Artificial intelligence-based beef cattle breeding planning method, device, equipment and medium

Abstract

The application relates to an artificial intelligence-based beef cattle breeding planning method, device, equipment and medium, the method comprising the following steps: acquiring multi-dimensional breeding basic data of a target population, taking each target breeding individual as a node and different correlation relationships as correlation edges to construct a population breeding multi-relation heterogeneous graph; inputting the population breeding multi-relation heterogeneous graph into a heterogeneous graph attention network based on a time sequence causal attention mechanism to output an individual representation vector; inputting the individual representation vector into a multi-scenario trait prediction network based on Bayesian deep learning to obtain a target trait prediction value and a confidence interval; constructing and solving a multi-objective long-term planning optimization model based on the target trait prediction value and the confidence interval, generating an optimal population mating sequence and an environmental management strategy sequence, decoupling and structuring in a time dimension, and generating a dynamic beef cattle breeding planning scheme. The method can fuse multi-dimensional breeding data and quantize trait prediction uncertainty, and supports stable and efficient intelligent beef cattle breeding decision-making.

Description

Technical Field

[0001] This invention belongs to the field of animal husbandry, and in particular relates to a method, device, equipment and medium for beef cattle breeding planning based on artificial intelligence. Background Technology

[0002] With the deepening application of artificial intelligence and big data technologies in livestock breeding, livestock breeding is gradually shifting from traditional experience-based decision-making to data-driven intelligent decision-making. This has led to the emergence of neural network-based methods for predicting beef cattle traits. For example, by constructing a two-stage model combining a fully connected neural network and a Long Short-Term Memory (LSTM) network, the reproductive trait performance of offspring at different age stages can be predicted using age-dependent trait data from breeding bulls and cows. Such methods can, to some extent, compensate for the shortcomings of traditional evaluation methods in time-series modeling and multi-period prediction, exhibiting good real-time performance and applicability.

[0003] However, existing machine learning-based breeding optimization methods still have several key problems: First, most of their models only predict single or a few traits, without fully considering the fusion and utilization of multi-dimensional data such as genomic information, pedigree structure, and spatiotemporal environment; Second, existing methods lack a systematic optimization mechanism for genetic diversity, long-term genetic risks, and multi-generation breeding goals; Third, most models do not introduce quantitative assessment of prediction uncertainty, making it difficult to support robust long-term breeding decisions; In addition, existing technologies are often limited to short-term trait prediction and have not yet formed a dynamic breeding planning system covering mating, environment, and management. Summary of the Invention

[0004] Therefore, it is necessary to provide an artificial intelligence-based method, device, equipment, and medium for beef cattle breeding planning to address the aforementioned technical problems.

[0005] Firstly, this application provides an artificial intelligence-based method for beef cattle breeding planning, including:

[0006] S101. Obtain multi-dimensional basic breeding data of target breeding individuals in the target population;

[0007] S102. Based on multi-dimensional breeding basic data, each target breeding individual in the target population is taken as an individual node, and different association edges are established between individual nodes to obtain a multi-relationship heterogeneous graph of population breeding.

[0008] S103. Input the heterogeneous graph of multiple relationships in population breeding into a heterogeneous graph attention network based on temporal causal attention mechanism, and output individual representation vectors; wherein, the individual representation vectors are used to represent the structure, attributes and temporal context information of the target breeding individual in the heterogeneous graph of multiple relationships in population breeding.

[0009] S104. Input the individual representation vector into a multi-scenario trait prediction network based on Bayesian deep learning mechanism to obtain the predicted values ​​and confidence intervals of the target traits of each target breeding individual in multiple key breeding cycles in the future.

[0010] S105. Based on the predicted values ​​and confidence intervals of the target traits of individuals in each target population, construct a multi-objective long-term programming optimization model, solve the multi-objective long-term programming optimization model, and generate the optimal population mating sequence and the corresponding environmental management strategy sequence.

[0011] S106. Decouple and structure the optimal population mating sequence and environmental management strategy sequence according to the time dimension to generate a dynamic beef cattle breeding plan.

[0012] Secondly, this application also provides an artificial intelligence-based beef cattle breeding planning device, comprising:

[0013] The data acquisition module is used to acquire multi-dimensional basic breeding data of individuals in the target population;

[0014] The relationship graph construction module is used to obtain a heterogeneous multi-relationship graph of population breeding by taking each target breeding individual in the target population as an individual node based on multi-dimensional breeding basic data and establishing association edges with different association relationships between individual nodes.

[0015] The attention representation module is used to input the heterogeneous graph of multiple relationships in population breeding into a heterogeneous graph attention network based on a temporal causal attention mechanism, and output individual representation vectors. The individual representation vectors are used to represent the structure, attributes and temporal context information of the target breeding individual in the heterogeneous graph of multiple relationships in population breeding.

[0016] The trait prediction module is used to input individual representation vectors into a multi-scenario trait prediction network based on a Bayesian deep learning mechanism to obtain the predicted values ​​and confidence intervals of the target traits of each target breeding individual in multiple key breeding cycles in the future.

[0017] The optimization decision module is used to construct a multi-objective long-term programming optimization model based on the predicted values ​​and confidence intervals of the target traits of individuals in each target population, and solve the multi-objective long-term programming optimization model to generate the optimal population mating sequence and the corresponding environmental management strategy sequence.

[0018] The scheme generation module is used to decouple and structure the optimal population mating sequence and environmental management strategy sequence according to the time dimension to generate a dynamic beef cattle breeding plan scheme.

[0019] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0020] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0021] The aforementioned AI-based beef cattle breeding planning methods, devices, equipment, and media, through the systematic integration of multi-dimensional data including genome, phenotype, pedigree, and environment, lay the foundation for a comprehensive digital characterization of target breeding individuals. By constructing a heterogeneous graph of population breeding relationships that integrates pedigree, genetic association, and environmental co-occurrence, complex breeding relationship networks can be formalized. Based on this, a heterogeneous graph attention network based on a temporal causal attention mechanism performs deep reasoning on the heterogeneous graph of population breeding relationships, generating robust individual representation vectors that include structure, attributes, and evolutionary context. Subsequently, a multi-contextual trait prediction network based on Bayesian deep learning enables accurate prediction of future multi-periodic traits. This method predicts and quantifies uncertainty, providing a risk perception basis for decision-making. By constructing and solving a multi-objective long-term programming optimization model centered on maximizing expected genetic gain and minimizing cumulative predicted risk, it can generate mating sequences and synergistic environmental strategy sequences that optimally balance gains and risks. By decoupling and structuring the mating sequences and synergistic environmental strategy sequences into executable mating work orders and environmental management instruction sets organized by time phases, it can generate dynamic planning schemes to guide integrated breeding production. This enables a complete technical closed loop from multi-source data fusion, deep representation learning, and risk quantification prediction to long-term robust optimization and implementation, significantly improving the systematicness, accuracy, and operability of breeding decisions. This approach can integrate multi-dimensional breeding data and quantify trait prediction uncertainty, supporting robust and efficient intelligent beef cattle breeding decisions. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating an artificial intelligence-based beef cattle breeding planning method provided as an exemplary embodiment of this application;

[0024] Figure 2This is a schematic diagram of an artificial intelligence-based beef cattle breeding planning device provided as an exemplary embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0026] In one embodiment, such as Figure 1 As shown, an artificial intelligence-based beef cattle breeding planning method is provided. This embodiment illustrates the application of this method to a breeding planning terminal. It is understood that this method can also be applied to a breeding planning server, and further to a system including both a breeding planning terminal and a breeding planning server, and is implemented through the interaction between the two. In this embodiment, the method may include the following steps:

[0027] S101. Obtain multi-dimensional basic breeding data of the target breeding individuals in the target population.

[0028] Specifically, the breeding planning terminal can obtain multi-dimensional basic breeding data for each target breeding individual in the target population from multiple heterogeneous data sources.

[0029] Optionally, the sources of multi-dimensional breeding basic data may include, but are not limited to, professional gene sequencing platforms, breeding management systems for livestock farms, livestock population databases, and environmental monitoring equipment.

[0030] Furthermore, the data sources for multi-dimensional breeding foundational data can include, but are not limited to: Genomic Single Nucleotide Polymorphism Data (GSNPD) provided by high-throughput gene sequencing platforms. GSNPD can record the genotypes of each target breeding individual at hundreds of thousands to millions of pre-defined single nucleotide polymorphism (SNP) loci; historical multi-generational genotype data from ranch production management systems; official or internal enterprise population pedigree data; and spatiotemporal environmental data continuously collected by IoT sensor networks deployed in cattle sheds and the environment. Historical multi-generational genotype data is typically stored in structured tabular form and can include multiple production trait measurements and reproductive trait records of the target breeding individual and its ancestors at different growth stages. Population pedigree data typically exists in the form of parentage lists and pedigree charts, recording the ancestry and kinship among individuals. Spatiotemporal environmental data can be sequence data with timestamps and spatial location tags.

[0031] Optionally, the breeding planning terminal can standardize the format of the collected multi-dimensional basic breeding data.

[0032] S102. Based on multi-dimensional breeding basic data, each target breeding individual in the target population is taken as an individual node, and different association edges are established between individual nodes to obtain a multi-relationship heterogeneous graph of population breeding.

[0033] Optionally, the population breeding multi-relationship heterogeneous graph can be an attribute graph data structure, which can be used to characterize the complex relationship network of the target population in the dimensions of genetics, lineage and environment.

[0034] Specifically, the breeding planning terminal can construct individual nodes based on multi-dimensional breeding basic data. The breeding planning terminal can use the unique identifier of each target breeding individual as the node ID, and use the core features of the SNP loci of the target breeding individual, key indicators of historical phenotypes, pedigree identification information and breeding unit association information as node attributes to generate an initial set of individual nodes.

[0035] Furthermore, the breeding planning terminal can establish different types of association edges according to preset association construction rules: the breeding planning terminal can construct pedigree edges between individual nodes with direct parent-child relationships by matching the unique identifiers of the father, mother, and offspring based on the parent-child relationship records in the population pedigree data; the breeding planning terminal can calculate the genomic genetic similarity based on GSNPD data, and construct genetic association edges between individual nodes whose genomic genetic similarity exceeds the preset genomic genetic similarity threshold; the breeding planning terminal can identify individual node pairs in the same breeding unit within the same preset breeding time period based on the feeding group identifier and timestamp information in the spatiotemporal environment data, and construct environmental co-occurrence edges between these individual node pairs.

[0036] Furthermore, the breeding planning terminal can integrate all individual nodes, pedigree edges, genetic association edges, and environmental co-occurrence edges to generate a multi-relationship heterogeneous graph for population breeding.

[0037] S103. Input the heterogeneous graph of multiple relationships in population breeding into a heterogeneous graph attention network based on temporal causal attention mechanism, and output the individual representation vector.

[0038] Optionally, the individual representation vector can be used to represent the structure, attributes, and temporal context information of the target breeding individual in the multi-relationship heterogeneous graph of population breeding.

[0039] Optionally, heterogeneous graph attention networks can be an extension of graph attention networks, capable of handling heterogeneous graphs with various node and edge types. Temporal causal attention mechanisms can be introduced into heterogeneous graph attention networks.

[0040] For example, for different types of related edges in a heterogeneous graph, a heterogeneous graph attention network can learn different attention weight calculation functions, enabling it to distinguish the differences in the contribution of different relationships to the representation of the target node during information aggregation. For instance, when predicting the trait of a target breeding individual, the information weight of its parent-child relationship edge may be calculated differently than the weight of its edge for individuals in the same environment. More importantly, the temporal causal attention mechanism, when calculating attention weights, not only calculates the similarity of node features but can also embed constraints of temporal order and causal inference.

[0041] Furthermore, heterogeneous graph attention networks can learn more causally explanatory target breeding individual dependencies by analyzing timestamp sequences in attribute data and reducing confounding biases caused by factors such as environmental co-occurrence.

[0042] Furthermore, heterogeneous graph attention networks can generate low-dimensional, dense individual representation vectors for each node through multi-layered information propagation and aggregation. These individual representation vectors can deeply integrate the multi-dimensional attributes of the individual node itself, its structural position in the complex social network, and contextual information considering temporal causality.

[0043] Specifically, the breeding planning terminal can convert the input format of the heterogeneous graph of multi-relationship breeding, transforming node attributes, edge types, and temporal information into feature matrices, edge type matrices, and temporal matrices recognizable by the heterogeneous graph attention network, respectively. Secondly, the breeding planning terminal can input these recognizable feature matrices, edge type matrices, and temporal matrices into the heterogeneous graph attention network. Through multi-layer convolutional operations, it extracts the local structural features of nodes and adaptively allocates the contribution of different types of associated edges using an attention mechanism, obtaining the contextual information for each breeding cycle. The heterogeneous graph attention network can then perform feature fusion and nonlinear transformations on the contextual information from each breeding cycle, outputting individual representation vectors.

[0044] S104. Input the individual representation vector into a multi-scenario trait prediction network based on Bayesian deep learning mechanism to obtain the predicted values ​​and confidence intervals of the target traits of each target breeding individual in multiple key breeding cycles in the future.

[0045] Optionally, the multi-scenario trait prediction network can be constructed based on the Bayesian Deep Learning (BDL) mechanism. This network can model network parameters by introducing probability distributions, thus quantifying the uncertainty of prediction results. The multi-scenario trait prediction network can include a temporal prediction layer, a multi-scenario adaptation layer, and a probability output layer. The temporal prediction layer can employ a Recurrent Neural Network (RNN) structure, which can be used to learn the evolutionary patterns of the target trait over the breeding cycle. The multi-scenario adaptation layer can simulate potential scenarios such as different feeding environments and genetic backgrounds, improving the model's adaptability to complex breeding scenarios. The probability output layer can calculate the predicted value of the target trait and the corresponding confidence interval based on Bayesian inference principles. The width and variance of the confidence interval directly characterize the degree of uncertainty in the prediction results.

[0046] Optionally, the target trait prediction value can be the optimal estimate of a specific economic trait provided by the multi-scenario trait prediction network for each target breeding individual, under each preset future scenario, and in each critical breeding cycle.

[0047] Specifically, the breeding planning terminal can normalize the individual representation vector to eliminate the dimensional differences between different feature dimensions and obtain the normalized individual representation vector.

[0048] Furthermore, the breeding planning terminal can input the normalized individual representation vector into the multi-scenario trait prediction network, and through multiple rounds of forward propagation calculation, output the predicted value of the target trait and the confidence interval corresponding to the predicted value of the target trait for each target breeding individual in multiple key breeding cycles in the future.

[0049] S105. Based on the predicted values ​​and confidence intervals of the target traits of individuals in each target population, construct a multi-objective long-term programming optimization model, solve the multi-objective long-term programming optimization model, and generate the optimal population mating sequence and the corresponding environmental management strategy sequence.

[0050] Optionally, the optimal population mating sequence can be the final decision scheme selected from the Pareto optimal solution set of the multi-objective long-term programming optimization model, and the optimal population mating sequence can be the core reproductive instruction set that drives the population genetic structure to evolve along the optimized path.

[0051] Alternatively, the environmental management strategy sequence can be directly input into the ranch management execution system or a dynamic environmental management schedule that guides manual operations.

[0052] Specifically, the breeding planning terminal can obtain the core decision variables of the multi-objective long-term planning optimization model, which may include individual mating combination relationships and environmental management strategy parameters.

[0053] Optionally, the individual mating combinations may include, but are not limited to, the selection of mating individuals and the timing of mating, while the environmental management strategy parameters may include, but are not limited to, indicators for controlling the rearing environment and parameters related to the nutritional supply program. The individual mating combinations and environmental management strategy parameters can change dynamically over multiple key breeding cycles in the future.

[0054] Optionally, the objective function of the multi-objective long-term programming optimization model can be constructed based on maximizing the total expected gain of the target trait and minimizing the cumulative prediction risk. The breeding planning terminal can calculate the total expected gain of the target trait by integrating the predicted values ​​of the target trait from each breeding cycle. The breeding planning terminal can obtain the cumulative prediction risk based on the confidence interval width or variance quantification corresponding to each predicted value. The constraints of the multi-objective long-term programming optimization model can include genetic diversity constraints, mating resource constraints, and rearing capacity constraints. Among them, the breeding planning terminal can achieve the genetic diversity constraint by controlling the distribution of genetic similarity among target breeding individuals. The breeding planning terminal can use a multi-objective intelligent optimization algorithm to solve the multi-objective long-term programming optimization model. Through iterative operations such as population initialization, selection, crossover, and mutation, it searches for a solution that satisfies all constraints and optimizes the objective function, and finally outputs the optimal population mating sequence and the corresponding environmental management strategy sequence. The mating sequence can represent the mating combination of each breeding cycle, and the environmental management strategy sequence can represent the environmental regulation scheme that changes over time.

[0055] S106. Decouple and structure the optimal population mating sequence and environmental management strategy sequence according to the time dimension to generate a dynamic beef cattle breeding plan.

[0056] Optionally, the breeding planning terminal can integrate multi-dimensional breeding data and multi-objective optimization decision results based on artificial intelligence technology to obtain a dynamic beef cattle breeding planning scheme in a time-based structure.

[0057] Specifically, the breeding planning terminal can decouple the optimal population mating sequence in terms of time dimension. Based on the natural cycle of beef cattle breeding, the optimal population mating sequence is broken down into several consecutive time-stage mating plans. Each mating plan can include the core information of the breeding bulls, the screening results of the breeding cows, and the genetic potential analysis of the expected offspring within that time stage.

[0058] Furthermore, the breeding planning terminal can extract environmental control parameters and their corresponding future execution timestamps from the environmental management strategy sequence. The breeding planning terminal can align each breeding plan with the future execution timestamps corresponding to the environmental control parameters, and match the environmental control parameters to the corresponding feeding stages in each breeding plan based on the spatiotemporal environmental data mapping relationship, thereby generating a standardized set of environmental control instructions.

[0059] Optionally, the environmental control instruction set may include, but is not limited to, instructions for controlling the feeding environment and adjusting the nutritional formula for each feeding stage.

[0060] Furthermore, the breeding planning terminal can associate and encapsulate the mating plans for each time period and the corresponding environmental control instruction sets for each time period according to the time axis, generating a structured dynamic beef cattle breeding planning scheme.

[0061] The aforementioned AI-based beef cattle breeding planning method, by systematically integrating multi-dimensional data from genomes, phenotypes, pedigrees, and environment, lays the foundation for a comprehensive digital characterization of target breeding individuals. By constructing a heterogeneous graph of population breeding relationships that integrates pedigree, genetic associations, and environmental co-occurrence, it formalizes complex breeding relationship networks. Based on this, a heterogeneous graph attention network using a temporal causal attention mechanism performs deep reasoning on the heterogeneous graph, generating robust individual representation vectors that include structure, attributes, and evolutionary context. Subsequently, a multi-contextual trait prediction network based on Bayesian deep learning enables accurate prediction and quantitative analysis of future multi-periodic traits. This method mitigates uncertainty and provides a risk perception basis for decision-making. By constructing and solving a multi-objective long-term programming optimization model centered on maximizing expected genetic gain and minimizing cumulative predicted risk, it can generate mating sequences and synergistic environmental strategy sequences that optimally balance gains and risks. By decoupling and structuring the mating sequences and synergistic environmental strategy sequences into executable mating work orders and environmental management instruction sets organized by time phases, it can generate dynamic programming schemes to guide integrated breeding production. This enables a complete technical closed loop from multi-source data fusion, deep representation learning, and risk quantification prediction to long-term robust optimization and implementation, significantly improving the systematicness, accuracy, and operability of breeding decisions. This method can integrate multi-dimensional breeding data and quantify trait prediction uncertainty, supporting robust and efficient intelligent beef cattle breeding decisions.

[0062] In one embodiment, multidimensional breeding basic data may include genome single nucleotide polymorphism data, historical multi-type data, population pedigree data, and spatiotemporal environmental data;

[0063] Based on multi-dimensional breeding data, each target breeding individual in the target population is treated as an individual node, and association edges with different relationships are established between individual nodes to obtain a heterogeneous graph of multi-relationship breeding in the population. This may include the following steps:

[0064] S201. Based on pedigree record data, construct pedigree kinship edges between individual nodes with direct parent-child relationships.

[0065] Optionally, pedigree record data can be structured data generated during the livestock breeding process to trace the lineage of individuals. The sources of this pedigree record data may include farm breeding management systems, livestock industry population registration databases, and artificial breeding archives.

[0066] Optionally, pedigree record data may include, but is not limited to, a unique identifier for each target breeding individual in the target population, a unique identifier for the father, a unique identifier for the mother, the time of birth, and pedigree information.

[0067] Optionally, phylogenetic edges can be used to characterize direct parent-child relationships between individual nodes.

[0068] Optionally, the attributes of the pedigree edge may include, but are not limited to, kinship type and parent-child relationship confirmation.

[0069] For example, the breeding planning terminal can clean and standardize pedigree record data, removing data entries with ambiguous kinship and / or missing identifiers.

[0070] Furthermore, the breeding planning terminal can extract the paternal and maternal identifiers for each individual node, and construct pedigree edges between the target individual node and its corresponding paternal node using the paternal identifier, and between the target individual node and its corresponding maternal node using the maternal identifier. The breeding planning terminal can associate the attributes of the pedigree edges with the pedigree edge structure. Pedigree edges can represent the direct genetic inheritance relationship between target breeding individuals.

[0071] S202. Based on genomic single nucleotide polymorphism data, calculate the genomic genetic similarity between any two individual nodes, and construct genetic association edges between individual nodes whose genomic genetic similarity exceeds a preset genomic genetic similarity threshold.

[0072] Optionally, the breeding planning terminal can set a genomic genetic similarity threshold according to the breeding objectives. The genomic genetic similarity threshold can be used to screen individual node pairs with significant genetic associations.

[0073] For example, the breeding planning terminal can convert the alleles in the SNP data into numerical codes according to preset coding rules, and calculate the SNP site weight coefficient of each SNP site in combination with the secondary allele frequency of the target population.

[0074] Optionally, the breeding planning terminal can calculate the SNP weight coefficient of each SNP locus based on the numerical coding value and the secondary allele frequency of the target population, and calculate the genomic genetic similarity between any two individual nodes.

[0075] Furthermore, the breeding planning terminal can compare genomic genetic similarity with a genomic genetic similarity threshold. If the genomic genetic similarity exceeds the threshold, the breeding planning terminal can construct a genetic association edge between two individual nodes and use genomic genetic similarity as the core attribute of the genetic association edge. Genomic genetic similarity can be used to characterize the tightness of the genetic association between target breeding individuals.

[0076] S203. Based on the breeding group and timestamp information in the spatiotemporal environmental data, identify the pairs of individual nodes that are in the same breeding unit within the same preset breeding time period, and construct environmental co-occurrence edges between the pairs of individual nodes.

[0077] Optionally, the spatiotemporal environmental data can be multi-dimensional data collected by IoT sensors, feeding management systems, and manual recording devices deployed at the breeding site. The spatiotemporal environmental data can be used to reflect the breeding environment and spatiotemporal relationship of the target breeding individual.

[0078] Optionally, the breeding time period can be a time interval set by the breeding planning terminal based on the needs of beef cattle growth and development cycle and breeding stage division.

[0079] For example, the breeding planning terminal can extract the rearing group identifier, rearing unit information, and the timestamp sequence corresponding to the rearing unit information for each individual node from spatiotemporal environmental data. The timestamp sequence includes all spatiotemporal environmental data collection time records for the target breeding individual at different rearing stages.

[0080] Furthermore, the breeding planning terminal can standardize the timestamp sequence corresponding to the feeding unit information, converting the non-uniform time format generated by different acquisition devices into a consistent numerical time format.

[0081] Furthermore, the breeding planning terminal can compare the standardized timestamps with the start and end timestamps of the breeding period, and filter out the spatiotemporal environmental data whose timestamp values ​​are between the start and end timestamps, as valid data for the target breeding individual within the breeding period.

[0082] Furthermore, the breeding planning terminal can match feeding unit information, identify individual node pairs within the same feeding unit during the same breeding time period, and construct environmental co-occurrence edges between individual node pairs within the same feeding unit. These environmental co-occurrence edges can be used to characterize the environmental association features between target breeding individuals.

[0083] Optionally, the attributes of the environmental co-occurrence edge may include, but are not limited to, co-occurrence time period, rearing unit identifier, and rearing group information.

[0084] S204. Based on each individual node, pedigree edge, genetic association edge, and environmental co-occurrence edge, a multi-relationship heterogeneous graph of population breeding is obtained.

[0085] For example, the breeding planning terminal can integrate all standardized individual nodes. Subsequently, the breeding planning terminal can classify and identify the constructed pedigree edges, genetic association edges, and environmental co-occurrence edges, assigning a unique edge type identifier to each type of association edge. The edge type identifier can be used to distinguish association relationships in different dimensions.

[0086] Furthermore, the breeding planning terminal can map individual nodes to pre-constructed pedigree edges, genetic association edges, and environmental co-occurrence edges according to the graph data structure specifications, establish correspondences between nodes and edges, and between edges and edge attributes, and generate a population breeding multi-relationship heterogeneous graph including a set of individual nodes, a set of multiple types of associated edges, and a set of edge attributes.

[0087] In this embodiment, the breeding planning terminal integrates pedigree, genome, and environmental multi-source data and constructs a multi-relationship heterogeneous graph to achieve an integrated structured representation of the genetic, kinship, and environmental association networks within a population, laying the foundation for subsequent in-depth intelligent analysis based on associated data.

[0088] In one embodiment, the genomic genetic similarity between any two individual nodes is calculated based on genomic single nucleotide polymorphism data, including:

[0089] S301. Based on the single nucleotide polymorphism sites in the genome single nucleotide polymorphism data, the alleles of each target breeding individual are numerically encoded to obtain the allele count, and the allele count is used as the allele encoding value of the target breeding individual at the single nucleotide polymorphism site.

[0090] Optionally, the breeding program terminal can perform low-quality site filtering and missing value imputation on GSNPD.

[0091] Furthermore, the breeding planning terminal can perform operations according to preset coding rules: for each SNP locus, the breeding planning terminal can encode the homozygous dominant genotype as the baseline value of the allele count; the breeding planning terminal can encode the heterozygous genotype as the median value of the allele count; the breeding planning terminal can encode the homozygous recessive genotype as the maximum value of the allele count. The breeding planning terminal outputs the allele count of each target breeding individual at each SNP locus and uses the allele count as the allele coding value of the target breeding individual at the single nucleotide polymorphism site. This coding method can preserve the genetic dosage effect of genotype.

[0092] S302. Based on the allele coding values ​​of the target population at each single nucleotide polymorphism site, calculate the minor allele frequency of each single nucleotide polymorphism site.

[0093] Optionally, the minor allele frequency can be used to characterize the distribution of minor alleles at each single nucleotide polymorphism site in the target population.

[0094] For example, the breeding planning terminal can collect data on all target breeding individuals in the target population during the [number]th [period]. Allele coding values ​​at each SNP locus were obtained by reverse mapping according to the coding rules to obtain the allele coding values ​​of each target breeding individual at the [number]th [location]. Alleles at each SNP locus constitute the genome.

[0095] Furthermore, the breeding planning terminal can obtain the total number of alleles at the SNP locus and count the total number of minor alleles. The breeding planning terminal can then calculate the ratio of the total number of minor alleles to the total number of alleles to obtain the number of the first allele. Secondary allele frequencies of each SNP locus.

[0096] S303. Calculate the genomic genetic similarity between any two individual nodes based on allele coding values ​​and suballele frequencies;

[0097] Alternatively, the expression for genomic genetic similarity can be:

[0098]

[0099]

[0100] in, Indicates the target breeding individual With target breeding individuals Genomic genetic similarity between them This indicates the total number of single nucleotide polymorphism sites. Indicates the target breeding individual In the Allelic genotype coding values ​​at single nucleotide polymorphism sites, Indicates the target breeding individual In the Allelic genotype coding values ​​at single nucleotide polymorphism sites, Represents a similarity indicator function. Indicates the first Site weighting coefficients for each single nucleotide polymorphism site. Indicates the target population in the th Secondary allele frequencies at single nucleotide polymorphism sites.

[0101] For example, a similarity indicator function can be used to quantify target breeding individuals. and target breeding individuals In the The genotypic similarity of each SNP locus. The breeding planning terminal can, for any two individual nodes, traverse all SNP loci and calculate the genotypic similarity of each locus. and Multiply and sum the products to obtain the numerator; and calculate all sites. The sum of the numerator and denominator terms is used to obtain the denominator; finally, the genomic genetic similarity between the two target breeding individuals is obtained by the ratio of the numerator to the denominator. .

[0102] In this embodiment, the breeding planning terminal provides a reliable basis for constructing a breeding association map by standardizing and encoding genomic data, calculating site frequencies, and introducing frequency-weighted fusion.

[0103] In one embodiment, a multi-objective long-term programming optimization model is constructed based on the predicted values ​​and confidence intervals of the target traits of individuals in each target population. Solving the multi-objective long-term programming optimization model generates the optimal population mating sequence and a corresponding environmental management strategy sequence. This process may include the following steps:

[0104] S401. Define the individual mating combination relationship and environmental management strategy parameters for the multi-objective long-term planning optimization model.

[0105] Optionally, individual mating combinations and environmental management strategy parameters can change over time in multiple key breeding cycles.

[0106] Optionally, individual mating combinations can be used to characterize the combination rules of bulls and cows that can participate in mating in the target population during future breeding cycles.

[0107] Optionally, the individual mating combination relationship may include, but is not limited to, the screening criteria for mating individuals, combination priority, and mating time window information.

[0108] Optionally, environmental management strategy parameters can be used to characterize the environmental regulation dimensions that support the achievement of breeding objectives.

[0109] Optionally, environmental management strategy parameters may include, but are not limited to, the temperature and humidity control range of the rearing environment, the nutrient composition ratio of the nutritional formula, and the stocking density threshold.

[0110] For example, the breeding planning terminal can refer to the historical mating effect data of the target population, the results of genetic structure analysis, and the preset long-term breeding goals to define the basic screening conditions and environmental parameter benchmark ranges for mating combinations in each cycle.

[0111] Furthermore, the breeding planning terminal can combine the phase division of future breeding cycles to obtain the adjustment direction and correlation logic of parameters in each cycle, ensuring that the individual mating combination relationship and environmental management strategy parameters can be dynamically adapted with population evolution and breeding progress.

[0112] S402. Construct the objective function of a multi-objective long-term programming optimization model to maximize the total expected gain of the target trait and minimize the cumulative prediction risk determined by the confidence interval; wherein the cumulative prediction risk is calculated based on the width or variance of the individual confidence interval.

[0113] Schematic, the objective function of a multi-objective long-term programming optimization model can be constructed following multi-objective optimization theory, with the core being the balance between breeding benefits and decision-making risks. The objective function of a multi-objective long-term programming optimization model can include maximizing the expected gain of the target trait and minimizing the cumulative prediction risk.

[0114] Optionally, the term maximizing the expected gain of the target trait can be constructed by quantifying the genetic gain potential of long-term breeding based on the simulated predicted values ​​of the target trait for each target breeding individual in multiple key breeding cycles in the future.

[0115] Optionally, minimizing the cumulative prediction risk can be constructed based on the principle of uncertainty quantification. This can be achieved by using the confidence interval output by Bayesian deep learning and setting the cumulative prediction risk based on the interval width or variance. The interval width or variance reflects the degree of uncertainty in the prediction results; a wider confidence interval and a larger variance indicate a higher prediction risk. The cumulative prediction risk can be obtained by integrating the risk values ​​of each cycle and each target breeding individual.

[0116] For example, the breeding planning terminal can aggregate the predicted values ​​of the target traits of each target breeding individual over time according to the breeding cycle to calculate the total expected gain of the target trait. The expression for the total expected gain of the target trait can be:

[0117]

[0118] In the formula, This represents the total expected gain of the target trait. Indicates the number of breeding cycles. Indicates the target number of individuals for breeding. Indicates the first The target breeding individual in the first The predicted values ​​of target traits for each breeding cycle can be calculated using time-series aggregation to determine the long-term genetic gain potential.

[0119] Furthermore, the breeding planning terminal can calculate the cumulative predicted risk using a weighted summation method based on the width and variance of the confidence interval. The breeding planning terminal can integrate maximizing genetic gain and minimizing predicted risk into a unified solvable objective function through multi-objective optimization techniques.

[0120] Optionally, multi-objective optimization techniques may include, but are not limited to, weighted combination and Pareto optimization methods.

[0121] S403. Set the breeding planning constraints for the multi-objective long-term planning optimization model.

[0122] Optionally, breeding program constraints may include genetic diversity constraints to maintain the feasibility and sustainability of the breeding program, which can be set based on population genetics and breeding resource allocation theory.

[0123] Optionally, genetic diversity constraints can be used to prevent population genetic decline caused by inbreeding. Genetic diversity constraints can be obtained by setting the distribution of genomic genetic similarity among target breeding individuals in the target population.

[0124] Optionally, the constraints of the multi-objective long-term programming optimization model may include, but are not limited to, breeding resource constraints, feeding capacity constraints, and environmental regulation feasibility constraints.

[0125] For example, the breeding planning terminal can set an upper limit threshold for the overall genetic similarity of the population and a lower limit standard for the heterozygosity of key SNP sites for genetic diversity constraints; for other constraints, the breeding planning terminal can combine the actual resource conditions, facility carrying capacity and environmental control technology level of the farm to obtain the boundary range of each constraint.

[0126] Furthermore, the breeding planning terminal can embed all constraints into the target long-term planning optimization model in the form of mathematical expressions.

[0127] S404. A multi-objective intelligent optimization algorithm is used to solve the multi-objective long-term programming optimization model, and the optimal population mating sequence and environmental management strategy sequence are output, which satisfy the constraints and make the objective function optimal. Among them, the optimal population mating sequence is used to characterize the individual mating combination relationship in multiple key breeding cycles in the future, and the environmental management strategy sequence is used to characterize the environmental management strategy that changes over time.

[0128] Optionally, the selection of a multi-objective intelligent optimization algorithm can be based on the algorithm's global search capability and its ability to search for and filter Pareto optimal solutions for the multi-objective solution.

[0129] Optionally, the multi-objective intelligent optimization algorithm may include, but is not limited to, non-dominated sorting genetic algorithms and multi-objective evolutionary algorithms. The multi-objective intelligent optimization algorithm can search for a set of non-dominated solutions that can simultaneously optimize multiple objectives within the constraints by simulating biological evolution and swarm intelligence behavior.

[0130] For example, the breeding planning terminal can perform mathematical modeling transformation on the decision variables, objective functions and constraints of a multi-objective long-term planning optimization model to generate decision variables, objective functions and constraints in an input format that can be recognized by the algorithm.

[0131] Furthermore, the breeding planning terminal can use the feasible solutions of individual mating combinations and environmental management strategy parameters as the initial population individuals, and carry out multi-generation evolution of the initial population individuals through operations such as selection, crossover, and mutation. In each generation, non-dominated solutions are selected based on the objective function value and the satisfaction of constraints.

[0132] Furthermore, in the case of iteration termination, the breeding planning terminal can select the final solution from the final non-dominated solution set according to the specific preferences of the breeding objectives, and decode the final solution into the optimal population mating sequence and environmental management strategy sequence.

[0133] In this embodiment, the breeding planning terminal defines decision variables, constructs a dual-objective optimization function that integrates benefits and risks, and sets multi-dimensional constraints including genetic diversity. It then uses a multi-objective intelligent optimization algorithm to solve the problem and outputs the optimal population mating sequence and environmental strategy sequence that are co-optimized over a long time scale. This enables a closed loop from multi-period uncertainty prediction to executable breeding decisions, thereby improving the scientific nature of breeding planning.

[0134] In one embodiment, the optimal population mating sequence and environmental management strategy sequence are decoupled and structured according to the time dimension to generate a dynamic beef cattle breeding plan, which may include the following steps:

[0135] S501, A seeding plan that divides the optimal population mating sequence into several time phases.

[0136] Optionally, the breeding program may include information on breeding bulls, breeding cows, and expected offspring for a given period of time.

[0137] Optionally, the information on breeding bulls may include, but is not limited to, a unique identifier for the breeding bull, a summary of its genetic background, dominant traits, and a record of its reproductive capacity.

[0138] Optionally, information on breeding cows may include, but is not limited to, unique identifiers, health status assessment results, historical breeding records, and genetic fitness analysis.

[0139] Optionally, the expected progeny information can be calculated based on the genetic similarity between the breeding bulls and cows and the predicted values ​​of the target traits. This expected progeny information may include, but is not limited to, assessments of the dominant traits and genetic potential of the offspring.

[0140] For example, the breeding planning terminal can divide the continuous optimal population mating sequence into several independent and continuous time-stage mating plans based on preset time-stage division rules and in combination with the reproductive physiological characteristics of beef cattle and the production scheduling needs of the farm.

[0141] S502. Align the environmental control parameters in the environmental management strategy sequence according to the timestamp and seeding plan, and map the environmental control parameters to the corresponding feeding stage and cattle group in the spatiotemporal environmental data according to the preset spatiotemporal environmental data mapping relationship, and generate an environmental control instruction set; wherein, the environmental control parameters are used to control the target set value of at least one of the feeding environment temperature, humidity and nutrient formula.

[0142] Optionally, the spatiotemporal environmental data mapping relationship can be a predefined mapping rule for feeding stage, cattle herd, and environmental parameters.

[0143] Optionally, environmental control parameters may include, but are not limited to, the temperature control range of the rearing environment, the humidity control range of the rearing environment, and the ratio standards of protein, energy, and minerals in the nutritional formula.

[0144] For example, the breeding planning terminal can extract all environmental regulation parameters from the environmental management strategy sequence.

[0145] Furthermore, the breeding planning terminal can align the environmental control parameters with the corresponding time stages of the breeding plans based on the timestamps of each breeding plan, and match the environmental control parameters to specific feeding stages and cattle groups based on the spatiotemporal environmental data mapping relationship, thereby generating a standardized set of environmental control instructions.

[0146] S503. The seed breeding plan and environmental control instruction set are associated and encapsulated according to the time dimension to generate a dynamic beef cattle breeding plan.

[0147] For example, the breeding planning terminal can use the time dimension as the core thread to associate and bind the seed mating plan for each time period with the corresponding environmental control instruction set. During the association and encapsulation process, the breeding planning terminal can use a structured document format to present the seed mating plan and environmental control instruction set for each time period in chronological order.

[0148] Optionally, a dynamic beef cattle breeding program may include, but is not limited to, detailed breeding implementation rules, a list of target breeding individuals, environmental control parameters, and implementation verification requirements.

[0149] In this embodiment, the breeding planning terminal transforms the long-term and complex breeding optimization problem into specific tasks that can be executed in stages, which can ensure the precise coordination of environmental management and mating plans during the breeding process, thereby effectively improving breeding efficiency and overall results.

[0150] In one embodiment, after generating a dynamic beef cattle breeding program, the method may further include the following steps:

[0151] S601. Obtain dynamic incremental breeding data of newborn individuals.

[0152] Optionally, the dynamic incremental breeding data of newborn individuals can be multi-dimensional breeding-related data of newborn individuals in the target population. The data dimensions of the dynamic incremental breeding data of newborn individuals are consistent with the initial multi-dimensional breeding base data. The dynamic incremental breeding data of newborn individuals can be used to supplement the population's genetic and phenotypic information.

[0153] For example, the breeding planning terminal can acquire dynamic incremental breeding data of newborn individuals by combining real-time collection and batch verification, and perform format standardization, missing value completion and outlier removal on the dynamic incremental breeding data to obtain dynamic incremental breeding data that can be used for feature attribute extraction.

[0154] S602. Extract feature attributes from dynamic incremental breeding data, extract node attribute values ​​of newborn individuals, and construct individual nodes corresponding to newborn individuals in the population breeding multi-relationship heterogeneous graph based on node attribute values ​​to obtain an updated population breeding multi-relationship heterogeneous graph.

[0155] For example, the breeding planning terminal can extract core attributes from dynamic incremental breeding data to obtain a set of node attribute values ​​for newborn individuals.

[0156] Furthermore, the breeding planning terminal can use the unique identifier of the new individual as the node ID to add individual nodes to the original population breeding multi-relationship heterogeneous graph, and associate the node attribute values ​​of the new individual nodes.

[0157] Furthermore, the breeding planning terminal can construct pedigree edges between newborn individuals and their fathers and mothers based on pedigree data, and construct environmental co-occurrence edges with individuals from the same period and unit based on initial rearing environment data. The breeding planning terminal can generate and update the multi-relationship heterogeneous graph of the breeding population by aggregating newly added individual nodes and associated edges.

[0158] Optionally, core attributes may include, but are not limited to, information at the genetic, phenotypic, and association levels.

[0159] Optionally, the genetic level may include, but is not limited to, key SNP locus characteristics and allele coding values; the phenotypic level may include, but is not limited to, birth phenotypic indicators and health assessment results; and the association level may include, but is not limited to, unique paternal and maternal identifiers and rearing unit information.

[0160] S603. Input the updated population breeding multi-relationship heterogeneous graph into the heterogeneous graph attention network, adaptively update the weight parameters corresponding to the associated edges of the individual nodes corresponding to the newborn individuals, and output the full node optimized representation vector set.

[0161] For example, the breeding planning terminal can convert the updated population breeding multi-relationship heterogeneous graph into a feature matrix, edge type matrix and temporal matrix that can be recognized by the heterogeneous graph attention network, and input the identifiable feature matrix, edge type matrix and temporal matrix into the pre-trained heterogeneous graph attention network.

[0162] Furthermore, heterogeneous graph attention networks can initiate an adaptive weight update process based on the phylogenetic and environmental co-occurrence edges of newly generated individual nodes, and adjust the weight parameters of the associated edges through backpropagation.

[0163] Furthermore, heterogeneous graph attention networks can output a fully optimized representation vector set through multi-layer feature aggregation and nonlinear transformation.

[0164] S604. The optimized representation vector set of all nodes is used as a supervised fine-tuning sample and input into the multi-scenario trait prediction network. The time-series prediction layer parameters of the multi-scenario trait prediction network are iteratively optimized using a dynamic learning rate strategy. The learning results on the historical long-term trait evolution law are retained through the time-series causal constraint loss function, and the multi-scenario trait prediction network is fine-tuned.

[0165] For example, the breeding planning terminal can divide the full-node optimized representation vector set into a training sample set and a validation sample set, and input the training sample set and validation sample set into a pre-trained multi-scenario trait prediction network, outputting the target trait prediction value and confidence interval for each target breeding individual in the training sample set in the future key breeding cycle, and the target trait prediction value and confidence interval for each target breeding individual in the validation sample set in the future key breeding cycle.

[0166] Furthermore, the breeding planning terminal can obtain the prediction error of the target trait prediction value of the validation sample, and maintain the current learning rate when the prediction error continues to decrease, and reduce the learning rate when the prediction error tends to level off.

[0167] Furthermore, the breeding planning terminal can calculate the deviation between the prediction results and the historical trait evolution patterns through the temporal causal constraint loss function. Based on the deviation between the prediction results and the historical trait evolution patterns, the breeding planning terminal can calculate the temporal causal constraint loss function value, thereby driving the multi-scenario trait prediction network to retain its understanding of long-term breeding patterns while learning new population information.

[0168] Preferably, the expression for the time-series causal constraint loss function can be:

[0169]

[0170] In the formula, This represents the total loss function used in the fine-tuning phase to optimize the multi-scenario trait prediction network. This represents the predicted loss term. It can be used to measure the deviation between the network's predicted traits of the current full-node optimized representation vector set and the actual observed or high-confidence estimates. This represents a hyperparameter that is greater than or equal to 0. It can be used to adjust the weight of the temporal causality constraint loss term in the total loss, balancing the model's emphasis on learning new data and retaining historical knowledge. This represents the loss term due to temporal causality constraints.

[0171] Preferably, the expression for the time-series causality constraint loss term can be:

[0172]

[0173] In the formula, Represents the time-series causality constraint loss term. This represents the number of historical individual samples used for constraints. Indicates the number of historical breeding cycles. This indicates that the optimized multi-scenario trait prediction network parameters, which are currently being fine-tuned, are being used for the first... The historical target breeding individual in the first Predicted values ​​for a periodic trait, This represents the predicted values ​​of the same historical target breeder for the same trait in the same period, using the optimized multi-scenario trait prediction network parameters before fine-tuning. It can be used to represent the results of historical learning.

[0174] Furthermore, the breeding planning terminal can optimize the temporal prediction layer parameters of the multi-scenario trait prediction network by driving iterative training through multiple rounds, thereby obtaining a fine-tuned multi-scenario trait prediction network that can adapt to the updated population characteristics.

[0175] In this embodiment, the breeding planning terminal fine-tunes the prediction network by introducing a dynamic learning rate and a temporal causal constraint loss function. This allows the network to effectively retain long-term historical patterns while acquiring new population data, thus enabling the continuous evolution of the trait prediction capabilities of the multi-scenario trait prediction network.

[0176] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0177] Based on the same inventive concept, this application also provides an artificial intelligence-based beef cattle breeding planning device for implementing the aforementioned method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the artificial intelligence-based beef cattle breeding planning device provided below can be found in the limitations of the artificial intelligence-based beef cattle breeding planning method described above, and will not be repeated here.

[0178] In one exemplary embodiment, such as Figure 2 As shown, an artificial intelligence-based beef cattle breeding planning device 700 is provided, comprising:

[0179] Data acquisition module 701 is used to acquire multi-dimensional basic breeding data of individuals in the target population;

[0180] The relationship graph construction module 702 is used to obtain a multi-relationship heterogeneous graph of population breeding by taking each target breeding individual in the target population as an individual node based on multi-dimensional breeding basic data and establishing association edges with different association relationships between individual nodes.

[0181] The attention representation module 703 is used to input the heterogeneous graph of multiple relationships in population breeding into a heterogeneous graph attention network based on a temporal causal attention mechanism and output an individual representation vector; wherein, the individual representation vector is used to represent the structure, attributes and temporal context information of the target breeding individual in the heterogeneous graph of multiple relationships in population breeding.

[0182] The trait prediction module 704 is used to input the individual representation vector into a multi-scenario trait prediction network based on a Bayesian deep learning mechanism to obtain the predicted values ​​and confidence intervals of the target traits of each target breeding individual in multiple key breeding cycles in the future.

[0183] The optimization decision module 705 is used to construct a multi-objective long-term programming optimization model based on the predicted values ​​and confidence intervals of the target traits of individuals in each target population, and solve the multi-objective long-term programming optimization model to generate the optimal population mating sequence and the corresponding environmental management strategy sequence.

[0184] The scheme generation module 706 is used to decouple and structure the optimal population mating sequence and environmental management strategy sequence according to the time dimension to generate a dynamic beef cattle breeding plan scheme.

[0185] In one embodiment, the relationship graph construction module includes:

[0186] The pedigree edge construction unit is used to construct pedigree edges between individual nodes with direct parent-child relationships based on pedigree record data.

[0187] The genetic association edge construction unit is used to calculate the genomic genetic similarity between any two individual nodes based on genomic single nucleotide polymorphism data, and to construct genetic association edges between individual nodes whose genomic genetic similarity exceeds a preset genomic genetic similarity threshold;

[0188] The environmental co-occurrence edge construction unit is used to identify pairs of individual nodes in the same breeding unit within the same preset breeding time period based on the breeding group and timestamp information in the spatiotemporal environmental data, and to construct environmental co-occurrence edges between the pairs of individual nodes.

[0189] The heterogeneity graph generation unit is used to set up a multi-relationship heterogeneity graph for population breeding based on each individual node, pedigree edge, genetic association edge, and environmental co-occurrence edge.

[0190] In one embodiment, the genetic association edge construction unit includes:

[0191] The coding subunit is used to numerically encode the alleles of each target breeding individual based on the single nucleotide polymorphism sites in the single nucleotide polymorphism data of the genome, to obtain the allele count, and to use the allele count as the allele coding value of the target breeding individual at the single nucleotide polymorphism site.

[0192] The secondary allele frequency calculation subunit is used to calculate the secondary allele frequency at each single nucleotide polymorphism (SNP) site based on the allele coding values ​​at each SNP site in the target population; wherein, the secondary allele frequency is used to characterize the secondary allele frequency in the target population. Distribution of minor alleles at each single nucleotide polymorphism site;

[0193] The similarity calculation subunit is used to calculate the genomic genetic similarity between any two individual nodes based on allele coding values ​​and suballele frequencies.

[0194] In one embodiment, the optimization decision module includes:

[0195] The decision variable definition unit is used to set the individual mating combination relationship and environmental management strategy parameters of the multi-objective long-term planning optimization model; wherein, the individual mating combination relationship and environmental management strategy parameters change over time in multiple key breeding cycles in the future.

[0196] The objective function building unit is used to construct the objective function of a multi-objective long-term programming optimization model to maximize the total expected gain of the target trait and minimize the cumulative prediction risk determined by the confidence interval; wherein, the cumulative prediction risk is calculated based on the width or variance of the individual confidence interval;

[0197] The constraint setting unit is used to set the constraints of the multi-objective long-term programming optimization model; among which, the constraints include genetic diversity constraints.

[0198] The optimization and solution unit is used to solve the multi-objective long-term programming optimization model using a multi-objective intelligent optimization algorithm, and outputs the optimal population mating sequence and environmental management strategy sequence that satisfy the constraints and make the objective function optimal. The optimal population mating sequence is used to characterize the individual mating combination relationship in multiple key breeding cycles in the future, and the environmental management strategy sequence is used to characterize the environmental management strategy that changes over time.

[0199] In one embodiment, the scheme generation module includes:

[0200] The phase decoupling unit is used to divide the optimal population mating sequence into mating seed plans in several time phases; wherein, the mating seed plan includes information on mating bulls, mating cows, and expected offspring within the time phase;

[0201] The alignment and mapping unit is used to align the environmental control parameters in the environmental management strategy sequence according to the timestamp and seeding plan, and to map the environmental control parameters to the corresponding feeding stage and cattle group in the spatiotemporal environmental data according to the preset spatiotemporal environmental data mapping relationship, thereby generating an environmental control instruction set; wherein, the environmental control parameters are used to control the target set value of at least one of the feeding environment temperature, humidity and nutrient formula.

[0202] The scheme encapsulation unit is used to associate and encapsulate the seed breeding plan and environmental control instruction set according to the time dimension to generate a dynamic beef cattle breeding plan scheme.

[0203] In one embodiment, the apparatus may further include:

[0204] The data acquisition module is used to acquire dynamic incremental breeding data of newborn individuals;

[0205] The incremental data acquisition module is used to extract feature attributes from dynamic incremental breeding data, extract node attribute values ​​of newborn individuals, and construct individual nodes corresponding to newborn individuals in the population breeding multi-relationship heterogeneous graph based on node attribute values ​​to obtain an updated population breeding multi-relationship heterogeneous graph.

[0206] The graph structure update module is used to input the updated population breeding multi-relationship heterogeneous graph into the heterogeneous graph attention network, adaptively update the weight parameters corresponding to the associated edges of the individual nodes corresponding to the newborn individuals, and output the full node optimized representation vector set.

[0207] The representation optimization module is used to input the optimized representation vector set of all nodes as supervised fine-tuning samples into the multi-scenario trait prediction network. It uses a dynamic learning rate strategy to iteratively optimize the temporal prediction layer parameters of the multi-scenario trait prediction network, and retains the learning results of the historical long-term trait evolution law through the temporal causal constraint loss function to fine-tune the multi-scenario trait prediction network.

[0208] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the artificial intelligence-based beef cattle breeding planning method described above.

[0209] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0210] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0211] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for beef cattle breeding planning based on artificial intelligence, characterized in that, The method includes: S101. Obtain multi-dimensional basic breeding data of target breeding individuals in the target population; S102. Based on the multi-dimensional breeding basic data, each target breeding individual in the target population is taken as an individual node, and different association edges are established between the individual nodes to obtain a multi-relationship heterogeneous graph of population breeding. S103. Input the heterogeneous graph of the population breeding multi-relationship into a heterogeneous graph attention network based on a temporal causal attention mechanism, and output an individual representation vector; wherein, the individual representation vector is used to represent the structure, attributes and temporal context information of the target breeding individual in the heterogeneous graph of the population breeding multi-relationship; S104. Input the individual representation vector into a multi-scenario trait prediction network based on Bayesian deep learning mechanism to obtain the target trait prediction values ​​and confidence intervals for each target breeding individual in multiple key breeding cycles in the future. S105. Based on the predicted values ​​of the target traits and the confidence intervals of each target breeding individual, a multi-objective long-term programming optimization model is constructed, and the multi-objective long-term programming optimization model is solved to generate the optimal population mating sequence and the environmental management strategy sequence corresponding to the optimal population mating sequence. S106. Decouple and structure the optimal population mating sequence and the environmental management strategy sequence according to the time dimension to generate a dynamic beef cattle breeding plan.

2. The method according to claim 1, characterized in that, The multidimensional breeding basic data includes genome single nucleotide polymorphism data, historical multi-type data, population pedigree data, and spatiotemporal environmental data; Based on the multi-dimensional breeding data, each target breeding individual in the target population is taken as an individual node, and association edges with different relationships are established between the individual nodes to obtain a heterogeneous multi-relationship graph of population breeding, including: S201. Based on pedigree record data, construct pedigree kinship edges between the individual nodes that have a direct parent-child relationship; S202. Based on the single nucleotide polymorphism data of the genome, calculate the genomic genetic similarity between any two individual nodes, and construct a genetic association edge between individual nodes whose genomic genetic similarity exceeds a preset genomic genetic similarity threshold; S203. Based on the breeding group and timestamp information in the spatiotemporal environment data, identify individual node pairs that are in the same breeding unit within the same preset breeding time period, and construct environmental co-occurrence edges between the individual node pairs. S204. Based on each individual node, the pedigree edge, the genetic association edge, and the environmental co-occurrence edge, a multi-relationship heterogeneous graph of the population breeding is set up.

3. The method according to claim 2, characterized in that, The calculation of genomic genetic similarity between any two individual nodes based on the genomic single nucleotide polymorphism data includes: S301. Based on the single nucleotide polymorphism sites in the genome single nucleotide polymorphism data, the alleles of each target breeding individual are numerically encoded to obtain the allele count, and the allele count is used as the allele encoding value of the target breeding individual at the single nucleotide polymorphism site. S302. Based on the allele coding values ​​at each single nucleotide polymorphism (SNP) site in the target population, calculate the minor allele frequency at each SNP site; wherein the minor allele frequency is used to characterize the genotype of the target population. Distribution of minor alleles at each single nucleotide polymorphism site; S303. Based on the allele coding value and the suballele frequency, calculate the genomic genetic similarity between any two individual nodes; The expression for the genomic genetic similarity is: in, Indicates the target breeding individual With target breeding individuals The genetic similarity between the genomes, This indicates the total number of single nucleotide polymorphism sites. Indicates the target breeding individual In the Allelic genotype coding values ​​at single nucleotide polymorphism sites, Represents a similarity indicator function. Indicates the first Site weighting coefficients for each single nucleotide polymorphism site. Indicates that the target population is in the first... Secondary allele frequencies at single nucleotide polymorphism sites.

4. The method according to claim 1, characterized in that, Based on the predicted values ​​of the target traits and the confidence intervals of the target breeding individuals in each target population, a multi-objective long-term programming optimization model is constructed, and the multi-objective long-term programming optimization model is solved to generate an optimal population mating sequence and an environmental management strategy sequence corresponding to the optimal population mating sequence, including: S401. Set the individual mating combination relationship and environmental management strategy parameters of the multi-objective long-term planning optimization model; wherein, the individual mating combination relationship and the environmental management strategy parameters change over time in multiple key breeding cycles in the future; S402. Construct the objective function of the multi-objective long-term programming optimization model to maximize the total expected gain of the target trait and minimize the cumulative prediction risk determined by the confidence interval; wherein the cumulative prediction risk is calculated based on the width or variance of the individual confidence interval. S403. Set the constraints of the multi-objective long-term programming optimization model; wherein, the constraints include genetic diversity constraints; S404. A multi-objective intelligent optimization algorithm is used to solve the multi-objective long-term programming optimization model, and the optimal population mating sequence and the environmental management strategy sequence are output, which satisfy the constraints and make the objective function optimal; wherein, the optimal population mating sequence is used to characterize the individual mating combination relationship in multiple key breeding cycles in the future, and the environmental management strategy sequence is used to characterize the environmental management strategy that changes over time.

5. The method according to any one of claims 1 to 4, characterized in that, The step of decoupling and structuring the optimal population mating sequence and the environmental management strategy sequence according to the time dimension to generate a dynamic beef cattle breeding plan includes: S501. Divide the optimal population mating sequence into several time-phase mating plans; wherein, the mating plan includes information on mating bulls, mating cows, and expected offspring within the time phase; S502. Align the environmental control parameters in the environmental management strategy sequence according to the timestamp and the seeding plan, and map the environmental control parameters to the corresponding feeding stage and cattle group in the spatiotemporal environmental data according to the preset spatiotemporal environmental data mapping relationship to generate an environmental control instruction set; wherein, the environmental control parameters are used to control the target set value of at least one of the feeding environment temperature, humidity and nutrient formula. S503. The seed breeding plan and the environmental control instruction set are associated and encapsulated according to the time dimension to generate the dynamic beef cattle breeding plan.

6. The method according to claim 1, characterized in that, After generating the dynamic beef cattle breeding program, the method further includes: S601. Obtain dynamic incremental breeding data of newborn individuals; S602. Extract feature attributes from the dynamic incremental breeding data to obtain the node attribute values ​​of the newborn individuals, and construct the individual nodes corresponding to the newborn individuals in the population breeding multi-relationship heterogeneous graph based on the node attribute values ​​to obtain the updated population breeding multi-relationship heterogeneous graph. S603. Input the updated population breeding multi-relationship heterogeneous graph into the heterogeneous graph attention network, adaptively update the weight parameters corresponding to the associated edges of the individual nodes corresponding to the newborn individuals, and output the full node optimized representation vector set. S604. The full-node optimized representation vector set is used as a supervised fine-tuning sample and input into the multi-scenario trait prediction network. The time-series prediction layer parameters of the multi-scenario trait prediction network are iteratively optimized using a dynamic learning rate strategy. The learning results on the historical long-term trait evolution law are retained through the time-series causal constraint loss function, and the multi-scenario trait prediction network is fine-tuned.

7. An artificial intelligence-based beef cattle breeding planning device, characterized in that, The device includes: The data acquisition module is used to acquire multi-dimensional basic breeding data of target breeding individuals in the target population; The relationship graph construction module is used to take each target breeding individual in the target population as an individual node based on the multi-dimensional breeding basic data, and establish association edges with different association relationships between the individual nodes to obtain a multi-relationship heterogeneous graph of population breeding. The attention representation module is used to input the heterogeneous graph of the population breeding multi-relationship into a heterogeneous graph attention network based on the temporal causal attention mechanism and output an individual representation vector; wherein, the individual representation vector is used to represent the structure, attributes and temporal context information of the target breeding individual in the heterogeneous graph of the population breeding multi-relationship; The trait prediction module is used to input the individual representation vector into a multi-context trait prediction network based on a Bayesian deep learning mechanism to obtain the target trait prediction values ​​and confidence intervals for each target breeding individual in multiple key breeding cycles in the future. The optimization decision module is used to construct a multi-objective long-term programming optimization model based on the predicted values ​​of the target traits and the confidence intervals of each target breeding individual, and to solve the multi-objective long-term programming optimization model to generate the optimal population mating sequence and the environmental management strategy sequence corresponding to the optimal population mating sequence. The scheme generation module is used to decouple and structure the optimal population mating sequence and the environmental management strategy sequence according to the time dimension to generate a dynamic beef cattle breeding plan scheme.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images