Animal physiological feature data acquisition system and analysis method based on Internet of Things

Abstract

The invention relates to the technical field of the Internet of Things, and discloses an animal physiological feature data acquisition system and analysis method based on the Internet of Things, and the method comprises the following steps: cleaning and standardizing original sensor data, including filling missing data, eliminating noise interference through a dynamic filtering technology, unifying the dimension of multi-source data, and analyzing the data; the method comprises the following steps of: acquiring a Beidou time service time reference, eliminating redundant features, interpolating and aligning asynchronous sampling data to the Beidou time service time reference, correcting a time sequence phase difference of physiological and behavior data through a dynamic path matching technology, and correcting GPS / Beidou positioning drift in combination with Kalman filtering. Through multi-source sensing fusion and an edge-fog-cloud cooperative computing architecture, the effective collection rate of animal physiological data is achieved, abnormal response is delayed and compressed, the flexible energy supply technology and dynamic communication optimization are combined, the endurance and communication success rate is maintained in a pasture, and the false alarm rate is reduced compared with a traditional method; common diseases such as abnormal body temperature, digestive system diseases and the like are accurately warned and covered.

Description

Technical Field

[0001] This invention relates to the field of the Internet of Things (IoT), and more specifically, to an IoT-based system and method for collecting and analyzing animal physiological characteristic data. Background Technology

[0002] The global livestock industry is shifting from traditional extensive management to intensive farming, using Internet of Things (IoT) technology to monitor the health of individual animals, improve production efficiency, and reduce the risk of disease.

[0003] Data loss during animal vigorous exercise, coupled with the high latency bottleneck of cloud-centralized architecture and timing misalignment caused by differences in sampling rates among multiple sensors. Summary of the Invention

[0004] This invention provides an animal physiological characteristic data acquisition system and analysis method based on the Internet of Things, solving the technical problems in related technologies.

[0005] This invention provides a method for analyzing animal physiological characteristic data based on the Internet of Things, comprising the following steps: S100 cleans and standardizes the raw sensor data, including filling in missing data, eliminating noise interference through dynamic filtering technology, unifying the dimensions of multi-source data, and removing redundant features. S200 interpolates and aligns asynchronously sampled data to the BeiDou time reference, corrects the temporal phase difference between physiological and behavioral data through dynamic path matching technology, and combines Kalman filtering to correct GPS / BeiDou positioning drift. S300 maps spatiotemporally aligned data to a high-dimensional space, calculates the interaction weights between physiological indicators and motion trajectories through a spatial awareness attention mechanism, and fuses original features and attention features in a hierarchical manner, ultimately reducing the dimensionality to generate a 128-dimensional global feature vector. The S400 establishes a multidimensional normal distribution baseline model based on historical data, calculates the current feature Mahalanobis distance in real time, determines the anomaly level by combining the chi-square distribution dynamic threshold, traces the anomaly to the specific sensor category, and optimizes the baseline parameters online. The S500 uses a gated recurrent network to model historical feature sequences, recursively generates multi-step prediction results, quantifies prediction uncertainty through Monte Carlo random sampling, and inverts the predicted values ​​to the original sensor physical dimensions. S600 generates a graded response strategy, calculates a dynamic risk score by combining the abnormal deviation degree and the predicted trend slope, divides the warning into three levels and triggers corresponding contingency plans, plans the optimal inspection path, generates disposal plans by reasoning from the knowledge graph, and continuously optimizes the decision-making model through reinforcement learning.

[0006] Furthermore, S100 specifically includes the following steps: S110, Multi-source data acquisition: Aggregates data from distributed sensor nodes; S120, Missing value handling: Filling in missing data based on temporal proximity; S130, Outlier Correction: Data correction based on the Raida Criterion; S140, Statistical Calculation: Calculate the characteristic statistics of the current batch of data; S150, Z-score normalization: Normalizes the corrected data.

[0007] Furthermore, S200 specifically includes the following steps: S210, Time Reference Synchronization: Unifying the timestamps of multiple sensors to the BeiDou time synchronization system; S220, Dynamic Time Warping: Aligns the temporal phase difference between physiological indicators and behavioral data; S230, Spatial Coordinate Correction: Integrates GPS / BeiDou data with motion trajectory.

[0008] Furthermore, in S300, the specific steps include: S310, Feature Embedding Representation: Mapping spatiotemporally aligned data to a high-dimensional space; S320, Cross-modal attention computation: Calculates the interaction weights between physiological indicators and spatial features; S330, Hierarchical Feature Fusion: Fusion of spatiotemporal attention and original features; S340, Feature Dimensionality Reduction: Extracting a 128-dimensional joint representation.

[0009] Furthermore, in S310, the specific steps are as follows: Input feature matrix: ; in It represents 7-dimensional physiological characteristics + 2-dimensional spatial coordinates (longitude) ,latitude ), Indicates the length of the time series. This represents the feature matrix after spatiotemporal alignment. Embedded functions: ; in Indicates the hidden layer dimension. Represents the embedding weight matrix. This represents the embedding bias vector. Represents embedded features, This represents the alignment feature at time t; In S320, the specific steps are as follows: Query / key / value vector generation: ; Space enhances attention: ; in Indicates the first Location coordinates of the time point , Indicates the first Location coordinates of the time point , Both represent time series indexes. , These represent the transformation matrices for the query, key, and value, respectively. Indicates attention weights, This represents the transpose of the query vector at time point u.

[0010] Furthermore, in S330 and S340, the specific steps are as follows: Gating fusion mechanism: ; ; ; in This represents attention aggregation features. Represents the gate weight matrix. , This represents the gating bias vector. Indicates the gating coefficient. Indicates the characteristics after fusion. This represents the sigmoid activation function. It represents the Hadamah accumulation. This indicates the average pooling operation. This represents the value vector at time point v. Dimensionality reduction formula: ; in Represents the dimensionality reduction weight matrix. , This represents the dimension reduction bias vector. , This represents the eigenvectors after dimensionality reduction. This represents the fusion characteristics at time t.

[0011] Furthermore, in S400, the specific steps include: S410, Health Baseline Modeling: Constructing a multidimensional normally distributed baseline based on historical data; S420, Real-time Mahalanobis distance calculation: Calculates the deviation of the current feature vector from the baseline; S430, Dynamic threshold determination: Set dynamic alarm thresholds based on chi-square distribution; S440, Anomaly Type Tracing: Decompose the anomaly contribution to locate the root cause sensor; S450, Baseline Adaptive Update: Gradually optimizes the baseline model through exponential smoothing.

[0012] Furthermore, the S500 specifically includes the following steps: S510, Temporal Window Reconstruction: Reconstructing fused features into a sliding window input; S520, LSTM multi-step prediction: Modeling temporal dependencies through a third-order gating mechanism; S530, Multi-step prediction unfolding: Recursively generate the maximum prediction step size for future prediction; S540, Uncertainty Quantification: Calculating Confidence Intervals via Monte Carlo Dropout; S550, Prediction Alignment: Maps predicted features back to the original sensor space.

[0013] Furthermore, the S600 specifically includes the following steps: S610, Comprehensive Risk Quantification: Integrating abnormal confidence levels and trend predictions to generate risk scores; S620, multi-level early warning classification: response levels are divided based on standards; S630, Spatial Path Planning: Generate the optimal inspection path for abnormal individuals; S640, Intervention Strategy Generation: Generate treatment plans through knowledge graph reasoning; S650, Online Policy Optimization: Updating the Decision Model Based on Q-learning.

[0014] This invention also proposes an Internet of Things (IoT)-based animal physiological characteristic data acquisition system for acquiring data from the aforementioned IoT-based animal physiological characteristic data analysis method, including: Multi-source sensor module: Real-time collection of 7 types of physiological indicators, including body temperature, activity level, rumination duration, food intake, breathing frequency, longitude and latitude, and simultaneous acquisition of GPS / BeiDou dual-mode positioning data; Built-in three-axis accelerometer monitors movement behavior. Edge preprocessing module: Performs preliminary calculations such as data denoising, missing value imputation, and sliding window statistics on the terminal device to reduce the amount of data transmitted wirelessly; Spatiotemporal alignment gateway: It unifies the time reference of each sensor through the Beidou timing chip, and uses an adaptive interpolation algorithm to align data streams with different sampling rates and correct geofence coordinate offsets; Computational fusion node: Deploys an attention mechanism fusion algorithm to integrate physiological indicators and motion trajectory data, generating a 128-dimensional joint feature vector; Cloud-based analytics engine: Performs deep computations for Mahalanobis distance anomaly detection, LSTM trend prediction, and dynamic risk assessment, and maintains the health baseline model; Decision execution controller: Automatically triggers ventilation adjustment and feeder start / stop equipment based on the warning level, and generates inspection paths to push to the execution equipment; Low-power communication unit: Supports LoRa / NB-IoT dual-mode transmission to achieve full coverage of the ranch and ensure good data connectivity in harsh environments.

[0015] The beneficial effects of this invention are as follows: This invention achieves an effective acquisition rate of animal physiological data through multi-source sensor fusion and edge-fog-cloud collaborative computing architecture, compresses abnormal response delays, and combines flexible power supply technology and dynamic communication optimization to maintain continuous operation and communication success rate in pastures. The false alarm rate is reduced compared to traditional methods, and accurate early warning covers seven common diseases such as abnormal body temperature and digestive system diseases. Attached Figure Description

[0016] Figure 1 This is a flowchart of an animal physiological characteristic data analysis method based on the Internet of Things proposed in this invention; Figure 2 This is a structural block diagram of an animal physiological characteristic data acquisition system based on the Internet of Things proposed in this invention.

[0017] In the diagram: 101, Multi-source sensing module; 102, Edge preprocessing module; 103, Spatiotemporal alignment gateway; 104, Computation fusion node; 105, Cloud analytics engine; 106, Decision execution controller; 107, Low-power communication unit. Detailed Implementation

[0018] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0019] like Figure 1 As shown, an animal physiological characteristic data analysis method based on the Internet of Things includes the following steps: S100, Data Preprocessing: Cleaning and standardizing the raw sensor data, including filling in missing data (such as body temperature and heart rate), eliminating noise interference through dynamic filtering technology, unifying the dimensions of multi-source data (such as normalizing the perceived temperature to 0-1), and removing redundant features. S110, Multi-source data acquisition: Aggregates data from distributed sensor nodes; Input matrix: ; in This represents the number of samples within the time window. , Indicates sensor dimensions (body temperature / activity level / rumination duration / feed intake / breathing frequency / longitude / latitude). This represents the original value of the j-th dimension of the i-th sample. This represents the original data input matrix; S120, Missing value handling: Filling in missing data based on temporal proximity; Interpolation formula: ; in Indicates missing value markers, This represents the data value after filling in the blanks. This represents the original value at the previous moment. This represents the original value at the next moment; S130, Outlier Correction: Data correction based on the Raida criterion (3σ principle); Rule revision: ; in This represents the mean of the historical data for the j-th dimension indicator. This represents the standard deviation of the historical data for the j-th dimension indicator. This indicates the corrected data value; S140, Statistical Calculation: Calculate the characteristic statistics of the current batch of data; Mean: ; Standard deviation: ; in This represents the mean of the current batch of the j-th dimension index. This represents the standard deviation of the j-th dimension index for the current batch; S150, Z-score standardization: normalizes the corrected data; Standardized formula: ; Reference benchmark selection: ; in This represents the standardized data value. Indicates the reference mean. Indicates the reference standard deviation; S200, spatiotemporal alignment: interpolates and aligns asynchronous sampled data to the BeiDou time reference, corrects the temporal phase difference between physiological and behavioral data through dynamic path matching technology, and combines Kalman filtering to correct GPS / BeiDou positioning drift. In one embodiment of the present invention, the following steps are specifically included: S210, Time Reference Synchronization: Unifying the timestamps of multiple sensors to the BeiDou time synchronization system; Interpolation formula (for non-equal interval sampled data): ; in Represents the k-th original sampling time point, where , This represents the (k+1)th original sampling time point. Indicates the target synchronization time point (interval) ), This represents the synchronized data value. Indicates time Standardized data values; S220, Dynamic Time Warping (DTW): Aligns the temporal phase difference between physiological indicators and behavioral data; Local distance matrix: ; in Representing time series of physiological indicators (such as body temperature, length) ), Represents time series of behavioral data (such as activity level, length) ), This represents the Euclidean distance between two time points p and q. This represents the data values ​​after the physiological indicators have been synchronized. This represents the data value after the behavior data has been synchronized. Represents the Euclidean norm; Cumulative distance matrix: ; in This represents the minimum cumulative distance from the starting point to the point (p, q). This indicates the distance from the starting point to the point (p). The minimum cumulative distance of ,q) Indicates from the starting point to the point ( The minimum cumulative distance, Indicates from the starting point to the point ( The minimum cumulative distance; S230, Spatial Coordinate Correction: Integrating GPS / BeiDou data with motion trajectory; State vector: ; in Represents the true value of latitude. Represents the true value of longitude. , These represent velocities in latitude and longitude, respectively. Represents the state vector; Kalman filter correction: predict: ; ; renew: ; ; ; in Represents the state transition matrix. Represents the process noise covariance ( degree²), Represents the observation noise covariance ( degree²), Indicates Kalman gain, Represents the observation vector. Represents the observation matrix. This represents the state estimation covariance matrix. This represents the prior state estimate. This indicates a priori estimation of the covariance. This represents the state vector from the previous time step. This represents the state estimation covariance matrix at the previous time step. Represents the identity matrix; S300, multimodal fusion: maps spatiotemporally aligned data to a high-dimensional space, calculates the interaction weights of physiological indicators and motion trajectories through a spatial awareness attention mechanism, and fuses original features and attention features in a hierarchical manner, finally reducing the dimensionality to generate a 128-dimensional global feature vector. In one embodiment of the present invention, the following steps are specifically included: S310, Feature Embedding Representation: Mapping spatiotemporally aligned data to a high-dimensional space; Input feature matrix: ; in It represents 7-dimensional physiological characteristics + 2-dimensional spatial coordinates (longitude) ,latitude ), Indicates the length of the time series. This represents the feature matrix after spatiotemporal alignment. Embedded functions: ; in Indicates the hidden layer dimension. Represents the embedding weight matrix. This represents the embedding bias vector. Represents embedded features, This represents the alignment feature at time t; S320, Cross-modal attention computation: Calculates the interaction weights between physiological indicators and spatial features; Query / key / value vector generation: ; Space enhances attention: ; in Indicates the first Location coordinates of the time point ( , ), Indicates the first Location coordinates of the time point ( , ), Both represent time series indexes. , These represent the transformation matrices for the query, key, and value, respectively. Indicates attention weights, This represents the transpose of the query vector at time point u. S330, Hierarchical Feature Fusion: Fusion of spatiotemporal attention and original features; Gating fusion mechanism: ; ; ; in This represents attention aggregation features. Represents the gate weight matrix. , This represents the gating bias vector. Indicates the gating coefficient. Indicates the characteristics after fusion. This represents the sigmoid activation function. It represents the Hadamah accumulation. This indicates the average pooling operation. This represents the value vector at time point v. S340, Feature Dimensionality Reduction: Extracting a 128-dimensional joint representation; Dimensionality reduction formula: ; in Represents the dimensionality reduction weight matrix. , This represents the dimension reduction bias vector. , This represents the eigenvectors after dimensionality reduction. This represents the fusion characteristics at time t; S400, Anomaly Detection: Based on historical data, a multidimensional normal distribution baseline model is established, the current feature Mahalanobis distance is calculated in real time, and the anomaly level is determined by combining the chi-square distribution dynamic threshold. The anomaly is traced to the specific sensor category, and the baseline parameters are optimized online. In one embodiment of the present invention, the following steps are specifically included: S410, Health Baseline Modeling: Constructing a multidimensional normally distributed baseline based on historical data; Input data: ; in This represents the number of samples in the training set (normal state data). Indicates the first The feature vector of each sample ; Parameter calculation: Mean vector: ; in This represents the mean vector of the baseline model. ; Covariance matrix: ; in Represents the covariance matrix of the baseline model. ; S420, Real-time Mahalanobis distance calculation: Calculates the deviation of the current feature vector from the baseline; Distance formula: ; in This represents the 128-dimensional features at the current time. Indicates Mahalanobis distance; S430, Dynamic threshold determination: Set dynamic alarm thresholds based on chi-square distribution; Threshold formula: ; in hour, ; when When, an exception flag is triggered. ; in This represents the threshold based on the chi-square distribution. Indicates the significance level. Indicates an abnormality flag (0 indicates normal, 1 indicates abnormal). The 1-α quantile represents the chi-square distribution with 128 degrees of freedom; S440, Anomaly Type Tracing: Decompose the anomaly contribution to locate the root cause sensor; Contribution calculation: ; in This represents the anomalous contribution of the k-th dimension feature. Indicates the feature dimension index. , The baseline mean of the k-th dimension feature is represented. This represents the k-th diagonal element of the baseline covariance matrix. This represents the k-th eigenvalue at the current time. Source tracing results: ; in This represents the anomalous contribution of the k-th dimension feature. Represents the feature subset corresponding to the j-th type of original sensor (e.g. (corresponding to body temperature-related characteristics) Indicates the sensor index of the abnormal source; S450, Baseline Adaptive Update: Stepwise optimization of the baseline model through exponential smoothing; Online formula update: ; ; in , Indicates the historical data retention factor (only when...) (Updated regularly) S500, Trend Prediction: It uses a gated recurrent network to model historical feature sequences, recursively generates multi-step prediction results, quantifies prediction uncertainty through Monte Carlo random sampling, and inverts the predicted values ​​to the original sensor physical dimensions; In one embodiment of the present invention, the following steps are specifically included: S510, Temporal Window Reconstruction: Fusing Features Refactored into a sliding window input; Input tensor: ; in Indicates the length of the input window (default). ), Represents the feature vector at the current time. This represents the input tensor of the sliding window; S520, LSTM multi-step prediction: Modeling temporal dependencies through a third-order gating mechanism; Gating calculation: Forgotten Gate: ; Input Gate: ; ; Status Update: ; Output gate: ; ; in Represents the weight matrix, i.e. , , and These represent the first, second, third, and fourth weight matrices, respectively. , Indicates the hidden layer dimension (default) ), Indicates the output is hidden. Indicates cell state, These represent the forget gate, input gate, and output gate, respectively. Represents the bias vector, i.e. , , and These represent the first, second, third, and fourth bias vectors, respectively. It represents the Hadamah accumulation. Indicates the state of candidate cells. This indicates the cell state at the previous moment. This indicates the hidden state at the previous moment. This represents the input features at the current moment. Indicates the time step index. This represents a vector concatenation operation; S530, Multi-Step Prediction Unfolding: Recursive Generation of the Future Step prediction; Iteration formula: ; in Represents the projection matrix. , Indicates projection offset. Indicates the maximum prediction step size (default) (i.e., 72 hours) Indicates the first The predicted value of the step; S540, Uncertainty Quantification: Calculating Confidence Intervals via Monte Carlo Dropout; Probability sampling: ; ; in This represents the Dropout mask. Indicates the number of samples. This represents the predicted value from the k-th sample. Indicates projection offset; Confidence interval: ; ; ; in Indicates the confidence interval. This represents the predicted mean. Indicates the standard deviation of the forecast; S550, Prediction Alignment: Maps predicted features back to the original sensor space; Inverse transform formula: in Indicates the reference mean. Indicates the reference standard deviation. , Represents the original sensor dimension index. Let represent the predicted value of the j-th dimension sensor at time t+τ; S600, Decision Output: Generate a graded response strategy, calculate a dynamic risk score by combining the abnormal deviation degree and the predicted trend slope, divide into three levels of early warning and trigger corresponding contingency plans, plan the optimal inspection path, generate disposal solutions from knowledge graph reasoning, and continuously optimize the decision model through reinforcement learning. In one embodiment of the present invention, the following steps are specifically included: S610, Comprehensive Risk Quantification: Integrating abnormal confidence levels and trend predictions to generate risk scores; Dynamic weight formula: ; ; in Indicates the weighting factor for abnormal deviation. Indicates the predicted feature gradient. This represents the overall risk score; S620, multi-level early warning classification: response levels are divided based on the ISO20976 standard; Threshold determination: ; in and These represent the first and second warning thresholds, respectively. , , When the emergency stop protocol is triggered, Indicates the warning level (1-Normal, 2-Warning, 3-Emergency); S630, Spatial Path Planning: Generate the optimal inspection path for abnormal individuals; Energy minimization model: ; in Represents a node Communication delay to base station This represents the priority adjustment coefficient. Represents a node Location coordinates, Represents a node Exception markers, Indicates the path length; S640, Intervention Strategy Generation: Generate treatment plans through knowledge graph reasoning; Atlas search: ; in Representation of domain knowledge graph (node ​​set) Relationship set ), This represents the recommended set of actions. Represents relationships in a knowledge graph. Indicates the action to be taken; S650, Online Policy Optimization: Updating the Decision Model Based on Q-learning; Q value update: ; in Indicates the learning rate. This represents the discount factor for future earnings. Indicates the current state. Indicates the current action. Indicates an immediate reward. Represents a state-action value function.

[0020] like Figure 2 As shown, in one embodiment of the present invention, according to the above-described method for analyzing animal physiological characteristic data based on the Internet of Things (IoT), the data to be analyzed is collected through an IoT-based animal physiological characteristic data acquisition system, which includes the following modules: Multi-source sensor module 101: Real-time collection of seven physiological indicators, including body temperature, activity level, rumination duration, food intake, breathing frequency, longitude and latitude, and simultaneous acquisition of GPS / BeiDou dual-mode positioning data; Built-in three-axis accelerometer to monitor movement behavior. Edge preprocessing module 102: Performs preliminary calculations such as data denoising, missing value imputation, and sliding window statistics on the terminal device to reduce the amount of data transmitted wirelessly; Spatiotemporal Alignment Gateway 103: It unifies the time reference of each sensor through the Beidou timing chip, and uses an adaptive interpolation algorithm to align data streams with different sampling rates and correct geofence coordinate offsets; Computation fusion node 104: Deploys an attention mechanism fusion algorithm to integrate physiological indicators and motion trajectory data, generating a 128-dimensional joint feature vector; Cloud-based analytics engine 105: Performs deep computations such as Mahalanobis distance anomaly detection, LSTM trend prediction, and dynamic risk assessment to maintain a healthy baseline model; Decision execution controller 106: Automatically triggers equipment such as ventilation adjustment and feeder start / stop according to the warning level, and generates inspection paths and pushes them to the execution equipment; Low-power communication unit 107: Supports LoRa / NB-IoT dual-mode transmission, achieving full coverage of the ranch and ensuring good data connectivity in harsh environments; Meanwhile, the system uses flexible solar thin film for power supply, combined with supercapacitor energy storage modules, to ensure the equipment can continue operating even during continuous rainy weather.

[0021] Its working principle is as follows: Based on a three-tier architecture of "device-edge-cloud", it achieves closed-loop control from data collection to decision execution: Real-time physiological indicator sensing: Wearable sensors continuously collect data such as animal body temperature and heart rate, accelerometers monitor exercise intensity, and GPS / BeiDou modules report positioning coordinates every second; Dynamic data cleaning: The edge computing unit performs noise filtering and outlier removal locally, and compresses the original data by 80% before transmission; Cross-modal spatiotemporal calibration: The alignment gateway synchronizes the timestamps of each device through the BeiDou timing chip to perform dynamic phase matching between motion trajectories and physiological parameters; Intelligent feature extraction: Computing nodes fuse multi-dimensional data to generate a 128-dimensional feature vector with spatiotemporal correlation; Cloud-based deep decision-making: The analysis engine simultaneously calls anomaly detection models and trend prediction algorithms, and combines historical data to assess the risk level; Precise response execution: The decision controller coordinates with the environment to adjust the equipment according to the instructions, and drives the execution equipment (drone) to perform directional inspections through LoRa broadcast.

[0022] The embodiments of the present invention have been described above, but the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention, all of which are within the protection scope of the present invention.

Claims

1. A method for analyzing animal physiological characteristic data based on an Internet of Things, characterized by, The method comprises the following steps: S100, cleaning and standardizing the original sensor data, including filling in missing data, eliminating noise interference by dynamic filtering technology, unifying the dimensions of multi-source data, and removing redundant features; S200, interpolating and aligning non-synchronous sampling data to the Beidou time reference, correcting the time sequence phase difference of physiological and behavioral data by dynamic path matching technology, and combining Kalman filtering to correct the GPS / Beidou positioning drift; S300, mapping the spatio-temporally aligned data to a high-dimensional space, calculating the interaction weight of physiological indicators and motion trajectories by a spatial perception attention mechanism, hierarchically fusing original features and attention features, and finally reducing dimension to generate a 128-dimensional global feature vector; S400, based on historical data, a multi-dimensional normal distribution baseline model is established, the current feature Mahalanobis distance is calculated in real time, the abnormal level is determined by combining the chi-square distribution dynamic threshold, the anomaly is traced to the specific sensor category, and the baseline parameters are optimized online; S500, a gated recurrent network is used to model the historical feature sequence, a multi-step prediction result is recursively generated, the prediction uncertainty is quantified by Monte Carlo random sampling, and the prediction value is inverted to the original sensor physical dimension; S600, a hierarchical response strategy is generated, the dynamic risk score is calculated by combining the abnormal deviation and the prediction trend slope, the three-level early warning is divided, the corresponding preplan is triggered, the optimal inspection path is planned, the disposal scheme is generated by reasoning from the knowledge graph, and the decision model is continuously optimized by reinforcement learning. 2.The animal physiological characteristic data analysis method based on the Internet of Things according to claim 1, characterized in that, In S100, the following steps are specifically included: S110, multi-source data acquisition: aggregating distributed sensor node data; S120, missing value processing: filling in missing data based on time proximity; S130, outlier correction: correcting data based on the Lillie criterion; S140, statistical quantity calculation: calculating the feature statistics of the current batch of data; S150, Z-score standardization: normalizing the corrected data. 3.The animal physiological characteristic data analysis method based on the Internet of Things according to claim 2, characterized in that, In S200, the following steps are specifically included: S210, time reference synchronization: unifying the multi-sensor timestamps to the Beidou timing system; S220, dynamic time warping: aligning the time sequence phase difference of physiological indicators and behavioral data; S230, spatial coordinate correction: fusing GPS / Beidou data and motion trajectories. 4.The animal physiological characteristic data analysis method based on the Internet of Things according to claim 3, characterized in that, In S300, the following steps are specifically included: S310, feature embedding representation: mapping the spatio-temporally aligned data to a high-dimensional space; S320, cross-modal attention calculation: calculating the interaction weight of physiological indicators and spatial features; S330, hierarchical feature fusion: fusing spatio-temporal attention and original features; S340, feature dimension reduction: extracting a 128-dimensional joint representation. 5.The animal physiological characteristic data analysis method based on the Internet of Things according to claim 4, characterized in that, In S310, the specific steps are as follows: Input feature matrix: ; wherein represents a 7-dimensional physiological feature + 2-dimensional spatial coordinates (longitude , latitude ), represents the length of the time series, represents the feature matrix after spatio-temporal alignment; Embedding function: ； wherein denotes the hidden layer dimension, denotes the embedding weight matrix, denotes the embedding bias vector, denotes the embedding feature, denotes the aligned feature at time t; In S320, the specific steps are as follows: Query / key / value vector generation: ； Spatially enhanced attention: ； wherein denotes the position coordinates of the first time point, , ), denotes the position coordinates of the first time point, , ), each denotes a time series index, denotes a transformation matrix for the query / key / value, respectively, denotes an attention weight, denotes the query vector transpose of the u-th time point. 6.The animal physiological characteristic data analysis method based on the Internet of Things according to claim 5, characterized in that, In S330 and S340, the specific steps are as follows: Gated fusion mechanism: ； ； ； wherein denotes an attention aggregation feature, denotes a gating weight matrix, , denotes a gating bias vector, denotes a gating coefficient, denotes a fused feature, denotes a sigmoid activation function, denotes a Hadamard product, denotes an average pooling operation, denotes a value vector at the v-th time point; Dimension reduction formula: ； wherein denotes a dimension reduction weight matrix, , denotes a dimension reduction bias vector, , denotes a dimension reduced feature vector, denotes a fusion feature at time t. 7.The animal physiological characteristic data analysis method based on the Internet of Things according to claim 6, characterized in that, In S400, the following steps are specifically included: S410, health baseline modeling: constructing a multi-dimensional normal distribution baseline based on historical data; S420, real-time Mahalanobis distance calculation: calculating the deviation of the current feature vector from the baseline; S430, dynamic threshold determination: setting a dynamic alarm threshold based on chi-square distribution; S440, abnormal type tracing: decomposing abnormal contribution to locate the root sensor; S450, baseline adaptive update: gradually optimizing the baseline model through exponential smoothing. 8.The animal physiological characteristic data analysis method based on the Internet of Things according to claim 7, characterized in that, In S500, the following steps are specifically included: S510, time window reconstruction: reconstructing the fusion features into sliding window input; S520, LSTM multi-step prediction: modeling time dependence through a three-order gating mechanism; S530, multi-step prediction unfolding: recursively generating the maximum prediction step for prediction; S540, uncertainty quantification: calculating the confidence interval through Monte Carlo Dropout; S550, prediction result alignment: mapping the predicted features back to the original sensor space. 9.The animal physiological characteristic data analysis method based on the Internet of Things according to claim 8, characterized in that, In S600, the following steps are specifically included: S610, comprehensive risk quantification: generating risk scores by fusing abnormal confidence and trend prediction; S620, multi-level warning classification: dividing response levels based on standards; S630, spatial path planning: generating the optimal inspection path for individual abnormalities; S640, intervention strategy generation: generating treatment plans through knowledge graph reasoning; S650, online strategy optimization: updating the decision model based on Q-learning.

10. An Internet of Things-based animal physiological feature data acquisition system for acquiring data in an Internet of Things-based animal physiological feature data analysis method according to any one of claims 1-9, comprising: a multi-source sensing module (101): real-time acquisition of 7 types of physiological indicators including body temperature, activity amount, rumination duration, feed intake, panting frequency, longitude, and latitude, synchronous acquisition of GPS / Beidou dual-mode positioning data, and built-in three-axis accelerometer for monitoring motion behavior; an edge preprocessing module (102): performing preliminary calculations such as data denoising, missing value interpolation, and sliding window statistics on the terminal device to reduce wireless transmission data volume; a space-time alignment gateway (103): unifying the time reference of each sensor through a Beidou time chip, aligning different sampling rate data streams using an adaptive interpolation algorithm, and correcting the coordinate offset of the geographic fence; a computing fusion node (104): deploying an attention mechanism fusion algorithm to integrate physiological indicators and motion trajectory data and generating a 128-dimensional joint feature vector; a cloud analysis engine (105): performing deep calculations such as Mahalanobis distance anomaly detection, LSTM trend prediction, and dynamic risk assessment, and maintaining a health baseline model; a decision execution controller (106): automatically triggering ventilation adjustment, feeding machine start-stop equipment, and generating an inspection path to push to the execution device according to the warning level; a low-power communication unit (107): supporting LoRa / NB-IoT dual-mode transmission, achieving global coverage of the pasture, and ensuring good data communication rate in harsh environments.

Images