Ruminant Monitoring System Based on Rumen Capsule Sensors and Artificial Intelligence

Abstract

This invention discloses a ruminant monitoring system based on rumen capsule sensors and artificial intelligence, comprising: a rumen capsule sensing module, embedded in the rumen, for collecting multimodal physiological data; a wireless communication processing device for transmitting data according to the data acquisition status; a service platform for preprocessing the data, using a multimodal health prediction algorithm to determine the predicted health status, and identifying the appropriate response end; an early warning module for generating early warning messages and adjusting the transmission decision; and the service platform also adjusts the sensor acquisition mode based on feedback results. This invention achieves early disease prediction with a low false alarm rate by using rumen capsules to stably collect rumen internal physiological parameters and / or combine them with behavioral status data, performing fusion weight analysis and trend evolution prediction, and outputting potential disease types, risk probabilities, and lesion evolution trends. It also forms a closed-loop automated management system for data acquisition, analysis, early warning, and feedback, improving the accuracy and long-term reliability of ruminant health monitoring.

Description

Technical Field

[0001] This invention relates to the field of intelligent animal management technology, and in particular to a ruminant monitoring system based on rumen capsule sensors and artificial intelligence. Background Technology

[0002] With the development of large-scale farming, common diseases in ruminants such as cattle and sheep (such as acidosis, ketosis, heat stress, and digestive disorders) seriously affect farming efficiency. Currently, health monitoring of ruminants mainly relies on external behavioral monitoring, manual inspections, and surface examinations; however, these methods have the following technical limitations:

[0003] They cannot accurately reflect the physiological state of ruminants. Surface sensors can only collect external signals such as skin temperature and activity level, but cannot obtain key digestive and metabolic indicators such as rumen temperature, pH value, rumination rhythm, rumen pressure, rumen humidity, and gas concentration, leading to a serious lag in disease diagnosis.

[0004] The data is unstable and easily affected by environmental factors. Data such as body surface temperature and activity level can fluctuate drastically due to external factors such as climate, humidity, water intake, and feed consumption, making it difficult to stably reflect the true health status of animals and resulting in low reliability of monitoring results.

[0005] Current monitoring methods lack multimodal data fusion and intelligent analysis capabilities, making it impossible to effectively combine in vivo physiological parameters with in vitro behavioral status data. Even in the few programs that use rumen capsules with a single physiological indicator, only simple threshold judgments can be made, failing to comprehensively utilize multidimensional information for early disease prediction.

[0006] The entire monitoring process relies heavily on manual operation, lacking fully automated closed-loop management from data collection, transmission, intelligent analysis to early warning feedback. Manual inspections and threshold setting are not only inefficient but also ignore the multifactorial causes and long-term dynamic evolution of diseases, leading to high false alarm and false negative rates, thus failing to achieve early intervention and long-term reliable monitoring.

[0007] Therefore, existing methods for monitoring the health of ruminants cannot provide accurate, reliable, and long-term effective monitoring results. There is an urgent need for a system that can achieve multimodal data fusion, intelligent analysis, and automated closed-loop management. Summary of the Invention

[0008] To address any or more of the above technical problems, embodiments of the present invention provide a ruminant monitoring system based on a rumen capsule sensor and artificial intelligence.

[0009] The ruminant monitoring system based on rumen capsule sensors and artificial intelligence provided in this invention includes:

[0010] The rumen capsule sensing module, built into the rumen of a ruminant, includes several sub-sensors and a controller. The sub-sensors are used to collect multimodal physiological data of the ruminant; the controller is used to adjust the data acquisition mode of each sub-sensor based on the real-time data acquisition status of each sub-sensor.

[0011] A wireless communication processing device is used to send the multimodal physiological data of the ruminant to a service platform according to the collected status of the multimodal physiological data.

[0012] The service platform is used to preprocess multimodal physiological data and / or behavioral state data of ruminants to obtain a set of effective physiological data and / or a set of effective behavioral state data; using a multimodal health prediction algorithm, the platform determines the health status prediction result of the ruminants based on the set of effective physiological data and / or the set of effective behavioral state data; and identifies the adaptive response end of the ruminants based on the health status prediction result.

[0013] The early warning module is used to generate early warning messages corresponding to the health status prediction results, and adjust the decision to send early warning messages to the adaptation response end based on the actual operation of the adaptation response end of the ruminant.

[0014] The service platform is also used to control the controller to adjust the data acquisition mode of the sub-sensors based on the feedback result of the received early warning message from the adapter response end.

[0015] Preferably, the plurality of sub-sensors includes at least two of the following: temperature sub-sensor, pH sub-sensor, acceleration sub-sensor, pressure sub-sensor, humidity sub-sensor, and gas concentration sub-sensor.

[0016] The controller is used to adjust the data acquisition mode of each sub-sensor based on the real-time data acquisition status of each sub-sensor, including:

[0017] An initial sampling mode is set for each sub-sensor, and the acquisition operation of each sub-sensor is controlled to collect multimodal physiological data of ruminants; wherein, the multimodal physiological data includes at least two of the following: rumen temperature data, rumen pH data, ruminant action data, rumen pressure data, rumen humidity data, and gas concentration data of ruminants.

[0018] Obtain the actual data acquisition load of each sub-sensor. Based on the actual data acquisition load and the data acquisition limit load of each sensor, determine whether each sub-sensor is in an overload operation state. If so, reduce the data acquisition frequency of the sub-sensor; otherwise, maintain the data acquisition frequency of the sub-sensor.

[0019] Preferably, the wireless communication processing device is used to send the multimodal physiological data of the ruminant to the service platform according to the collected status of the multimodal physiological data, including:

[0020] Based on the total amount of multimodal physiological data collected, determine whether data acquisition saturation has occurred, or whether the current time has reached the preset transmission time, or whether the data value collected by any sub-sensor exceeds the preset normal physiological value range corresponding to that sub-sensor;

[0021] If any judgment result is yes, then send the currently collected multimodal physiological data to the service platform;

[0022] If all judgments result in no, then multimodal physiological data will not be sent to the service platform for the time being.

[0023] Preferably, the service platform is used to preprocess multimodal physiological data and / or behavioral state data of ruminants to obtain a valid set of physiological data and / or a valid set of behavioral state data, including:

[0024] Missing and outlier data were removed from the multimodal physiological and / or behavioral data of ruminants, and timestamps were aligned to obtain a valid set of physiological and / or behavioral data.

[0025] Preferably, a multimodal health prediction algorithm is used to determine the health status prediction result of the ruminant based on the effective physiological data set and / or effective behavioral state data set, including:

[0026] A multimodal health prediction algorithm is used to perform multi-scale data fusion weight analysis on the effective physiological data set and / or the effective behavioral state data set to obtain the physiological state change trend of the ruminant; based on the physiological state change trend, a health status prediction result of the ruminant is generated; wherein, the health status prediction result includes any one or more of the ruminant's potential disease type, disease occurrence risk probability, and lesion evolution trend.

[0027] Preferably, a multimodal health prediction algorithm is used to determine the health status prediction result of the ruminant based on the effective physiological data set and / or effective behavioral state data set, including:

[0028] Step A: Perform dimensionless normalization on each type of physiological data at the current time t in the effective physiological data set to obtain the normalized value of each type of physiological data at the current time t; perform dimensionless normalization on each type of behavioral state data at the current time t in the effective behavioral state data set to obtain the normalized value of each type of behavioral state data at the current time t.

[0029] Step B: Using the hyperbolic tangent function, determine the physiological modal tension coefficient at the current time t based on the normalized value of each type of physiological data and the number of physiological data types. ;

[0030] The behavioral modal tension coefficient is determined based on the normalized value of each behavioral state data, the number of behavioral state data types, and the median of the normalized value of each behavioral state data within a preset time window before the current time t. ;

[0031] Step C: Based on the physiological modal tension coefficient The behavioral modal tension coefficient The variances of the normalized values ​​of all physiological data at time t and the variances of the normalized values ​​of all behavioral state data at time t are used to determine the physiological modality fusion weights. Weights for fusion with behavioral modalities ;

[0032] Step D1: Based on the physiological modality fusion weights Behavioral modality fusion weights The physiological modal tension coefficient The behavioral modal tension coefficient The rate of change of the physiological state trend scalar at the previous moment is used to determine the physiological state trend scalar at the current moment t. ;

[0033] Step D2, at the current moment Based on whether this moment is the first sampling moment after the system starts, determine respectively The possible values ​​of , where: if Then calculate all normalized physiological data in the initial period after implantation. absolute maximum value ;like Then let Otherwise set For positive constants less than 0.1; if Then, based on the severity index stored in the previous time step... and the scalar of the physiological state change trend at the current time t Determine the current severity index at time t. ;

[0034] Step E, for each potential disease According to the current severity index The scalar of the trend of the physiological state change Future predicted values, disease progression rate coefficient Disease initiation threshold and predicted duration and discrete time step By using discrete summation and exponential functions, future moments can be determined. Severity Index To characterize the evolution trend of the lesion.

[0035] Preferably, the method of using a multimodal health prediction algorithm to determine the health status prediction result of the ruminant based on the effective physiological data set and / or effective behavioral state data set further includes:

[0036] Step F: For each potential disease, scalarize the physiological state change trend. The severity index of the current k-th disease KL divergence between the distribution of normalized values ​​of current physiological data and the healthy baseline The probability of developing each disease is determined using the Logistic function. ;

[0037] Step G: The service platform outputs at least one disease with the highest probability of occurrence as the potential disease type of the ruminant, and according to... The prediction results generate the corresponding disease evolution trend.

[0038] Preferably, the service platform identifies the adaptive response end of the ruminant based on the health status prediction results, including:

[0039] Based on the health status prediction results, the critical time points for the disease progression of each potential disease in the ruminant are estimated; wherein, the critical time points refer to the time points at which the disease progression of each potential disease to several target disease stages occurs.

[0040] Based on the key time points, determine the urgency of all potential diseases in the ruminant; based on the urgency and the type of potential disease, identify the adaptive response end of the ruminant.

[0041] Preferably, the early warning module generates an early warning message corresponding to the health status prediction result, and adjusts the decision to send early warning messages to the ruminant's adaptive response terminal based on the actual operation of the ruminant's adaptive response terminal, including:

[0042] The health status prediction results are semantically extracted and transformed to identify potential disease types and their disease evolution trends, generating health status semantic text information; the semantic text information is then packaged and encapsulated to generate a warning message.

[0043] The real-time busy / idle status of the adaptive response terminal is obtained, and the allowed message receiving period of the adaptive response terminal is determined according to the real-time busy / idle status of the adaptive response terminal; the decision to send warning messages to the adaptive response terminal is adjusted according to the time domain distribution of the allowed message receiving period; wherein, the sending decision includes the time interval for sending warning messages to the adaptive response terminal.

[0044] Preferably, the service platform controls the controller to adjust the data acquisition mode of the sub-sensors based on the feedback result of the received warning message from the adaptive response terminal, including:

[0045] The credibility identification result of the warning content in the warning message is extracted from the feedback result of the adaptation response end. Based on the credibility identification result, the controller is controlled to adjust the data acquisition mode of the sub-sensor.

[0046] The beneficial effects of the above-mentioned technical solutions provided in the embodiments of the present invention include at least the following:

[0047] This invention, by embedding a rumen capsule inside the rumen of ruminants, enables long-term, stable collection of multimodal physiological parameters such as temperature, pH, rumination behavior, air pressure, humidity, and gas concentration. This avoids the susceptibility of surface sensors to environmental interference, providing a reliable in vivo data foundation for disease prediction. Based on this, by combining rumen internal data and / or behavioral state data, and utilizing multimodal fusion weight analysis and trend evolution prediction models, the invention can quantify physiological state changes and output potential disease types, probability of occurrence, or disease evolution trends. This allows for early identification of common diseases such as acidosis, ketosis, and heat stress, enabling proactive intervention and reducing breeding losses. The service platform automatically identifies and adapts response terminals based on risk probability and urgency, intelligently schedules early warning message sending strategies based on the response terminal's workload, and dynamically adjusts sensor acquisition modes based on the degree of physiological data anomalies. This forms a closed-loop automated management system from data acquisition, transmission, intelligent analysis to early warning feedback, improving system reliability and reducing manual inspection workload.

[0048] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0049] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0050] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This is a schematic diagram of the ruminant monitoring system provided in an embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram illustrating the implementation effect of the ruminant monitoring system provided in this embodiment of the invention;

[0053] Figure 3 This is a first field implementation diagram of the ruminant monitoring system provided in this embodiment of the invention;

[0054] Figure 4 This is a second field implementation diagram of the ruminant monitoring system provided in this embodiment of the invention. Detailed Implementation

[0055] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0056] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," "outer," "far," "near," "front," and "rear," etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the accompanying drawings and are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0057] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0058] Please see Figure 1 , 2As shown, an embodiment of the present invention provides a ruminant monitoring system, comprising:

[0059] The rumen capsule sensing module, built into the rumen of a ruminant, includes several sub-sensors and a controller. The sub-sensors are used to collect multimodal physiological data of the ruminant; the controller is used to adjust the data acquisition mode of each sub-sensor based on the real-time data acquisition status of each sub-sensor.

[0060] A wireless communication processing device is used to send the multimodal physiological data of the ruminant to a service platform according to the collected status of the multimodal physiological data.

[0061] The service platform is used to preprocess multimodal physiological data and / or behavioral state data of ruminants to obtain a set of effective physiological data and / or a set of effective behavioral state data; using a multimodal health prediction algorithm, the platform determines the health status prediction result of the ruminants based on the set of effective physiological data and / or the set of effective behavioral state data; and identifies the adaptive response end of the ruminants based on the health status prediction result.

[0062] The early warning module is used to generate early warning messages corresponding to the health status prediction results, and adjust the decision to send early warning messages to the adaptation response end based on the actual operation of the adaptation response end of the ruminant.

[0063] The service platform is also used to control the controller to adjust the data acquisition mode of the sub-sensors based on the feedback result of the received early warning message from the adapter response end.

[0064] In this embodiment of the invention, the adapter response terminal refers to an external terminal device associated with ruminants and identified by the service platform based on the health status prediction results, including but not limited to veterinary terminals, rancher mobile terminals, automatic feeding control terminals, environmental regulation equipment control terminals, etc.

[0065] Specifically, the rumen capsule sensing module is configured to be implanted inside the rumen of ruminants (such as cattle and sheep). Its outer shell is made of acid- and corrosion-resistant biocompatible materials to ensure that long-term residence does not affect the animal's health. This module is an integrated device, with several sub-sensors and a controller integrated within the shell. The sub-sensors include at least two or more of the following: temperature sub-sensor, pH sub-sensor, acceleration sub-sensor, pressure sub-sensor, humidity sub-sensor, and gas concentration (such as methane) sub-sensor. These sub-sensors are used to collect multimodal physiological data of the ruminant in real time, specifically including rumen temperature data, rumen pH data, rumination movement data, rumen pressure data, rumen humidity data, and gas concentration data. The controller has a wireless communication connection with a wireless communication processing device installed in the breeding environment.

[0066] Behavioral data collection devices, such as machine vision cameras, electronic ear tags, and pedometers, can be deployed in the farm environment to collect behavioral data of ruminants, including but not limited to activity level, rumination duration, number of steps taken, and milk production.

[0067] The wireless communication processing device is installed in the breeding environment to acquire multimodal physiological data sent by the controller. Based on the collected status of the multimodal physiological data, it sends the data to the service platform. Simultaneously, it can also forward data from behavioral status data acquisition devices to the service platform.

[0068] The service platform is the data processing and analysis center of the entire system. It can be deployed in the cloud or on a local server to receive all data and perform preprocessing, health prediction, and adaptation of response terminal identifiers.

[0069] The early warning module communicates with the service platform, is responsible for generating and sending early warning messages, and adjusts the sending strategy according to the actual status of the responding end.

[0070] Figure 3 , 4 The diagram shown is a field implementation illustration of the technical solution of this invention.

[0071] The aforementioned technical solution, by constructing a complete closed-loop system comprising a rumen capsule sensing module, wireless communication, a service platform, and an early warning module, achieves fully automated management of the entire process from multimodal data acquisition, transmission, intelligent analysis to early warning feedback. This system, combining in vivo physiological parameters and / or in vitro behavioral status data, lays a data foundation for subsequent health prediction and enhances the systematic nature and reliability of ruminant health monitoring.

[0072] In one embodiment, the plurality of sub-sensors includes at least two of the following: a temperature sub-sensor, a pH sub-sensor, an acceleration sub-sensor, a pressure sub-sensor, a humidity sub-sensor, and a gas concentration sub-sensor.

[0073] The controller is used to adjust the data acquisition mode of each sub-sensor based on the real-time data acquisition status of each sub-sensor, including:

[0074] An initial sampling mode is set for each sub-sensor, and the acquisition operation of each sub-sensor is controlled to collect multimodal physiological data of ruminants; wherein, the multimodal physiological data includes at least two of the following: rumen temperature data, rumen pH data, ruminant action data, rumen pressure data, rumen humidity data, and gas concentration data of ruminants.

[0075] Obtain the actual data acquisition load of each sub-sensor. Based on the actual data acquisition load and the data acquisition limit load of each sensor, determine whether each sub-sensor is in an overload operation state. If so, reduce the data acquisition frequency of the sub-sensor; otherwise, maintain the data acquisition frequency of the sub-sensor.

[0076] In this embodiment, after the rumen capsule sensing module is fed into the rumen of a ruminant, the controller first sets an initial sampling mode for each sub-sensor. For example, the temperature sub-sensor samples once every 15 minutes, the pH sub-sensor samples once every hour, and the acceleration sub-sensor continuously collects data at a frequency of 50Hz.

[0077] The controller monitors the actual data acquisition load of each sub-sensor in real time (e.g., the amount of data generated in the past hour) and compares it with the preset data acquisition limit load (determined by the sensor's processing power and buffer size). The controller performs the following judgments:

[0078] If the actual load on a sub-sensor exceeds its limit (for example, acceleration data surges due to vigorous animal movement), it is determined to be operating under overload conditions. In this case, the controller reduces the data acquisition frequency of that sub-sensor (e.g., reducing the acceleration sampling frequency to 25Hz) to prevent data loss or system crash. If the actual load does not exceed the limit, the current acquisition frequency is maintained to ensure data integrity.

[0079] Through this adaptive adjustment, the rumen capsule sensing module can achieve long-term, stable, and reliable data acquisition under different physiological and activity states.

[0080] In this embodiment, the controller dynamically adjusts the sampling frequency according to the actual load of each sub-sensor, preventing data loss or system failure caused by sub-sensors operating under overload conditions. This adaptive mechanism ensures the long-term stable operation of the rumen capsule sensing module under different physiological and activity states of animals, extends the effective service life of the device, avoids invalid data redundancy, and improves the efficiency and reliability of data acquisition.

[0081] In one embodiment, the wireless communication processing device is used to send the multimodal physiological data of the ruminant to a service platform according to the acquired status of the multimodal physiological data, including:

[0082] Based on the total amount of multimodal physiological data collected, determine whether data acquisition saturation has occurred, or whether the current time has reached the preset transmission time, or whether the data value collected by any sub-sensor exceeds the preset normal physiological value range corresponding to that sub-sensor;

[0083] If any judgment result is yes, then send the currently collected multimodal physiological data to the service platform;

[0084] If all judgments are negative, multimodal physiological data will not be sent to the service platform until any judgment is positive, at which point the multimodal physiological data will be sent to the service platform.

[0085] In this embodiment, the wireless communication processing device does not transmit all data in real time, but makes judgments based on preset conditions to save energy consumption and communication bandwidth. The specific judgment logic is as follows:

[0086] Check whether the total amount of multimodal physiological data collected but not transmitted has reached or exceeded a preset buffer threshold (e.g., 80% of the buffer capacity). Alternatively, check whether the current time has reached a preset periodic transmission time point (e.g., transmission every 6 hours). Or, check whether the latest data value collected by any sub-sensor exceeds the preset normal physiological value range corresponding to that sensor.

[0087] If any of the above judgments is true, the wireless communication processing device will package and send all currently collected multimodal physiological data to the service platform. If all judgments are false, the data will not be sent and will continue to be cached locally.

[0088] In this embodiment, a multi-condition triggered non-real-time transmission strategy (data saturation, timing, and anomaly triggering) is adopted, which reduces the wireless communication frequency and power consumption, and extends the battery life of the rumen capsule. At the same time, the anomaly data triggering mechanism ensures that key anomaly information can be transmitted in a timely manner, avoiding missing the best intervention opportunity due to delayed transmission, thus balancing energy saving and real-time requirements.

[0089] In one embodiment, the service platform is used to preprocess multimodal physiological data and / or behavioral state data of ruminants to obtain a valid set of physiological data and / or a valid set of behavioral state data, including:

[0090] Missing and outlier data were removed from the multimodal physiological and / or behavioral data of ruminants, and timestamps were aligned to obtain a valid set of physiological and / or behavioral data.

[0091] Specifically, null and outlier values ​​(e.g., negative pH values ​​or values ​​greater than 14) caused by sensor noise or communication errors are detected and removed. Since different sub-sensors and behavior monitoring devices may have different sampling frequencies, the service platform uses linear interpolation or nearest neighbor interpolation methods based on the timestamps embedded in the data to unify all data onto the same time grid (e.g., at 0.25-hour intervals). The processed data is then stored in the valid physiological data set and the valid behavioral state data set, respectively.

[0092] In this embodiment, by removing missing and outlier values ​​and aligning timestamps, noise and interference in the original data are eliminated, ensuring the quality of the data input into the health prediction model. A high-quality, effective dataset directly improves the accuracy of subsequent multimodal fusion analysis and disease prediction, reducing false positive and false negative rates.

[0093] In one embodiment, a multimodal health prediction algorithm is used to determine the predicted health status of the ruminant based on the effective physiological data set and / or the effective behavioral state data set, including:

[0094] A multimodal health prediction algorithm is used to perform multi-scale data fusion weight analysis on the effective physiological data set and / or the effective behavioral state data set to obtain the physiological state change trend of the ruminant; based on the physiological state change trend, a health status prediction result of the ruminant is generated; wherein, the health status prediction result includes any one or more of the ruminant's potential disease type, disease occurrence risk probability, and lesion evolution trend.

[0095] Specifically, the service platform utilizes a pre-trained multimodal health prediction algorithm to analyze effective physiological data sets and / or effective behavioral state data sets. The algorithm first performs multi-scale fusion weight analysis on multi-source data to capture and quantify the physiological state change trends of ruminants. Based on this trend, the algorithm generates a comprehensive health status prediction result, which specifically includes:

[0096] Potential disease types: such as acidosis, ketosis, heat stress, etc.

[0097] Risk of disease occurrence: For example, the risk of developing acidosis within the next 24 hours is 75%.

[0098] Disease progression trend: For example, if the current condition continues, the ketosis severity index is predicted to increase by 0.4 after 3 days.

[0099] In this embodiment, a multimodal health prediction algorithm is used to analyze fused data, outputting specific and quantitative prediction results, including disease type, risk probability, and evolution trend. This multidimensional prediction not only informs farmers what diseases might occur but also provides crucial information such as the probability of occurrence and future development, thus advancing the disease intervention timeline and facilitating the development of precise prevention and treatment plans.

[0100] In another embodiment, determining the predicted health status of the ruminant based on the multimodal health prediction algorithm, the effective physiological data set, and / or the effective behavioral state data set can also be implemented as follows:

[0101] Step A: Perform dimensionless normalization on each type of physiological data at the current time t in the effective physiological data set to obtain the normalized value of each type of physiological data at the current time t; perform dimensionless normalization on each type of behavioral state data at the current time t in the effective behavioral state data set to obtain the normalized value of each type of behavioral state data at the current time t.

[0102] Specifically, the normalized value of the i-th type of physiological data at the current time t can be determined according to the following formula:

[0103]

[0104] in, This represents the original value of the i-th physiological data at the current time t; and These are the mean and standard deviation of the i-th physiological data during the healthy period of ruminants (usually the first 3 days after rumen capsule implantation), which are pre-calculated and stored by the service platform; This is the first zero-prevention constant, used to ensure that the denominator is not zero so that the calculation is meaningful. Its specific value can be set by those skilled in the art according to the actual calculation accuracy requirements, so as not to affect the accuracy of the calculation results. This is the normalized value of the i-th physiological data at the current time t;

[0105] Specifically, the normalized value of the j-th behavior state data at the current time t can be determined according to the following formula:

[0106]

[0107] in, This represents the original value of the j-th behavior state data at the current time t; and , respectively, are the mean and standard deviation of the j-th behavioral state data during the healthy period of ruminants; This is the normalized value of the j-th behavior state data at the current time t; This is the second zero-prevention constant, used to ensure that the denominator is not zero so that the calculation is meaningful. Its specific value can be set by those skilled in the art according to the actual calculation accuracy requirements, so as not to affect the accuracy of the calculation results.

[0108] The above formula converts raw data of different dimensions (temperature, pH value, activity level, etc.) into standardized scores centered on the mean of the healthy period, eliminating dimensional differences and ensuring that all subsequent calculations are performed on a dimensionless scale with similar numerical ranges, thus preventing a single data source from dominating the analysis results due to excessively large values.

[0109] Step B: Using the hyperbolic tangent function, determine the physiological modal tension coefficient at the current time t based on the normalized value of each type of physiological data and the number of physiological data types. ;

[0110] Specifically, The calculation formula can be:

[0111]

[0112] in, The number of physiological data types (its value is greater than or equal to 2); , which represents the change in the normalized value of physiological data at adjacent time points; The sampling time interval (determined by the sampling clock of the rumen capsule, for example, 0.25 hours); Used as a reference time constant, for example, 1 hour, to dimensionlessly measure the rate of change over time; To collect the first The personalized sensitivity coefficients of the sub-sensors for various physiological data range from 0.5 to 2.0. These values ​​are preset by the service platform based on the correlation between the sensor indicators and common diseases. For example, the temperature sub-sensor has a value of 1.2 (temperature changes are sensitive to diseases), the pH sub-sensor has a value of 1.5 (pH is a key indicator of acidosis, and changes need to be amplified), the acceleration sub-sensor (rumination) has a value of 0.8 (the amplitude of movement changes is small, so it should be appropriately compressed), and the pressure, humidity, and gas concentration sub-sensors all have a value of 1.0 (neutral). It is a hyperbolic tangent function, and its absolute value guarantees that the result is non-negative; The larger the value, the more serious the deviation of physiological data from a healthy state;

[0113] In the above formula, the squared term amplifies the degree of deviation, the hyperbolic tangent performs nonlinear compression on the rate of change, and the product reflects the overall abnormality of the current value and its changing trend. This formula can quantify the degree to which physiological data deviates from a healthy state in both amplitude and rate of change; a larger value indicates a more abnormal physiological condition inside the rumen.

[0114] The behavioral modal tension coefficient is determined based on the normalized value of each behavioral state data, the number of behavioral state data types, and the median of the normalized value of each behavioral state data within a preset time window before the current time t. ;

[0115] Specifically, it is determined according to the following formula. :

[0116]

[0117] in, The number of categories of behavioral state data (value equal to or greater than 1); For the first The median of the normalized value of the behavioral state data within a preset time window (e.g., a past 6-hour window) before the current time t is calculated by the service platform. The third zero-prevention constant is used to ensure that the denominator is not zero so that the calculation is meaningful. Its specific value can be set by those skilled in the art according to the actual calculation accuracy requirements, so as not to affect the accuracy of the calculation results. , where is the second-order difference (discrete acceleration) of the normalized values ​​of the behavioral state data, and . This is the normalized value of the j-th behavior state data from the previous time t-1. The normalized value of the j-th behavior state data at time t-2; For the first The decay coefficient for behavioral status data ranges from 0.5 to 2.0. It is obtained by the service platform based on the correlation between historical behavioral abnormalities and diseases. For example, the value of activity level and rumination duration is 1.2 (behavioral abnormalities are sensitive to diseases), and the value of walking steps and milk production is 0.8 (the changes are relatively gradual). If there is no need for detailed subdivision, it is also acceptable to take 1.0 for all behaviors. A higher value indicates more intense or abnormal behavior.

[0118] In the above formula, the first term measures the deviation of the current value from the historical median, and the second term uses an exponential function to convert acceleration into a factor between 0 and 1; the greater the acceleration, the closer the factor is to 1. This formula quantifies the degree of abnormal fluctuation in behavioral state data, especially sudden escalation or cessation of action.

[0119] Step C: Based on the physiological modal tension coefficient The behavioral modal tension coefficient The variances of the normalized values ​​of all physiological data at time t and the variances of the normalized values ​​of all behavioral state data at time t are used to determine the physiological modality fusion weights. Weights for fusion with behavioral modalities ;

[0120] Specifically, the calculation formula can be:

[0121]

[0122] ;

[0123] in, , The variance of all normalized physiological data at the current moment. The variance of all normalized behavioral state data at the current moment; The scaling factor is used to prevent the fusion weight from becoming too extreme when the tension coefficients are not significantly different; its value ranges from 1.5 to 4. And the sum is 1;

[0124] The above formula uses the proportion of variance in physiological data. After weighting the tension coefficient, the weights are assigned using a softmax-like function. A larger variance indicates more discrete physiological data, and the physiological modality receives a higher weight. This formula dynamically adjusts the contribution ratio of the physiological and behavioral modalities in the final trend scalar, enabling the model to adapt to the reliability of information under different states.

[0125] Step D1: Based on the physiological modality fusion weights Behavioral modality fusion weights The physiological modal tension coefficient The behavioral modal tension coefficient The rate of change of the physiological state trend scalar at the previous moment is used to determine the physiological state trend scalar at the current moment t. ;

[0126] Specifically, the calculation formula is as follows:

[0127]

[0128] in, ; The rate of change of the scalar measure of the trend of physiological state changes at the previous moment; This is the trend inertia coefficient, which ensures that historical trends have a moderate influence on the current value, avoiding excessive smoothing or oscillation. It is dimensionless and ranges from 0.1 to 0.5. It is the hyperbolic tangent function; A higher value indicates a greater likelihood of deterioration in health; scalar initial value of physiological state change trend. Both can be set to a fixed value, such as 0.5 (this value is the natural center point of the Sigmoid function when the input is zero, corresponding to a healthy neutral state), or set to any other equal value (such as 0.2), since the recursive formula only uses the difference between the two. When the two are equal, the initial absolute value does not affect any subsequent time step. ( The calculation results are only used as temporary historical placeholders for difference calculations; setting equal values ​​can avoid introducing non-zero inertial interference into the initial difference terms, thereby preventing the initial risk from being unreasonably increased.

[0129] The above formula integrates the tension coefficients of the two modes and adds the inertia correction of historical trends (tanh term). Finally, it is mapped to the (0,1) interval through the Sigmoid function to generate a health deterioration tendency index between 0 and 1. The closer the value is to 1, the more likely the current state is to develop in the direction of disease.

[0130] Step D2, at the current moment Based on whether this moment is the first sampling moment after the system starts, determine respectively The possible values ​​of are:

[0131] like Then calculate all normalized physiological data in the initial period after implantation (e.g., the first hour). absolute maximum value ;like Then let The value is 0.1; otherwise, set to 0.1. The value is a normal number less than 0.1, preferably a very small value such as 0.001, to ensure the numerical continuity of the recursive calculation, and this very small value does not have a substantial impact on the calculation of the risk probability in a healthy state.

[0132] like Then, based on the severity index stored in the previous time step... and the scalar of the physiological state change trend at the current time t Determine the current severity index at time t. ;

[0133] Specifically, it can be determined according to the following formula.

[0134]

[0135] in, For the first The progression rate coefficient (in units such as 1 / day) for each disease is obtained by fitting historical disease course data by the service platform. For example, acidosis (acute, rapid progression) is taken as 0.12 day⁻¹, ketosis (subacute) as 0.08 day⁻¹, heat stress (moderate) as 0.10, and other diseases (such as parasitic infections, chronic) as 0.06. That is, based on common disease course experience, the acute disease coefficient is high and the chronic disease coefficient is low. All diseases can also be uniformly taken as 0.10. For the first The initiation threshold for each disease is dimensionless and ranges from [0.5, 0.7]. It can be preset by the disease knowledge base based on the breed and age of the ruminant (e.g., acidosis 0.6, heat stress 0.55). Heat stress is usually triggered by ambient temperature, and it may be triggered by a slight change in physiological trend, so the threshold is low. Acidosis and ketosis require a more explicit deviation to be triggered. The fourth zero-prevention constant is used to ensure that the denominator is not zero so that the calculation is meaningful. Its specific value can be set by those skilled in the art according to the actual calculation accuracy requirements, so as not to affect the accuracy of the calculation results.

[0136] Truncation operation ensure Always in Within the range;

[0137] This step uniformly handles the determination of the severity index at the initial and subsequent times, ensuring that at each current time... The system can obtain a valid current severity index. This is used for future predictions in subsequent step E.

[0138] Step E, for each potential disease According to the current severity index The scalar of the trend of the physiological state change Future predicted values, disease progression rate coefficient Disease initiation threshold and predicted duration and discrete time step By using discrete summation and exponential functions, future moments can be determined. Severity Index To characterize the evolution trend of the lesion;

[0139] Specifically, The calculation formula is:

[0140]

[0141] in, For disease indexing; This represents the severity index of the current k-th disease, with a value range of [0,1].

[0142] The forecast duration (in days) is preset by the user (e.g., 3 days). Set the time step for discrete summation (in units such as days), for example, 0.1 days, to ensure computational accuracy; To find the number of terms in the summation; The predicted trend scalar value can be estimated by linear extrapolation or by holding the current value. If estimated by linear extrapolation, it can be determined as follows: take the current time... and the previous moment of Value, calculate the rate of change per unit time Then the future The predicted value at time is To prevent extrapolated values ​​from exceeding a reasonable range, they can be limited to a reasonable interval.

[0143] This is a severity index for prediction; the higher the value, the more severe the illness will be in the future.

[0144] In actual numerical calculations, the above prediction results can be truncated: To ensure that the severity index remains at a certain level Within the range.

[0145] The above formula applies when the predicted trend scalar exceeds the disease initiation threshold. At that time, according to the rate of progress Exponential accumulation occurs, otherwise growth ceases. The duration of each potential disease can be predicted in the future. The changes in the severity of the disease within the body provide a quantitative basis for determining the timing of disease intervention.

[0146] Step F: For each potential disease, scalarize the physiological state change trend. The severity index of the current k-th disease KL divergence between the distribution of normalized values ​​of current physiological data and the healthy baseline The probability of developing each disease is determined using the Logistic function. ;

[0147] Specifically, the calculation formula is as follows:

[0148]

[0149] in, The first logistic regression coefficient (range 1.0–5.0), the second logistic regression coefficient (range 1.0–6.0), and the third logistic regression coefficient (range 0.5–4.0) were obtained by the service platform through training on historical case data. The Kullback-Leibler divergence between the normalized distribution of current physiological data and the distribution of the healthy baseline is used to quantify the overall degree of deviation. In this embodiment, assuming that each physiological data point is independent and follows a Gaussian distribution, the KL divergence can be calculated using the following simplified formula:

[0150]

[0151] in, Let i be the normalized value of the i-th type of physiological data. The standard deviation of the healthy period, The standard deviation of this physiological data within a preset sliding time window (e.g., the past hour) before the current time t. The fifth zero-prevention constant is used to ensure that the denominator is not zero so that the calculation is meaningful. Its specific value can be set by those skilled in the art according to the actual calculation accuracy requirements, so as not to affect the accuracy of the calculation results.

[0152] Since the simplified KL divergence formula described above may produce a small negative value in numerical calculations due to distribution approximation errors, and the physical definition of KL divergence requires it to be non-negative, the larger of its value and zero is taken in actual calculations, i.e., the final value is... Those skilled in the art can also use other known KL divergence estimation methods, such as histogram-based or kernel density-based methods, all of which fall within the scope of protection of this invention.

[0153] A higher value indicates a higher probability of the occurrence of the k-th disease;

[0154] The above formula linearly combines the current trend scalar, the current severity index, and the overall distribution deviation (KL divergence), and then converts them into probability values ​​through the Logistic function. It outputs the probability of occurrence of each disease, with higher values ​​indicating a greater likelihood of disease. This is used to ultimately output potential disease types and warnings.

[0155] Finally, the service platform outputs at least one disease with the highest probability of occurrence as the potential disease type of the ruminant, and based on... The prediction results generate corresponding disease evolution trends (e.g., describing the change curve of severity index over time through linear fitting) and output them.

[0156] In this embodiment, the system acquires multiple physiological parameters inside the rumen through a rumen capsule sensor, combines them with behavioral state data, and utilizes multimodal fusion weight analysis and trend evolution prediction to output risk probability and severity index change curves before obvious clinical symptoms appear. This allows for earlier identification of diseases such as acidosis, ketosis, and heat stress compared to traditional methods, while also reducing the false alarm rate.

[0157] In one implementation, the service platform identifies the appropriate response endpoint of the ruminant based on the health status prediction results, which can be implemented as follows:

[0158] Based on the health status prediction results, the critical time points for the disease progression of each potential disease in the ruminant are estimated; wherein, the critical time points refer to the time points at which the disease progression of each potential disease to several target disease stages occurs.

[0159] Based on the key time points, determine the urgency of all potential diseases in the ruminant; based on the urgency and the type of potential disease, identify the adaptive response end of the ruminant.

[0160] Specifically, based on health status predictions, the service platform estimates key time points in the progression of potential diseases. For example, it predicts acidosis will progress to a moderate stage after 24 hours, and heat stress to a severe stage after 6 hours. The platform then determines the urgency level based on these key time points. Clearly, heat stress is more urgent. Depending on the urgency level and disease type, the platform searches for suitable response devices in a pre-defined list. For example, a high-urgency heat stress warning is sent to the rancher's mobile app and the on-site environmental controller (for automatically activating sprinklers and fans), while a moderate-urgency acidosis warning is sent to the veterinarian's computer terminal and the control panel of the feeding system.

[0161] In this embodiment, different response terminals (such as rancher's mobile phone, veterinarian's terminal, environmental controller, etc.) are intelligently matched according to the urgency and type of the disease, realizing precise hierarchical and categorized alarms and resource allocation. High-urgency diseases can reach the most critical execution terminals immediately to ensure rapid response; low-urgency diseases notify the corresponding roles for routine handling. This optimizes the utilization efficiency of ranch manpower and equipment resources and avoids information overload or important alarms being ignored.

[0162] In one embodiment, the early warning module generates an early warning message corresponding to the health status prediction result, and adjusts the decision to send the early warning message to the adaptive response end based on the actual operation of the ruminant's adaptive response end. This can be implemented as follows:

[0163] The health status prediction results are semantically extracted and transformed to identify potential disease types and their disease evolution trends, generating health status semantic text information; the semantic text information is then packaged and encapsulated to generate a warning message.

[0164] The real-time busy / idle status of the adaptive response terminal is obtained, and the allowed message receiving period of the adaptive response terminal is determined according to the real-time busy / idle status of the adaptive response terminal; the decision to send warning messages to the adaptive response terminal is adjusted according to the time domain distribution of the allowed message receiving period; wherein, the sending decision includes the time interval for sending warning messages to the adaptive response terminal.

[0165] Specifically, for example, the early warning module first converts predictions such as "potential disease type is acidosis, risk probability 85%, severity index is expected to double in 72 hours" into semantic text "Animal ID#1234, high risk of acidosis, please check and adjust diet immediately". Then, it encapsulates this text to generate an early warning message.

[0166] Simultaneously, the module obtains the real-time operational status of the corresponding response terminal (such as a veterinary terminal). If the veterinary terminal is outside of its operating hours (e.g., 2 AM), the module's sending decision is to postpone the message to a preset allowed receiving period (e.g., 8 AM), and only send a short message to the emergency contact's terminal first. If the response terminal is idle, the complete warning message is sent immediately.

[0167] Transforming abstract prediction results into easily understandable natural language alert messages lowers the barrier to understanding for users. Simultaneously, adjusting the sending strategy based on the real-time busy / idle status of the response end (e.g., avoiding non-working hours) prevents untimely message interruptions, increases user acceptance and willingness to respond to alert messages, and ensures that critical information is effectively processed at the most appropriate time.

[0168] In one embodiment, the service platform controls the controller to adjust the data acquisition mode of the sub-sensors based on the feedback result of the adaptive response terminal to the received warning message. Specifically, this can be implemented as follows:

[0169] The credibility identification result of the warning content in the warning message is extracted from the feedback result of the adaptation response end. Based on the credibility identification result, the controller is controlled to adjust the data acquisition mode of the sub-sensor.

[0170] Specifically, for example, after viewing the warning message through the adapter response terminal, the veterinarian deems the warning content highly credible and promptly verifies the risk of acidosis through clinical examination of the ruminant. The veterinarian confirms the validity of the warning on the adapter response terminal, and this confirmation information is sent back to the service platform as feedback. The service platform extracts the credibility identification result from the feedback result. In this case, the credibility is considered high, and the service platform, based on this, sends an instruction to the controller of the rumen capsule sensor module to maintain the current sampling mode of each sub-sensor to keep the current monitoring stability unchanged. Conversely, if the veterinarian reports low credibility of the warning, that is, if the warning content is considered inconsistent with clinical observation, the service platform will send an instruction to the controller of the rumen capsule sensor module to increase the sampling frequency of relevant sub-sensors (such as pH and temperature sub-sensors) to obtain more refined data for model correction. It should be noted that the "warning content" mentioned in this embodiment refers to the health status prediction conclusion (including potential disease types, risk probabilities, and disease evolution trends, etc.) carried in the warning message, and the credibility identification result is the identification result of whether the prediction conclusion is consistent with the actual clinical situation. In addition to the warning content, the "early warning message" may also include auxiliary information such as the warning level and recommended measures. Those skilled in the art can understand that the core of the credibility identification result lies in judging the credibility of the warning content.

[0171] Through this closed-loop feedback, the system can continuously optimize itself, forming long-term effective and accurate monitoring and management.

[0172] The above embodiments form a closed-loop control chain of perception, analysis, early warning, feedback, and optimization by sending manual confirmation results (feedback) back to the service platform and adjusting the data acquisition mode of the rumen capsule accordingly. When an early warning is confirmed as true, the system maintains stable monitoring; when an early warning is a false positive, the system automatically increases the sampling frequency to optimize the model. This self-learning and adaptive mechanism enables the system to continuously evolve, maintain high accuracy and reliability over the long term, and avoid model degradation.

[0173] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. This disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims. Thus, if these modifications and variations of the invention fall within the scope of the claims of the invention and their equivalents, the invention is also intended to include these modifications and variations.

Claims

1. A rumen capsule sensor and artificial intelligence based ruminant monitoring system characterized in that, include: The rumen capsule sensing module, built into the rumen of a ruminant, includes several sub-sensors and a controller. The sub-sensors are used to collect multimodal physiological data of the ruminant; the controller is used to adjust the data acquisition mode of each sub-sensor based on the real-time data acquisition status of each sub-sensor. A wireless communication processing device is used to send the multimodal physiological data of the ruminant to a service platform according to the collected status of the multimodal physiological data. The service platform is used to preprocess multimodal physiological data and / or behavioral state data of ruminants to obtain a valid set of physiological data and / or a valid set of behavioral state data. Using a multimodal health prediction algorithm, the health status prediction result of the ruminant is determined based on the effective physiological data set and / or effective behavioral state data set; Based on the health status prediction results, the adaptive response end of the ruminant is identified; The early warning module is used to generate early warning messages corresponding to the health status prediction results, and adjust the decision to send early warning messages to the adaptation response end based on the actual operation of the adaptation response end of the ruminant. The service platform is also used to control the controller to adjust the data acquisition mode of the sub-sensors based on the feedback result of the received early warning message from the adapter response end.

2. The ruminant monitoring system as described in claim 1, characterized in that: The plurality of sub-sensors includes at least two of the following: temperature sub-sensor, pH sub-sensor, acceleration sub-sensor, pressure sub-sensor, humidity sub-sensor, and gas concentration sub-sensor; The controller is used to adjust the data acquisition mode of each sub-sensor based on the real-time data acquisition status of each sub-sensor, including: An initial sampling mode is set for each sub-sensor, and the acquisition operation of each sub-sensor is controlled to collect multimodal physiological data of ruminants; wherein, the multimodal physiological data includes at least two of the following: rumen temperature data, rumen pH data, ruminant action data, rumen pressure data, rumen humidity data, and gas concentration data of ruminants. Obtain the actual data acquisition load of each sub-sensor. Based on the actual data acquisition load and the data acquisition limit load of each sensor, determine whether each sub-sensor is in an overload operation state. If so, reduce the data acquisition frequency of the sub-sensor; otherwise, maintain the data acquisition frequency of the sub-sensor.

3. The ruminant monitoring system as described in claim 1, characterized in that: The wireless communication processing device is used to send the multimodal physiological data of the ruminant to the service platform according to the collected status of the multimodal physiological data, including: Based on the total amount of multimodal physiological data collected, determine whether data acquisition saturation has occurred, or whether the current time has reached the preset transmission time, or whether the data value collected by any sub-sensor exceeds the preset normal physiological value range corresponding to that sub-sensor; If any judgment result is yes, then send the currently collected multimodal physiological data to the service platform; If all judgments result in no, then multimodal physiological data will not be sent to the service platform for the time being.

4. The ruminant monitoring system as described in claim 1, characterized in that: The service platform is used to preprocess multimodal physiological and / or behavioral state data of ruminants to obtain a valid set of physiological and / or behavioral state data, including: Missing and outlier data were removed from the multimodal physiological and / or behavioral data of ruminants, and timestamps were aligned to obtain a valid set of physiological and / or behavioral data.

5. The ruminant monitoring system as described in claim 1, characterized in that: Using a multimodal health prediction algorithm, the health status prediction result of the ruminant is determined based on the effective physiological data set and / or effective behavioral state data set, including: A multimodal health prediction algorithm is used to perform multi-scale data fusion weight analysis on the effective physiological data set and / or the effective behavioral state data set to obtain the physiological state change trend of the ruminant; based on the physiological state change trend, a health status prediction result of the ruminant is generated; wherein, the health status prediction result includes any one or more of the ruminant's potential disease type, disease occurrence risk probability, and lesion evolution trend.

6. The ruminant monitoring system of claim 5, wherein, Using a multimodal health prediction algorithm, the health status prediction result of the ruminant is determined based on the effective physiological data set and / or effective behavioral state data set, including: Step A: Perform dimensionless normalization on each type of physiological data at the current time t in the effective physiological data set to obtain the normalized value of each type of physiological data at the current time t; perform dimensionless normalization on each type of behavioral state data at the current time t in the effective behavioral state data set to obtain the normalized value of each type of behavioral state data at the current time t. Step B, using the hyperbolic tangent function, according to the normalized value of each physiological data at the current time t and the number of physiological data types, determine the physiological modal tension coefficient at the current time t ; According to each behavior state data normalized value, the number of behavior state data categories, the median of each behavior state data normalized value within a preset time window before the current time t, a behavior modality tension coefficient is determined ; Step C, the physiological modality tension coefficient , the behavior modality tension coefficient , the variance of the normalized value of all physiological data at the current time t and the variance of the normalized value of all behavior state data at the current time t, to determine the physiological modality fusion weight and the behavior modality fusion weight ; Step D1, determining a physiological modality fusion weight according to the physiological modality , a behavior modality fusion weight , the physiological modality tension coefficient , the behavior modality tension coefficient , a change rate of the physiological state change trend scalar of the previous moment, determining a physiological state change trend scalar of the current moment t ; Step D2, at the current moment Based on whether this moment is the first sampling moment after the system starts, determine respectively The possible values ​​of , where: if Then calculate all normalized physiological data in the initial period after implantation. absolute maximum value ;like Then let Otherwise set For positive constants less than 0.1; if Then, based on the severity index stored in the previous time step... and the scalar of the physiological state change trend at the current time t Determine the current severity index at time t. ; Step E, for each potential disease According to the current severity index The scalar of the trend of the physiological state change Future predicted values, disease progression rate coefficient Disease initiation threshold and predicted duration and discrete time step By using discrete summation and exponential functions, future moments can be determined. Severity Index To characterize the evolution trend of the lesion.

7. The ruminant monitoring system as described in claim 6, characterized in that, Using a multimodal health prediction algorithm, the health status prediction result of the ruminant is determined based on the effective physiological data set and / or effective behavioral state data set, and further includes: Step F: For each potential disease, scalarize the physiological state change trend. The severity index of the current k-th disease KL divergence between the distribution of normalized values ​​of current physiological data and the healthy baseline The probability of developing each disease is determined using the Logistic function. ; Step G: The service platform outputs at least one disease with the highest probability of occurrence as the potential disease type of the ruminant, and according to... The prediction results generate the corresponding disease evolution trend.

8. The ruminant monitoring system as described in claim 1, characterized in that: Based on the health status prediction results, the service platform identifies the adaptive response endpoints of the ruminant, including: Based on the health status prediction results, the critical time points for the disease progression of each potential disease in the ruminant are estimated; wherein, the critical time points refer to the time points at which the disease progression of each potential disease to several target disease stages occurs. Based on the key time points, determine the urgency of all potential diseases in the ruminant; based on the urgency and the type of potential disease, identify the adaptive response end of the ruminant.

9. The ruminant monitoring system as described in claim 1, characterized in that: The early warning module generates an early warning message corresponding to the health status prediction result, and adjusts the decision to send early warning messages to the ruminant's adaptive response terminal based on the actual operation of the ruminant's adaptive response terminal, including: The health status prediction results are semantically extracted and transformed to identify potential disease types and their disease evolution trends, generating health status semantic text information; the semantic text information is then packaged and encapsulated to generate a warning message. The real-time busy / idle status of the adaptive response terminal is obtained, and the allowed message receiving period of the adaptive response terminal is determined according to the real-time busy / idle status of the adaptive response terminal; the decision to send warning messages to the adaptive response terminal is adjusted according to the time domain distribution of the allowed message receiving period; wherein, the sending decision includes the time interval for sending warning messages to the adaptive response terminal.

10. The ruminant monitoring system as described in claim 1, characterized in that: Based on the feedback from the adaptive response end regarding the received warning message, the service platform controls the controller to adjust the data acquisition mode of the sub-sensors, including: The credibility identification result of the warning content in the warning message is extracted from the feedback result of the adaptation response end. Based on the credibility identification result, the controller is controlled to adjust the data acquisition mode of the sub-sensor.

Images