. For example, if the system detects that the THI value in a certain area is continuously above 78 (which is generally considered the threshold at which cows begin to experience heat stress), the warning report may recommend increasing ventilation or cooling measures in that area.

[0151] Through the above steps, the method of the present invention can comprehensively and accurately monitor the health status of livestock, timely warn of potential disease risks, and provide powerful decision support for breeding management. This method based on environmental data fusion not only improves the accuracy of early warning, but also provides a scientific basis for the optimization of breeding environment, and has important practical application value.

[0152] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto; any person familiar with the art who, within the scope disclosed by the present invention, makes substitutions or changes based on the scheme and improved concepts of the present invention shall be covered within the protection scope of the present invention.

Claims

1. Livestock health monitoring and disease early warning system based on environmental data fusion, characterized by: include: Data acquisition module for: Collect environmental data of livestock breeding space and livestock physiological data; transmitting the environmental data and the livestock physiological data to a data analysis module; A data analysis module is connected to the data acquisition module for: Receiving environmental data and livestock physiological data sent by the data acquisition module; constructing a livestock health signature based on the environmental data and the livestock physiological data; Analyzing the livestock health characteristics using an improved graph neural network model; The early warning module is connected to the data analysis module for: Receiving the analysis result of the data analysis module; Based on the analysis results, a livestock disease risk warning report is generated; The application end is connected to the early warning module for: Receiving the disease risk warning report generated by the warning module; The disease risk warning report is displayed to the user.

2. The system according to claim 1, characterized in that The data acquisition module comprises: Temperature and humidity sensors are used to collect temperature and humidity data of livestock breeding spaces; CO2 sensor, used to collect CO2 concentration data in livestock breeding spaces; Cameras to collect data on livestock’s body temperature, shape, posture, and behavior; A data converter is connected to the temperature and humidity sensor, the CO2 sensor and the camera, and is used to convert the collected data into a processable format.

3. The system according to claim 1, characterized in that The data analysis module includes: A data cleaning unit, used for cleaning the environmental data and the livestock physiological data; A feature construction unit, used to construct livestock health features based on the cleaned data; A graph neural network analysis unit is used to analyze the health characteristics of the livestock using an improved graph neural network model.

4. The system according to claim 3, characterized in that The graph neural network analysis unit comprises: A graph structure building module, used to build a graph structure based on the livestock health characteristics; A multi-layer graph self-attention network module, used for extracting features from the graph structure; A global-local graph convolutional neural network module is used to combine global information and local features for analysis.

5. The system according to claim 1, characterized in that The early warning module comprises: A conditional random field model is used to perform sequence labeling on the analysis results of the data analysis module; A Bayesian inference unit, used for inferring disease risk based on the output results of the conditional random field model; The threshold adjustment unit is used to dynamically adjust the disease risk warning threshold based on historical data.

6. The system according to claim 1, characterized in that The application end includes: Visual display module, used to display disease risk warning results in the form of charts or text; Event analysis module, used to track and analyze warning events; The early warning management module is used to set system alarm thresholds and manage user permissions.

7. The system according to claim 1, characterized in that The data analysis module and the early warning module transmit data via an encrypted communication protocol.

8. The system according to claim 1, characterized in that The system also includes a data storage module for storing the environmental data, the livestock physiological data, the livestock health characteristics and the disease risk warning report.

9. The system according to claim 1, characterized in that The improved graph neural network model includes adding additional edges on the basis of the original graph, and the additional edges are used to connect each node and its most adjacent node.

10. A livestock health monitoring and disease early warning method based on environmental data fusion using the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Collect environmental data of livestock breeding space and livestock physiological data; Cleaning the environmental data and the livestock physiological data; construct livestock health signatures based on the cleaned data; Analyzing the livestock health characteristics using an improved graph neural network model; Based on the analysis results, sequence labeling is performed using a conditional random field model; Generate early warning reports on livestock disease risks using a Bayesian inference framework; The disease risk warning report is displayed to the user.

Images