A system and method for herbivore feed formulation optimization based on heterogeneous graphs

Abstract

The application provides a herbivorous animal feed formula optimization system and method based on a heterogeneous graph. The synergistic effect correlation matrix between the remaining feed raw material nodes after each masking is determined by performing multiple masking processes on the heterogeneous graph during the construction of the herbivorous animal feed formula, and then the compatibility degree vector of each feed raw material during the raw material compatibility is determined. The reference nutritional requirement characteristics are extracted from the heterogeneous graph, the nutritional requirement satisfaction rate at each growth stage is determined based on the initial growth stage information of the target herbivorous animal, and then the nutritional requirement parameter vector of the target herbivorous animal is generated. The nutritional adaptation degree of different feed raw materials to the target herbivorous animal is determined through the compatibility degree vector and the nutritional requirement parameter vector, and the optimal feed formula scheme of the corresponding feed of the target herbivorous animal variety is determined based on all the nutritional adaptation degrees. By adopting the application, the feed of the herbivorous animal can be formulaed in combination with the synergistic relationship of each feed raw material to the nutritional requirement of the herbivorous animal.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ingredient technology, and more particularly to a herbivorous animal feed formula optimization system and method based on a heterogeneous graph. BACKGROUND

[0002] Feed ingredient refers to the process of scientifically screening, proportionally adjusting and optimally combining multiple raw materials according to the core needs of target herbivorous animals such as species, growth stage and physiological state, combined with the nutritional components, physicochemical properties and compatibility rules of various feed raw materials.

[0003] In the traditional field of feed ingredient, due to the diversity of target herbivorous animal species, dynamic changes in growth stages, and complex nutritional synergies or antagonisms between different feed raw materials, existing methods often fail to accurately balance the adaptability of herbivorous animal species-specific nutritional needs and dynamic changes in growth stages, and at the same time, the potential correlation rules between raw materials are easily ignored in raw material compatibility optimization, resulting in either the inability to fully meet the accurate nutritional supply of herbivorous animals at different growth stages or the unreasonable combination of raw materials causing nutritional waste or low absorption efficiency. Therefore, how to combine the synergistic relationship between various feed raw materials and the nutritional needs of herbivorous animals to formulate feed for herbivorous animals has become a problem in the industry. SUMMARY

[0004] The present application provides a herbivorous animal feed formula optimization system and method based on a heterogeneous graph, which can combine the synergistic relationship between various feed raw materials and the nutritional needs of herbivorous animals to formulate feed for herbivorous animals.

[0005] In a first aspect, the present application provides a herbivorous animal feed formula optimization method based on a heterogeneous graph, wherein a plurality of feed raw materials are used to formulate feed for a target herbivorous animal, and the method is characterized in that it comprises the following steps:

[0006] Constructing a heterogeneous graph for herbivorous animal feed ingredient with target herbivorous animal species, herbivorous animal growth stages, feed raw materials and nutritional components as nodes;

[0007] Performing multiple masking processes on the feed raw material nodes in the heterogeneous graph, and then determining the correlation matrix of the synergistic effect between the remaining feed raw material nodes after each masking, and calculating the fit degree vector of each feed raw material and other feed raw materials when the raw materials are compatible according to all correlation matrices through a mutual information maximization loss function;

[0008] extract a benchmark nutrient demand feature of a nutrient component node in the heterogeneous graph when transmitting nutrients to a herbivorous animal growth stage node, determine a nutrient demand satisfaction rate of the target herbivorous animal at each growth stage based on the benchmark nutrient demand feature and initial growth stage information of the target herbivorous animal, fuse and analyze all the nutrient demand satisfaction rates and specific basic data corresponding to the target herbivorous animal breed to generate a nutrient demand parameter vector of the target herbivorous animal at different growth stages;

[0009] determine a nutrient adaptation degree of different feed raw materials to the target herbivorous animal by combining all the nutrient component characteristics corresponding to the feed raw materials with the fitting degree vector and the nutrient demand parameter vector, and perform combination optimization on all the feed raw materials in the feed formula based on all the nutrient adaptation degrees to obtain an optimal feed ingredient scheme corresponding to the feed of the target herbivorous animal breed.

[0010] In some embodiments, the feed raw material nodes in the heterogeneous graph are subjected to multiple masking processes, and then a correlation matrix of the synergistic action between the remaining feed raw material nodes after each masking is determined, which specifically includes:

[0011] determining a masking proportion of the feed raw material nodes in the heterogeneous graph;

[0012] According to the masking proportion, the feed raw material nodes in the heterogeneous graph are subjected to multiple masking processes, and the remaining heterogeneous graphs after each masking are recorded;

[0013] selecting one remaining heterogeneous graph as a selected remaining heterogeneous graph, and determining the association information between each feed raw material node and other feed raw material nodes in the selected remaining heterogeneous graph;

[0014] determining a correlation matrix of the synergistic action between the feed raw material nodes in the selected remaining heterogeneous graph according to all the association information;

[0015] continue to determine the correlation matrix of the synergistic action between the feed raw material nodes in the remaining heterogeneous graph.

[0016] In some embodiments, and according to all the correlation matrices, the fitting degree vector of each feed raw material and other feed raw materials when the raw materials are matched is calculated by a mutual information maximization loss function, which specifically includes:

[0017] standardizing all the correlation matrices to obtain each standardized correlation matrix;

[0018] selecting one feed raw material as a selected feed raw material, and extracting a sequence of association values between the selected feed raw material and other feed raw materials from all the standardized correlation matrices;

[0019] construct a mutual information maximization loss function;

[0020] inputting the sequence of correlation values between the selected feedstock and other feedstocks into the loss function of the mutual information maximization, and outputting a fitting degree vector of the selected feedstock and other feedstocks when the feedstocks are matched;

[0021] continuing to determine the fitting degree vector of the remaining feedstock and other feedstocks when the feedstocks are matched.

[0022] In some embodiments, the reference nutrient requirement feature of the nutrient component node passing nutrients to the herbivorous animal growth stage node is extracted from the heterogeneous graph and specifically includes:

[0023] extracting a direct correlation path of the nutrient component node and the herbivorous animal growth stage node from the heterogeneous graph;

[0024] determining the required amount of each nutrient component of the herbivorous animal growth stage node according to the heterogeneous graph;

[0025] determining the reference nutrient requirement feature of the nutrient component node passing nutrients to the herbivorous animal growth stage node according to all the required amounts corresponding to the herbivorous animal growth stage node and the direct correlation path.

[0026] In some embodiments, determining the nutrient requirement satisfaction rate of the target herbivorous animal at each growth stage based on the reference nutrient requirement feature and initial growth stage information of the target herbivorous animal specifically includes:

[0027] obtaining initial growth stage information of the target herbivorous animal;

[0028] determining a nutrient component feature vector of the growth stage node according to the reference nutrient requirement feature;

[0029] performing feature fusion on the initial growth stage information and the nutrient component feature vector, and then determining the nutrient requirement satisfaction rate of the target herbivorous animal at each growth stage.

[0030] In some embodiments, performing fusion analysis on all nutrient requirement satisfaction rates and specific basic data corresponding to the target herbivorous animal breed to generate a nutrient requirement parameter vector of the target herbivorous animal at different growth stages specifically includes:

[0031] obtaining specific basic data of the target herbivorous animal breed;

[0032] associating and aggregating all nutrient requirement satisfaction rates and the specific basic data of the target herbivorous animal breed to obtain aggregation information between breed characteristics and nutritional adaptation degree of the target herbivorous animal;

[0033] generating a nutrient requirement parameter vector of the target herbivorous animal at different growth stages according to the aggregation information.

[0034] In some embodiments, determining the nutritional adaptation degree of each feed raw material to the target herbivorous animal based on the matching degree vector and the nutritional requirement parameter vector and the nutritional component characteristics of all feed raw materials specifically comprises:

[0035] determining the nutritional component characteristics of all feed raw materials;

[0036] determining the matching degree of the nutritional component of each feed raw material to the herbivorous animal species-specific requirement according to the nutritional requirement parameter vector and the nutritional component characteristics;

[0037] determining the nutritional adaptation degree of the corresponding feed raw material to the target herbivorous animal according to the matching degree of the nutritional component of each feed raw material to the herbivorous animal species-specific requirement and the matching degree vector.

[0038] In some embodiments, the combination optimization of all feed raw materials in the feed formula based on all nutritional adaptation degrees obtains an optimal feed formulation scheme of the feed corresponding to the target herbivorous animal species specifically comprises:

[0039] obtaining an initial feed formulation scheme of all feed raw materials in the feed formula;

[0040] optimizing and adjusting the initial feed formulation scheme according to all nutritional adaptation degrees to obtain an optimal feed formulation scheme of the feed corresponding to the target herbivorous animal species.

[0041] In some embodiments, the initial growth stage information of the target herbivorous animal is obtained from a breeding management system database.

[0042] In a second aspect, the present application provides a heterogeneous graph-based herbivorous animal feed formula optimization system, comprising:

[0043] a construction module for constructing a heterogeneous graph in the process of herbivorous animal feed formulation with the target herbivorous animal species, the growth stage of the herbivorous animal, the feed raw material and the nutritional component as nodes;

[0044] a processing module for performing multiple masking processes on the feed raw material nodes in the heterogeneous graph, thereby determining the correlation matrix of the synergistic action between the remaining feed raw material nodes after each masking, and calculating the matching degree vector of each feed raw material and other feed raw materials in the raw material compatibility according to all correlation matrices through a mutual information maximization loss function;

[0045] The processing module is further configured to extract reference nutrient requirement characteristics of the nutrient component nodes when transmitting nutrients to the herbivorous animal growth stage nodes from the heterogeneous graph, determine a nutrient requirement satisfaction rate of the target herbivorous animal at each growth stage based on the reference nutrient requirement characteristics and initial growth stage information of the target herbivorous animal, perform fusion analysis on all the nutrient requirement satisfaction rates and specific basic data corresponding to the target herbivorous animal breed, and generate a nutrient requirement parameter vector of the target herbivorous animal at different growth stages.

[0046] The execution module is configured to determine a nutrient adaptation degree of different feed raw materials to the target herbivorous animal by combining all nutrient component characteristics of the feed raw materials with the fitting degree vector and the nutrient requirement parameter vector, and perform combination optimization on all feed raw materials in the feed formula based on all the nutrient adaptation degrees, to obtain an optimal feed ingredient scheme of the feed corresponding to the target herbivorous animal breed.

[0047] The technical scheme provided by the embodiments disclosed in the application has the following beneficial effects:

[0048] In the herbivorous animal feed formula optimization system and method based on a heterogeneous graph provided by the application, a heterogeneous graph for herbivorous animal feed ingredient is first constructed by taking a target herbivorous animal breed, a herbivorous animal growth stage, a feed raw material and a nutrient component as nodes; the feed raw material nodes in the heterogeneous graph are subjected to multiple masking processes, and then a correlation matrix of the synergistic effect between the remaining feed raw material nodes after each masking is determined, and a fitting degree vector of each feed raw material and other feed raw materials in raw material compatibility is calculated based on all the correlation matrices through a mutual information maximization loss function; reference nutrient requirement characteristics of the nutrient component nodes when transmitting nutrients to the herbivorous animal growth stage nodes are extracted from the heterogeneous graph, a nutrient requirement satisfaction rate of the target herbivorous animal at each growth stage is determined based on the reference nutrient requirement characteristics and initial growth stage information of the target herbivorous animal, fusion analysis is performed on all the nutrient requirement satisfaction rates and specific basic data corresponding to the target herbivorous animal breed, and a nutrient requirement parameter vector of the target herbivorous animal at different growth stages is generated; a nutrient adaptation degree of different feed raw materials to the target herbivorous animal is determined by combining all nutrient component characteristics of the feed raw materials with the fitting degree vector and the nutrient requirement parameter vector, and combination optimization is performed on all feed raw materials in the feed formula based on all the nutrient adaptation degrees, to obtain an optimal feed ingredient scheme of the feed corresponding to the target herbivorous animal breed.

[0049] It can be seen that, in the feed ingredient process, first, the target herbivorous animal species, growth stage, feed raw material and nutrient composition are taken as nodes to construct a heterogeneous graph, which can build a multi-element correlated visualization framework, break the limitation of scattered and isolated elements in traditional ingredient, make the internal relationship between species characteristics, growth demand and raw material nutrition clearer, and lay a comprehensive data foundation for subsequent collaborative analysis; the feed raw material nodes in the heterogeneous graph are masked multiple times and the fitting degree vector is determined through the correlation matrix and mutual information loss function, which can accurately mine the synergistic or antagonistic rules between different raw materials, avoid nutrient waste or conflict caused by blind collocation by quantifying the adaptation degree of raw material combination, and improve the scientificity of raw material synergistic utilization; the growth stage benchmark nutrient demand characteristics are extracted from the heterogeneous graph and the species specificity basic data are fused to generate the nutrient demand parameter vector, which can realize accurate positioning of nutrient demand, taking into account the dynamic changes of different growth stages of herbivorous animals and meeting the unique physiological needs of species, solving the problem of insufficient adaptability of general nutrition scheme; the nutrient adaptation degree is determined and the raw material combination is optimized through the fitting degree vector, the nutrient demand parameter vector and the raw material nutrient composition characteristics, which can achieve the dual optimization of nutrient satisfaction and raw material synergy, and finally obtain an ingredient scheme that can ensure the best synergistic effect between raw materials and accurately meet the specific nutrient demand of the target herbivorous animal, significantly improving the breeding effect and economic value of the feed. By using the above scheme, the feed of the herbivorous animal can be collocated in combination with the synergistic relationship of each feed raw material to the nutrient demand of the herbivorous animal. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 is an exemplary flowchart of a heterogeneous graph-based herbivorous animal feed formula optimization method according to some embodiments of the present application;

[0051] Figure 2 is an exemplary flowchart of determining a correlation matrix according to some embodiments of the present application;

[0052] Figure 3 is an exemplary flowchart of determining a nutrient demand parameter vector according to some embodiments of the present application;

[0053] Figure 4 is a structural schematic diagram of a heterogeneous graph-based herbivorous animal feed formula optimization system according to some embodiments of the present application;

[0054] Figure 5 is a structural schematic diagram of a computer device for implementing a heterogeneous graph-based herbivorous animal feed formula optimization method according to some embodiments of the present application. DETAILED DESCRIPTION

[0055] In order to better understand the technical solutions of the present application, the technical solutions of the present application will be described in detail below in combination with the drawings in the specification and specific embodiments.

[0056] Referring to Figure 1 The figure is an exemplary flow chart of a heterogeneous graph-based herbivore feed formula optimization method according to some embodiments of the present application, which mainly includes the following steps:

[0057] In step 101, a heterogeneous graph for herbivore feed formulation is constructed with target herbivore species, herbivore growth stages, feed raw materials, and nutritional components as nodes.

[0058] In implementation, when constructing the heterogeneous graph for herbivore feed formulation, first, the specific division of four types of core nodes needs to be clarified: the target herbivore species node covers the specific species type of the herbivore to be formulated, each species is an independent node and is labeled with biological characteristic information of the species; the herbivore growth stage node is divided into multiple key stages according to the growth cycle of the target herbivore, and each stage node is associated with the basic data of the growth indicators and physiological needs of the stage; the feed raw material node contains all single raw materials to be selected, and each raw material node records its name, origin, and basic ingredient content range attributes; the nutritional component node covers all types of nutrients required for the growth of herbivores, and each nutritional component node clearly indicates its chemical composition, physiological function, and recommended intake amount information; then, the association between nodes is established: between the herbivore species node and the growth stage node, according to the specific growth cycle of the species, the corresponding connection is established, and the growth parameters of different species in each stage are labeled, wherein the growth parameters of each stage represent the core indicators for quantitatively describing the growth and development state, physiological function level, and production-related characteristics of the herbivore species in the stage of its growth cycle, including average daily gain, stage-end body weight, and feed conversion ratio reflecting growth rate and feed conversion efficiency, body height, body length, chest circumference, and other body size indicators reflecting body condition and organ development state, rumen volume, rumen microbial number, and other digestive organ development parameters, daily milk yield, milk fat rate during the lactation period of dairy cows, carcass rate, backfat thickness during the fattening period of beef cattle, and other production performance-related parameters that link growth and target production needs, and blood calcium, blood phosphorus, blood glucose, and other blood biochemical indicators that ensure the physiological basis of growth, as well as health and metabolism parameters such as the incidence of nutrition-related diseases; between the feed raw material node and the nutritional component node, a weighted connection is established according to the actual detected nutritional component content of the raw material, and the weight value corresponds to the content proportion of the specific nutritional component in the raw material; between the herbivore growth stage node and the nutritional component node, an association is established based on the nutritional requirement standard of the stage, and the demand threshold of each nutritional component is labeled; at the same time, the association between feed raw material nodes can be established according to actual compounding experience or known synergistic / antagonistic effects, thereby forming a heterogeneous graph containing multiple types of nodes and rich association information; in other embodiments, other ways of construction can also be used, which are not limited here.

[0059] It should be noted that the heterogeneous graph in this application represents a structured data model of multiple different types of nodes and specific relationships between nodes corresponding to the target herbivores. It takes the target herbivore species, herbivore growth stage, feed ingredients and nutrients as four core nodes. Each node carries detailed attribute information of the corresponding field. At the same time, different nodes are linked together through targeted connection relationships, intuitively presenting the complex relationship between herbivore species, growth stage, feed ingredients and nutrients, and providing comprehensive structured data support for feed formulation analysis.

[0060] In step 102, the feed ingredient nodes in the heterogeneous graph are masked multiple times to determine the correlation matrix of the synergistic effect between the remaining feed ingredient nodes after each masking. Based on all the correlation matrices, the compatibility vector of each feed ingredient with other feed ingredients in the raw material matching is calculated by using a loss function that maximizes mutual information.

[0061] In some embodiments, reference Figure 2 As shown, this diagram is an exemplary flowchart for determining the association matrix in some embodiments of this application. In this embodiment, the feed ingredient nodes in the heterogeneous graph are subjected to multiple masking processes, and the association matrix of the synergistic effect among the remaining feed ingredient nodes after each masking can be determined by the following steps:

[0062] First, in step 1021, the shading ratio of the feed ingredient nodes in the heterogeneous graph is determined;

[0063] Secondly, in step 1022, the feed raw material nodes in the heterogeneous graph are masked multiple times according to the masking ratio, and the remaining heterogeneous graph after each masking is recorded.

[0064] Then, in step 1023, a remaining heterogeneous graph is selected as the selected remaining heterogeneous graph, and the association information between each feed ingredient node and other feed ingredient nodes in the selected remaining heterogeneous graph is determined.

[0065] Therefore, in step 1024, the association matrix of the synergistic effect between feed raw material nodes in the selected remaining heterogeneous graph is determined based on all the association information;

[0066] Finally, in step 1025, the correlation matrix of the synergistic effects between feed ingredient nodes in the remaining heterogeneous graph is determined.

[0067] In a specific implementation, when determining the masking proportion of the feed raw material nodes in the heterogeneous graph, the total number and diversity characteristics of the feed raw material nodes are combined, and a masking proportion range of 10%-30% is usually set. For example, based on experimental verification or industry experience, when the number of raw material nodes is 50-100, a masking proportion of 20% is selected, which ensures that the remaining nodes after each masking can still reflect the core association between raw materials, generates a sufficient number of node combination scenarios through multiple masking, and at the same time avoids the situation that the remaining node association information is incomplete due to a too high masking proportion or the difference between different masking scenarios is insufficient due to a too low masking proportion. Secondly, when performing multiple masking processing on the feed raw material nodes in the heterogeneous graph according to the masking proportion, nodes corresponding to the proportion are selected from all feed raw material nodes as masking objects. For example, a random sampling method is used, and the random seed is reset before each masking to ensure sampling independence. The masking operation is realized by marking the node state as “invalid” (without deleting the node, only shielding its ability to participate in association calculation), and at the same time, the ID, original attributes, and connection relationship with other nodes of the remaining valid raw material nodes after each masking are recorded to form a remaining heterogeneous graph data file containing a node set, node attributes, and association structure. The masking operation is repeated 20-50 times to obtain multiple different remaining heterogeneous graphs. In other embodiments, other methods can also be used, which are not limited here.

[0068] In a specific implementation, when determining the association information between each feed raw material node and other feed raw material nodes in the selected remaining heterogeneous graph, all feed raw material nodes in the selected remaining heterogeneous graph are traversed first to extract the direct association information between the feed raw material nodes, including the known synergistic effect, antagonistic effect, and action intensity level between raw materials, and at the same time, the indirect association information such as the overlap degree of jointly associated nutrient component nodes (which can be quantified by the Jaccard coefficient), and the shortest path length in the heterogeneous graph are calculated. These information is classified and arranged into structured association information, wherein the association information represents the information of the association between the feed raw material nodes, and clearly indicates the association type and quantitative value of each pair of raw material nodes. In other embodiments, other methods can also be used to determine, which are not limited here.

[0069] In addition, in the implementation, when determining the association matrix of the synergistic action between the feedstuff nodes in the selected residual isomorphic graph according to all the association information, a square matrix is constructed with the residual feedstuff nodes in the selected residual isomorphic graph as the rows and columns, the matrix dimension is consistent with the number of residual nodes, for each element in the matrix, the direct association strength (synergy is positive, antagonism is negative, and the strength level is assigned as 1-5) and the indirect association degree (such as the degree of overlap of nutrients multiplied by a weight of 0.3) of the feedstuff and the feedstuff are integrated by weighted summation, the numerical value is mapped to the interval [-1, 1], for example, the maximum-minimum standardization can be used, so that the matrix elements intuitively reflect the direction and strength of the association between the feedstuffs, and finally the association matrix of the synergistic action between the feedstuff nodes in the selected residual isomorphic graph is formed; in other embodiments, other ways can also be used to determine, which is not limited here.

[0070] It should be noted that the association matrix in the present application represents the matrix of the association relationship between the feedstuff nodes in the feedstuff ingredients, and the positive and negative and size of the element value in the association matrix intuitively reflect the direction (synergy is positive, antagonism is negative) and strength of the association, so that the system presents the association mode between the residual feedstuff nodes.

[0071] In some embodiments, and according to all the association matrices, the compatibility degree vector of each feedstuff and other feedstuffs in the feedstuff compatibility can be realized by the following steps:

[0072] Standardizing all the association matrices to obtain each standardized association matrix;

[0073] Selecting a feedstuff as a selected feedstuff, and extracting the association value sequence between the selected feedstuff and other feedstuffs from all the standardized association matrices;

[0074] Constructing a loss function of mutual information maximization;

[0075] Inputting the association value sequence between the selected feedstuff and other feedstuffs into the loss function of mutual information maximization, and outputting the compatibility degree vector of the selected feedstuff and other feedstuffs in the feedstuff compatibility;

[0076] Continue to determine the compatibility degree vector of the residual feedstuff and other feedstuffs in the feedstuff compatibility.

[0077] In a specific implementation, when all the correlation matrices are normalized, the maximum-minimum normalization method is used to map the numerical values to [-1, 1] according to the distribution range of the element values in each correlation matrix (the formula is: normalized value = (original value-minimum value) / (maximum value-minimum value) x 2-1). For example, the correlation value of raw material A and B in a certain matrix is 3, the minimum value of the matrix element is -5, and the maximum value is 5. After normalization, it is (3+5) / (5+5) x 2-1 = 0.6. Then, each normalized correlation matrix is obtained, and the dimensional difference between different matrices is eliminated to ensure data comparability.

[0078] In addition, in a specific implementation, the correlation value sequence between the selected feed raw material and other feed raw materials can be extracted from all normalized correlation matrices in the following way: traverse all normalized correlation matrices, locate the row (or column) where the selected feed raw material is located in each matrix, extract the element values corresponding to other feed raw materials in the row, arrange the sequence according to the matrix number, for example, the correlation values between the selected feed raw material and other feed raw materials extracted from 20 matrices are 0.3, 0.25, …, 0.4 in turn, to form the correlation value sequence of the selected feed raw material and other feed raw materials. In other embodiments, other ways can also be used for extraction, which is not limited here.

[0079] In the loss function for maximizing mutual information, the correlation value sequence of the selected raw material and other raw materials is taken as a random variable, and the loss function is defined as 1 minus the sum of mutual information between each sequence (i.e. Loss = 1 -∑I(X,Y), where I(X,Y) is the mutual information between sequences X and Y). The goal is to maximize mutual information by minimizing the loss function. For example, if the sequence mutual information between the selected raw material and another raw material is 0.8 and with another raw material is 0.6, the initial loss is 1-(0.8+0.6)=-0.4, which needs to be further optimized to improve mutual information. A differentiable loss function is built through a tensor flow framework, where the mutual information between each sequence is calculated by the following method: the correlation value sequence is preprocessed, including removing outliers, standardizing or discretizing according to the data type, and ensuring that the data meets the basic requirements for mutual information calculation. Data types include continuous and discrete types. Based on the data type, the mutual information calculation method is selected to calculate the mutual information between each sequence. For example, if the sequence is discrete, the joint probability distribution of the two sequences and the marginal probability distribution of each sequence are constructed by counting the frequency of each value in the sequence, and then the mutual information definition formula is used for calculation. If the sequence is continuous, the probability density function of the sequence is estimated by kernel density estimation, and then the mutual information is calculated by integral operation or directly using the mutual information estimation tool supported by the tensor flow framework for differentiable calculation. The preprocessed sequence is modeled and learned, and finally the mutual information value I(X,Y) between the correlation sequence of the selected raw material and other raw materials is output. In other embodiments, other methods can be used to construct the loss function, which is not limited here.

[0080] In addition, when inputting the correlation value sequence of the selected feed raw material into the loss function, the values in the sequence are taken as the input features of the model parameters, and the gradient descent optimization algorithm is used to iteratively adjust the parameters to make the loss function converge to a minimum value (i.e. mutual information maximization). For example, for the correlation sequence of the selected feed raw material, the loss stabilizes at-1.2 after 500 iterations. At this time, each element in the output parameter vector corresponds to the compatibility of the selected feed raw material with other raw materials, forming a compatibility vector of the selected feed raw material with other feed raw materials in raw material compatibility. In other embodiments, other methods can be used to achieve this, which is not limited here.

[0081] It should be noted that the compatibility vector in this application represents a vector of the adaptation degree of one feed raw material with all other feed raw materials in raw material compatibility, which can be used to reflect the correlation pattern between raw materials under different masking scenarios, and provide a quantitative basis for subsequent screening of reasonable feed raw material combinations and formulating scientific ingredient schemes.

[0082] In step 103, the baseline nutrient requirement characteristics of the nutrient component nodes when transmitting nutrients to the herbivorous animal growth stage nodes are extracted from the heterogeneous graph, the nutrient requirement satisfaction rates of the target herbivorous animal at each growth stage are determined based on the baseline nutrient requirement characteristics and the initial growth stage information of the target herbivorous animal, all the nutrient requirement satisfaction rates are fused and analyzed with the specific basic data corresponding to the target herbivorous animal breed to generate the nutrient requirement parameter vector of the target herbivorous animal at different growth stages.

[0083] In some embodiments, the baseline nutrient requirement characteristics of the nutrient component nodes when transmitting nutrients to the herbivorous animal growth stage nodes can be achieved by the following steps:

[0084] The direct association path between the nutrient component nodes and the herbivorous animal growth stage nodes is extracted from the heterogeneous graph.

[0085] The required amount of each nutrient component of the herbivorous animal growth stage node is determined according to the heterogeneous graph.

[0086] The baseline nutrient requirement characteristics of the nutrient component nodes when transmitting nutrients to the herbivorous animal growth stage nodes are determined according to all the required amounts corresponding to the herbivorous animal growth stage nodes and the direct association path.

[0087] In specific implementation, when the direct association path between the nutrient component nodes and the herbivorous animal growth stage nodes is extracted from the heterogeneous graph, the edge set of the heterogeneous graph needs to be traversed, the edges with the source node type of "nutrient component" and the target node type of "herbivorous animal growth stage" are screened out, the connection relationship (such as "vitamin A-chick stage" and "protein-fattening herbivorous animal stage") and the attributes (such as association strength and action type) of the edges are recorded, and the paths connected indirectly through other nodes (such as feed raw materials) are excluded to obtain the direct association path between each nutrient component node and each herbivorous animal growth stage node, and then the direct association path between the nutrient component node and the herbivorous animal growth stage node is obtained. This step can be realized by querying the graph database, wherein the direct association path represents the connection path between the nutrient component node and the herbivorous animal growth stage node without passing through other type nodes; in other embodiments, other ways can be used for extraction, which is not limited here.

[0088] In addition, in the implementation, when determining the required amount of each nutrient component of the herbivorous animal growth stage node according to the heterogeneous graph, the attribute field of the target growth stage node can be queried first, such as obtaining the original data stored in the form of key-value pairs (such as {"protein requirement amount": "20%", "calcium requirement amount": "0.8%"} through a graph database access tool; then the key-value pairs are traversed, the nutrient component type is extracted by using string truncation (taking the core name "protein" before "requirement amount") for each key (such as "protein requirement amount"), and the requirement amount value is extracted by using string replacement (removing "%") and type conversion for the corresponding value (such as "20%"), and thus the required amount of each nutrient component of each herbivorous animal growth stage node is obtained, wherein the required amount represents the parameter of the requirement degree of each nutrient component required by the herbivorous animal growth stage node. In other embodiments, other ways can also be used to determine, which are not limited here.

[0089] In addition, in the implementation, when determining the reference nutrient requirement feature according to the required amount of the herbivorous animal growth stage node and the direct association path, the required amount is taken as the core parameter of the propagation weight, and the direct association path is calculated by weighting: for each herbivorous animal growth stage node, if the required amount of the herbivorous animal growth stage node for the nutrient component node A is 20% (weight 0.2), and the association strength of A to the stage is 0.8, then the propagation contribution value of the direct association path corresponding to the nutrient component node of the herbivorous animal growth stage node is 0.2*0.8=0.16; the propagation contribution values of all nutrient component nodes to the growth stage through the direct association path are summarized to form a feature vector containing the propagation strength of each nutrient component node to the herbivorous animal growth stage node, and then the feature vector of the propagation strength of each nutrient component node to each herbivorous animal growth stage node is obtained. This step can be realized by means of the message passing mechanism in the graph neural network, such as using the DeepGraph Library (DGL) to pass the nutrient component feature by weighting the required amount, or assigning attention weights to different paths through the graph attention network, and finally aggregating all the feature vectors to obtain the reference nutrient requirement feature when the nutrient component node transmits nutrients to the herbivorous animal growth stage node. In other embodiments, other ways can also be used to determine, which are not limited here.

[0090] It should be noted that the reference nutrient requirement feature in the present application represents the comprehensive feature of the nutrient component node transmitting the nutrient requirement ability to the herbivorous animal growth stage node, and reflects the actual requirement, adaptation degree and association closeness of different nutrient components to a specific growth stage, which is a key bridge connecting the nutrient component and the growth requirement of the herbivorous animal.

[0091] In some embodiments, determining the nutrition requirement satisfaction rate of the target herbivorous animal at each growth stage based on the baseline nutrition requirement feature in combination with the initial growth stage information of the target herbivorous animal can be achieved by the following steps:

[0092] Obtaining initial growth stage information of the target herbivorous animal;

[0093] Determining a nutrition component feature vector of the growth stage node according to the baseline nutrition requirement feature;

[0094] Feature fusion of the initial growth stage information and the nutrition component feature vector, and further determining the nutrition requirement satisfaction rate of the target herbivorous animal at each growth stage.

[0095] In obtaining the initial growth stage information of the target herbivorous animal, the initial growth stage information of the target herbivorous animal needs to be extracted from the breeding management system database, sensor records or manually entered data, and the initial growth stage information includes herbivorous animal breed, current growth stage name, starting age, initial weight, health status label; wherein the initial growth stage information represents a core information set of the basic characteristics and state of the target herbivorous animal at the initial stage of a specific growth cycle, and is mainly used to define the growth background and basic requirement baseline of the herbivorous animal.

[0096] In determining the nutrition component feature vector of the growth stage node according to the baseline nutrition requirement feature, the propagation intensity, requirement matching degree and adaptation direction of each nutrition component are quantified and integrated into vector elements after the propagation characteristics of each nutrition component are quantified, with nutrition component types (such as protein, calcium, lysine, vitamins) as dimensions; for example, according to a pre-set nutrition component list, the first dimension corresponds to the propagation intensity of protein (such as 0.18), the second dimension corresponds to the requirement matching degree of calcium (such as 15%), and the third dimension corresponds to the adaptation direction of lysine (positive adaptation is recorded as 1, and negative adaptation is recorded as -1); a fixed-length nutrition component feature vector is constructed through a numerical Python array, wherein the nutrition component feature vector represents a vector of the structured and numerical propagation characteristics of each type of nutrition component associated with the growth stage node, the vector length is consistent with the number of nutrition component types, and the feature vector structure of different growth stages is unified; in other embodiments, other ways can also be used to determine the nutrition component feature vector, which is not limited here.

[0097] In addition, in specific implementation, when fusing the initial growth stage information and the nutrient component feature vector, the categorical features (such as breed and health status) in the initial information are first converted into numerical vectors, such as by one-hot encoding (e.g., "white-feathered broiler" is encoded as [1,0,0]) or embedding encoding. The numerical initial information (such as age and weight) is standardized to the [0,1] interval, for example by standardization using the maximum-minimum value. The processed initial information vector (e.g., length 5) is combined with the nutrient component feature vector (e.g., length 10) to form a fused feature vector of length 15, such as by feature concatenation or by weighted summation. Finally, the fused feature vector is used as input and imported into a pre-trained prediction model (e.g., based on random forest). The model (using regression models, graph neural networks, or rule-based reasoning engines) learns and integrates the correlation patterns contained in the feature vectors (e.g., "broiler chicken + 30 days old + protein transmission intensity 0.18" corresponds to "protein supply 22%)", outputs the recommended supply amount, adaptation level, and constraints (e.g., calcium-phosphorus ratio range) of each nutrient component for each growth stage, and calculates the nutritional requirement satisfaction rate of the target herbivore at each growth stage based on all recommended supply amounts. The above calculation can be implemented using regression analysis or analysis of variance in mathematical statistics. For example, the matching ratio of each recommended supply amount to the baseline nutritional requirement can be used as the nutritional requirement satisfaction rate for the corresponding growth stage. Other methods can also be used to determine this in other embodiments, which are not limited here.

[0098] It should be noted that the nutritional requirement satisfaction rate in this application represents the parameter value of the degree to which the individual nutritional requirements of the target herbivores are met during their growth stages. In essence, it is a precise definition and scientific recommendation of the nutrients required by herbivores at specific growth stages, used to guide feed formulation optimization or feeding management, and to ensure an efficient match between the nutritional requirements of herbivores and the feed supply capacity.

[0099] In some embodiments, reference Figure 3 As shown, this figure is an exemplary flowchart of determining the nutritional requirement parameter vector in some embodiments of this application. In this embodiment, the nutritional requirement satisfaction rate of all nutritional requirements is fused and analyzed with the specific basic data corresponding to the target herbivore species to generate the nutritional requirement parameter vector of the target herbivore at different growth stages. This can be achieved by the following steps:

[0100] First, in step 1031, specific basic data of the target herbivorous animal species are obtained;

[0101] Secondly, in step 1032, all nutritional requirement satisfaction rates are correlated and aggregated with the specific basic data of the target herbivorous animal species to obtain aggregated information between the species characteristics and nutritional suitability of the target herbivorous animal.

[0102] Finally, in step 1033, the nutritional requirement parameter vector of the target herbivore at different growth stages is generated according to the aggregation information.

[0103] In a specific implementation, first, the specific basic data of the target herbivore breed is extracted from the breed database corresponding to the target herbivore, wherein the specific basic data of the target herbivore breed represents a set of basic information of the unique biological characteristics, nutritional requirement rules, physiological metabolism characteristics and breeding adaptability of the target herbivore breed accumulated in long-term breeding practice and scientific research, including breed genetic characteristics (such as the high lean meat rate gene of "tall and large herbivore"), breed standard nutritional requirement threshold (such as the standard value of crude protein during the fattening period of this breed is 16%), metabolic characteristic parameters (such as energy conversion rate 0.85); secondly, when all the nutritional requirement satisfaction rates are associated and aggregated with the target herbivore breed, all the nutritional requirement satisfaction rates are associated and calculated with the specific basic data, for example, the deviation rate of the nutritional requirement satisfaction rate from the standard nutritional supply threshold of the breed (such as the nutritional requirement satisfaction rate recommended is 16.5%, and the deviation rate from the standard 16% is 3.125%), the matching coefficient of the adaptation level and the breed requirement (such as "high adaptation" corresponds to a coefficient of 1.0, and "moderate adaptation" corresponds to a coefficient of 0.7), and the integration of the breed-specific nutritional sensitivity components (such as the high demand weight of lysine for this breed), and finally these data are integrated into a structured aggregation information table containing "breed name-nutritional component deviation rate-adaptation matching coefficient-sensitivity component weight", and this structured aggregation information table is used as the aggregation information between the breed characteristics and the nutritional adaptation degree of the target herbivore, for example, the above steps can be realized by merging the function of panda in the programmable language to associate the nutritional requirement satisfaction rate with the breed characteristic data table, wherein the aggregation information represents the result after the information between the breed characteristics and the nutritional adaptation degree of the target herbivore is aggregated; in other embodiments, other ways can also be used to realize it, which is not limited here.

[0104] In a specific implementation, when generating the nutritional requirement parameter vector according to the aggregated information, the vector dimension needs to be determined first: the variety-specific dimension (such as variety genetic characteristic encoding value, metabolic characteristic quantitative value), the growth stage dynamic requirement dimension (such as the deviation rate of each nutritional component, the matching coefficient), and the sensitive component dimension (such as the weight value of the sensitive nutritional component); then, the various types of data in the aggregated information are quantified and standardized, for example, the variety genetic characteristics are converted into one-hot encoding (such as “du long big herbivore” is encoded as [1, 0, 0]), and the deviation rate and the matching coefficient numerical data are standardized to the [0, 1] interval, such as by the maximum-minimum standardization method; finally, the standardized quantitative values are combined into a vector in the preset dimension order, and the vector is taken as the nutritional requirement parameter vector containing the variety-specific and growth stage dynamic requirements of the herbivore, for example, the vector of a certain du long big herbivore in the fattening period may be [1, 0.85, 0.03125, 1.0, 0.9], wherein the first position is the variety code, the second position is the metabolic characteristic value, the third position is the crude protein deviation rate, the fourth position is the matching coefficient, and the fifth position is the lysine sensitivity weight. The vector can be constructed by a numerical array to ensure fixed dimension and unified dimension.

[0105] It should be noted that the nutritional requirement parameter vector in the present application represents the parameter vector of the nutritional requirements of the target herbivore at different growth stages, reflecting the dynamic adjustment of the requirements in the growth stage transition, and providing structured input data for subsequent feed formula optimization, nutritional supply dynamic decision-making, intelligent regulation of the breeding system, and the like, to realize the transformation support from “experience feeding” to “data-driven precise nutrition”.

[0106] In step 104, the nutritional adaptation degrees of different feed raw materials to the target herbivore are determined by combining the matching degree vector and the nutritional requirement parameter vector with the nutritional component characteristics corresponding to all the feed raw materials, and all the feed raw materials in the feed formula are combined and optimized based on all the nutritional adaptation degrees, to obtain the optimal feed formula scheme of the feed corresponding to the target herbivore variety.

[0107] In some embodiments, the determination of the nutritional adaptation degrees of different feed raw materials to the target herbivore by combining the matching degree vector and the nutritional requirement parameter vector with the nutritional component characteristics corresponding to all the feed raw materials can be implemented by the following steps:

[0108] Determine the nutritional component characteristics corresponding to all the feed raw materials;

[0109] Determine the matching degrees of the nutritional components of each feed raw material to the variety-specific requirements of the herbivore according to the nutritional requirement parameter vector and the nutritional component characteristics;

[0110] According to the matching degree of the nutritional components of each feed raw material and the specific needs of the herbivorous animal species and the matching degree vector, the nutritional adaptation degree of the corresponding feed raw material to the target herbivorous animal is determined.

[0111] In a specific implementation, when determining the nutritional component features corresponding to all feed raw materials, the core nutritional data of each raw material is extracted from a feed component database, a raw material detection report, or an industry standard, and all the core nutritional data is taken as the nutritional component features corresponding to all feed raw materials. The core nutritional data includes the content (unit: %, g / kg) of crude protein, amino acids (such as lysine and methionine), minerals (such as calcium and phosphorus), and vitamins.

[0112] In addition, in a specific implementation, when determining the matching degree according to the nutritional requirement parameter vector and the nutritional component features, the nutritional component features of the feed raw material are first converted into a feature vector of the same dimension as the nutritional requirement parameter vector (for example, the standardized content of each component in the raw material is filled in the order of the nutritional components in the vector); and then the similarity (such as cosine similarity or Pearson correlation coefficient) or weighted deviation value (such as a higher weight is given to components sensitive to specific needs of the species, and the weighted sum of the absolute deviation of the raw material components from the requirements is calculated) of the feature vector and the nutritional requirement parameter vector is calculated to determine the matching degree of the nutritional components of each feed raw material to the specific needs of the herbivorous animal species. For example, if the weight of the protein dimension in the nutritional perception vector is 0.3 and the protein content of a certain raw material deviates from the requirement by 2%, the contribution of the deviation of this component is 0.006, and after normalizing the total component deviation, the matching degree in the interval of 0-1 (1 for complete matching) is obtained, wherein the matching degree is a parameter of the matching degree of the nutritional components of the feed raw material to the specific needs of the herbivorous animal species. In other embodiments, other ways of determining can also be used, which are not limited here.

[0113] In addition, in a specific implementation, when determining the nutritional adaptation degree according to the matching degree and the matching degree vector, the two are weighted and fused: for each feed raw material, the matching degree of the raw material to the requirements of the herbivorous animal is taken as the basis (such as accounting for 60%), and the average adaptation value of the raw material to other potential compatible raw materials in the matching degree vector is combined (such as accounting for 40%, to avoid the imbalance of the overall nutrition caused by the adaptation of a single raw material but the conflict with other raw materials), and the comprehensive score is calculated through the formula "nutritional adaptation degree = matching degree x 0.6 + average matching degree x 0.4"; the score is standardized (such as mapping to 0-100 points), and the adaptation level is divided according to the score (such as 80 points or more for "high adaptation"), and finally the nutritional adaptation degree of each feed raw material to the target herbivorous animal is obtained. The above scheme can be batch calculated through the application function of Panda, or the multi-raw material parallel processing can be realized by numerical matrix operation; in other embodiments, other ways of processing can also be used, which are not limited here.

[0114] It should be noted that the nutritional adaptation degree in the present application represents the parameter value of the adaptation degree of the feed raw material to the target herbivorous animal in nutrition, which can intuitively judge the rationality of a certain raw material or formula in nutritional supply, provide data support for precise feed formula design and nutritional regulation in the breeding process, reduce the problem of nutritional waste or deficiency, and improve the breeding efficiency.

[0115] In some embodiments, the combination optimization of all feed raw materials in the feed formula based on all nutritional adaptation degrees can be realized by the following steps to obtain the optimal feed ingredient scheme of the feed corresponding to the target herbivorous animal variety:

[0116] Obtain the initial feed ingredient scheme of all feed raw materials in the feed formula;

[0117] Adjust and optimize the initial feed ingredient scheme according to all nutritional adaptation degrees to obtain the optimal feed ingredient scheme of the feed corresponding to the target herbivorous animal variety.

[0118] In specific implementation, the initial feed ingredient scheme of all feed raw materials in the feed formula is obtained from the feed database corresponding to the target herbivorous animal, wherein the initial feed ingredient scheme is the basic blueprint of feed combination optimization, mainly including the core information of the target herbivorous animal (such as variety, growth stage, nutritional demand standard, such as 16%-18% crude protein demand of laying hen in egg laying period), the list of available feed raw materials (including raw material name, category, such as energy raw material corn, protein raw material soybean meal, mineral additive calcium hydrogen phosphate, etc.), the initial addition proportion of each raw material (expressed in percentage or weight ratio, such as corn 55%, soybean meal 20%, bran 10%), the basic attribute data of the raw material (such as nutritional ingredient content, unit cost, procurement channel and supply quantity limitation), the estimated total amount of nutritional ingredients (such as whether the total content of crude protein, calcium and lysine in the scheme is close to the demand threshold), and the preliminary cost estimation (such as the total cost of raw materials per ton of feed), the cost upper limit constraint and the compatibility preliminary screening record (such as whether to avoid obvious raw material antagonistic combination).

[0119] In addition, in the implementation, when the initial feed ingredient scheme is optimized and adjusted according to all the nutritional adaptation degrees, the nutritional adaptation degrees of the raw materials are first converted into weight coefficients (high-adaptation-degree raw materials are given high weights), which can be realized by linear conversion method or piecewise weighting method, and the weighted sum of all the nutritional adaptation degrees is calculated, and an optimization model is established with the core indexes of "nutritional requirement compliance rate", "raw material compatibility score" and "adaptation degree weighted sum", which can be established by multi-objective genetic algorithm or linear weighted sum method; the raw material proportion is iteratively adjusted through the optimization model (for example, the proportion of raw materials with adaptation degree ≥0.8 is increased by 5%-10%, and the proportion of raw materials with nutritional adaptation degree ≤0.5 is reduced or replaced by high-adaptation-degree alternative raw materials), and the genetic algorithm in the prior art can be used as the iteration engine of the optimization model, and the safety threshold of the raw materials commonly used in the feed industry and the cost upper limit constraint involved in the initial feed ingredient scheme are introduced, so as to simultaneously check whether the nutritional components of the adjusted scheme meet the target requirements (for example, whether the measured value of protein is within the standard range) and whether the combination of raw materials has compatibility taboo (for example, whether the coexistence proportion of antagonistic raw materials is out of limit through the compatibility matrix); the initial feed ingredient scheme is optimized for multiple rounds based on the adjustment of the raw material proportion, which can be realized by means of heuristic algorithm (such as genetic algorithm) or professional feed formula software, until the scheme meets the requirements of "nutritional requirement compliance rate ≥95%", "compatibility score ≥90 points" and "adaptation degree weighted sum maximum", and finally outputs the optimal feed ingredient scheme containing the optimized raw material proportion, measured value of nutritional components, adaptation degree total score and compatibility evaluation, which is used as the optimal feed ingredient scheme of the target herbivorous animal species.

[0120] It should be noted that the optimal feed ingredient scheme in the present application represents a feed formula scheme that meets the specific requirements of the target herbivorous animal and realizes the balance of multiple dimensions such as nutritional adaptation, raw material compatibility, economy and safety through scientific raw material screening, proportion adjustment and combination optimization, which can be used to determine the feed ingredient of the target herbivorous animal, and can improve the production performance (such as weight gain, egg production and milk production) and health level of the herbivorous animal while maximizing the breeding benefit.

[0121] In addition, another aspect of the present application, in some embodiments, the present application provides a heterogeneous graph-based herbivorous animal feed formula optimization system, which is described with reference to Figure 4 The figure is a structural schematic diagram of the heterogeneous graph-based herbivorous animal feed formula optimization system according to some embodiments of the present application, which includes a construction module 401, a processing module 402 and an execution module 403, which are described as follows:

[0122] The construction module 401 is mainly used to construct a heterogeneous graph of grass-eating animal feed ingredients, taking target grass-eating animal species, grass-eating animal growth stages, feed raw materials and nutritional components as nodes.

[0123] The processing module 402 is used to perform multiple masking processes on the feed raw material nodes in the heterogeneous graph, to determine the correlation matrix of the synergistic effect between the remaining feed raw material nodes after each masking, and to calculate the compatibility degree vector of each feed raw material and other feed raw materials in the raw material matching according to all the correlation matrices through a loss function of mutual information maximization.

[0124] It should be noted that the processing module 402 is also used to extract the reference nutritional demand characteristics of the nutritional component nodes when delivering nutrition to the grass-eating animal growth stage nodes from the heterogeneous graph, to determine the nutritional demand satisfaction rate of the target grass-eating animal in each growth stage based on the reference nutritional demand characteristics and the initial growth stage information of the target grass-eating animal, to fuse and analyze all the nutritional demand satisfaction rates with the specificity base data corresponding to the target grass-eating animal species, and to generate the nutritional demand parameter vector of the target grass-eating animal in different growth stages.

[0125] The execution module 403 is mainly used to determine the nutritional adaptation degree of different feed raw materials to the target grass-eating animal by combining the compatibility degree vector and the nutritional demand parameter vector with the nutritional component characteristics of all feed raw materials, and to perform combination optimization on all feed raw materials in the feed formula based on all the nutritional adaptation degrees, to obtain the optimal feed ingredient scheme of the feed corresponding to the target grass-eating animal species.

[0126] In addition, the present application also provides a computer device, which comprises a memory and a processor, the memory stores codes, and the processor is configured to acquire the codes and execute the above-mentioned grass-eating animal feed formula optimization method based on a heterogeneous graph.

[0127] In some embodiments, with reference to Figure 5 The figure is a structural schematic diagram of a computer device for implementing the grass-eating animal feed formula optimization method based on a heterogeneous graph according to some embodiments of the present application. The grass-eating animal feed formula optimization method based on a heterogeneous graph in the above-mentioned embodiments can be implemented by the computer device shown in Figure 5 The computer device 500 comprises at least one processor 501, a communication bus 502, a memory 503 and at least one communication interface 504.

[0128] The processor 501 can be a general central processing unit (CPU), or an application-specific integrated circuit (ASIC).

[0129] The communication bus 502 can be used to transmit information between the above-mentioned components.

[0130] The memory 503 can be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disk storage, a magnetic disk storage or other magnetic storage devices, or any other medium capable of storing desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to this. The memory 503 can exist independently and be connected to the processor 501 through the communication bus 502. The memory 503 can also be integrated with the processor 501.

[0131] The memory 503 is used to store program code for executing the scheme of the present application, and the processor 501 is used to control the execution. The processor 501 is used to execute the program code stored in the memory 503. The program code can include one or more software modules. The method used in the above-mentioned embodiments can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0132] The communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as an Ethernet, a radio access network (RAN), a wireless local area network (WLAN), etc.

[0133] In a particular implementation, as one embodiment, the computer device can include multiple processors, each of which can be a single-CPU processor or a multi-CPU processor. A processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0134] The computer device described above can be a general-purpose computer device or a special-purpose computer device. In a particular implementation, the computer device can be a desktop computer, a laptop computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of the present application do not limit the type of the computer device.

[0135] In addition, the present application also provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the above-mentioned herbivorous animal feed formula optimization method based on a heterogeneous graph.

[0136] Although the preferred embodiments of the present application have been described, those skilled in the art who understand the basic inventive concept can make additional changes and modifications to the embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0137] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A method for optimizing feed formulations for herbivores based on heterogeneous graphs, wherein, The method of formulating feed for target herbivores using multiple feed ingredients is characterized by comprising the following steps: A heterogeneous graph of herbivore feed formulation is constructed using the target herbivore species, herbivore growth stage, feed ingredients and nutrient composition as nodes. The feed ingredient nodes in the heterogeneous graph are masked multiple times to determine the correlation matrix of the synergistic effect between the remaining feed ingredient nodes after each masking. Based on all the correlation matrices, the compatibility vector of each feed ingredient with other feed ingredients in the raw material matching is calculated by a loss function that maximizes mutual information. The baseline nutrient requirement features of the nutrient component nodes when transferring nutrients to the growth stage nodes of the herbivores are extracted from the heterogeneous graph. Based on the baseline nutrient requirement features and the initial growth stage information of the target herbivores, the nutrient requirement satisfaction rate of the target herbivores at each growth stage is determined. All nutrient requirement satisfaction rates are fused and analyzed with the specific basic data corresponding to the target herbivores species to generate nutrient requirement parameter vectors of the target herbivores at different growth stages. By combining the fit vector and the nutrient requirement parameter vector with the nutrient composition characteristics of all feed ingredients, the nutritional suitability of different feed ingredients for the target herbivores is determined. Based on all the nutritional suitability, all feed ingredients in the feed formula are combined and optimized to obtain the optimal feed formulation scheme for the target herbivorous species.

2. The method as described in claim 1, characterized in that, The process of performing multiple masking operations on the feed ingredient nodes in the heterogeneous graph, and then determining the correlation matrix of the synergistic effects among the remaining feed ingredient nodes after each masking, specifically includes: Determine the occlusion ratio of the feed ingredient nodes in the heterogeneous graph; The feed ingredient nodes in the heterogeneous graph are masked multiple times according to the masking ratio, and the remaining heterogeneous graph after each masking is recorded. Select a remaining heterogeneous graph as the selected remaining heterogeneous graph, and determine the association information between each feed ingredient node and other feed ingredient nodes in the selected remaining heterogeneous graph; Based on all the association information, determine the association matrix of the synergistic effects between feed ingredient nodes in the selected remaining heterogeneous graph; Continue to determine the correlation matrix of the synergistic effects among the feed ingredient nodes in the remaining heterogeneous graph.

3. The method as described in claim 1, characterized in that, Based on all the correlation matrices, the compatibility vector between each feed ingredient and other feed ingredients in ingredient formulation is calculated using a loss function that maximizes mutual information. Specifically, this includes: Standardize all the correlation matrices to obtain the standardized correlation matrices. Select one feed ingredient as the selected feed ingredient, and extract the correlation numerical sequence between the selected feed ingredient and other feed ingredients from all standardized correlation matrices; Construct a loss function that maximizes mutual information; Input the numerical sequence of the correlation between the selected feed ingredient and other feed ingredients into the loss function that maximizes mutual information, and output the compatibility vector of the selected feed ingredient and other feed ingredients when the ingredients are matched. Continue to determine the compatibility vector between the remaining feed ingredients and other feed ingredients when formulating them.

4. The method as described in claim 1, characterized in that, The baseline nutrient requirement characteristics extracted from the heterogeneous graph for nutrient transfer from nutrient component nodes to herbivore growth stage nodes specifically include: The direct association paths between nutrient nodes and herbivore growth stage nodes are extracted from the heterogeneous graph. The required amounts of each nutrient at each stage of the growth process of herbivores are determined based on the heterogeneous diagram. Based on all the demand corresponding to the herbivore growth stage nodes and the directly associated paths, determine the baseline nutrient demand characteristics when nutrient component nodes transfer nutrients to herbivore growth stage nodes.

5. The method as described in claim 1, characterized in that, Based on the aforementioned baseline nutritional requirements characteristics and combined with the initial growth stage information of the target herbivores, the nutritional requirement satisfaction rate of the target herbivores at each growth stage is determined, specifically including: Obtain information on the initial growth stage of the target herbivore; The nutrient component feature vectors of the growth stage nodes are determined based on the aforementioned baseline nutrient requirement characteristics. The initial growth stage information and the nutrient component feature vector are fused to determine the nutrient requirement satisfaction rate of the target herbivore at each growth stage.

6. The method as described in claim 1, characterized in that, By fusing and analyzing all nutritional requirement fulfillment rates with specific baseline data corresponding to the target herbivore species, a vector of nutritional requirement parameters for the target herbivore at different growth stages is generated, specifically including: Obtain specific basic data on the target herbivorous animal species; By correlating and aggregating all nutritional requirement satisfaction rates with the specific basic data of the target herbivorous animal species, we can obtain aggregated information on the relationship between the species characteristics and nutritional suitability of the target herbivorous animal. Based on the aggregated information, a vector of nutritional requirements parameters for the target herbivore at different growth stages is generated.

7. The method as described in claim 1, characterized in that, Determining the nutritional suitability of different feed ingredients for the target herbivore by combining the fit vector and the nutritional requirement parameter vector with the nutritional component characteristics of all feed ingredients specifically includes: Determine the nutritional characteristics of all feed ingredients; The matching degree between the nutritional components of each feed ingredient and the specific needs of herbivorous animal breeds is determined based on the nutritional requirement parameter vector and the nutritional component characteristics. The nutritional suitability of each feed ingredient for the target herbivore is determined based on the matching degree between the nutritional composition of each feed ingredient and the specific needs of the herbivore species, and the fit vector.

8. The method as described in claim 1, characterized in that, Based on all nutritional compatibility factors, the combination and optimization of all feed ingredients in the feed formulation are performed to obtain the optimal feed formulation for the target herbivore species. Specifically, this includes: Obtain the initial feed formulation plan for all feed ingredients in the feed formula; The initial feed formulation scheme is optimized and adjusted based on all nutritional compatibility factors to obtain the optimal feed formulation scheme for the target herbivore species.

9. The method as described in claim 1, characterized in that, Obtain initial growth stage information of the target herbivores from the breeding management system database.

10. A system for optimizing feed formulation for herbivores based on heterogeneous graphs, characterized in that, include: The module is used to construct a heterogeneous graph for herbivore feed formulation, with the target herbivore species, herbivore growth stage, feed ingredients and nutrients as nodes. The processing module is used to perform multiple masking processes on the feed ingredient nodes in the heterogeneous graph, thereby determining the correlation matrix of the synergistic effect between the remaining feed ingredient nodes after each masking, and calculating the compatibility vector of each feed ingredient with other feed ingredients in the raw material matching based on all the correlation matrices and through the loss function that maximizes mutual information. The processing module is also used to extract the baseline nutritional requirement features when the nutrient component nodes transfer nutrients to the growth stage nodes of the herbivores from the heterogeneous graph, determine the nutritional requirement satisfaction rate of the target herbivores at each growth stage based on the baseline nutritional requirement features and the initial growth stage information of the target herbivores, and fuse all the nutritional requirement satisfaction rates with the specific basic data corresponding to the target herbivores species to generate a nutritional requirement parameter vector of the target herbivores at different growth stages. The execution module is used to determine the nutritional suitability of different feed ingredients for the target herbivorous animal by combining the fit degree vector and the nutritional requirement parameter vector with the nutritional component characteristics of all feed ingredients, and to optimize the combination of all feed ingredients in the feed formula based on all nutritional suitability to obtain the optimal feed formulation scheme for the target herbivorous animal species.

Images