A method for constructing an animal disease diagnosis model based on multimodal data analysis

Abstract

The invention relates to the technical field of data processing, and discloses a method for constructing an animal disease diagnosis model based on multimodal data analysis, comprising the following steps: collecting historical multimodal data of beef cattle; constructing graph structure data according to the historical multimodal data of beef cattle; inputting the graph structure data into a disease prediction model, wherein the disease prediction model comprises a processing layer and a prediction layer, the processing layer outputs an updated node feature vector, the prediction layer inputs the updated node feature vector, and outputs a corresponding disease prediction probability vector of the beef cattle; constructing a feed strategy model, inputting the multimodal feature vector and the disease prediction probability vector of the beef cattle into the feed strategy model, and outputting a feed strategy; the invention monitors and analyzes the biological characteristics, environmental conditions, behavior patterns and historical disease records of beef cattle in real time, so that the system can identify early signs of the disease, realize disease early warning, and is conducive to preventing the spread of the disease, reducing treatment costs and improving cure rates.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and more specifically, to a method for constructing an animal disease diagnosis model based on multimodal data analysis. Background Art

[0002] At present, beef cattle breeding has shown a trend of large-scale development. In the process of beef cattle breeding, relevant personnel need to focus on the prevention and treatment of beef cattle diseases, especially during the high incidence period of the disease. After the occurrence of beef cattle diseases, not only the output of beef cattle is reduced, but also serious economic losses are caused to farmers, affecting the stability of the beef market. In order to improve the prevention and treatment level of beef cattle diseases, relevant industry personnel need to classify diseases and study effective diagnostic methods. Traditional animal disease diagnosis usually relies on veterinary experience and laboratory tests, which is inefficient and difficult to achieve early warning and large-scale screening. In addition, most farms use fixed feed formulas and feed amounts, and cannot flexibly adjust feed amounts and formulas according to changes in individual health conditions, resulting in feed waste and increased breeding costs. Summary of the invention

[0003] The present invention provides a method for constructing an animal disease diagnosis model based on multimodal data analysis, which solves the technical problems raised in the above-mentioned background technology.

[0004] The present invention provides a method for constructing an animal disease diagnosis model based on multimodal data analysis, comprising the following steps:

[0005] Step 101, collecting historical multimodal data of beef cattle, the multimodal data including biometric data, environmental data, behavioral data and historical disease records;

[0006] Step 102, constructing graph structure data based on historical multimodal data of beef cattle;

[0007] Step 103, inputting the graph structure data into a disease prediction model, the disease prediction model includes a processing layer and a prediction layer, the processing layer outputs an updated node feature vector, the prediction layer inputs the updated node feature vector, and outputs a corresponding disease prediction probability vector of beef cattle;

[0008] Step 104, constructing a feed strategy model, inputting the encoded historical multimodal data and the disease prediction probability vector of beef cattle into the feed strategy model, and outputting a feed strategy, wherein the feed strategy includes a corresponding feed combination and feed combination dosage for each beef cattle.

[0009] In a preferred embodiment, the graph structure data includes nodes and edges, a node represents a beef cattle, and the condition for the existence of an edge between two nodes is:

[0010] The pens where the two beef cattle were kept were adjacent;

[0011] The two beef cattle are related by blood;

[0012] There were more than a threshold number of social interactions between the two beef cattle;

[0013] Both beef cattle had suffered from the same disease.

[0014] In a preferred embodiment, the graph structure data also includes a multimodal feature vector encoded from the historical multimodal data as an initial node feature vector corresponding to each node.

[0015] In a preferred embodiment, the calculation formula of the treatment layer is:

[0016]

[0017] in, represents the node feature vector after the update of the i-th node, represents the initial node feature vector of the i-th node, represents the initial vector of the jth node, represents the set of neighbor nodes of the ith node, represents the weight matrix of the processing layer, Represents the ReLU function;

[0018] The calculation formula of the prediction layer is:

[0019]

[0020]

[0021] in, represents the predicted probability vector of the disease category distribution of the i-th beef cattle, represents the weight matrix of the prediction layer, represents the node feature vector after the update of the i-th node, express function, Indicates the number of disease types.

[0022] In a preferred embodiment, the feed strategy model includes: a feature combination layer, a hidden layer and an output layer, the feature combination layer inputs a multimodal feature vector of beef cattle and a disease prediction probability vector, and transmits the combined features of the two to the hidden layer, the hidden layer inputs the combined features and outputs the hidden features, the hidden features are input to the output layer, the output layer outputs the Q value of each optional feed strategy, and selects the feed strategy with the highest Q value as the output result;

[0023] The calculation formula of the feature combination layer is:

[0024]

[0025] in, represents the combined features, represents the multimodal feature vector, represents the disease prediction probability vector, CONCAT represents vector concatenation;

[0026] The calculation formula of the hidden layer is as follows:

[0027]

[0028] in, represents hidden features, represents the combined features, represents the weight matrix of the hidden layer, represents the bias coefficient of the hidden layer, Represents the ReLU function;

[0029] The calculation formula of the output layer is as follows:

[0030]

[0031] in, represents the Q value of each optional feeding strategy, represents hidden features, represents the ReLU function, represents the weight matrix of the output layer, Represents the bias parameter of the output layer.

[0032] In a preferred embodiment, the state space of the feed strategy model includes beef cattle's biometric data, environmental data, behavioral data, historical disease records, and disease prediction data;

[0033] Action space of feed strategy model: Combine the feed combination and feed combination amount corresponding to each beef cattle to generate the action space;

[0034] One action includes the feed combination and feed amount of each beef cow.

[0035] In a preferred embodiment, the feed strategy model The feedback is calculated one week after the feeding of the feed strategy is completed during the feeding period after the feed strategy is adjusted. Including health balance rewards, disease prevention penalty rewards and growth efficiency rewards;

[0036] The calculation formula for health balance bonus is as follows:

[0037]

[0038] in, represents the health balance reward of the i-th beef cattle, Indicates the average health level of the beef cattle population. represents the health level of the i-th beef cattle, Indicates the maximum health level of the beef cattle population, Indicates the threshold value of poor health level of beef cattle, represents the weight coefficient of health balance;

[0039] The calculation formula for disease prevention penalty reward is as follows:

[0040]

[0041] in, represents the disease prevention penalty reward for the i-th beef cattle, represents the maximum value of the disease prediction probability, represents the disease prediction probability of the i-th beef cattle, represents the threshold of disease prediction probability, represents the weight coefficient of disease prevention;

[0042] The calculation formula for growth efficiency bonus is as follows:

[0043]

[0044] in, represents the growth efficiency reward of the i-th beef cattle, is the average growth rate of the population, represents the average feed cost of the group, represents the growth rate of the i-th cow, represents the maximum growth rate, represents the feed cost of the i-th beef cattle, represents the maximum feed cost, , represent the weight coefficients of growth efficiency and feed cost respectively;

[0045] The calculation formula for feedback rewards is:

[0046]

[0047] in, Represents feedback reward, , , They represent the health balance reward, disease prevention penalty reward and growth efficiency reward of the i-th beef cattle respectively.

[0048] In a preferred embodiment, the historical multimodal data includes biometric data, environmental data, behavioral data and historical disease records. The biometric data and environmental data are encoded using a z-score standardized encoding method, and the behavioral data and historical disease records are encoded using a one-hot encoding method.

[0049] A system for constructing an animal disease diagnosis model based on multimodal data analysis, including the following modules:

[0050] Data collection and preprocessing module, used to collect historical multimodal data of beef cattle;

[0051] A graph structure building module, used to build graph structure data based on historical multimodal data of beef cattle;

[0052] The disease prediction model is used to input graph structure data and output the corresponding disease prediction probability vector of beef cattle;

[0053] The feed strategy model is used to input a multimodal feature vector and a disease prediction probability vector of beef cattle, and output a feed strategy.

[0054] A storage medium stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, the steps in the method for constructing an animal disease diagnosis model based on multimodal data analysis as described above can be executed.

[0055] The beneficial effects of the present invention are:

[0056] 1. The present invention monitors and analyzes the biological characteristics, environmental conditions, behavioral patterns and historical disease records of beef cattle in real time. The system can identify early signs of disease and achieve disease early warning, which is conducive to preventing the spread of disease, reducing treatment costs and improving cure rates;

[0057] 2. The present invention generates a customized feed strategy based on the health status, disease risk and nutritional needs of each cow, which can not only meet the nutritional needs of each cow, but also improve feed conversion efficiency and reduce breeding costs by optimizing feed combination and dosage. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flow chart of the method for constructing an animal disease diagnosis model based on multimodal data analysis of the present invention.

[0059] Figure 2 It is a block diagram of a system for constructing an animal disease diagnosis model based on multimodal data analysis of the present invention.

[0060] In the figure: 100, data collection and preprocessing module; 200, graph structure building module; 300, disease prediction model; 400, feed strategy model. DETAILED DESCRIPTION

[0061] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0062] like Figure 1 As shown, a method for constructing an animal disease diagnosis model based on multimodal data analysis includes the following steps:

[0063] Step 101, collecting historical multimodal data of beef cattle, the multimodal data including biometric data, environmental data, behavioral data and historical disease records;

[0064] Biometric data includes the cattle’s weight, temperature, heart rate, respiratory rate, blood oxygen saturation, blood pressure, and blood sugar levels;

[0065] Biometric data is obtained through sensors installed in the farm or through regular veterinary examinations. For example, body weight can be measured by a scale, and body temperature and heart rate can be obtained through regular veterinary examinations.

[0066] Environmental data include temperature (average temperature in the pen of each beef cow), humidity (average humidity in the pen of each beef cow), light intensity, air quality (such as average concentration of carbon dioxide), and noise level in the farm;

[0067] Environmental data is collected in real time through environmental monitoring sensors placed in the farms;

[0068] Behavioral data include cattle activity levels, sleep patterns, eating and drinking frequency, and social behavior, which can be collected in real time through video surveillance equipment installed in the farm;

[0069] Historical disease records include the name of previous diseases, onset date, symptom description, treatment plan and recovery status;

[0070] Historical disease records are collected and maintained through veterinary records, farm health management software, or electronic medical record systems.

[0071] Step 102, constructing graph structure data based on historical multimodal data of beef cattle;

[0072] Graph structure data includes nodes and edges. A node represents a beef cattle. The condition for the existence of an edge between two nodes is:

[0073] First, the fences where the two beef cattle were kept were adjacent;

[0074] Second, there is a blood relationship between the two beef cattle, for example, the two beef cattle are born in the same litter;

[0075] 3. There is more than a threshold number of social interactions between the two beef cattle;

[0076] Fourth, both beef cattle had suffered from the same disease;

[0077] It should be noted that the threshold number of social interactions can be manually set by the staff. For example, if the contact time is at least more than 20 minutes, it is considered that there is an effective social interaction between two beef cattle. The threshold number of effective social interactions is set to 5 times. When there are more than 5 social interactions between two beef cattle, it means that there is an edge between the two nodes.

[0078] The graph structure data also includes a multimodal feature vector encoded from the historical multimodal data, which serves as the initial node feature vector corresponding to each node;

[0079] Historical multimodal data includes biometric data, environmental data, behavioral data, and historical disease records. Biometric data and environmental data are encoded using z-score standardized encoding, behavioral data sleep patterns and eating data are encoded using one-hot encoding, and other behavioral data are also encoded using z-score standardized encoding. Historical disease records are encoded using one-hot encoding;

[0080] It should be noted that the encoded historical multimodal data are concatenated in the order of biometric data, environmental data, behavioral data, and historical disease records to form a multimodal feature vector;

[0081] Step 103, inputting the graph structure data into a disease prediction model, the disease prediction model includes a processing layer and a prediction layer, the processing layer outputs an updated node feature vector, the prediction layer inputs the updated node feature vector, and outputs a corresponding disease prediction probability vector of beef cattle;

[0082] It should be noted that the corresponding disease prediction probability vector of beef cattle includes the probability values ​​of the beef cattle suffering from different types of diseases;

[0083] The calculation formula of the processing layer is:

[0084]

[0085] in, represents the node feature vector after the update of the i-th node, represents the initial node feature vector of the i-th node, represents the initial vector of the jth node, represents the set of neighbor nodes of the ith node, represents the weight matrix of the processing layer, Represents a nonlinear activation function, such as the ReLU function;

[0086] The calculation formula of the prediction layer is:

[0087]

[0088]

[0089] in, represents the predicted probability vector of the disease category distribution of the i-th beef cattle, represents the weight matrix of the prediction layer, represents the node feature vector after the update of the i-th node, express function, Indicates the number of disease types;

[0090] In one embodiment of the present invention, the disease prediction model is trained by common methods such as back propagation and gradient descent, and the loss function is a cross entropy loss function. The data set is formed by collecting historical multimodal data of beef cattle and disease data marked by veterinarians on the beef cattle, and the data set is divided into a test set, a training set and a validation set for training the above-mentioned disease prediction model;

[0091] Step 104, constructing a feed strategy model, inputting the multimodal feature vector and the disease prediction probability vector of the beef cattle into the feed strategy model, and outputting a feed strategy, wherein the feed strategy includes a corresponding feed combination and feed combination dosage for each beef cattle;

[0092] The feed strategy model includes: a feature combination layer, a hidden layer and an output layer. The feature combination layer inputs the multimodal feature vector of beef cattle and the disease prediction probability vector, and passes the combined features of the two to the hidden layer. The hidden layer inputs the combined features and outputs the hidden features. The hidden features are input to the output layer. The output layer outputs the Q value of each optional feed strategy and selects the feed strategy with the highest Q value as the output result.

[0093] The calculation formula of the feature combination layer is:

[0094]

[0095] in, represents the combined features, represents the multimodal feature vector, represents the disease prediction probability vector, CONCAT represents vector concatenation;

[0096] The calculation formula of the hidden layer is as follows:

[0097]

[0098] in, represents hidden features, represents the combined features, represents the weight matrix of the hidden layer, represents the bias coefficient of the hidden layer, Represents a nonlinear activation function such as the ReLU function;

[0099] The calculation formula of the output layer is as follows:

[0100]

[0101] in, represents the Q value of each optional feeding strategy, represents hidden features, Represents a nonlinear activation function such as the ReLU function, represents the weight matrix of the output layer, Represents the bias parameter of the output layer;

[0102] It should be noted that before each feeding (or weekly, which can be understood as a weekly feed adjustment, and in special circumstances, a daily feed adjustment can be made, such as when the beef cattle are sick), the beef cattle's biological characteristic data, environmental data, behavioral data, historical disease records, and disease prediction data need to be updated to regenerate the feed strategy;

[0103] The training of the feed strategy model consists of the following steps:

[0104] Step 1: Initialize the experience pool, randomly initialize the feed strategy model, and initialize network, whose parameters are the same as those of the feed strategy model;

[0105] Step 2: reset the environment and obtain the first state;

[0106] Step 3, use The policy generates an action: The probability of randomly selecting an action is the exploration mode; in other cases, , select In the state The biggest action is the experience mode. Calculated by feed strategy model;

[0107] Get feedback based on the interaction between actions and the environment , next state and whether the termination condition done is triggered, the termination condition is the end of the feeding cycle of the beef cattle;

[0108] Experience Deposit into the experience pool, one experience is a four-tuple or ;

[0109] Step 4: Get an experience randomly from the experience pool;

[0110]

[0111] Indicates status Down The network's maximum output;

[0112] according to and Find the loss and use the gradient descent method to update the feed strategy model;

[0113]

[0114] Indicates status and actions The output Q value of the feed strategy model is as follows;

[0115] is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.5;

[0116] Step 5: Update every fixed number of network updates. network, making its parameters the same as those of the current dropper control strategy network;

[0117] Step 6, until After the network converges or the number of updates reaches the set value, the step is terminated;

[0118] Among them, the feedback The feedback is calculated one week after the feeding of the feed strategy is completed during the feeding period after the feed strategy is adjusted. Including health balance rewards, disease prevention penalty rewards and growth efficiency rewards;

[0119] The calculation formula for health balance bonus is as follows:

[0120]

[0121] in, represents the health balance reward of the i-th beef cattle, Indicates the average health level of the beef cattle population. represents the health level of the i-th beef cattle, Indicates the maximum health level of the beef cattle population, Indicates the threshold value of poor health level of beef cattle, represents the weight coefficient of health balance;

[0122] The calculation formula for disease prevention penalty reward is as follows:

[0123]

[0124] in, represents the disease prevention penalty reward for the i-th beef cattle, represents the maximum value of the disease prediction probability, represents the disease prediction probability of the i-th beef cattle, represents the threshold of disease prediction probability, represents the weight coefficient of disease prevention;

[0125] The calculation formula for growth efficiency bonus is as follows:

[0126]

[0127] in, represents the growth efficiency reward of the i-th beef cattle, is the average growth rate of the population, represents the average feed cost of the group, represents the growth rate of the i-th cow, represents the maximum growth rate, represents the feed cost of the i-th beef cattle, represents the maximum feed cost, , represent the weight coefficients of growth efficiency and feed cost respectively;

[0128] The calculation formula for feedback rewards is:

[0129]

[0130] in, Represents feedback reward, , , They represent the health balance reward, disease prevention penalty reward and growth efficiency reward of the i-th beef cattle respectively;

[0131] It should be noted that the weight and health status of all beef cattle were collected one week after a feed strategy adjustment.

[0132] The purpose of designing such rewards is to promote the overall health and balance of beef cattle herds, effectively prevent diseases, and optimize growth efficiency and feed costs to achieve the best breeding results;

[0133] The state space includes the beef cattle’s biological data, environmental data, behavioral data, historical disease records, and disease prediction data;

[0134] Action space: Combine the feed combination and feed combination amount corresponding to each beef cow to generate the action space;

[0135] One action includes the feed combination and feed amount of each beef cow.

[0136] like Figure 2 As shown, a system for constructing an animal disease diagnosis model based on multimodal data analysis includes the following modules:

[0137] The data collection and preprocessing module 100 is used to collect historical multimodal data of beef cattle;

[0138] Biometric data collection: weight, temperature, heart rate, etc., obtained through sensors and veterinary examinations;

[0139] Environmental data collection: Use environmental monitoring sensors to collect data such as temperature, humidity, light intensity, and air quality;

[0140] Behavioral data collection: video surveillance equipment was used to record activity levels, sleeping patterns, eating and drinking frequency, and social behavior;

[0141] Historical disease record collection: extracted from veterinary records, health management software or electronic medical records;

[0142] A graph structure building module 200, for building graph structure data based on historical multimodal data of beef cattle;

[0143] Node definition: Each beef cow is regarded as a node.

[0144] Edge definition: The existence of an edge is determined based on fence proximity, blood relationship, number of social interactions, and shared disease experience;

[0145] Node feature vector: Z-score standardization and One-Hot encoding are used to encode multimodal data to generate multimodal feature vectors;

[0146] The disease prediction model 300 is used to input graph structure data and output a corresponding disease prediction probability vector of beef cattle;

[0147] Processing layer: update node feature vector;

[0148] Prediction layer: predict the disease probability of each beef cattle based on the updated node feature vector;

[0149] A feed strategy model 400 is used to input a multimodal feature vector and a disease prediction probability vector of beef cattle, and output a feed strategy;

[0150] Feature combination layer: combine multimodal features and disease probability features;

[0151] Hidden layer: Process the combined features to generate hidden features;

[0152] Output layer: calculate the Q value of each feeding strategy and select the optimal strategy;

[0153] The system operation process of the present invention is as follows:

[0154] Step 1: Automatically or periodically collect all relevant data;

[0155] Step 2: Clean, standardize and code the data;

[0156] Step 3: Build a graph model based on the processed data;

[0157] Step 4: Use the disease prediction model to analyze the health status of each beef cattle;

[0158] Step 5: Develop the best feed strategy based on the prediction results;

[0159] Step 6. Implement the feeding strategy according to the optimal feeding strategy, continuously monitor its effect, and adjust the model parameters.

[0160] The system of the present invention can dynamically respond to health changes of beef cattle and timely adjust feeding strategies to promote the overall health and production efficiency of the beef cattle herd.

[0161] A storage medium stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, the steps in the method for constructing an animal disease diagnosis model based on multimodal data analysis as described above can be executed.

[0162] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A method for constructing an animal disease diagnosis model based on multimodal data analysis, characterized in that: The following steps are involved: Step 101, collecting historical multimodal data of beef cattle, the historical multimodal data including biometric data, environmental data, behavioral data and historical disease records; the biometric data includes the weight, body temperature, heart rate, respiratory rate, blood oxygen saturation, blood pressure and blood sugar level of beef cattle; the environmental data includes the temperature, humidity, light intensity, air quality and noise level in the farm; the behavioral data includes the activity level, sleep pattern, eating, drinking frequency and social behavior of beef cattle; the historical disease records include the name of the previous disease, date of onset, symptom description, treatment plan and recovery status; Step 102, constructing graph structure data based on historical multimodal data of beef cattle; Graph structure data includes nodes and edges. A node represents a beef cattle. The condition for the existence of an edge between two nodes is: The pens where the two beef cattle were kept were adjacent; The two beef cattle are related by blood; There were more than a threshold number of social interactions between the two beef cattle; Both beef cattle had suffered from the same disease; The graph structure data also includes a multimodal feature vector encoded from the historical multimodal data, which serves as the initial node feature vector corresponding to each node; The sleep patterns, eating data, and historical disease records of behavioral data were encoded using one-hot encoding, and the other data of biometric data, environmental data, and behavioral data were encoded using z-score standardized encoding. The encoded historical multimodal data are spliced ​​in the order of biometric data, environmental data, behavioral data, and historical disease records to form a multimodal feature vector; Step 103, inputting the graph structure data into a disease prediction model, the disease prediction model includes a processing layer and a prediction layer, the processing layer outputs an updated node feature vector, the prediction layer inputs the updated node feature vector, and outputs a corresponding disease prediction probability vector of beef cattle; Step 104, constructing a feed strategy model, inputting the encoded historical multimodal data and the disease prediction probability vector of the beef cattle into the feed strategy model, and outputting a feed strategy, wherein the feed strategy includes a corresponding feed combination and feed combination dosage for each beef cattle; The feed strategy model includes: a feature combination layer, a hidden layer and an output layer. The feature combination layer inputs the multimodal feature vector of beef cattle and the disease prediction probability vector, and passes the combined features of the two to the hidden layer. The hidden layer inputs the combined features and outputs the hidden features. The hidden features are input to the output layer. The output layer outputs the Q value of each optional feed strategy and selects the feed strategy with the highest Q value as the output result. The calculation formula of the feature combination layer is: ； in, represents the combined features, represents the multimodal feature vector, represents the disease prediction probability vector, CONCAT represents vector concatenation; The calculation formula of the hidden layer is as follows: ； in, represents hidden features, represents the combined features, represents the weight matrix of the hidden layer, represents the bias coefficient of the hidden layer, Represents the ReLU function; The calculation formula of the output layer is as follows: ； in, represents the Q value of each optional feeding strategy, represents hidden features, represents the ReLU function, represents the weight matrix of the output layer, Represents the bias parameter of the output layer; The training of the feed strategy model consists of the following steps: Step 1: Initialize the experience pool, randomly initialize the feed strategy model, and initialize network, whose parameters are the same as those of the feed strategy model; Step 2: reset the environment and obtain the first state; Step 3, use The policy generates an action: The probability of randomly selecting an action is the exploration mode; in other cases, , select In the state The biggest action is the experience mode. Calculated by feed strategy model; Get feedback based on the interaction between actions and the environment , next state and whether the termination condition done is triggered, the termination condition is the end of the feeding cycle of the beef cattle; Experience Deposit into the experience pool, one experience is a four-tuple or ; Step 4: Get an experience randomly from the experience pool; ； Indicates status Down The network's maximum output; according to and Find the loss and use the gradient descent method to update the feed strategy model; ； Indicates status and actions The output Q value of the feed strategy model is as follows; is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.5; Step 5: Update every fixed number of network updates. network, making its parameters the same as those of the current drip control strategy network; Step 6, until After the network converges or the number of updates reaches the set value, the step is terminated; Among them, the feedback It is calculated one week after the feeding of the feed strategy is completed during the feeding period after the feed strategy is adjusted. Including health balance rewards, disease prevention penalty rewards and growth efficiency rewards; The state space of the feed strategy model includes the beef cattle’s biological characteristic data, environmental data, behavioral data, historical disease records, and disease prediction data; Action space of feed strategy model: Combine the feed combination and feed combination amount corresponding to each beef cow to generate the action space; One action includes the feed combination and feed amount of each beef cow; The calculation formula for health balance bonus is as follows: ； in, represents the health balance reward of the i-th beef cattle, Indicates the average health level of the beef cattle population. represents the health level of the i-th beef cattle, Indicates the maximum health level of the beef cattle population, Indicates the threshold value of poor health level of beef cattle, represents the weight coefficient of health balance; The calculation formula for disease prevention penalty reward is as follows: ； in, represents the disease prevention penalty reward for the i-th beef cattle, represents the maximum value of the disease prediction probability, represents the disease prediction probability of the i-th beef cattle, represents the threshold of disease prediction probability, represents the weight coefficient of disease prevention; The calculation formula for growth efficiency bonus is as follows: ； in, represents the growth efficiency reward of the i-th beef cattle, is the average growth rate of the group, represents the average feed cost of the group, represents the growth rate of the i-th cow, represents the maximum growth rate, represents the feed cost of the i-th beef cattle, represents the maximum feed cost, , represent the weight coefficients of growth efficiency and feed cost respectively; The calculation formula for feedback rewards is: ； in, Represents feedback reward, , , They represent the health balance reward, disease prevention penalty reward and growth efficiency reward of the i-th beef cattle respectively; The purpose of designing this reward is to promote the overall health and balance of the beef cattle herd, effectively prevent diseases, and optimize growth efficiency and feed costs to achieve the best breeding results.

2. The method for constructing an animal disease diagnosis model based on multimodal data analysis according to claim 1, characterized in that: The calculation formula of the processing layer is: ； in, represents the node feature vector after the update of the i-th node, represents the initial node feature vector of the i-th node, represents the initial vector of the jth node, represents the set of neighbor nodes of the ith node, represents the weight matrix of the processing layer, Represents the ReLU function; The calculation formula of the prediction layer is: ； ； in, represents the predicted probability vector of the disease category distribution of the i-th beef cattle, represents the weight matrix of the prediction layer, represents the node feature vector after the update of the i-th node, express function, Indicates the number of disease types.

3. A system for executing the method for constructing an animal disease diagnosis model based on multimodal data analysis as claimed in claim 2, characterized in that: Includes the following modules: Data collection and preprocessing module, used to collect historical multimodal data of beef cattle; A graph structure building module, used to build graph structure data based on historical multimodal data of beef cattle; The disease prediction model is used to input graph structure data and output the corresponding disease prediction probability vector of beef cattle; The feed strategy model is used to input a multimodal feature vector and a disease prediction probability vector of beef cattle, and output a feed strategy.

4. A storage medium storing non-transitory computer-readable instructions, characterized in that: When the non-transitory computer-readable instructions are executed by a computer, the steps in the method for constructing an animal disease diagnosis model based on multimodal data analysis as described in claim 2 can be executed.

Images