A Multidimensional Prediction Method for Animal Health Status in Smart Ranches

Abstract

This invention relates to the field of data analysis technology, specifically to a multi-dimensional prediction method for the health status of animals in a smart ranch, solving the technical problem of insufficient timeliness and accuracy in judging the health status of ranch animals based on disease symptoms in existing technologies. The method includes: acquiring multiple types of health monitoring data of animals and identifying abnormal nodes in each type of health monitoring data; constructing a symptom manifestation function for each type of disease based on historical disease records, and determining the position with the largest slope variance between the symptom manifestation function of each type of disease and the symptom manifestation functions of other diseases as the optimal identification node for each type of disease; monitoring the health status of ranch animals in real time, calculating the cumulative function value of the abnormal values ​​of the already occurring abnormal nodes in the symptom manifestation function; and determining that the animal suffers from the target disease when the distance between the cumulative function value and the symptom manifestation function value at the optimal identification node of the target disease is less than a preset distance threshold.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, and particularly relates to a multi-dimensional prediction method for animal health status of a smart pasture. BACKGROUND

[0002] The pasture animal health status monitoring based on big data analysis mainly analyzes various data (including physiological data, behavior data and environmental data) collected by various sensor devices, can realize real-time monitoring and early warning, and can also propose optimization management measures and intervention strategies through data analysis results, so as to help breeders improve production efficiency and reduce disease risk.

[0003] At present, when the pasture animal health status is monitored, the type of disease is usually judged according to the specific symptoms that have already appeared, but the symptom manifestation of animal disease has hysteresis and will only appear when the disease has deteriorated to a certain extent. Therefore, relying on the symptom manifestation of the disease to judge the disease will lead to insufficient reaction time. In addition, different types of disease symptoms may have high overlap, which still lacks accuracy in monitoring the health status of pasture animals. SUMMARY

[0004] In order to solve the technical problems of insufficient timeliness and accuracy of judging the health status of pasture animals by relying on the symptom manifestation of the disease in the prior art, the purpose of the present application is to provide a multi-dimensional prediction method for the health status of smart pasture animals, and the technical solution adopted is as follows:

[0005] In a first aspect, a multi-dimensional prediction method for the health status of smart pasture animals is provided, comprising: acquiring multiple types of health monitoring data of animals, and determining an abnormal node of each type of health monitoring data; the abnormal node is a time point at which the health monitoring data exceeds a preset safety threshold; based on historical disease records, constructing a symptom manifestation function of each disease, and determining the position with the maximum slope variance of the symptom manifestation function of each disease and the symptom manifestation function of other diseases as the best recognition node of each disease; the symptom manifestation function is used to describe the cumulative effect of the abnormal value of each abnormal node on the symptom manifestation in the disease deterioration process; real-time monitoring of the health status of pasture animals, calculating the cumulative function value of the abnormal value of the appeared abnormal node in the symptom manifestation function; when the distance between the cumulative function value and the symptom manifestation function value at the best recognition node of the target disease is less than a preset distance threshold, it is determined that the animal has the target disease.

[0006] Based on the above technical scheme, in the multi-dimensional prediction method for the health state of the intelligent pasture animals provided by the application, the abnormal nodes are obtained to provide comprehensive and accurate abnormal signal input for health state analysis; the symptom appearance function is constructed based on the historical disease records, and the best recognition node is determined to effectively highlight the differentiated characteristics between different diseases and avoid misjudgment caused by overlapping disease symptoms, thereby improving the accuracy of disease recognition; meanwhile, by monitoring the health state of the pasture animals in real time, calculating the cumulative function value, and determining that the animal has the target disease when the cumulative function value is close to the best recognition node of the target disease, the key abnormal accumulation signal can be captured in the early stage of disease deterioration, rather than relying on the typical symptoms that appear after the disease worsens, thereby greatly shortening the lag time of disease recognition, reserving more sufficient intervention reaction time for the breeding managers, and finally realizing more timely and accurate prediction of the health state of the pasture animals.

[0007] In combination with the first aspect, in a possible implementation manner, the method for constructing the symptom appearance function of each disease based on the historical disease records specifically includes: arranging multiple types of health monitoring data according to the sequence in which the abnormal nodes of each disease appear in the historical disease records to obtain a symptom observation sequence of each disease; analyzing the predictability of the abnormal nodes in each disease based on the historical disease records; the predictability is used to represent the prediction ability of the abnormal nodes in the development of the disease; and constructing the symptom appearance function of each disease based on the predictability, the abnormal value of the abnormal nodes, and the symptom observation sequence.

[0008] In combination with the first aspect, in a possible implementation manner, the method for analyzing the predictability of the abnormal nodes in each disease based on the historical disease records specifically includes: analyzing the credibility weight of the abnormal nodes based on the historical disease records; the credibility weight is used to represent the reliability of the abnormal nodes in the diagnosis of the disease; determining the predictability according to the credibility weight and the time sequence relationship between the abnormal nodes in the symptom observation sequence; and the predictability is used to represent the indication strength of the abnormal nodes of the next type of health monitoring data after the abnormal nodes of the previous type of health monitoring data appear.

[0009] In combination with the first aspect, in a possible implementation manner, the method for analyzing the credibility weight of the abnormal nodes based on the historical disease records specifically includes: counting the occurrence rate and the error rate of the abnormal nodes of each type of health monitoring data in multiple historical disease records; the occurrence rate represents the frequency of the abnormal nodes of each type of health monitoring data; the error rate represents the frequency of the abnormal nodes of each type of health monitoring data being recorded but not having the corresponding symptoms confirmed by diagnosis; analyzing the distribution of the abnormal nodes of each type of health monitoring data in the symptom observation sequence of the multiple historical disease records; and determining the credibility weight according to the occurrence rate, the error rate, and the distribution.

[0010] In a possible implementation of the first aspect, the method of analyzing the distribution of the abnormal nodes of each type of health monitoring data in the symptom observation sequence of the plurality of historical illness records, specifically comprises: determining the relative time position of the abnormal nodes in the symptom observation sequence of a single historical illness record, and analyzing the dispersion degree of the relative time position in the plurality of historical illness records.

[0011] In a possible implementation of the first aspect, the method of determining the predictability according to the credibility weight and the time sequence relationship between the abnormal nodes in the symptom observation sequence, specifically comprises: calculating the time sequence correlation between the abnormal value of each abnormal node in the symptom observation sequence and the abnormal value of the previous abnormal node adjacent in time sequence according to the plurality of historical illness records; and calculating the predictability according to the credibility weight and the time sequence correlation.

[0012] In a possible implementation of the first aspect, the method of constructing the symptom manifestation function of each type of disease based on the predictability and the abnormal value of the abnormal node and the symptom observation sequence, specifically comprises: comparing the difference of the sequence position of each type of health monitoring data in the symptom observation sequence of each type of disease and the symptom observation sequence of other diseases, to determine the recognition weight of each type of health monitoring data in each type of disease; weighting the standardized value of the abnormal value of each type of health monitoring data according to the predictability and the recognition weight of each type of health monitoring data, to obtain the weighted abnormal value of each type of health monitoring data; and accumulating the weighted abnormal values of the plurality of types of health monitoring data according to the order of the symptom observation sequence, to obtain the symptom manifestation function.

[0013] In a possible implementation of the first aspect, before the method of calculating the accumulated function value of the abnormal value of the appeared abnormal node in the symptom manifestation function, the method further comprises: obtaining real-time health monitoring data, and identifying abnormal nodes to obtain an abnormal node set; and determining at least one type of disease with a similarity to the symptom observation sequence higher than a preset similarity threshold value as a candidate disease set.

[0014] In a possible implementation of the first aspect, the method of calculating the accumulated function value of the abnormal value of the appeared abnormal node in the symptom manifestation function, specifically comprises: for each type of disease in the candidate disease set, calculating the accumulation sum of the weighted abnormal values corresponding to the abnormal node set as the accumulated function value according to the symptom manifestation function.

[0015] In a possible implementation manner of the first aspect, the method for acquiring the multi-type health monitoring data of the animal specifically comprises: collecting, by a sensor device, data for representing a physiological state, a behavior state or an environmental state of the animal; and collecting, by an image collection device, data for representing an external state of the animal or an interaction state of a group of animals that needs to be determined by vision.

[0016] In a second aspect, a multi-dimensional prediction device for animal health states in a smart ranch is provided, which comprises a processor and a storage medium, and the storage medium comprises instructions, and the processor is configured to execute the instructions to implement the actions described in the first aspect and any possible implementation manner of the first aspect. The multi-dimensional prediction device for animal health states in the smart ranch can be an electronic device or a chip in an electronic device.

[0017] In a third aspect, a computer-readable storage medium is provided, and the computer-readable storage medium stores instructions, and when the instructions are executed on the multi-dimensional prediction device for animal health states in the smart ranch, the multi-dimensional prediction device for animal health states in the smart ranch performs the actions described in the first aspect and any possible implementation manner of the first aspect.

[0018] In a fourth aspect, a computer program product is provided, and the computer program product stores instructions, and when the instructions are executed on the multi-dimensional prediction device for animal health states in the smart ranch, the multi-dimensional prediction device for animal health states in the smart ranch performs the actions described in the first aspect and any possible implementation manner of the first aspect.

[0019] The present application has the following beneficial effects:

[0020] By acquiring the abnormal nodes, comprehensive and accurate abnormal signal inputs are provided for health state analysis; by constructing a symptom appearance function based on historical disease records and determining the best recognition nodes, the differentiated features between different diseases are effectively highlighted, the misjudgment problem caused by overlapping of disease symptoms is avoided, and the accuracy of disease recognition is improved; at the same time, by monitoring the health state of the ranch animal in real time, calculating the cumulative function value, and determining that the animal has the target disease when the cumulative function value approaches the best recognition node of the target disease, the key abnormal accumulation signal in the early stage of disease deterioration can be captured, instead of relying on the typical symptoms that appear after the disease worsens, which greatly shortens the lag time of disease recognition, leaves more sufficient intervention reaction time for the breeder, and finally realizes more timely and accurate prediction of the health state of the ranch animal. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, the following will briefly introduce the drawings needed in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the field, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A method flowchart of a multi-dimensional prediction method of a smart pasture animal health state provided by an embodiment of the present application;

[0023] Figure 2 A method flowchart of another multi-dimensional prediction method of a smart pasture animal health state provided by an embodiment of the present application;

[0024] Figure 3 A method flowchart of another multi-dimensional prediction method of a smart pasture animal health state provided by an embodiment of the present application;

[0025] Figure 4 A method flowchart of another multi-dimensional prediction method of a smart pasture animal health state provided by an embodiment of the present application;

[0026] Figure 5 A method flowchart of another multi-dimensional prediction method of a smart pasture animal health state provided by an embodiment of the present application;

[0027] Figure 6 A hardware structure schematic diagram of a multi-dimensional prediction device of a smart pasture animal health state provided by an embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined purposes, the following describes a multi-dimensional prediction method of a smart pasture animal health state according to the present application, its specific implementation, structure, features and effects in detail, with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0030] The following specifically describes a specific scheme of a multi-dimensional prediction method of a smart pasture animal health state provided by the present application, with reference to the accompanying drawings.

[0031] Please refer toFigure 1 FIG. 1 shows a flow chart of a method for multi-dimensional prediction of animal health status in a smart pasture according to an embodiment of the present application. The method comprises:

[0032] S1, acquiring multi-type health monitoring data of the animal, and determining an abnormal node of each type of health monitoring data.

[0033] The abnormal node is a time point at which the health monitoring data exceeds a preset safety threshold. If a type of health monitoring data exceeds the preset safety threshold at multiple time points, the time point at which the health monitoring data first exceeds the preset safety threshold is determined as the abnormal node.

[0034] In some implementations, data for representing the physiological state, behavioral state, or environmental state of the animal is collected by a multi-source sensor device; data for representing the appearance state of the animal or the group interaction state that needs to be judged by vision is collected by an image collection device.

[0035] Each type of disease usually corresponds to multiple symptoms. For example, in the pig foot-and-mouth disease, the following common symptoms are included: blisters and ulcers, drooling, increased heart rate, lameness, reduced exercise, fever, decreased appetite, less water intake, growth retardation, weight loss, etc. Each type of symptom corresponds to a type of health monitoring data, which can be obtained from an Internet of Things composed of multi-source sensor devices.

[0036] The multi-source sensor device can include a temperature and humidity sensor, a camera, an intelligent ear tag (positioning, exercise amount, blood pressure, respiration, galvanic skin response, etc.), a weight sensor, etc., for monitoring data such as animal weight, body temperature, heart rate, respiratory rate, environmental temperature, humidity, air quality, exercise amount, food intake, sleep time, etc.

[0037] The data collected by the image collection device can include fine hair state, posture, excrement, feed amount, wound, social behavior, etc.

[0038] The above data can be further checked and supplemented by artificial means to improve data accuracy.

[0039] S2, based on historical disease records, constructing a symptom appearance function of each type of disease, and determining the position with the maximum slope variance of the symptom appearance function of each type of disease and the symptom appearance function of other diseases as the best recognition node of each type of disease.

[0040] The symptom appearance function is used to describe the cumulative effect of the abnormal value of each abnormal node on the symptom performance in the disease degradation process.

[0041] Specifically, for each time sequence position of the target disease symptom appearance function, the slope of the target disease function at this position is calculated, which can reflect the cumulative change rate corresponding to the time sequence position, and then the slopes of the symptom appearance functions of other diseases at the corresponding time sequence positions are calculated. Then, for each time sequence position, the dispersion degree (i.e., slope variance) of all slope differences at this position is calculated. The greater the slope variance, the more significant the difference between the cumulative change rates of the target disease and other diseases at the time sequence position, and the higher the recognition degree of the target disease. Therefore, the time sequence position with the maximum slope variance is selected as the best recognition node of the target disease, and the function value at this node can be used as a key reference benchmark for disease determination.

[0042] In some implementations, based on historical disease records, the symptom appearance function of each type of disease is constructed, which can be realized by a symptom appearance function fitting method based on a supervised learning model. Specifically, the abnormal node time sequence features in the historical disease records are taken as input, and the disease degradation stage is taken as label. A regression model is trained to directly fit the symptom appearance function. This method is suitable for scenarios with sufficient historical sample size and complex abnormal value time sequence features.

[0043] In some other implementations, based on historical disease records, the symptom appearance function of each type of disease is constructed, which can be realized by a symptom appearance function construction method based on knowledge graph association reasoning. Specifically, a knowledge graph of diseases, symptoms, and health monitoring data is constructed, and the association strength is verified using historical disease records. The abnormal value contribution degree is determined based on association reasoning, and then the function is constructed. This method is suitable for scenarios with less historical sample size but rich domain knowledge.

[0044] S3, real-time monitoring of the health status of the animals in the pasture, and calculating the cumulative function value of the abnormal value of the appeared abnormal node in the symptom appearance function.

[0045] In some implementations, the health monitoring data is synchronously obtained through preset sensor devices, image acquisition devices, and manual recording channels, and then the data is preprocessed: the garbled data caused by device failure is removed, the data repeatedly submitted by manual recording is removed, and the missing data caused by temporary disconnection is completed by linear interpolation method to ensure the integrity and effectiveness of real-time data. Then, the same function construction method as in S2 is used to calculate the cumulative function value of the appeared abnormal node.

[0046] S4, when the distance between the cumulative function value and the symptom appearance function value at the best recognition node of the target disease is less than a preset distance threshold, it is determined that the animal has the target disease.

[0047] In some implementations, the distance between the cumulative function value and the best recognition node function value is calculated for each disease, and the numerical absolute difference can be used as the distance measurement standard (i.e., distance = |cumulative function value - node function value|), which is simple in calculation logic and can intuitively reflect the closeness of the two.

[0048] If the symptom appearance function values of different diseases differ greatly in magnitude (for example, the node function values of swine flu are generally 5-8, and the node function values of skin diseases are generally 2-4), the calculated distance needs to be standardized. Specifically, the standard deviation of the node function values in the historical disease records of the disease (for example, the standard deviation of the node function values of swine flu is 1.2) is used as a benchmark, the absolute difference is divided by the standard deviation, and the standardized distance (for example, the absolute difference is 0.8, and the standardized distance = 0.8 / 1.2 ≈ 0.67, and the preset distance threshold can be 0.5) is obtained, so that the distances of different diseases have a unified comparison dimension, and the threshold adaptation problem caused by the magnitude difference of the function values is avoided.

[0049] Further, if the standardized distances of two or more diseases are less than the preset distance threshold, the standardized distances of all candidate diseases need to be sorted from small to large, and the disease with the smallest distance is selected as the high-priority target disease.

[0050] Based on the above technical solutions, the abnormal node is obtained, which provides comprehensive and accurate abnormal signal input for health status analysis; the symptom appearance function is constructed based on the historical disease records, and the best recognition node is determined, which effectively highlights the differentiated features between different diseases and avoids misjudgment caused by overlapping symptoms of diseases, thereby improving the accuracy of disease recognition; at the same time, by monitoring the health status of the animals in the pasture in real time, calculating the cumulative function value, and determining that the animal has the target disease when the cumulative function value is close to the best recognition node of the target disease, the key abnormal accumulation signal can be captured in the early stage of disease deterioration, rather than relying on the typical symptoms that appear after the disease worsens, which greatly shortens the lag time of disease recognition, leaves more sufficient intervention time for the breeder, and ultimately realizes more timely and accurate prediction of the health status of the animals in the pasture.

[0051] In one possible implementation, the method for constructing the symptom appearance function of each disease based on the historical disease records in S2 can be implemented by the following S21 to S23, which will be described in detail below. Figure 1 As shown in FIG. 2, the method for constructing the symptom appearance function of each disease based on the historical disease records in S2 can be implemented by the following S21 to S23, which will be described in detail below. Figure 2 As shown in FIG. 2, the method for constructing the symptom appearance function of each disease based on the historical disease records in S2 can be implemented by the following S21 to S23, which will be described in detail below.

[0052] S21, arrange the multiple types of health monitoring data according to the order in which the abnormal nodes of each disease appear in the historical disease records to obtain a symptom observation sequence of each disease.

[0053] For each type of disease, based on its historical records of illness, the abnormal nodes of each type of health monitoring data are counted in the order of disease symptom appearance. For example, in the historical records of swine flu, the abnormal node of body temperature is the first abnormal node in 80% of the records, and the abnormal node of cough-related respiratory rate is the second abnormal node in 75% of the records. Therefore, the fixed position of the body temperature monitoring data is set to 1, and the fixed position of the respiratory rate monitoring data is set to 2.

[0054] After determining the fixed positions of all health monitoring data according to the rule, the data are arranged in the fixed positions from the first to the last to form a symptom observation sequence of the target disease, which provides an ordering basis for the subsequent time sequence accumulation of the symptom appearance function.

[0055] S22, based on the historical records of illness, the predictability of abnormal nodes in each type of disease is analyzed.

[0056] Among them, the predictability is used to represent the prediction ability of the abnormal node in the development of the disease.

[0057] In some implementations, the time sequence relationship can be quantitatively analyzed first according to the correlation of adjacent abnormal nodes in the historical records of illness. Then, the reliability of the abnormal node in the diagnosis of the disease is combined as a weight to calculate the predictability. Finally, 10% of the historical samples are selected as a verification set. By comparing the actual appearance probability of the subsequent abnormality after the appearance of the previous abnormality with the calculated predictability, if the error is within 10%, the predictability analysis result is confirmed to be effective; otherwise, the correlation calculation method (such as mutual information) or the weight proportion is adjusted until the verification requirement is met.

[0058] In other implementations, the predictability analysis can also be based on a machine learning classification model. The features of the previous abnormal nodes are used as input to train the classification model to predict whether the subsequent abnormal nodes appear, and the model performance index is used to quantify the predictability. It is suitable for scenarios where the relationship between abnormal nodes is complex (nonlinear, multi-factor correlation).

[0059] In other implementations, the predictability of different lag periods can be quantified by analyzing the lag correlation between the occurrence time of the previous abnormal event and the occurrence time of the subsequent abnormal event (such as the appearance probability of the subsequent abnormality within 12 hours or 24 hours after the occurrence of the previous abnormality). It is suitable for diseases with large time differences between abnormal nodes (such as chronic diseases).

[0060] S23, based on the predictability and the abnormal value of the abnormal node, and the symptom observation sequence, the symptom appearance function of each type of disease is constructed.

[0061] In some implementations, since there are differences in the dimension of the abnormal value of different types of health monitoring data (such as the abnormal value of body temperature in ℃, and the abnormal value of exercise in steps / day), direct accumulation will cause calculation deviation, and the abnormal value needs to be standardized first. Then, in order to highlight the role of abnormal nodes with high prediction value and high discrimination, the standardized value of the abnormal value needs to be weighted by combining the predictability and the recognition weight determined based on the symptom observation sequence, to obtain the weighted abnormal value. Finally, since the symptom manifestation function needs to reflect the cumulative effect of the abnormal node on the symptom in the disease deterioration process, the weighted abnormal value needs to be accumulated in time sequence according to the order of the symptom observation sequence, to obtain the symptom manifestation function.

[0062] In other implementations, the function can also be constructed based on the dynamic weighting of the disease deterioration stage. Specifically, the weighting coefficient is adjusted according to the disease deterioration stage (early stage, middle stage, and late stage), to highlight the contribution of early abnormal nodes to early warning (early abnormalities are more conducive to timely intervention). This is suitable for scenarios with high demand for early detection of diseases (such as breeding of breeding stock and young stock).

[0063] In other implementations, the function can also be constructed based on a nonlinearly accumulated symptom manifestation function. Specifically, a nonlinear function (such as an exponential function or a power function) is used to accumulate the weighted abnormal value, to highlight the accelerated accumulation characteristics of late-stage abnormalities, in line with the deterioration law that the deterioration speed of symptoms in the late stage of acute diseases (such as foot-and-mouth disease and swine flu) is accelerated.

[0064] Based on the above technical solutions, by constructing a symptom observation sequence, the time sequence framework of symptom manifestation in the disease deterioration process is determined, the predictability of abnormal nodes is further analyzed, the prediction ability of abnormal nodes in the disease development is quantitatively represented, key abnormal nodes with high indication value for subsequent disease progression are effectively screened out, and the interference of low-prediction-value nodes is weakened. Finally, the construction of the symptom manifestation function can directly reflect the cumulative effect of each abnormal node on the manifestation of the corresponding symptom in the disease deterioration process, and convert discrete abnormal data into continuous disease deterioration quantitative representation. This provides a structured time sequence analysis basis and a quantitative disease deterioration model for disease health state prediction, and effectively improves the problems of traditional methods, such as dependence on late-stage typical symptoms, difficulty in early detection of disease signals, and inability to distinguish diseases with overlapping symptoms.

[0065] In one possible implementation, the method of S22 can be implemented by combining the following S221 to S222, which will be described in detail below. Figure 2 As shown in FIG. 2, the method of S22 can be implemented by combining the following S221 to S222, which will be described in detail below. Figure 3 S221, based on the historical disease record, analyze the credibility weight of the abnormal node.

[0066] S221, based on the historical disease record, analyze the credibility weight of the abnormal node.

[0067] The credibility weight is used to represent reliability of the abnormal node in disease diagnosis.

[0068] In some implementations, based on historical illness records, the method of analyzing the credibility weight of the abnormal node can include: counting occurrence rate and error rate of the abnormal node of each type of health monitoring data in multiple historical illness records; the occurrence rate represents frequency of occurrence of the abnormal node of each type of health monitoring data; the error rate represents frequency of confirmation of non-occurrence of corresponding symptoms after the abnormal node of each type of health monitoring data is recorded and diagnosed; analyzing distribution of the abnormal node of each type of health monitoring data in symptom observation sequences of the multiple historical illness records; and determining the credibility weight according to the occurrence rate, the error rate and the distribution.

[0069] In some implementations, the method of analyzing the distribution of the abnormal node of each type of health monitoring data in symptom observation sequences of the multiple historical illness records can include: determining relative time positions of the abnormal node in symptom observation sequences of single historical illness records, and analyzing discrete degree of the relative time positions in the multiple historical illness records.

[0070] In some implementations, the method of determining the relative time position of the abnormal node in the symptom observation sequence of the single historical illness record can include: determining a first abnormal node in the symptom observation sequence of the single historical illness record, and determining the relative time position of the abnormal node according to a time difference between a first occurrence time of the first abnormal node and a last occurrence time of the last abnormal node. In some implementations, the method of determining the relative time position of the abnormal node in the symptom observation sequence of the single historical illness record can include: determining a first abnormal node in the symptom observation sequence of the single historical illness record, and determining the relative time position of the abnormal node according to a time difference between a first occurrence time of the first abnormal node and a last occurrence time of the last abnormal node. In some implementations, the method of determining the relative time position of the abnormal node in the symptom observation sequence of the single historical illness record can include: determining a first abnormal node in the symptom observation sequence of the single historical illness record, and determining the relative time position of the abnormal node according to a time difference between a first occurrence time of the first abnormal node and a last occurrence time of the last abnormal node. In some implementations, the calculation formula of the relative time position of the abnormal node in the symptom observation sequence of the single historical illness record can be:

[0071]

[0072] In the formula, t is the relative time position of the abnormal node in the symptom observation sequence of the single historical illness record, t is the first occurrence time of the first abnormal node in the symptom observation sequence of the single historical illness record, and t is the last occurrence time of the last abnormal node in the symptom observation sequence of the single historical illness record. In the formula, t is the first occurrence time of the first abnormal node in the symptom observation sequence of the single historical illness record. In the formula, t is the last occurrence time of the last abnormal node in the symptom observation sequence of the single historical illness record. In the formula, t is the first occurrence time of the first abnormal node in the symptom observation sequence of the single historical illness record.

[0073] In the formula, t is the last occurrence time of the last abnormal node in the symptom observation sequence of the single historical illness record. In the formula, t is the first occurrence time of the first abnormal node in the symptom observation sequence of the single historical illness record.

[0074] In the formula, t is the last occurrence time of the last abnormal node in the symptom observation sequence of the single historical illness record. In the formula, t is the first occurrence time of the first abnormal node in the symptom observation sequence of the single historical illness record.

[0075] In the formula, t is the first occurrence time of the first abnormal node in the symptom observation sequence of the single historical illness record. In the formula, t is the first occurrence time of the first abnormal node in the symptom observation sequence of the single historical illness record. In the formula, t is the first occurrence time of the first abnormal node in the symptom observation sequence of the single historical illness record. In the formula, t is the first occurrence time of the first abnormal node in the symptom observation sequence of the single historical illness record. In the formula, t is the first occurrence time of the first abnormal node in the symptom observation sequence of the single historical illness record.

[0076] ​Further, calculate the abnormal nodes in all historical records of each disease The standard deviation of relative time position is denoted as The time randomness of abnormal nodes is quantified by standard deviation, i.e. the degree of dispersion.

[0077] Abnormal nodes The reliability weight of abnormal nodes The calculation formula can be:

[0078]

[0079] In the formula, The error rate of abnormal nodes , which represents subjectivity. The occurrence rate of abnormal nodes , which represents sampling frequency.

[0080] The relative error rate of abnormal nodes , the greater the value, the more errors the abnormal node has, and the less trustworthy it is.

[0081] It represents the comprehensive impact factor combining position randomness, sampling frequency, and subjectivity. The greater the product, the lower the sampling frequency of abnormal nodes of this type of data, the stronger the subjectivity, and the poorer the reliability.

[0082] The comprehensive impact factor is mapped to [0, 1] by deviation normalization , and then converted inversely by (1-normalized value) to obtain the reliability weight of abnormal nodes . Among them The nodes of are completely untrustworthy nodes and will be excluded from analysis.

[0083] In other implementations, the method of analyzing the distribution of abnormal nodes of each type of health monitoring data in the symptom observation sequence of multiple historical records of the disease can also convert the position of abnormal nodes in the symptom observation sequence (such as sequence rank, normalized time position) into a continuous probability density curve, and intuitively present the concentrated area (peak position) and dispersion degree of its distribution. It is suitable for analyzing whether abnormal nodes tend to appear at a specific stage (such as early or middle) of the sequence.

[0084] In other implementations, the method of analyzing the reliability weight can also introduce the health of the data acquisition device as a core indicator, and evaluate the reliability of abnormal nodes through the running state of the device. It is suitable for large-scale farms with dense sensor networks and real-time monitoring of device status.

[0085] In other implementations, the method for analyzing credibility weights can also use whether an abnormal node is a true abnormality as the classification objective, train the model to predict the credibility probability of the abnormal node, and use the probability value as the credibility weight. This is suitable for scenarios where health monitoring data is diverse and the causes of abnormalities are complex.

[0086] S222. Based on the confidence weight and the temporal relationship between abnormal nodes in the symptom observation sequence, determine the predictability.

[0087] Predictability is used to characterize the strength of an indication that a subsequent type of health monitoring data will also show anomalies after an anomaly occurs in a previous type of health monitoring data.

[0088] In some implementations, methods for determining predictability based on confidence weights and temporal relationships between anomalous nodes in the symptom observation sequence may include: calculating the temporal correlation between the outlier value of each anomalous node in the symptom observation sequence and the outlier value of the temporally adjacent preceding anomalous node based on multiple historical disease records; and calculating predictability based on confidence weights and temporal correlation.

[0089] In some implementations, abnormal nodes Time series correlation The calculation formula can be:

[0090]

[0091] In the formula, Abnormal node The sequence of deterioration values ​​in all historical disease records, where deterioration values ​​are the standardized result of the difference between outliers and healthy standard values.

[0092] Abnormal node The next anomalous node in the temporally adjacent sequence of all historical disease records The deterioration value sequence.

[0093] and The degradation values ​​are all sorted in chronological order according to all historical disease records.

[0094] Indicates abnormal nodes with abnormal nodes The absolute value of the Pearson correlation coefficient between the deterioration value sequences can measure the synchronicity of the deterioration of the two types of data, reflecting the temporal correlation strength of the subsequent occurrence of the former anomaly. The value ranges from [0, 1]. The larger the value, the higher the synchronicity of the deterioration of the two types of data, and the stronger the possibility that the former anomaly will be followed by the latter.

[0095] Furthermore, based on credibility weights and temporal relevance... Calculate predictability The calculation formula can be:

[0096]

[0097] In the formula, Indicates abnormal nodes with abnormal nodes The ratio of the credibility weights is such that the higher the ratio, the greater the credibility of the former type of health monitoring data is compared to the latter type, and the former has a higher degree of identification of the occurrence of diseases.

[0098] The larger the product of temporal correlation and reliability identification, the more likely it is to be an anomalous node. The more likely it is to be a key identification point for a disease, the higher its predictability.

[0099] In some implementations, predictability is determined based on credibility weights and the temporal relationship between abnormal nodes in the symptom observation sequence. Another method is to calculate predictability based on a weighted average of time differences. Specifically, this involves quantifying the time interval between preceding and subsequent abnormal nodes and combining their credibility weights to measure the strength of the temporal correlation (the shorter the time interval, the stronger the predictability).

[0100] Based on the above technical solution, the reliability differences of health monitoring data from different sources are quantified by using credibility weights to screen high-credibility nodes and avoid interference from low-credibility data. Then, predictability is determined by combining the credibility weights with the temporal relationship of abnormal nodes. This considers both the strength of the temporal correlation between abnormal nodes and enhances the predictive value of high-credibility nodes, accurately identifying key nodes that are highly indicative of disease development.

[0101] In one possible implementation, combining Figure 3 ,like Figure 4 As shown, the method in S23 described above can be specifically implemented through the following steps S231 to S233, which are explained in detail below:

[0102] S231. Compare the symptom observation sequence of each disease with the symptom observation sequence of other diseases to determine the identification weight of each type of health monitoring data in each disease.

[0103] In some implementations, the Spearman rank correlation coefficient (by comparing the ranks of variables) between the symptom observation sequences of the target disease and other diseases is first calculated and sorted. Diseases that are most easily confused with the target disease in terms of their order are then selected to ensure that subsequent order difference calculations focus on diseases with high confusion risk, thereby improving the targeting of identification weights.

[0104] In some implementations, the abnormal node In the target disease The recognition weight The calculation formula is:

[0105]

[0106] In the formula, is the abnormal node In the target disease The rank in the symptom observation sequence.

[0107] is the abnormal node In other diseases The rank in the symptom observation sequence.

[0108] is the number of all diseases of the abnormal node Participate in the recognition of health monitoring data .

[0109] Indicates the hyperbolic tangent function, which is used to normalize the value to the range of [0, 1).

[0110] Calculate the average of the rank difference, which is used to quantify the degree of rank difference of the abnormal node Corresponding to the target disease and other diseases, the greater the value, the more difficult the abnormal node Health monitoring data Be confused, so its recognition weight is higher.

[0111] Finally, the difference degree is mapped to the interval [0, 1) by using the hyperbolic tangent function, realizing the standardization of the recognition weight, and the monotonicity of the function ensures that the greater the difference, the closer the weight to 1, which is convenient for highlighting the role of high recognition data in subsequent weighted operation.

[0112] S232, according to the predictability and recognition weight of each type of health monitoring data, the standardized value of the abnormal value of each type of health monitoring data is weighted, and the weighted abnormal value of each type of health monitoring data is obtained.

[0113] In some implementations, in the symptom appearance function of the target disease , according to the predictability And the recognition weight , the standardized value of the abnormal value Weighted, the calculation formula of the weighted abnormal value

[0114] ​

[0115] In the formula, is a standardized value obtained by processing the abnormal value.

[0116] In some implementations, the way of standardizing the abnormal value can be any one of standard deviation standardization, median absolute deviation standardization, etc.

[0117] S233, according to the order of the symptom observation sequence, the weighted abnormal value of the multi-class health monitoring data is accumulated to obtain a symptom manifestation function.

[0118] In some implementations, according to the order of the symptom observation sequence, the total number of abnormal nodes is obtained by the time when the abnormal node appears.

[0119]

[0120] In the formula, is the total number of abnormal nodes in the target disease by the time when the abnormal node appears.

[0121] Based on the above technical solution, the recognition weight is determined by comparing the sequence difference of health monitoring data between diseases, which effectively quantifies the value of data in distinguishing the disease type. Then, the standardized abnormal value is weighted by combining the predictability and the recognition weight, which not only strengthens the influence of data with high indication on disease development, but also highlights the role of data with distinguishing significance on disease type, and improves the accuracy of abnormal value. Finally, the symptom manifestation function is obtained by accumulating the weighted abnormal value in time sequence, which converts the discrete abnormal node data into a continuous disease deterioration quantitative model, and intuitively reflects the symptom accumulation effect of the disease from early stage to late stage.

[0122] In one possible implementation, in combination with Figure 4 , as shown in Figure 5 , before S3, the method can further include the following S31 to S32, which are described in detail as follows:

[0123] S31, obtain real-time health monitoring data, and identify abnormal nodes to obtain a set of abnormal nodes that have appeared.

[0124] S32, determine at least one disease with a similarity between the set of abnormal nodes and the symptom observation sequence higher than a preset similarity threshold as a candidate disease set.

[0125] In some implementations, the symptom observation sequences of all diseases are loaded, and the health monitoring data type set is extracted from the set of abnormal nodes that have appeared. Then, the comprehensive similarity (including type overlap and sequence consistency) between the features of the abnormal node set and the disease subsequence is calculated. At least one disease with a comprehensive similarity higher than a preset similarity threshold (such as 80%) is selected to form a candidate disease set.

[0126] Based on the above technical solution, by capturing abnormal animal health signals in real time and comparing them with symptom observation sequences, a set of candidate diseases with similarity higher than a preset threshold is selected, which effectively narrows the scope of disease judgment, reduces invalid calculations, and improves the efficiency and accuracy of subsequent disease judgment.

[0127] One possible implementation is, such as Figure 5 As shown, the method S3 described above can be implemented through the following S33, which will be explained in detail below:

[0128] S33. For each type of disease in the candidate disease set, calculate the cumulative sum of the weighted outlier values ​​corresponding to the abnormal node set according to the symptom manifestation function, and use it as the cumulative function value.

[0129] Based on the above technical solution, for each type of disease in the candidate disease set, the cumulative function value is calculated based on the symptom manifestation function, thus achieving precise adaptation of the calculation process to the characteristics of each candidate disease.

[0130] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0131] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0132] In this embodiment of the invention, the multi-dimensional prediction device for animal health status in a smart ranch can be divided into functional units according to the above method example. For example, each function can be divided into its own functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0133] This invention also provides a schematic diagram of the hardware structure of a multi-dimensional prediction device for animal health status in a smart ranch, see [link / reference]. Figure 6 The smart ranch animal health status multidimensional prediction device 600 includes a processor 601, and optionally, a memory 602 connected to the processor 601.

[0134] In the first possible implementation, see Figure 6 The multi-dimensional prediction device 600 for animal health status in a smart ranch also includes a transceiver 603. The processor 601, memory 602, and transceiver 603 are connected via a bus. The transceiver 603 is used to communicate with other devices or communication networks. Optionally, the transceiver 603 may include a transmitter and a receiver. The device in the transceiver 603 that implements the receiving function can be considered as a receiver, which is used to perform the receiving steps in the embodiments of the present invention. The device in the transceiver 603 that implements the transmitting function can be considered as a transmitter, which is used to perform the transmitting steps in the embodiments of the present invention.

[0135] Based on the first possible implementation method Figure 6 The structural diagram shown can be used to illustrate the structure of the multi-dimensional prediction device for the health status of animals in the smart ranch involved in the above embodiments.

[0136] in, Figure 6 This can also be illustrated by the system chip in the multi-dimensional prediction device for animal health status in a smart ranch. In this case, the actions performed by the aforementioned multi-dimensional prediction device for animal health status in a smart ranch can be implemented by this system chip. The specific actions performed can be found above and will not be repeated here.

[0137] In implementation, each step of the method provided in this embodiment can be completed by integrated logic circuits in the processor or by instructions in software form. The steps of the method disclosed in this embodiment can be directly manifested as being executed by a hardware processor, or being executed by a combination of hardware and software modules in the processor.

[0138] The processor in this invention may include, but is not limited to, at least one of the following: a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a microcontroller unit (MCU), or an artificial intelligence processor, etc., which are various computing devices that run software. Each computing device may include one or more cores for executing software instructions to perform calculations or processing. The processor may be a standalone semiconductor chip or integrated with other circuits into a single semiconductor chip. For example, it may be integrated with other circuits (such as encoding / decoding circuits, hardware acceleration circuits, or various bus and interface circuits) to form a System-on-a-Chip (SoC), or it may be integrated as a built-in processor within an ASIC. The ASIC with the integrated processor may be packaged separately or together with other circuits. In addition to the cores for executing software instructions to perform calculations or processing, the processor may further include necessary hardware accelerators, such as field-programmable gate arrays (FPGAs), programmable logic devices (PLDs), or logic circuits that implement dedicated logic operations.

[0139] The memory in the embodiments of the present invention may include at least one of the following types: read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions; random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions; or electrically erasable programmable read-only memory (EEPROM). In some scenarios, the memory may also be a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0140] This invention also provides a computer-readable storage medium including instructions that, when run on a computer, cause the computer to perform any of the methods described above.

[0141] This invention also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform any of the methods described above.

[0142] This invention also provides a chip, which includes a processor and an interface circuit. The interface circuit is coupled to the processor. The processor is used to run computer programs or instructions to implement the above-described method. The interface circuit is used to communicate with other modules outside the chip.

[0143] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0144] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings and the disclosure, will understand and implement other variations of the disclosed embodiments in carrying out the claimed invention. In this invention, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several of the functions listed in this invention.

[0145] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the invention and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications of the invention fall within the scope of the invention and its equivalents, the invention is also intended to include such modifications and modifications.

Claims

1. A multi-dimensional prediction method for animal health status in a smart ranch, characterized in that, The application comprises: acquiring multiple types of health monitoring data of an animal, and determining an abnormal node of each type of health monitoring data; the abnormal node is a time point at which the health monitoring data exceeds a preset safety threshold; based on historical disease records, constructing a symptom appearance function of each disease, and determining a position with the maximum slope variance of the symptom appearance function of each disease and the symptom appearance functions of other diseases as an optimal recognition node of each disease; the symptom appearance function is used to describe the cumulative effect of abnormal values of each abnormal node on symptom appearance in the disease deterioration process; constructing the symptom appearance function of each disease comprises: comparing the difference between the sequence position of each type of health monitoring data in the symptom observation sequence of each disease and the symptom observation sequence of other diseases to determine the recognition weight of each type of health monitoring data in each disease; the symptom observation sequence refers to the time sequence of the occurrence of abnormal nodes in each disease based on historical disease records; according to the predictability of each type of health monitoring data and the recognition weight, the standardized value of the abnormal value of each type of health monitoring data is weighted to obtain the weighted abnormal value of each type of health monitoring data; the predictability is used to represent the indication strength of the occurrence of abnormal values of the next type of health monitoring data after the occurrence of abnormal values of the previous type of health monitoring data; the weighted abnormal values of the multiple types of health monitoring data are accumulated in the order of the symptom observation sequence to obtain the symptom appearance function; real-time monitoring of the health status of the animals in the pasture, and calculating the cumulative function value of the abnormal value of the abnormal node in the symptom appearance function; when the distance between the cumulative function value and the symptom appearance function value at the optimal recognition node of the target disease is less than a preset distance threshold, it is determined that the animal has the target disease.

2. The multi-dimensional prediction method of claim 1, wherein, The method for obtaining the predictability comprises: based on the historical disease records, analyzing the credibility weight of the abnormal node; the credibility weight is used to represent the reliability of the abnormal node in disease diagnosis; according to the credibility weight and the time sequence relationship between the abnormal nodes in the symptom observation sequence, the predictability is determined.

3. The multi-dimensional prediction method of claim 2, wherein, The method for analyzing the credibility weight of the abnormal node based on the historical disease records comprises: statistically analyzing the occurrence rate and the error rate of the abnormal node of each type of health monitoring data in multiple historical disease records; the occurrence rate represents the frequency of the occurrence of the abnormal node of each type of health monitoring data; the error rate represents the frequency of the confirmation of the non-occurrence of the corresponding symptoms after the abnormal node of each type of health monitoring data is recorded and diagnosed; analyzing the distribution of the abnormal node of each type of health monitoring data in the symptom observation sequence of multiple historical disease records; according to the occurrence rate, the error rate and the distribution, the credibility weight is determined.

4. The multi-dimensional prediction method of claim 3, wherein, The method for analyzing the distribution of the abnormal node of each type of health monitoring data in the symptom observation sequence of multiple historical disease records comprises: determining the relative time position of the abnormal node in the symptom observation sequence of a single historical disease record, and analyzing the dispersion degree of the relative time position in multiple historical disease records.

5. The multi-dimensional prediction method of claim 4, wherein, The determining the predictability comprises: According to a plurality of historical disease records, calculating a time correlation between an abnormal value of each abnormal node in the symptom observation sequence and an abnormal value of a previous abnormal node adjacent in time; According to the credibility weight and the time correlation, calculating the predictability.

6. The multi-dimensional prediction method of claim 1, wherein, The calculating the cumulative function value of the abnormal value of the appeared abnormal node in the symptom appearance function further comprises: Obtaining real-time health monitoring data and identifying abnormal nodes to obtain a set of appeared abnormal nodes; Determining, as a candidate disease set, at least one type of disease whose similarity to the symptom observation sequence is higher than a preset similarity threshold.

7. The multi-dimensional prediction method of claim 6, wherein, The calculating the cumulative function value of the abnormal value of the appeared abnormal node in the symptom appearance function comprises: For each type of disease in the candidate disease set, calculating, according to the symptom appearance function, a cumulative sum of weighted abnormal values corresponding to the set of abnormal nodes as the cumulative function value.

8. The multi-dimensional prediction method of claim 1, wherein, The obtaining a plurality of types of health monitoring data of the animal comprises: Collecting, by a sensor device, data for representing a physiological state, a behavioral state or an environmental state of the animal; Collecting, by an image collection device, data for representing an external state of the animal or a group interaction state which needs to be determined by vision.

Images