Cow breeding early warning system and method based on multi-feature fusion analysis

Abstract

The invention relates to the technical field of cow breeding early warning, and discloses a cow breeding early warning system and method based on multi-feature fusion analysis, and the system comprises a data collection module, a feature fusion analysis module, an abnormity early warning module, a decision and suggestion module, and a breeding data display module. The feature fusion analysis module is used for analyzing the health condition, the production performance and the environmental comfort of the dairy cow based on the physiological data, the feeding data and the colony house environment data of the dairy cow; and the decision suggestion module is used for making a dairy cow breeding optimization suggestion according to the feature fusion analysis result, and the dairy cow breeding optimization suggestion comprises optimization of feed distribution and dynamic division of activity areas. The system effectively solves the problems of health management lagging, resource waste, difficulty in recognition of environmental hidden dangers, unreasonable region division and the like in traditional cow breeding, remarkably improves the intelligent level of pasture management and breeding efficiency, and has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of dairy cattle breeding early warning, and in particular to a dairy cattle breeding early warning system and method based on multi-feature fusion analysis. Background Art

[0002] With the rapid development of modern animal husbandry and the continuous innovation of technology, dairy farming is gradually moving towards intensive, large-scale and intelligent development. This transformation has significantly improved the efficiency of dairy farming and made management more systematic, but it has also brought many new challenges. In the process of large-scale dairy farming, the difficulties faced by ranch management have become more prominent, mainly in terms of dairy cow health management, production efficiency optimization, resource utilization improvement, and pen environment management. If these problems are not effectively solved, they will directly affect the overall breeding benefits and sustainable development of the ranch.

[0003] First, insufficient health monitoring and disease warning for dairy cows is one of the main challenges of modern ranch management. As a high-yield economic animal, the health status of dairy cows directly affects milk production, life cycle and breeding costs. Under the traditional breeding model, ranches usually rely on manual observation or simple physiological monitoring equipment for health management. However, this method has major limitations: manual observation is highly subjective, has poor real-time performance, and is difficult to detect potential health problems in a timely manner; the monitoring dimensions of simple sensors are limited and cannot capture subtle changes in the physiological indicators of dairy cows, resulting in insufficient early warning capabilities for diseases. Once an infectious disease occurs in a dairy cow herd, it is often impossible to quickly and effectively contain it, causing large-scale dairy cow health problems, affecting milk production and ranch benefits.

[0004] Secondly, production performance management and identification of inefficient cows are difficult, resulting in waste of resources and overall low breeding efficiency. Feed costs in modern ranches account for a major part of breeding costs, but feeding management often lacks refinement and data support, making it difficult to dynamically adjust feed allocation according to individual differences and production performance of dairy cows. At the same time, due to the lack of effective production data analysis methods, it is difficult for ranch managers to accurately identify inefficient cows, that is, those with low milk production but high resource consumption. This situation leads to a waste of feed resources, and the overall milk production efficiency cannot be optimized, which increases the operating costs of the ranch in the long run. In addition, the life cycle management of dairy cows is not in place, and an efficient milk production cycle optimization and elimination mechanism has not been achieved, further limiting the improvement of production performance.

[0005] Furthermore, environmental factors in pens affect the comfort and health of dairy cows, which is also a key issue that needs to be addressed in large-scale dairy farming. The physiological state and behavioral habits of dairy cows are highly sensitive to the environment. Adverse factors such as temperature and humidity changes, excessive ammonia concentrations, poor air quality, and slippery ground can have a significant impact on the health and milk production of dairy cows. For example, a hot and humid environment can easily cause heat stress in dairy cows, which in turn affects their feed intake and milk production; excessive ammonia concentrations can easily cause respiratory diseases in dairy cows. Traditional environmental management methods lack accurate monitoring and analysis tools, and cannot achieve dynamic adjustment of the pen environment and identification of risks and hidden dangers. In addition, there is a complex interactive relationship between the activity trajectory and behavior distribution of dairy cows and the layout of the pen, but ranch managers often find it difficult to intuitively grasp these relationships, making it impossible to optimize the pen environment in a timely manner to ensure the comfort of dairy cows.

[0006] In addition, the functional area division of the ranch lacks a dynamic optimization mechanism, which further aggravates the problems of uneven resource utilization and inefficient management. In traditional ranches, the division of functional areas such as rest areas, feeding areas, and drinking water areas is mostly statically planned based on experience, lacking the ability to adjust dynamically. With the expansion of the size of the dairy cow herd and the differentiation of individual needs, static area division often leads to regional crowding and uneven resource utilization. For example, there may be excessive gathering of dairy cows in the feed trough area, resulting in resource competition and stress reactions between individuals, while some areas may have idle resources and low overall utilization. In addition, the activity habits and behavioral patterns of dairy cows will also change with time and environmental changes. Fixed functional areas are difficult to meet the dynamic needs of dairy cows, further affecting the management efficiency of the ranch.

[0007] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0008] 1. Technical issues to be resolved

[0009] In view of the shortcomings of the prior art, the present invention provides a dairy cattle breeding early warning system and method based on multi-feature fusion analysis, which can effectively solve the problems of lagging health management, waste of resources, difficulty in identifying environmental hazards and unreasonable regional division in traditional dairy cattle breeding, significantly improve the intelligence level of pasture management and breeding efficiency, have the advantage of broad application prospects, and thus solve the problems in the background technology.

[0010] (II) Technical solution

[0011] In order to achieve the above-mentioned advantages of effectively solving the problems of lagging health management, waste of resources, difficulty in identifying environmental hazards and unreasonable regional division in traditional dairy farming, significantly improving the intelligent level of pasture management and breeding efficiency, and having broad application prospects, the specific technical solutions adopted by the present invention are as follows:

[0012] According to one aspect of the present invention, there is provided a dairy cattle breeding early warning system based on multi-feature fusion analysis, including a data acquisition module, a feature fusion analysis module, an abnormal warning module, a decision suggestion module and a breeding data display module;

[0013] The data acquisition module is used to obtain the physiological data, feeding data and pen environment data of dairy cows;

[0014] The feature fusion analysis module is used to analyze the health status, production performance and environmental comfort of dairy cows based on the physiological data, feeding data and pen environment data of dairy cows;

[0015] The abnormal warning module is used to determine the abnormal situation of dairy cattle breeding and the corresponding warning level according to the feature fusion analysis results, and to issue abnormal warning according to the warning level;

[0016] The decision-making suggestion module is used to formulate dairy cattle breeding optimization suggestions based on the feature fusion analysis results, and the dairy cattle breeding optimization suggestions include optimizing feed distribution and dynamically dividing activity areas;

[0017] The farming data display module is used to present the collected data, analysis results, early warning signals and decision suggestions to farming users in an intuitive manner.

[0018] Furthermore, the feature fusion analysis module includes a data preprocessing module, a health status analysis module, a production performance analysis module and an environmental comfort analysis module;

[0019] The data preprocessing module is used to clean, standardize and extract features from the acquired physiological data, feeding data and pen environment data of dairy cows;

[0020] The health status analysis module is used to analyze the objective health status and potential health status of the dairy cows based on the physiological data, feeding data and pen environment data of the dairy cows;

[0021] The production performance analysis module is used to predict and analyze the average daily milk production, cumulative milk production, feed utilization efficiency and overall breeding efficiency of dairy cows, and identify inefficient dairy cows based on the cumulative milk production and overall breeding efficiency;

[0022] The environmental comfort analysis module is used to analyze the comfort of the dairy cow breeding environment, environmental hazards and the impact of the environment on dairy cow breeding based on the physiological data, feeding data and pen environmental data of the dairy cows.

[0023] Furthermore, the health status analysis module includes an objective health status analysis module and a potential health status analysis module;

[0024] The objective health status determination module is used to determine the standard physiological data threshold according to the breed and age stage of the dairy cow, and determine the health status of the dairy cow by comparing the dairy cow's physiological data with the standard physiological data threshold;

[0025] The potential health status analysis module is used to construct a dynamic group relationship graph based on the location, interaction frequency and behavioral similarity between cows, extract implicit relationship features between cows using a graph neural network algorithm based on behavioral patterns, and determine potential abnormal cows in the cow group based on a graph embedded isolation forest algorithm.

[0026] Furthermore, the potential health status analysis module constructs a dynamic group relationship graph according to the positions, interaction frequencies and behavior similarities between cows, extracts implicit relationship features between cows using a graph neural network algorithm based on behavior patterns, and determines potential abnormal cows in the cow group based on a graph embedded isolation forest algorithm, including:

[0027] Obtain location data, interaction frequency, and behavior similarity data of dairy cows in the breeding area, and pre-process the acquired data;

[0028] Based on the position similarity, interaction frequency and behavior similarity between cows, the comprehensive edge weight is calculated, and the dynamic group relationship graph is constructed by combining the position, interaction frequency and behavior similarity between cows. The nodes in the dynamic group relationship graph are represented by cows, and the edges represent the relationship between cows.

[0029] The graph neural network is used to embed nodes in the constructed dynamic group relationship graph, compress the high-dimensional features of each cow into a low-dimensional space, and extract the implicit relationship features between cows;

[0030] The node embedding vector output by the graph neural network is input into the isolation forest model to calculate the anomaly score of each node during the isolation process; based on the comparison result between the anomaly score and the preset anomaly score threshold, the potential abnormal cows in the dairy cow group are determined;

[0031] Among them, the calculation formula of the comprehensive edge weight is:

[0032] W(i,j)=α·W pos (i,j)+β·W int (i,j)+γ·W beh (i,j)

[0033]

[0034] In the process of embedding nodes in the constructed dynamic group relationship graph using graph neural network, the features of node i in the l+1th layer aggregating its neighbor node j are:

[0035]

[0036] The calculation formula for the anomaly score is:

[0037]

[0038] Where W(i,j) represents the comprehensive edge weight, W pos (i,j) represents the position similarity weight of cows i and j, W int (i,j) represents the interaction frequency weight between cows i and j, W beh (i, j) represents the behavioral similarity weights of cows i and j, α, β, and γ represent the position similarity weight parameter, interaction frequency weight parameter, and behavioral similarity weight parameter, respectively. ij represents the Euclidean distance between cows i and j, σ represents the smoothing parameter of the Gaussian kernel, and f ij Indicates the number of contacts, max(f ij ) represents the maximum number of contacts, B i represents the behavior vector of cow i, B j represents the behavior vector of cow j, ||B i || represents the behavior vector B i The Euclidean norm of ||B j || represents the behavior vector B j The Euclidean norm of , represents the feature vector of node i in the l+1 layer, μ represents the activation function, represents the set of neighbor nodes of node i, AGG(·) represents the aggregation function, represents the feature vector of node j in layer l, s(Z i ,n) represents the abnormal score of node i, Z i represents the node embedding vector, n is the total number of samples, E(h(Z i )) represents the node embedding vector Z i The average path length in the isolation forest, c(n) represents the average path length of a binary tree, T represents the total number of trees in the forest, and h t (Z i ) represents the node embedding vector Z i The path length in the tth tree, L represents the number of layers in the graph neural network.

[0039] Furthermore, the production performance analysis module includes a milk production analysis module, a feed utilization efficiency analysis module and a dairy cow life cycle management module;

[0040] The milk production analysis module is used to use a machine learning model to combine the physiological data, age stage and historical milk production of dairy cows to predict the average daily milk production of each dairy cow during its life cycle, and calculate the cumulative milk production based on the average daily milk production;

[0041] The feed utilization efficiency analysis module is used to analyze the feed utilization efficiency of each dairy cow based on the feed intake and average daily milk production of the dairy cow;

[0042] The dairy cow life cycle management module is used to evaluate the breeding efficiency according to the cumulative milk production and feed conversion rate of each dairy cow, and identify inefficient dairy cows based on the breeding efficiency.

[0043] Furthermore, the dairy cow life cycle management module includes: evaluating the breeding efficiency according to the cumulative milk production and feed conversion rate of each dairy cow, and identifying inefficient dairy cows based on the breeding efficiency:

[0044] Obtain the cumulative milk production and feed conversion rate of each cow, and calculate the comprehensive efficiency score of each cow based on the cumulative milk production and feed conversion rate of each cow; determine whether the comprehensive efficiency score of each cow meets the preset scoring standard. If yes, continue to breed the cow; if not, mark the cow as an inefficient cow;

[0045] The calculation formula for the breeding efficiency of each dairy cow is:

[0046]

[0047] In the formula, S i represents the comprehensive efficiency score of the i-th cow, P i represents the feed conversion rate of the i-th dairy cow, C i represents the cumulative milk production of the i-th cow, C max represents the maximum cumulative milk production in the dairy cow herd, ω1 and ω2 represent the weight of feed conversion rate, cumulative milk production and the weight of maximum cumulative milk production, respectively.

[0048] Furthermore, the environmental comfort analysis module includes an environmental data analysis module and an environmental hazard analysis module;

[0049] The environmental data analysis module is used to determine the comfort level of the dairy cow breeding environment based on the comparison results between the environmental data of the breeding area and the preset standard environmental threshold;

[0050] The environmental hazard analysis module is used to analyze potential environmental hazard areas using the pen distribution data and the cow movement trajectory.

[0051] Furthermore, the environmental hazard analysis module includes the following steps when analyzing potential environmental hazard areas using the pen distribution data and the cow movement trajectory:

[0052] Obtain the spatial layout data of the pasture pens, the movement trajectory data of the cows and the environmental index data, and perform exception processing, data synchronization and spatial grid division on the acquired data;

[0053] Calculate the residence time of each cow in each grid area, integrate the residence time of all cows, generate a heat map of the entire pen area, and draw the movement paths of all cows to identify areas with high path overlap;

[0054] Determine regional indicators of potential environmental hazards based on pen layout and cow behavior, including high residence time areas, abnormally low movement areas, and high track intersection areas;

[0055] Spatial matching of the pen environment data and the cow trajectory data was performed to identify areas with abnormal temperature and humidity, areas with poor air quality, and areas with slippery weather.

[0056] Based on the results of spatial matching of regional indicators of potential environmental hazards, pen environment data and cow trajectory data, the hazard areas are marked on the pasture map, where the hazard areas include high-risk areas, medium-risk areas and low-risk areas.

[0057] Further, the decision-making suggestion module includes a feed distribution optimization module and an activity area division module;

[0058] The feed distribution optimization module is used to intelligently adjust the feed distribution amount according to the needs, age stage and health status of the cows;

[0059] The activity area division module is used to dynamically adjust the functional division of different areas in the pasture by using a swarm intelligence algorithm in combination with the activity data, demand data and pen environment data of the dairy cows;

[0060] The activity area division module uses a swarm intelligence algorithm to dynamically adjust the functional division of different areas in the pasture in combination with the activity data, demand data and pen environment data of the dairy cows, including:

[0061] Obtain cow activity data, demand data and pen environment data, and analyze the acquired data to explore the cow's activity patterns and demand preferences;

[0062] According to the average distance of cows to the target area, the regional crowding and resource utilization, the objective function of dynamic functional division is determined, and the pasture is divided into several functional areas in the initial state, and the location and functional label of the current area are recorded;

[0063] The expression of the objective function of dynamic function partitioning is:

[0064] min F=λ1·D dist +λ2·D cong -λ3·S util

[0065] In the formula, F represents the comprehensive target value, D distrepresents the average distance of the cows to reach the target area, D cong represents the regional congestion, S util represents resource utilization, λ1, λ2, and λ3 represent the weight of the average distance of the cows to the target area, the weight of the regional crowding, and the weight of the resource utilization, respectively;

[0066] The ant colony algorithm is used to iteratively optimize the regional division, and the functional division of different areas of the pasture is adjusted according to the optimization results to obtain the best functional regional division result.

[0067] According to another aspect of the present invention, a dairy cattle breeding early warning method based on multi-feature fusion analysis is provided, the method comprising the following steps:

[0068] S1. Obtaining the physiological data, feeding data and pen environment data of dairy cows;

[0069] S2. Analyze the health status, production performance and environmental comfort of dairy cows based on their physiological data, feeding data and pen environment data;

[0070] S3. Determine the abnormal situation of dairy cattle breeding and the corresponding warning level according to the feature fusion analysis results, and issue abnormal warning according to the warning level;

[0071] S4. Formulate optimization suggestions for dairy farming based on the results of feature fusion analysis, and the optimization suggestions for dairy farming include optimizing feed distribution and dynamically dividing activity areas;

[0072] S5. Present the collected data, analysis results, warning signals and decision-making suggestions to the breeding users in an intuitive manner.

[0073] (III) Beneficial effects

[0074] Compared with the prior art, the present invention provides a dairy cattle breeding early warning system and method based on multi-feature fusion analysis, which has the following beneficial effects:

[0075] (1) The present invention constructs an intelligent dairy cattle breeding early warning system integrating data collection, health assessment, production performance analysis, environmental comfort assessment and dynamic decision-making suggestions through multi-feature data fusion, machine learning and swarm intelligence algorithm. It can not only realize real-time monitoring of the health status of dairy cows and early abnormal warning, as well as accurate assessment of production performance and optimal resource allocation, but also realize intelligent identification of environmental hazards in pens and dynamic functional area optimization, effectively solving the problems of lagging health management, waste of resources, difficulty in identifying environmental hazards and unreasonable regional division in traditional dairy cattle breeding, significantly improving the intelligence level of pasture management and breeding efficiency, and has broad application prospects.

[0076] (2) The present invention can not only analyze the health status of dairy cows by collecting physiological data, feeding data and pen environment data of dairy cows, but also use graph neural network and isolation forest algorithm based on behavior patterns to combine the location, interaction frequency and behavior similarity of dairy cows to construct a dynamic group relationship diagram, accurately identify potential abnormal dairy cows, so as to monitor the health status of dairy cows in real time, warn of health abnormalities in advance, reduce the risk of disease transmission, and improve breeding efficiency.

[0077] (3) The present invention can not only evaluate the production performance of dairy cows based on their average daily milk production and feed utilization efficiency and identify inefficient dairy cows, but also predict the milk production performance of dairy cows during their life cycle through machine learning models, and provide data support to optimize feed distribution, thereby improving the accuracy of resource allocation, reducing feed waste, and maximizing breeding efficiency.

[0078] (4) The present invention can not only combine the activity data, demand data and pen environment data of dairy cows and use the swarm intelligence algorithm to dynamically adjust the functional division of different areas in the pasture, but also reduce the movement cost of dairy cows, alleviate crowding, and improve the utilization rate of resources (such as feed troughs, water troughs, and rest areas) through optimized functional division.

[0079] (5) The present invention can not only identify high residence time areas, abnormal low movement areas and high trajectory intersection areas through the fusion analysis of cow movement trajectories and environmental indicator data, and locate potential environmental hazards, but also mark and warn of abnormal temperature and humidity, poor air quality and slippery areas, thereby ensuring the comfort of the pen environment and reducing the stress response and health risks of cows.

[0080] (6) The present invention can not only intuitively display data collection results, health status analysis, environmental comfort assessment and early warning signals through the breeding data display module, but also provide intelligent decision-making suggestions, including feed distribution optimization and dynamic activity area division, to help breeding managers make scientific decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0082] Figure 1 is a structural block diagram of a dairy cattle breeding early warning system based on multi-feature fusion analysis according to an embodiment of the present invention;

[0083] Figure 2It is a flow chart of a dairy cattle breeding early warning method based on multi-feature fusion analysis according to an embodiment of the present invention.

[0084] In the figure:

[0085] 1. Data collection module; 2. Feature fusion analysis module; 3. Abnormal warning module; 4. Decision-making recommendation module; 5. Breeding data display module. DETAILED DESCRIPTION

[0086] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0087] According to an embodiment of the present invention, a dairy cattle breeding early warning system and method based on multi-feature fusion analysis are provided.

[0088] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a dairy cattle breeding early warning system based on multi-feature fusion analysis is provided, including a data acquisition module 1, a feature fusion analysis module 2, an abnormal warning module 3, a decision suggestion module 4 and a breeding data display module 5;

[0089] The data acquisition module 1 is used to obtain the physiological data, feeding data and pen environment data of dairy cows, and specifically includes:

[0090] Physiological data collection: Physiological data such as body temperature, heart rate, activity, sound, rumination behavior, and milk production are very critical, especially through wearable sensors (such as smart ear tags and collars) to monitor the status of dairy cows in real time. The technology for collecting this data is mature and can be transmitted through wireless sensor networks.

[0091] Feeding data collection: The collection of data such as feed type, intake, feeding time, etc. is particularly important for subsequent production performance analysis, especially for the analysis of feed utilization efficiency;

[0092] Collection of pen environmental data: Pen temperature and humidity, ammonia concentration, light intensity and other environmental parameters are important factors affecting the health and production performance of dairy cows. Environmental data collection can be achieved through temperature and humidity sensors, gas detectors, etc., which are suitable for real-time monitoring;

[0093] The feature fusion analysis module 2 is used to analyze the health status, production performance and environmental comfort of dairy cows based on the physiological data, feeding data and pen environmental data of dairy cows;

[0094] Specifically, the feature fusion analysis module 2 includes a data preprocessing module, a health status analysis module, a production performance analysis module and an environmental comfort analysis module;

[0095] The data preprocessing module is used to clean, standardize and extract features from the acquired physiological data, feeding data and pen environment data of dairy cows, and specifically includes:

[0096] Data cleaning: Identify and correct data anomalies caused by sensor noise or failure;

[0097] Data standardization: unify the formats of different data sources and eliminate dimensional differences;

[0098] Feature extraction: extract key indicators from multidimensional data to provide input for subsequent models;

[0099] The health status analysis module is used to analyze the objective health status and potential health status of the dairy cow based on the physiological data, feeding data and pen environment data of the dairy cow; the health status analysis module includes an objective health status analysis module and a potential health status analysis module;

[0100] The objective health status determination module is used to determine the standard physiological data threshold according to the breed and age stage of the dairy cow, and determine the health status of the dairy cow by comparing the dairy cow's physiological data with the standard physiological data threshold;

[0101] The potential health status analysis module is used to construct a dynamic group relationship graph based on the location, interaction frequency and behavior similarity between cows, extract implicit relationship features between cows using a graph neural network algorithm based on behavior patterns, and determine potential abnormal cows in the cow group based on a graph embedded isolation forest algorithm, specifically including:

[0102] 1) Obtain the location data, interaction frequency and behavior similarity data of dairy cows in the breeding area, and pre-process the acquired data, including:

[0103] Location data: Get the real-time coordinate data of the cows through GPS, RFID or UWB (ultra-wideband) positioning system;

[0104] Interaction frequency: Count the number of contacts between cows based on a time window to measure the frequency of their social interactions;

[0105] Behavioral data: Use wearable devices (such as smart ear tags and collars) to collect cow behavioral data (such as rumination, activity level, standing time, etc.) to form the cow's behavioral feature vector;

[0106] Data preprocessing includes:

[0107] Standardization: normalize data of different dimensions (position, interaction frequency, behavioral characteristics);

[0108] Noise removal: remove outliers from positioning data;

[0109] Time window division: segment the data into fixed time intervals to generate dynamic data streams;

[0110] 2) Based on the position similarity, interaction frequency and behavior similarity among cows, the comprehensive edge weight is calculated, and a dynamic group relationship graph is constructed by combining the position, interaction frequency and behavior similarity among cows;

[0111] Specifically, the cows are regarded as nodes in the graph, and the relationships between cows (position, interaction frequency, and behavior similarity) are regarded as edges, and a dynamic group relationship graph is constructed, in which each node represents a cow, the node features include behavior vectors and position information, and the edge weights between nodes represent the strength of the relationship between cows;

[0112] The calculation formula of comprehensive edge weight is:

[0113] W(i,j)=α·W pos (i,j)+β·W int (i,j)+γ·W beh (i,j)

[0114]

[0115] Where W(i,j) represents the comprehensive edge weight, W pos (i,j) represents the position similarity weight of cows i and j, W int (i,j) represents the interaction frequency weight between cows i and j, W beh (i, j) represents the behavioral similarity weights of cows i and j, α, β, and γ represent the position similarity weight parameter, the interaction frequency weight parameter, and the behavioral similarity weight parameter, respectively, which can be adjusted according to actual conditions. In this embodiment, α=0.5 (behavioral similarity is more important), β=0.3 (spatial dependence has a moderate impact), and γ=0.2 (social interaction has a relatively low impact). ij represents the Euclidean distance between cows i and j, σ represents the smoothing parameter of the Gaussian kernel, which controls the similarity decay speed, and f ij Indicates the number of contacts, max(f ij ) represents the maximum number of contacts, Bi represents the behavior vector of cow i, B j represents the behavior vector of cow j, ||B i || represents the behavior vector B i The Euclidean norm of ||B j || represents the behavior vector B j The Euclidean norm of ;

[0116] 3) Use graph neural networks to embed nodes in the constructed dynamic group relationship graph, compress the high-dimensional features of each cow into a low-dimensional space, and extract the implicit relationship features between cows;

[0117] Specifically, the graph neural network learns the embedded representation of nodes through the information propagation mechanism and extracts the implicit relationship features between cows. The embedding steps include:

[0118] Initialize node features: The initial feature of each node is its behavior vector

[0119] Information propagation and aggregation: At layer l+1, node i aggregates the features of its neighbor node j as follows:

[0120]

[0121] In the formula, represents the feature vector of node i in the l+1 layer, μ represents the activation function, such as ReLU, represents the set of neighbor nodes of node i, AGG(·) represents the aggregation function, such as average or weighted sum, Represents the feature vector of node j in layer l;

[0122] Node embedding result: After L layers of propagation, the final embedding representation of the node is obtained

[0123] 4) Input the node embedding vector output by the graph neural network into the isolation forest model to calculate the abnormality score of each node in the isolation process; determine the potential abnormal cows in the dairy cow group based on the comparison result between the abnormality score and the preset abnormality score threshold;

[0124] Specifically, embed the node vector Z output by the graph neural network i Input to the Isolation Forest algorithm to identify potential abnormal cows; the Isolation Forest algorithm uses random trees to split data and quickly "isolate" abnormal data:

[0125] Build an isolation tree: randomly select a feature and split point, embedding vector Z i Constantly dividing;

[0126] Path length calculation: Abnormal data is usually segmented earlier and the path length is shorter;

[0127] Anomaly score calculation:

[0128]

[0129] In the formula, s(Z i ,n) represents the abnormal score of node i, Z i represents the node embedding vector, n is the total number of samples, E(h(Z i )) represents the node embedding vector Z i The average path length in the isolation forest, c(n) represents the average path length of a binary tree, T represents the total number of trees in the forest, and h t (Z i ) represents the node embedding vector Z i The path length in the tth tree, L represents the number of layers in the graph neural network;

[0130] Abnormal node detection: Set the abnormal score threshold θ, when s(Z i ,n)>θ, node i is marked as a potential abnormal cow.

[0131] The production performance analysis module is used to predict and analyze the average daily milk production, cumulative milk production, feed utilization efficiency and overall breeding efficiency of dairy cows, and identify inefficient dairy cows based on the cumulative milk production and overall breeding efficiency;

[0132] Specifically, the production performance analysis module includes a milk production analysis module, a feed utilization efficiency analysis module and a dairy cow life cycle management module;

[0133] The milk production analysis module is used to use a machine learning model to combine the physiological data, age stage and historical milk production of dairy cows to predict the average daily milk production of each dairy cow during its life cycle, and calculate the cumulative milk production based on the average daily milk production;

[0134] Specifically, the machine learning model is combined with the physiological data, age stage and historical milk production of dairy cows to predict the average daily milk production of each dairy cow during its life cycle, and the cumulative milk production is calculated based on the average daily milk production. The following steps can be followed:

[0135] 1) Data collection and preprocessing:

[0136] Physiological data of dairy cows: weight, body temperature, mammary gland health, feed intake, blood indicators (such as lactic acid, ketone bodies, calcium-phosphorus ratio, etc.); genetic data (such as whether there are high-yield genes or certain disease genes, etc.);

[0137] Cow age stage data: cow age (such as age at first calving, days after calving, etc.), fertility history (such as pregnancy cycle, delivery time, etc.);

[0138] Historical milk production data: the milk production of each cow in the past several periods (such as monthly or annually);

[0139] Data preprocessing:

[0140] Clean data: handle missing values ​​and outliers to ensure data quality;

[0141] Feature engineering: Generate the age and lactation stage of the cow (such as dry period, early lactation period, stable lactation period, late lactation period, etc.); standardize or normalize physiological data (such as weight, body temperature); use the lagged value of historical milk production as input features, such as the average milk production in the past 7 days or the past month;

[0142] Time series data processing: If the data is recorded on a daily, weekly or monthly basis, time series decomposition methods can be used to separate trends, seasonality and residuals;

[0143] 2) Feature selection and modeling:

[0144] Feature selection: Based on the collected data, select features that can affect the milk production of dairy cows. Generally speaking, the following features are closely related to milk production: physiological data (weight, body temperature, mammary gland health, feed intake, etc. of dairy cows), age stage (the age of dairy cows, such as the age of first milk production, the number of days after calving, etc., and various stages in the life cycle), historical milk production (the past milk production records of dairy cows, including periodic changes);

[0145] Modeling steps:

[0146] Prepare training data: organize the cow's physiological data, historical milk production, age stage and other data into a feature matrix and target variable (average daily milk production);

[0147] Select a model: Select a suitable machine learning model (for example, random forest regression, support vector regression, etc.); divide the data into training set and test set, and perform model training;

[0148] Training and evaluation: Use the training set to train the model and evaluate the model's performance; use cross-validation (such as K-fold) to reduce the risk of overfitting; use evaluation indicators such as mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) to evaluate the model effect;

[0149] Predicting average daily milk production: Once the model is trained, it can be used to predict the average daily milk production of each cow at different points in its life cycle; each prediction can be based on the cow’s current physiological data, age stage, and historical milk production;

[0150] Calculate cumulative milk production: Use the trained model to predict the average daily milk production of each cow at different stages of its life cycle. For example, predict the average daily milk production at each time point from first calving to the end of lactation. For long-term predictions (such as the entire life cycle), the average daily milk production of each stage (such as each month) can be accumulated to obtain the monthly cumulative milk production. In actual applications, as time goes by, the model can be continuously updated based on new data (such as new milk production records, changes in health status, etc.) for periodic re-prediction.

[0151] The feed utilization efficiency analysis module is used to analyze the feed utilization efficiency of each dairy cow according to the feed intake and average daily milk production of the dairy cow. In this embodiment, the feed utilization efficiency of each dairy cow is calculated by the ratio of the average daily milk production (numerator) of the dairy cow to the average daily feed consumption (denominator) of the dairy cow;

[0152] The dairy cow life cycle management module is used to evaluate the breeding efficiency according to the cumulative milk production and feed conversion rate of each dairy cow, and identify inefficient dairy cows based on the breeding efficiency, specifically including:

[0153] The cumulative milk production and feed conversion rate of each cow are obtained, and the comprehensive efficiency score of each cow is calculated according to the cumulative milk production and feed conversion rate of each cow; it is determined whether the comprehensive efficiency score of each cow meets the preset scoring standard (in this embodiment, the standard for eliminating inefficient cows can be preferably the 25% quantile, or the lowest 10% or 20% of cows are eliminated). If yes, the cow is continued to be raised, if not, the cow is marked as an inefficient cow;

[0154] The calculation formula for the breeding efficiency of each dairy cow is:

[0155]

[0156] In the formula, S i represents the comprehensive efficiency score of the i-th cow, P i represents the feed conversion rate of the i-th dairy cow, C i represents the cumulative milk production of the i-th cow, C max represents the maximum cumulative milk production in the dairy cow herd, ω1 and ω2 represent the weight of feed conversion rate, the weight of cumulative milk production and the weight of maximum cumulative milk production, and in this embodiment, preferably ω1=0.6, ω2=0.4;

[0157] In addition, the present embodiment also includes regularly (such as monthly) re-evaluating the average daily milk production, feed conversion rate and cumulative milk production of dairy cows, updating the overall breeding efficiency and individual efficiency rankings; drawing distribution histograms of feed conversion rate and cumulative milk production, identifying inefficient individuals, and comparing the efficiency differences between efficient and inefficient dairy cows to assist management decisions; outputting the identification results of inefficient dairy cows as a report, providing specific data support, analyzing inefficient dairy cows, determining whether the inefficiency is caused by health, age, genetics and other factors, and taking corresponding measures (such as elimination or improvement of feeding methods); finally, based on the average daily milk production, feed conversion rate, cumulative milk production and overall breeding efficiency, the overall breeding efficiency level is evaluated, inefficient dairy cows are identified, and managers are assisted in making scientific decisions.

[0158] The environmental comfort analysis module is used to analyze the comfort of the dairy cow breeding environment, environmental hazards and the impact of the environment on dairy cow breeding based on the physiological data, feeding data and pen environmental data of dairy cows;

[0159] Specifically, the environmental comfort analysis module includes an environmental data analysis module and an environmental hazard analysis module;

[0160] The environmental data analysis module is used to determine the comfort level of the dairy cow breeding environment based on the comparison results between the environmental data of the breeding area and the preset standard environmental threshold;

[0161] The environmental hazard analysis module is used to analyze potential environmental hazard areas using the pen distribution data and the cow movement trajectory, specifically including:

[0162] 1) Obtain the spatial layout data of the pasture pens, the movement trajectory data of the cows and the environmental index data, and perform exception processing, data synchronization and spatial grid division on the acquired data;

[0163] Shed distribution data: Obtain the spatial layout of the ranch pens (map or coordinate data), including: pen boundary coordinates (x, y), facility distribution (such as water troughs, feed troughs, drainage outlets), and high-risk area identification (such as sewage discharge areas and poorly ventilated areas);

[0164] Cow movement trajectory data: Use GPS, RFID or UWB (ultra-wideband positioning) devices to obtain the movement trajectory coordinates of cows;

[0165] Environmental indicator data: environmental monitoring data such as temperature and humidity, ammonia concentration, and carbon dioxide concentration, marking the environmental status of a specific area;

[0166] Clean up abnormal data: remove erroneous positioning data (such as coordinate jumps or duplications);

[0167] Data synchronization: Match the pen distribution data with the cow movement trajectory data in time and space;

[0168] Spatial grid division: Divide the enclosure area into several grids (e.g. 1m 2 Units) to facilitate subsequent statistics and analysis;

[0169] 2) Calculate the residence time of each cow in each grid area, integrate the residence time of all cows, generate a heat map of the entire pen area, and draw the movement paths of all cows to identify areas with high path overlap;

[0170] Trajectory residence time analysis: Calculate the residence time of each cow in each grid area;

[0171] Aggregation heat map: The residence time of all cows is summed up to generate a heat map of the entire pen area. Grid areas with high residence time may have hidden dangers of crowding or abnormal environment;

[0172] Movement trajectory distribution: Plot the movement paths of all cows and identify areas with high path overlap, which may indicate that the area is a must-go for cows or has environmental problems;

[0173] 3) Determine regional indicators of potential environmental hazards based on pen layout and cow behavior, where regional indicators of potential environmental hazards include high residence time areas, abnormally low movement areas, and high track intersection areas;

[0174] High residence time area: If the residence time in a certain area is significantly higher than that in other areas, possible reasons include poor ventilation (cows gather in high temperature areas), slippery ground (cows stay and are unwilling to move), and facility attraction (such as near the trough or feed trough, there may be crowding problems);

[0175] Abnormally low movement area: If the cows enter a certain area less frequently or the movement tracks are sparse, the possible reasons include cows avoiding the area (such as environmental pollution, sewage discharge area, slippery ground, etc.) and unsafe factors (obstacles, insufficient light, etc.);

[0176] High track intersection areas: If the tracks of cows frequently cross in a certain area, it may cause hidden dangers, such as crowding risk (cows are prone to collision or fighting in this area) and excessive trampling (the ground is prone to damage or become slippery, which may cause injuries);

[0177] 4) Spatial matching of the pen environment data and the cow trajectory data to identify areas with abnormal temperature and humidity, areas with poor air quality, and areas with slippery conditions;

[0178] Abnormal temperature and humidity areas: In the grid with high residence time, check the temperature and humidity data to determine whether they exceed the appropriate range;

[0179] Areas with poor air quality: Combine ammonia concentration and carbon dioxide concentration data to identify areas with air pollution;

[0180] Slippery area identification: If the residence time in a certain area is abnormally high and accompanied by an increase in humidity, it can be determined that the ground is slippery;

[0181] 5) Based on the results of spatial matching of regional indicators of potential environmental hazards, pen environment data and cow trajectory data, the potential hazard areas are marked on the pasture map. The potential hazard areas include high-risk areas, medium-risk areas and low-risk areas, for example:

[0182] Red area: high risk (abnormal environmental indicators, high cow retention);

[0183] Yellow area: medium risk (trajectory crossings are frequent);

[0184] Blue areas: low risk (trajectory sparse or avoid areas).

[0185] The abnormal warning module 3 is used to determine the abnormal situation of dairy cattle breeding and the corresponding warning level according to the feature fusion analysis results, and to issue abnormal warning according to the warning level;

[0186] The decision-making suggestion module 4 is used to formulate dairy cattle breeding optimization suggestions according to the feature fusion analysis results, and the dairy cattle breeding optimization suggestions include optimizing feed distribution and dynamically dividing activity areas;

[0187] Specifically, the decision-making suggestion module 4 includes a feed distribution optimization module and an activity area division module;

[0188] The feed distribution optimization module is used to intelligently adjust the feed distribution amount according to the needs, age stage and health status of the cows;

[0189] The activity area division module is used to dynamically adjust the functional division of different areas in the pasture by using a swarm intelligence algorithm in combination with the activity data, demand data and pen environment data of the dairy cows;

[0190] The activity area division module uses a swarm intelligence algorithm to dynamically adjust the functional division of different areas in the pasture in combination with the activity data, demand data and pen environment data of the dairy cows, including:

[0191] 1) Obtain cow activity data, demand data and pen environment data, and analyze the acquired data to mine the cow's activity patterns and demand preferences;

[0192] Cow activity data: cow movement trajectory (real-time location coordinates obtained through GPS, RFID or UWB), residence time (calculation of the cumulative residence time of cows in various areas), activity preference (statistics of the frequency of cows' use of different areas, such as water troughs and feed troughs);

[0193] Demand data: feeding demand (recording the frequency and time distribution of cows’ eating), drinking demand (recording the time cows stay in the trough area and the frequency of drinking), rest demand (identifying the time and distribution of cows’ stay in comfort areas, such as beds);

[0194] Environmental data: environmental indicators such as temperature and humidity, ammonia concentration, and ground slipperiness, as well as facility distribution and resource usage (such as the location of feed troughs and water troughs);

[0195] Mining the activity patterns and demand preferences of dairy cows includes:

[0196] Activity heat map: Analyze the movement trajectory of dairy cows and generate activity heat maps of dairy cows in different areas of the pasture; identify high-frequency use areas (such as drinking water, feeding, and resting areas);

[0197] Time period behavior preference: Based on time division (such as day, night, feeding time), analyze the activity preferences and needs of dairy cows at different times;

[0198] Demand clustering: Use clustering algorithms (such as K-means) to classify the needs of dairy cows (drinking water, resting, feeding, etc.) and identify high-demand groups and areas;

[0199] 2) According to the average distance of the cows to the target area, the regional crowding and resource utilization, the objective function of dynamic functional division is determined, and the pasture is divided into several functional areas in the initial state, and the location and functional label of the current area are recorded;

[0200] The expression of the objective function of dynamic function partitioning is:

[0201] minF=λ1·D dist +λ2·D cong -λ3·S util

[0202] In the formula, F represents the comprehensive target value, D dist It indicates the average distance that cows need to reach the target area (such as water trough, feed trough), that is, the convenience of the area. cong represents the regional crowding, which means the uneven distribution caused by too many cows. utilIndicates resource utilization, which refers to the utilization efficiency of facilities such as feed troughs and water troughs. λ1, λ2, and λ3 respectively represent the weight of the average distance of the cows to the target area, the weight of the area crowding, and the weight of the resource utilization. Preferably, in this embodiment, λ1=0.6, λ2=0.3, and λ3=0.1;

[0203] 3) Use the ant colony algorithm to iteratively optimize the regional division, and adjust the functional division of different areas of the pasture according to the optimization results to obtain the best functional regional division results, including:

[0204] Initialize area division: In the initial state, the pasture is divided into several functional areas (such as rest area, feeding area, drinking water area), and the location and function label of the current area are recorded;

[0205] Iterative optimization of regional division: Initialize pheromone concentration on each grid unit; simulate ants moving between different areas based on cow activity data and update pheromone concentration; update pheromone based on cow demand intensity and regional usage; initialize particle position (coordinates of each functional area), calculate particle fitness based on objective function, and update particle speed and position; iterate repeatedly until the objective function converges or reaches the preset optimization condition;

[0206] The farming data display module 5 is used to present the collected data, analysis results, warning signals and decision suggestions to the farming users in an intuitive manner.

[0207] According to another embodiment of the present invention, Figure 2 As shown, a dairy cattle breeding early warning method based on multi-feature fusion analysis is provided, and the method comprises the following steps:

[0208] S1. Obtaining the physiological data, feeding data and pen environment data of dairy cows;

[0209] S2. Analyze the health status, production performance and environmental comfort of dairy cows based on their physiological data, feeding data and pen environment data;

[0210] S3. Determine the abnormal situation of dairy cattle breeding and the corresponding warning level according to the feature fusion analysis results, and issue abnormal warning according to the warning level;

[0211] S4. Formulate optimization suggestions for dairy farming based on the results of feature fusion analysis, and the optimization suggestions for dairy farming include optimizing feed distribution and dynamically dividing activity areas;

[0212] S5. Present the collected data, analysis results, warning signals and decision-making suggestions to the breeding users in an intuitive manner.

[0213] To sum up, with the help of the above-mentioned technical scheme of the present invention, the present invention constructs an intelligent dairy cattle breeding early warning system integrating data collection, health assessment, production performance analysis, environmental comfort assessment and dynamic decision-making suggestions through multi-feature data fusion, machine learning and swarm intelligence algorithm. It can not only realize real-time monitoring of the health status of dairy cows and early abnormal warning, as well as accurate evaluation of production performance and optimal allocation of resources, but also realize intelligent identification of environmental hazards in pens and dynamic functional area optimization, effectively solving the problems of lagging health management, waste of resources, difficult identification of environmental hazards and unreasonable regional division in traditional dairy cattle breeding, significantly improving the intelligence level of pasture management and breeding efficiency, and has broad application prospects.

[0214] At the same time, the present invention can not only analyze the health status of dairy cows by collecting and fusing the physiological data, feeding data and pen environment data of dairy cows, but also use graph neural networks and isolation forest algorithms based on behavioral patterns, combined with the location, interaction frequency and behavioral similarity of dairy cows to construct a dynamic group relationship diagram, accurately identify potentially abnormal dairy cows, and thus monitor the health status of dairy cows in real time, warn of health abnormalities in advance, reduce the risk of disease transmission, and improve breeding efficiency.

[0215] At the same time, the present invention can not only evaluate the production performance of dairy cows based on their average daily milk production and feed utilization efficiency and identify inefficient cows, but also predict the milk production performance of dairy cows during their life cycle through machine learning models, and provide data support to optimize feed distribution, thereby improving the accuracy of resource allocation, reducing feed waste, and maximizing breeding efficiency.

[0216] At the same time, the present invention can not only combine the activity data, demand data and pen environment data of dairy cows, and use the swarm intelligence algorithm to dynamically adjust the functional division of different areas in the pasture, but also reduce the movement cost of dairy cows, alleviate crowding, and improve the utilization rate of resources (such as feed troughs, water troughs, and rest areas) through optimized functional division.

[0217] At the same time, the present invention can not only identify high residence time areas, abnormal low movement areas and high trajectory intersection areas and locate potential environmental hazards through the fusion analysis of cow movement trajectories and environmental indicator data, but also mark and warn of abnormal temperature and humidity, poor air quality and slippery areas, thereby ensuring the comfort of the pen environment and reducing the stress response and health risks of cows.

[0218] At the same time, the present invention can not only intuitively display data collection results, health status analysis, environmental comfort assessment and early warning signals through the breeding data display module, but also provide intelligent decision-making suggestions, including feed distribution optimization and dynamic activity area division, to help breeding managers make scientific decisions.

[0219] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A dairy cattle breeding early warning system based on multi-feature fusion analysis, characterized in that: It includes a data collection module (1), a feature fusion analysis module (2), an abnormal warning module (3), a decision-making suggestion module (4) and a breeding data display module (5); The data acquisition module (1) is used to obtain the physiological data, feeding data and pen environment data of the dairy cow; The feature fusion analysis module (2) is used to analyze the health status, production performance and environmental comfort of the dairy cows based on the physiological data, feeding data and pen environmental data of the dairy cows; The abnormal warning module (3) is used to determine the abnormal situation of dairy cattle breeding and the corresponding warning level according to the feature fusion analysis result, and to issue an abnormal warning according to the warning level; The decision-making suggestion module (4) is used to formulate dairy cattle breeding optimization suggestions based on the feature fusion analysis results, and the dairy cattle breeding optimization suggestions include optimizing feed distribution and dynamically dividing activity areas; The farming data display module (5) is used to present the collected data, analysis results, warning signals and decision suggestions to farming users in an intuitive manner.

2. The dairy cattle breeding early warning system based on multi-feature fusion analysis according to claim 1 is characterized in that: The feature fusion analysis module (2) includes a data preprocessing module, a health status analysis module, a production performance analysis module and an environmental comfort analysis module; The data preprocessing module is used to clean, standardize and extract features from the acquired physiological data, feeding data and pen environment data of dairy cows; The health status analysis module is used to analyze the objective health status and potential health status of the dairy cows based on the physiological data, feeding data and pen environment data of the dairy cows; The production performance analysis module is used to predict and analyze the average daily milk production, cumulative milk production, feed utilization efficiency and overall breeding efficiency of dairy cows, and identify inefficient dairy cows based on the cumulative milk production and overall breeding efficiency; The environmental comfort analysis module is used to analyze the comfort of the dairy cow breeding environment, environmental hazards and the impact of the environment on dairy cow breeding based on the physiological data, feeding data and pen environmental data of the dairy cows.

3. The dairy cattle breeding early warning system based on multi-feature fusion analysis according to claim 2 is characterized in that: The health status analysis module includes an objective health status analysis module and a potential health status analysis module; The objective health status determination module is used to determine the standard physiological data threshold according to the breed and age stage of the dairy cow, and determine the health status of the dairy cow by comparing the dairy cow's physiological data with the standard physiological data threshold; The potential health status analysis module is used to construct a dynamic group relationship graph based on the location, interaction frequency and behavior similarity between cows, extract implicit relationship features between cows using a graph neural network algorithm based on behavior patterns, and determine potential abnormal cows in the cow group based on a graph embedded isolation forest algorithm.

4. The dairy cattle breeding early warning system based on multi-feature fusion analysis according to claim 3 is characterized in that: The potential health status analysis module constructs a dynamic group relationship graph according to the positions, interaction frequencies and behavior similarities between cows, extracts implicit relationship features between cows using a graph neural network algorithm based on behavior patterns, and determines potential abnormal cows in the cow group based on an isolation forest algorithm based on graph embedding, including: Obtain location data, interaction frequency, and behavior similarity data of dairy cows in the breeding area, and pre-process the acquired data; Based on the position similarity, interaction frequency and behavior similarity between cows, the comprehensive edge weight is calculated, and the dynamic group relationship graph is constructed by combining the position, interaction frequency and behavior similarity between cows. The nodes in the dynamic group relationship graph are represented by cows, and the edges represent the relationship between cows. The graph neural network is used to embed nodes in the constructed dynamic group relationship graph, compress the high-dimensional features of each cow into a low-dimensional space, and extract the implicit relationship features between cows; The node embedding vector output by the graph neural network is input into the isolation forest model to calculate the anomaly score of each node during the isolation process; based on the comparison result between the anomaly score and the preset anomaly score threshold, the potential abnormal cows in the dairy cow group are determined; Among them, the calculation formula of the comprehensive edge weight is: W(i,j)=α·W pos (i,j)+β·W int (i,j)+γ·W beh (i,j) In the process of embedding nodes in the constructed dynamic group relationship graph using graph neural network, the features of node i in the l+1th layer aggregating its neighbor node j are: The calculation formula for the anomaly score is: Where W(i,j) represents the comprehensive edge weight, W pos (i,j) represents the position similarity weight of cows i and j, W int (i,j) represents the interaction frequency weight between cows i and j, W beh (i, j) represents the behavioral similarity weights of cows i and j, α, β, and γ represent the position similarity weight parameter, interaction frequency weight parameter, and behavioral similarity weight parameter, respectively. ij represents the Euclidean distance between cows i and j, σ represents the smoothing parameter of the Gaussian kernel, and f ij Indicates the number of contacts, max(f ij ) represents the maximum number of contacts, B i represents the behavior vector of cow i, B j represents the behavior vector of cow j, ||B i || represents the behavior vector B i The Euclidean norm of ||B j || represents the behavior vector B j The Euclidean norm of , represents the feature vector of node i in the l+1 layer, μ represents the activation function, represents the set of neighbor nodes of node i, AGG(·) represents the aggregation function, represents the feature vector of node j in layer l, s(Z i ,n) represents the abnormal score of node i, Z i represents the node embedding vector, n is the total number of samples, E(h(Z i )) represents the node embedding vector Z i The average path length in the isolation forest, c(n) represents the average path length of a binary tree, T represents the total number of trees in the forest, and h t (Z i ) represents the node embedding vector Z i The path length in the tth tree, L represents the number of layers in the graph neural network.

5. The dairy cattle breeding early warning system based on multi-feature fusion analysis according to claim 1 is characterized in that: The production performance analysis module includes a milk production analysis module, a feed utilization efficiency analysis module and a dairy cow life cycle management module; The milk production analysis module is used to use a machine learning model to combine the physiological data, age stage and historical milk production of dairy cows to predict the average daily milk production of each dairy cow during its life cycle, and calculate the cumulative milk production based on the average daily milk production; The feed utilization efficiency analysis module is used to analyze the feed utilization efficiency of each dairy cow based on the feed intake and average daily milk production of the dairy cow; The dairy cow life cycle management module is used to evaluate the breeding efficiency according to the cumulative milk production and feed conversion rate of each dairy cow, and identify inefficient dairy cows based on the breeding efficiency.

6. The dairy cattle breeding early warning system based on multi-feature fusion analysis according to claim 5 is characterized in that: The dairy cow life cycle management module evaluates the breeding efficiency according to the cumulative milk production and feed conversion rate of each dairy cow, and identifies inefficient dairy cows based on the breeding efficiency, including: Obtain the cumulative milk production and feed conversion rate of each cow, and calculate the comprehensive efficiency score of each cow based on the cumulative milk production and feed conversion rate of each cow; determine whether the comprehensive efficiency score of each cow meets the preset scoring standard. If yes, continue to breed the cow; if not, mark the cow as an inefficient cow; The calculation formula for the breeding efficiency of each dairy cow is: In the formula, S i represents the comprehensive efficiency score of the i-th cow, P i represents the feed conversion rate of the i-th dairy cow, C i represents the cumulative milk production of the i-th cow, C max represents the maximum cumulative milk production in the dairy cow herd, ω1 and ω2 represent the weight of feed conversion rate, cumulative milk production and the weight of maximum cumulative milk production, respectively.

7. The dairy cattle breeding early warning system based on multi-feature fusion analysis according to claim 1 is characterized in that: The environmental comfort analysis module includes an environmental data analysis module and an environmental hazard analysis module; The environmental data analysis module is used to determine the comfort level of the dairy cow breeding environment based on the comparison results between the environmental data of the breeding area and the preset standard environmental threshold; The environmental hazard analysis module is used to analyze potential environmental hazard areas using the pen distribution data and the cow movement trajectory.

8. The dairy cattle breeding early warning system based on multi-feature fusion analysis according to claim 7 is characterized in that: The environmental hazard analysis module includes the following steps when analyzing potential environmental hazard areas using the pen distribution data and the cow movement trajectory: Obtain the spatial layout data of the pasture pens, the movement trajectory data of the cows and the environmental index data, and perform exception processing, data synchronization and spatial grid division on the acquired data; Calculate the residence time of each cow in each grid area, integrate the residence time of all cows, generate a heat map of the entire pen area, and draw the movement paths of all cows to identify areas with high path overlap; Determine regional indicators of potential environmental hazards based on pen layout and cow behavior, including high residence time areas, abnormally low movement areas, and high track intersection areas; Spatial matching of the pen environment data and the cow trajectory data was performed to identify areas with abnormal temperature and humidity, areas with poor air quality, and areas with slippery weather. Based on the results of spatial matching of regional indicators of potential environmental hazards, pen environment data and cow trajectory data, the hazard areas are marked on the pasture map, where the hazard areas include high-risk areas, medium-risk areas and low-risk areas.

9. The dairy cattle breeding early warning system based on multi-feature fusion analysis according to claim 1, characterized in that: The decision-making suggestion module (4) includes a feed distribution optimization module and an activity area division module; The feed distribution optimization module is used to intelligently adjust the feed distribution amount according to the needs, age stage and health status of the cows; The activity area division module is used to dynamically adjust the functional division of different areas in the pasture by using a swarm intelligence algorithm in combination with the activity data, demand data and pen environment data of the dairy cows; The activity area division module uses a swarm intelligence algorithm to dynamically adjust the functional division of different areas in the pasture in combination with the activity data, demand data and pen environment data of the dairy cows, including: Obtain cow activity data, demand data and pen environment data, and analyze the acquired data to explore the cow's activity patterns and demand preferences; According to the average distance of cows to the target area, the regional crowding and resource utilization, the objective function of dynamic functional division is determined, and the pasture is divided into several functional areas in the initial state, and the location and functional label of the current area are recorded; The expression of the objective function of dynamic function partitioning is: min F=λ1·D dist +λ2·D cong -λ3·S util In the formula, F represents the comprehensive target value, D dist represents the average distance of the cows to reach the target area, D cong represents the regional congestion, S util represents resource utilization, λ1, λ2, and λ3 represent the weight of the average distance of the cows to the target area, the weight of the regional crowding, and the weight of the resource utilization, respectively; The ant colony algorithm is used to iteratively optimize the regional division, and the functional division of different areas of the pasture is adjusted according to the optimization results to obtain the best functional regional division result.

10. A dairy cattle breeding early warning method based on multi-feature fusion analysis, which is implemented based on the dairy cattle breeding early warning system based on multi-feature fusion analysis according to any one of claims 1 to 9, characterized in that: The method comprises the following steps: S1. Obtaining the physiological data, feeding data and pen environment data of dairy cows; S2. Analyze the health status, production performance and environmental comfort of dairy cows based on their physiological data, feeding data and pen environment data; S3. Determine the abnormal situation of dairy cattle breeding and the corresponding warning level according to the feature fusion analysis results, and issue abnormal warning according to the warning level; S4. Formulate optimization suggestions for dairy farming based on the results of feature fusion analysis, and the optimization suggestions for dairy farming include optimizing feed distribution and dynamically dividing activity areas; S5. Present the collected data, analysis results, warning signals and decision-making suggestions to the breeding users in an intuitive manner.

Images