Intelligent health monitoring and precise intervention support system for dairy cow in perinatal period
By using multimodal data acquisition and intelligent decision-making algorithms, the problems of lag and subjectivity in peripartum health monitoring of dairy cows have been solved, enabling early disease warning and precise intervention, and improving the accuracy and efficiency of dairy cow health management.
Patent Information
- Application Number
- CN202511312304.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-15
- Publication Date
- 2025-12-16
AI Technical Summary
Existing technologies cannot fully reflect the dynamic changes in the health of dairy cows during the peripartum period. Human judgment is lagging and highly subjective, lacking real-time data processing and intelligent decision-making, resulting in limited accuracy in disease prediction and individualized intervention capabilities, making it difficult to adapt to individual differences.
By employing multimodal data acquisition, standardization and smoothing, combined with logistic regression risk assessment and multi-objective optimization algorithms, individualized intervention plans are generated. The system adaptability is optimized through a closed-loop feedback mechanism to achieve early disease warning and precise intervention.
It enables real-time monitoring and analysis of multi-source health data of dairy cows during the peripartum period, improving the accuracy of disease risk prediction and the targeting of interventions, and reducing disease incidence and economic losses.
Smart Images
Figure CN121148744A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of intelligent livestock breeding, in particular to an intelligent health monitoring and precise intervention support system for dairy cows in the peripartum period. BACKGROUND
[0002] The peripartum period (usually referring to three weeks before and after delivery) is a critical stage for the health management of dairy cows. During this period, the metabolism, immune system and endocrine state of dairy cows will change significantly, and the energy demand will increase dramatically, while the feed intake is often insufficient, which can easily lead to negative energy balance, and further cause a series of metabolic and infectious diseases, such as ketosis, hypocalcemia, mastitis, retained placenta, etc. These diseases not only affect the lactation performance and reproductive ability, but also increase the culling rate and treatment cost, which has a significant impact on the economic benefits of the farm.
[0003] Currently, the management methods commonly used by farms are still mainly based on manual inspection and regular detection. Some large-scale farms have introduced single-type monitoring devices, such as electronic rumination counters, automatic feed intake meters or environmental sensors. However, these single data sources cannot fully reflect the dynamic changes of dairy cow health, and the manual experience judgment is lagging and subjective, often discovering the disease when it has entered the clinical stage, missing the best intervention opportunity. Although some studies have attempted to integrate multiple sensor data, there is a lack of mature real-time data processing and intelligent decision-making mechanism, and the prediction accuracy, individualized intervention ability and closed-loop management ability of the system are limited.
[0004] In addition, the health status of dairy cows in the peripartum period changes rapidly and fluctuates greatly, and the abnormality of a single indicator cannot reliably indicate the disease risk, which must rely on comprehensive analysis and trend prediction of multi-modal data. At the same time, different dairy cows have differences in constitution, production stage, feeding environment, etc., and fixed threshold judgment method is difficult to adapt to individualized needs.
[0005] Therefore, there is an urgent need for an intelligent health monitoring and precise intervention support system for dairy cows in the peripartum period to solve the above problems. SUMMARY
[0006] The present application aims to solve the technical problems raised in the background art, and provides an intelligent health monitoring and precise intervention support system for dairy cows in the peripartum period. The above purpose of the present application is achieved as follows:
[0007] The intelligent health monitoring and precise intervention support system for dairy cows in the peripartum period comprises a data acquisition module: for acquiring physiological parameters, behavior parameters and environmental parameters of dairy cows, the physiological parameters include body temperature T bheart rate HR, respiratory rate RR; the behavior parameters include rumination time RT, feed intake FI, step count ST, lying time LT; the environmental parameters include ambient temperature T e , relative humidity H e , ammonia concentration NH3
[0008] The data preprocessing and feature construction module is used for constructing a standardized feature vector z k (t) and a smooth feature M k (t) based on the collected multi-modal data and extracting a health-related feature set φ(t);
[0009] The health index calculation module is used for calculating a comprehensive health index HI(t) based on the features, wherein the health index HI(t) contains weighted scores of each physiological and behavior sub-index and environmental factor scores, and a weight vector is w = [w1, …, w K , e n];
[0010] The risk assessment module is used for calculating a disease risk probability p(t) based on the health index HI(t) and its variation features;
[0011] The intervention decision module is used for generating an intervention vector u(t) based on the risk probability p(t); wherein u N is a nutritional adjustment amount, u D is a drug dosage, and u E is an environmental control amount;
[0012] The execution and feedback module is used for delivering instructions corresponding to the intervention vector to an execution device and updating parameters of the health index calculation module and the risk assessment module based on intervention effects;
[0013] The user interaction module is used for displaying an individual health record, a real-time monitoring curve, alarm information and intervention suggestions.
[0014] As a preferred technical solution of the present application, the data preprocessing and feature construction module obtains the following by individual baseline calibration and standardization calculation:
[0015]
[0016] wherein: x k (t) is an original value of the kth collected parameter at time t; μ k is a mean value of the parameter in the prenatal baseline period; σ k is a standard deviation of the parameter in the prenatal baseline period; z k (t) is a standardized feature value of the kth parameter.
[0017] As a preferred technical solution of the present application, the smooth feature M k(t) the standardized feature z k (t) is obtained by sliding average:
[0018]
[0019] wherein: M k (t): the smoothed feature value of the kth channel at time t; L: sliding window length; z k (t-i): the standardized feature value at time t-i.
[0020] As a preferred technical solution of the present application, the health index calculation module converts the smoothed feature M k (t) of each channel into a sub-score f k (t), and obtains a weighted sum:
[0021]
[0022] wherein: HI(t): comprehensive health index; K: number of physiological and behavioral channels participating in calculation; w k : weight coefficient of the kth channel; w e : weight of environmental factors; f k (t): health score of the kth channel, f k (t) ranges from 0 to 1; f e (T e , H e , NH3): environmental score calculated based on environmental temperature, humidity and ammonia concentration, f e (T e , H e , NH3) ranges from 0 to 1.
[0023] As a preferred technical solution of the present application, the single-channel health score f k (t) adopts a Gaussian-type penalty function:
[0024]
[0025] wherein: r k : reference value of the kth channel; γ k : sensitivity parameter; M k (t): smoothed feature value of the kth channel.
[0026] As a preferred technical solution of the present application, the disease risk probability p(t) of the risk assessment module is calculated by a logistic regression model:
[0027]
[0028] Wherein: p(t): disease risk probability (0-1 interval); θ0: logistic regression bias term; θ: logistic regression weight vector;
[0029] φ(t): feature vector
[0030] ΔHI(t) = HI(t) - HI(t-1): health index change amount;
[0031] Health index change rate.
[0032] As a preferred technical solution of the application, the optimal intervention vector u of the intervention decision module is Determined by normalizing the multi-objective optimization model:
[0033]
[0034] Wherein: Normalized predicted risk probability; First-order approximation of risk probability; p max = 1: maximum risk probability; Normalized intervention cost; Normalized welfare score; B: intervention sensitivity matrix; λ1, λ2, λ3: trade-off coefficients; τ: prediction time span.
[0035] As a preferred technical solution of the application, the nutrition adjustment amount u N Calculated from energy balance:
[0036] ΔNE(t) = NEL req (t) - FI(t)·d NE
[0037]
[0038] Wherein: ΔNE(t): net energy gap (MJIB); NEL req (t): net energy required for lactation and maintenance; FI(t): feed intake; d NE : net energy density of the ration; e add : additive net energy value; ρ add : energy conversion coefficient; u N : nutrition adjustment amount.
[0039] As a preferred technical solution of the application, the execution and feedback module adopts online parameter update:
[0040] θ (t+1) = θ (t) + η [y(t) - p(t)] φ(t)
[0041]
[0042] where: θ(t): the logistic regression parameter vector; η: learning rate; ·y(t): true label (0 or 1); w (t) : health index weight vector; η w : weight update learning rate; loss function; Π Δ : projection operator to simplex set Δ; Δ: weight constraint set {w≥0,∑w k +w e =1}.
[0043] As a preferred technical solution of the present application, the user interaction module allows setting a health index threshold H th and a risk probability threshold p th , both of which are determined by maximizing the Youden index:
[0044]
[0045] where: H th : health index alarm threshold; p th : risk probability alarm threshold; TPR: true rate; FPR: false positive rate: optimal threshold pair.
[0046] Beneficial effects: the present application can realize real-time collection and fusion analysis of perinatal cow multi-source health data, through the synergistic monitoring of physiological and behavioral indicators such as body temperature, heart rate, respiratory rate, rumination time, feed intake, step count, and lying time, as well as environmental parameters such as environmental temperature, humidity, and ammonia concentration, using standardization and smoothing processing to reduce noise interference, and quantifying the individual health status based on the weighted health index, thereby overcoming the problems of relying on a single indicator, monitoring lag, and strong subjectivity of artificial judgment in the prior art, and realizing early warning of disease risk.
[0047] The present application introduces a logistic regression risk assessment model and a normalized multi-objective optimization algorithm, uses the health index, change trend, and standardized feature vector as risk prediction input, outputs the disease risk probability, and combines the intervention sensitivity matrix and the energy balance formula to generate an individualized intervention plan covering the nutritional adjustment amount, drug dosage, and environmental control amount, so that the intervention measures are matched with individual differences and real-time health status, improving the pertinence and effectiveness of the intervention and reducing unnecessary resource consumption.
[0048] The application constructs a closed-loop architecture of "perception-analysis-decision-implementation-feedback", collects effect feedback data after intervention implementation, updates risk assessment model parameters and health index weights online, continuously optimizes prediction and intervention accuracy, and realizes system adaptive iteration. This closed-loop optimization mechanism enables the system to continuously improve the adaptability and generalization ability to different individuals and different environments in long-term operation, and reduce the incidence of perinatal diseases and economic losses of the ranch. Figure 1 The system block diagram of the application. DETAILED DESCRIPTION
[0049] In order to make the purpose, technical scheme and advantages of the application more clear and explicit, the application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application and do not limit the application.
[0050] Example one: the application provides an intelligent health monitoring and precise intervention support system for dairy cows in perinatal period. The system can collect multi-modal health data in real time for dairy cows about three weeks before and after delivery, calculate a comprehensive health index, predict disease risk, and generate a precise intervention plan. Through implementation and feedback, the system realizes adaptive optimization, thereby reducing the occurrence of common diseases in perinatal period and improving the production performance and health level of dairy cows.
[0051] The system mainly consists of a data acquisition module, a data preprocessing and feature construction module, a health index calculation module, a risk assessment module, an intervention decision module, an implementation and feedback module, and a user interaction module. The modules form a closed-loop structure of perception-analysis-decision-implementation-feedback according to data flow and control flow.
[0052] In actual application, the data acquisition module includes temperature and heart rate sensors, accelerometers and gyroscopes worn on the neck of dairy cows, and weight sensors installed in the feeding trough and environmental sensors installed in the cowshed. The sensors synchronously collect body temperature T b , heart rate HR, respiratory rate RR, rumination time RT, feed intake FI, step count ST, lying time LT, environmental temperature T e , environmental humidity H e , and ammonia concentration NH3, and transmit them to the data preprocessing module at a sampling interval Δt.
[0053] The data preprocessing and feature construction module first calculates the baseline mean μ k and standard deviation σ k of each parameter during the baseline period before delivery (e.g. 14 days before delivery) to standardize the collected data and obtain
[0054]
[0055] where x k (t) is the original value at time t, z k (t) is the normalized feature value. Then, to reduce the impact of short-term fluctuations, the normalized features are smoothed by a moving average, resulting in
[0056]
[0057] where L is the length of the moving window, for example, L = 6 means smoothing the data within 30 minutes.
[0058] The health index calculation module converts the smoothed feature M k (t) of each channel into a health score f k (t) using a Gaussian-type penalty function:
[0059]
[0060] where r k is the reference value of the index, γ k is the sensitivity parameter, and the environmental parameters are converted into environmental scores by the function f e (T e , H e , NH3).
[0061] The comprehensive health index is obtained by weighted summation:
[0062]
[0063] The weights w = [w1, …, w K , w e ] satisfy The health index HI(t) varies between 0 and 1, and the lower the value, the worse the health status.
[0064]
[0065] The risk assessment module constructs a feature vector
[0066]
[0067] where ΔHI(t) = HI(t) - HI(t-1), The logistic regression model is used to calculate the disease risk probability:
[0068]
[0069] When p(t) is higher than the threshold p th , the system determines that the cow is in a high-risk state. The intervention decision module optimizes the intervention vector where u N is the nutritional adjustment amount, u D is the drug dosage, u E is the environmental control amount. The optimization problem is defined as:
[0070]
[0071] where,
[0072]
[0073] λ1, λ2, λ3 are trade-off coefficients, and B is the intervention sensitivity matrix.
[0074] where the nutritional adjustment amount u N is calculated by the energy balance model:
[0075] ΔNE(t) = NEL req (t) - FI(t) · d NE
[0076]
[0077] NEL req (t) is the net energy required for lactation and maintenance, d N E is the net energy density of the diet, e add is the net energy value of the additive, and ρ add is the energy conversion coefficient.
[0078] The execution and feedback module sends the intervention vector command to the automatic feeding equipment, the spraying system or the drug dispenser, collects feedback data after execution, and is used for online updating of the logistic regression parameters and the health index weight:
[0079] θ (t+1) = θ (t) + η [y(t) - p(t)] φ(t)
[0080]
[0081] where y(t) is the true label (healthy / diseased), Π Δ is the projection operator.
[0082] The user interaction module displays the health index curve, risk probability trend, intervention record and effect feedback of each cow on the interface, and allows setting alarm thresholds (H th , p th ), which are determined by maximizing the Youden index:
[0083]
[0084] In actual operation, for example, when the system detects that the health index of a dairy cow has dropped from 0.82 to 0.65 for three consecutive days, accompanied by a rise in p(t) to 0.72, which is higher than the set p... th If the value is 0.7, the intervention decision module will output a value containing u. N =1.5lg / day additive adjustment, u D For oral electrolyte rehydration solution, u E As an intervention vector to enhance ventilation, after the implementing device completes the intervention, the feedback data is used to update the model parameters to improve the accuracy of the next prediction and decision.
[0085] Example 2: In a modern dairy farm with 500 cows, a peripartum cow 12 days before calving was selected as the monitoring subject. The system deployed wearable neck sensors, feed trough weight sensors, and temperature, humidity, and ammonia concentration sensors in the cowshed. Body temperature T was collected at fixed sampling intervals Δt. b Heart rate (HR), respiratory rate (RR), rumination time (RT), food intake (FI), steps (ST), lying down time (LT), ambient temperature (T) e Humidity H e Multimodal data, including ammonia concentration (NH3).
[0086] At a certain sampling time, the cow's body temperature was 39.2℃, heart rate was 80 beats / min, respiratory rate was 30 breaths / min, rumination time was 420 minutes / day, feed intake was 18.0 kg / day, steps were 2200 steps / day, lying time was 12.5 hours / day, ambient temperature was 25.0℃, humidity was 65%, and ammonia concentration was 12 ppm.
[0087] The cow's average body temperature in historical data from the pre-partum baseline period (past fourteen days) was... The temperature was 38.8℃, with a standard deviation of 100°C. It is 0.3℃.
[0088] This is used to calculate standardized features. Similarly, standardized values z were calculated for parameters such as feed intake, rumination duration, heart rate, and respiratory rate. k (t).
[0089] Subsequently, these standardized values were averaged, assuming that the body temperature was at the past five sampling times. The smoothing values are 1.10, 1.20, 1.25, 1.35, and 1.30 respectively.
[0090] The system converts smooth features into health scores, and the body temperature channel uses... Reference value Sensitivity parameters The score is approximately 0.285.
[0091] Scores for feed intake, rumination duration, and heart rate were calculated sequentially, with assumed results of 0.750, 0.800, and 0.600, respectively. Environmental parameters were then processed using the function f. e (T e H e The calculated score for NH3 is 0.850.
[0092] The scores of each channel are weighted. w FI =0.25, w RT =0.25, w RT =0.25, w HR =0.15, w e The weighted summation of 0.10 yields a comprehensive health index HI(t)≈0.634, indicating that the health of the dairy cow has shown a downward trend.
[0093] The risk assessment module constructs feature vectors:
[0094]
[0095] Where ΔHI(t) = 0.634 - 0.680 = -0.046, (Time unit is two hours).
[0096] Substitute the feature vectors into the logistic regression formula Assuming model parameters θ0 = -1.0 and θ = [-2.0, -1.0, -0.5, 0.2, 0.2, 0.1, 0.1, 0.1], the calculated disease risk probability p(t) ≈ 0.721, which exceeds the threshold p. th =0.70
[0097] The intervention decision-making module uses a normalized multi-objective optimization model. Among them, predicting risks Normalized cost C n =0.40, Normalized Welfare Score Q n =0.80, with tradeoff coefficients λ1=0.5, λ2=0.3, λ3=0.2, resulting in J(u)≈0.3205. The intervention vector output by the optimizer is u. N = 1.5 kg / day of nutritional supplements, u D For 20 grams of oral electrolyte rehydration solution, u E Turn on the ventilation system for two hours.
[0098] Nutritional adjustment amounts are calculated using the energy balance formula, assuming the cow's net energy requirement (NEL) is...req = 160 MJ / day, daily feed intake FI(t) = 18.0 kg / day, daily ration net energy density d NE = 6.5 MJ / kg, additive net energy value e add = 10 MJ / kg, conversion factor p add = 0.95, then the net energy gap ANE(t) = 160 - 18 x 6.5 = 43 MJ / day, calculate u N = (43 / 10) x 0.95 ≈ 4.085 kg / day, due to the limitation of equipment capacity, the system will be clipped to 1.5 kg / day to match the implementable range of the execution module.
[0099] After the intervention is executed for twenty-four hours, the system re-collects the data of the cow, calculates the new HI(t+1) and p(t+1), and updates the parameters online θ (t+1) = θ (t) + η[y(t) - p(t)]φt) and so as to improve the accuracy of prediction and decision.
[0100] Finally, the user interaction module displays the health index curve, the change of risk probability and the intervention effect on the visual interface, and allows the user to set a new H th and p th to optimize the future alarm performance.
[0101] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. An intelligent health monitoring and precision intervention support system for dairy cows during the peripartum period, characterized in that, include: Data acquisition module: used to collect physiological, behavioral, and environmental parameters of dairy cows, including body temperature (T). b Heart rate (HR), respiratory rate (RR); behavioral parameters include rumination duration (RT), feed intake (FI), steps (ST), and recumbent time (LT); environmental parameters include ambient temperature (T). e Relative humidity H e Ammonia concentration (NH3); Data preprocessing and feature construction module: used to construct a standardized feature vector z based on the collected multimodal data. k (t) and smoothing feature M k (t), and extract the health-related feature set φ(t); Health Index Calculation Module: Used to calculate the comprehensive health index HI(t) based on the aforementioned features, which includes the weighted scores of each physiological and behavioral sub-indicator and environmental factor scores, with the weight vector being w = [w1, ..., w K ,w e ]; Risk assessment module: used to calculate the disease risk probability p(t) based on the health index HI(t) and its changing characteristics; Intervention decision module: used to generate intervention vectors based on risk probability p(t). Where u N For nutritional adjustment, u D For drug dosage, u E For environmental control quantities; Execution and feedback module: used to transmit the instructions corresponding to the intervention vector to the execution device, and update the parameters of the health index calculation module and the risk assessment module based on the intervention effect; User interaction module: used to display individual health records, real-time monitoring curves, alarm information and intervention suggestions.
2. The intelligent health monitoring and precision intervention support system for dairy cows during the peripartum period according to claim 1, characterized in that, The data preprocessing and feature construction module obtains the following through individual baseline calibration and standardization calculation: Where: x k (t): The original value of the k-th acquired parameter at time t; μ k : The mean of this parameter at the prenatal baseline period; σ k : The standard deviation of this parameter at the prenatal baseline period; z k (t): The standardized eigenvalue of the k-th parameter.
3. The intelligent health monitoring and precision intervention support system for dairy cows during the peripartum period according to claim 1, characterized in that, The smoothing feature M k (t) is derived from the standardized feature z k (t) is obtained through moving average: Where: M k (t): The smoothed feature value of the k-th channel at time t; L: The sliding window length; z k (ti): Standardized eigenvalue at time ti.
4. The intelligent health monitoring and precision intervention support system for dairy cows during the peripartum period according to claim 1, characterized in that, The health index calculation module calculates the health index by using the smoothing feature M of each channel. k (t) is converted to a sub-score f k (t), and by weighted summation, we obtain: Where: HI(t): comprehensive health index; K: number of physiological and behavioral channels involved in the calculation; w k : The weight coefficient of the k-th channel; w e : Environmental factor weights; f k (t): Health score of the k-th channel, f k (t) takes values in the range of 0-1; f e (T e H e NH3): Environmental score calculated based on ambient temperature, humidity, and ammonia concentration, f e (T e H e The value of NH3 is in the range of 0-1.
5. The intelligent health monitoring and precision intervention support system for dairy cows during the peripartum period according to claim 4, characterized in that, The single-channel health score f k (t) Using a Gaussian penalty function: Where: r k : Reference value for the k-th channel; γ k Sensitivity parameter; M k (t): The smoothed feature value of the k-th channel.
6. The intelligent health monitoring and precision intervention support system for dairy cows during the peripartum period according to claim 1, characterized in that, The disease risk probability p(t) of the risk assessment module is calculated by a logistic regression model: Where: p(t): disease risk probability (0-1 interval); θ0: logistic regression bias term; θ: logistic regression weight vector; φ(t): Eigenvector ΔHI(t) = HI(t) - HI(t-1): Change in health index; Rate of change in health index.
7. The intelligent health monitoring and precision intervention support system for dairy cows during the peripartum period according to claim 1, characterized in that, The optimal intervention vector of the intervention decision module Determined through a normalized multi-objective optimization model: in: Normalized prediction of risk probability; First-order approximation of risk probability; p max =1: Maximum risk probability; Normalized intervention costs; Normalized welfare score; B: Intervention sensitivity matrix; λ1, λ2, λ3: Trade-off coefficients; τ: Prediction time span.
8. The intelligent health monitoring and precision intervention support system for dairy cows during the peripartum period according to claim 1, characterized in that, The nutritional adjustment amount u N Calculated from energy balance: ΔNE(t)=NEL req (t)-FI(t)·d NE Where: ΔNE(t): Net energy gap (MJIB); NEL req (t): Net energy required for lactation and maintenance; FI(t): Feed intake; d NE Net energy density of the diet; e add Net energy value of additives; ρ add Energy conversion coefficient; u N Nutritional adjustment amount.
9. The intelligent health monitoring and precision intervention support system for dairy cows during the peripartum period according to claim 1, characterized in that, The execution and feedback module uses online parameter updates: i (t+1) =θ (t) +η[y(t)-p(t)]φ(t) Where: θ (t) : Logistic regression parameter vector; η: learning rate; ·y(t): true label (0 or 1); w (t) : Health index weight vector; η w : Weight update learning rate; Loss function; Π Δ : Projection operator to the simplex set Δ; Δ: Weight constraint set {w≥0,∑w k +w e =1}.
10. The intelligent health monitoring and precision intervention support system for dairy cows during the peripartum period according to claim 1, characterized in that, The user interaction module allows setting a health index threshold H. th With risk probability threshold p th The two are determined by maximizing the Youden exponent: Wherein: H th Health index alarm threshold; p th Risk probability alarm threshold; TPR: True Positive Rate; FPR: False Positive Rate Optimal threshold pair.