A method and system for collecting and analyzing data on the expansion of the Pingliang red cattle population

Abstract

The application discloses a kind of Pingliang red cattle population propagation data acquisition and analysis method and system, belong to the field of information technology of animal husbandry.The application includes: obtaining the multi-generation genome of Pingliang red cattle population, historical performance and blood record, construct original data set;Classify the multi-generation genetic path, determine the key genetic path diagram;Extract the implicit genomic characteristics of high genetic similarity node in the graph, construct offspring potential prediction model;Pretreat the feature set and fuse environmental variables to adjust weight, obtain the optimized genetic prediction vector;Extract the path dynamic parameters in the vector, determine the bias correction coefficient;Iterative optimization prediction model using the coefficient, obtain accurate offspring potential evaluation value;Accordingly generate population propagation recommendation sequence.The application corrects by dynamically analyzing multi-generation genetic path and fusing environmental factors, effectively solves the offspring potential prediction bias problem caused by ignoring the dynamic nature of genetic transmission in traditional methods, improves the accuracy of selection.

Description

Technical Field

[0001] This invention relates to the field of livestock information technology, specifically to a method and system for collecting and analyzing data on the population expansion of Pingliang Red Cattle. Background Technology

[0002] In the field of livestock breeding, especially for breeds with local characteristics and genetic conservation value such as Pingliang Red Cattle, the efficiency and accuracy of population propagation are directly related to the maintenance of superior breed traits and the economic benefits of the industry. Traditional breeding methods largely rely on farmers' experience and simple pedigree records. Predictions of offspring production performance (such as weight and meat quality) are mostly based on parental phenotypic data, lacking quantitative analysis of deep genetic laws.

[0003] With the development of information technology, some methods have begun to utilize genomic data for assisted selection. However, existing solutions mostly focus on the association analysis between genotype and phenotype in contemporary individuals, failing to effectively track and quantify the dynamic transmission of genetic information across generations. Blood relations are used only as static background information, rather than as a dynamic analytical object. This approach ignores the complex pathways of genetic effects' accumulation, interaction, and decay across generations, leading to significant biases in predictive models applied to actual multi-generational breeding. For example, predicting offspring based solely on superior genes from parents may fail due to skipped-generation inheritance or interruptions in key genetic pathways, resulting in inaccurate breeding decisions and wasted resources.

[0004] Therefore, the main problem with existing technologies is that they are limited to the analysis of static genomic data of contemporary individuals and fail to effectively track and quantify the dynamic transmission pathways of genetic information across generations, resulting in significant biases in the prediction of offspring performance. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for collecting and analyzing data on the population expansion of Pingliang Red Cattle, so as to solve the above-mentioned problems existing in the prior art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for collecting and analyzing data on the population expansion of Pingliang Red Cattle, including: Multigenerational genomic data, historical offspring performance data, and pedigree records of the Pingliang Red Cattle population were acquired to construct an original dataset; wherein the pedigree records were obtained by automatically associating the identification devices on the cattle with the data acquisition devices. Based on the original dataset, a blood relationship matrix was constructed, and a random forest algorithm was used to classify multi-generational genetic paths to determine the key genetic path graphs representing the gene transmission chain from ancestor to offspring. Extract node data from the key genetic path graph where the genetic similarity between nodes exceeds a preset similarity threshold, and extract their hidden genomic features through a deep learning network to form an input feature set; Based on the input feature set and the historical offspring performance data, a prediction model for offspring potential is constructed. The input feature set is preprocessed and environmental variables from breeding records are fused to adjust the feature weights, resulting in an optimized genetic prediction vector. Based on the optimized genetic prediction vector, multi-generational path dynamic parameters are extracted. The random forest regression algorithm is used to analyze the impact of the path dynamic parameters on the performance of offspring, and the bias correction coefficient is determined accordingly. The offspring potential prediction model is updated using the deviation correction coefficient, and the model is iteratively optimized to make the matching degree between the simulated offspring performance data output by the model and the historical offspring performance data reach a preset matching degree threshold, thereby obtaining an accurate offspring potential assessment value. Based on the precise offspring potential assessment values, a recommended sequence for population expansion is generated, and the breeding priority of high-potential cattle is determined.

[0007] Furthermore, the construction of the original dataset specifically includes: Based on the multi-generational genomic data, a clustering algorithm is used to divide the data into multiple genomic data groups according to individuals; Determine whether the kinship records between different individuals match the genetic consistency of their genomic groups, and determine the kinship chain when matching; Based on the bloodline, identify recurring genetic patterns in the original dataset; Based on the recurring gene inheritance pattern, regression analysis is used to predict genomic change trends. Based on the predicted genomic change trend, determine whether the newly acquired genomic data has deviated, and mark any deviations as abnormal gene points; The abnormal gene points are compared with the relevant kinship records. Based on the comparison results, the kinship records, genomic data and historical offspring performance data are integrated and corrected to construct the original dataset.

[0008] Furthermore, the determination of the key genetic pathway map representing the gene transfer chain from ancestor to offspring specifically includes: Kinship data is extracted from the original dataset, and kinship coefficients between individuals are calculated to construct the blood relationship matrix. The random forest classification algorithm is used to classify multiple genetic paths derived from the kinship matrix and identify key genetic path features; Based on the key genetic pathway features, a path graph structure is constructed with individuals as nodes and parent-child relationships with significant genetic contributions as directed edges. Calculate the connection strength of each node in the path graph structure, and determine the path with connection strength exceeding a preset strength threshold as the main transmission path; Based on the main transmission pathways and their associated individual attributes, the layout of the key genetic pathway map is completed.

[0009] Furthermore, the formation of the input feature set specifically includes: Calculate the genome-wide genetic similarity between individuals corresponding to any two nodes in the critical genetic pathway graph; If the genetic similarity between the two nodes exceeds the preset similarity threshold, the genomic data of the individuals corresponding to the two nodes are extracted from the original dataset and input into the convolutional neural network to extract high-dimensional hidden genomic features. All extracted latent genomic features were grouped using clustering methods to form grouped feature sets; The grouped feature set is normalized to obtain the input feature set.

[0010] Furthermore, the offspring potential prediction model employs a deep neural network or gradient boosting tree model trained based on the gradient descent algorithm.

[0011] Furthermore, the obtained optimized genetic prediction vector specifically includes: The input feature set is processed using the Z-score normalization method to obtain a normalized feature set; Identify a subset of environmental features related to environmental factors from the standardized feature set; If the environmental feature subset is not empty, then the mutual information value between each environmental feature and the remaining features is calculated; wherein, the environmental features are taken from the environmental feature subset, and the remaining features are features in the standardized feature set other than the environmental feature subset; Based on the calculated mutual information values, the initial weights of the remaining features corresponding to each value are adjusted, and the adjusted weights are fused with the standardized feature set using a linear weighting method to generate the optimized genetic prediction vector.

[0012] Furthermore, determining the deviation correction coefficient specifically includes: From the optimized genetic prediction vector, a set of dynamic parameters for multiple generations of paths is parsed out. The dynamic parameters include path strength, path stability, and path length calculated based on the key genetic path graph. Using the dynamic parameter set as input features and historical offspring performance data as target variables, a random forest regression model is constructed. Based on the importance ranking of the features output by the random forest regression model, parameters with importance higher than a preset importance threshold are marked as critical path parameters; Calculate the average difference between the model predictions and actual values ​​of the critical path parameters, and determine the deviation correction coefficient based on this difference.

[0013] Furthermore, the step of iteratively optimizing the model output to achieve a preset matching degree threshold between the simulated offspring performance data and the historical offspring performance data specifically includes: The output of the offspring potential prediction model is initially corrected using the deviation correction coefficient to obtain preliminary simulation data; The cosine similarity between the preliminary simulation data and the historical performance data of future generations is calculated as the matching degree. If the matching degree is lower than the matching degree threshold, the gradient descent algorithm is used to adjust the parameters of the offspring potential prediction model, and simulated data is regenerated for calculation until the matching degree reaches or exceeds the matching degree threshold.

[0014] Furthermore, the process of generating recommended sequences for population expansion and determining breeding priorities for high-potential cattle specifically includes: The precise offspring potential assessment values ​​of all candidate individuals in the population are obtained, and the assessment values ​​are used as input to perform secondary correction using a regression model to eliminate the population bias of the assessment values. All individuals are sorted in descending order based on the corrected potential values ​​to generate a preliminary candidate sequence; Individuals ranked higher in the preliminary candidate sequence were selected as high-potential cattle. The breeding priority score is calculated by combining the health index of the high-potential cattle with the estimated breeding cost. High-potential cattle are ranked according to the breeding priority score, a final population propagation recommendation sequence is generated, and the breeding priority of the high-potential cattle is determined.

[0015] Another objective of this invention is to provide a method and system for collecting and analyzing data on the population expansion of Pingliang Red Cattle, used to execute the aforementioned method for collecting and analyzing data on the population expansion of Pingliang Red Cattle, comprising: The data acquisition and preprocessing module is used to acquire multi-generational genomic data, historical offspring performance data, and bloodline records of the Pingliang Red Cattle population to form the original dataset; The genetic path analysis module is used to construct a kinship matrix based on the original dataset and to classify multi-generational genetic paths using a random forest algorithm to determine the key genetic path graph. The feature extraction and model building module is used to extract hidden genomic features from the key genetic pathway map to form an input feature set, and to build a progeny potential prediction model based on this set. The feature optimization module is used to preprocess the input feature set and perform environmental variable fusion processing to obtain the optimized genetic prediction vector. The deviation correction analysis module is used to determine the deviation correction coefficient based on the analysis of dynamic parameters of multiple generations of paths. The model optimization and evaluation module is used to update and iteratively optimize the offspring potential prediction model using the bias correction coefficient until an accurate offspring potential assessment value is obtained. The decision output module is used to generate a recommended sequence for population expansion based on the accurate offspring potential assessment value.

[0016] Compared with the prior art, the present invention has the following beneficial effects: I. This invention classifies and filters critical paths of multi-generational genetic pathways using the random forest algorithm, which can accurately locate the core genetic transmission chains that affect the performance of offspring, overcoming the shortcomings of traditional methods in providing general and imprecise genetic association analysis.

[0017] Second, by extracting latent genomic features among highly similar individuals through deep learning and integrating environmental variables to adjust the weights of the prediction model, the model can simultaneously characterize the complex intertwined effects between genetic and environmental factors, thereby significantly improving the generalization ability and robustness of the prediction model in actual variable aquaculture environments.

[0018] Third, by using dynamic parameters of the genetic pathway for bias correction and iterative model optimization, the accuracy of offspring potential assessment is significantly improved. This invention effectively solves the prediction bias problem caused by ignoring the dynamic transmission of genetic pathways across multiple generations, significantly optimizes the accuracy of breeding priority recommendations, and provides reliable technical support for the efficient and scientific propagation of the Pingliang Red Cattle population. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for collecting and analyzing data on the population expansion of Pingliang Red Cattle, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the feature optimization and genetic prediction vector generation process in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0021] This embodiment provides a method for collecting and analyzing data on the propagation of Pingliang Red Cattle populations. This method is based on a hardware system. Physically, the system consists of sensing layer devices deployed at the farm, a cloud platform server for data aggregation and computation, and a terminal application providing interaction. Logically, it executes the method described in this embodiment. The following is a combination of... Figure 1 The specific implementation process of this method will be explained in detail.

[0022] Step 1: Obtain multi-generational genomic data, historical offspring performance data, and pedigree records of the Pingliang Red Cattle population to construct the original dataset; wherein the pedigree records are obtained by automatically linking the identification devices on the cattle with the data acquisition devices.

[0023] This step is executed by the system's data acquisition and preprocessing module. Data governance controls prediction bias at the source, improving the quality of input data and laying the foundation for accurate analysis. First, data acquisition is performed. In this embodiment, at the Pingliang Red Cattle farm, each cow is fitted with a smart ear tag integrated with an RFID chip as a unique identification device. This ear tag embeds a miniature biosensor that can periodically, for example, monthly, non-invasively collect trace amounts of tissue fluid. Through a built-in CRISPR-based rapid sequencing unit, it scans 500,000 preset single nucleotide polymorphism (SNP) sites to obtain the individual's genomic SNP data. This process uses algorithms such as error correction models to ensure data accuracy. Simultaneously, data acquisition devices, such as fixed RFID readers and handheld PDAs, automatically record kinship events between cattle. For example, when the reader detects that cow number 001 and bull number 002 coexist in the breeding pen, and subsequently detects a newborn calf 003, the system automatically generates and encrypts the kinship record "001&002->003" in a blockchain database to ensure immutability. In addition, the system imports historical offspring performance data for each cow from the breeding management database, such as 24-month-old weight, eye muscle area, and marbling score.

[0024] After completing the initial acquisition of multi-generational genomic data, historical progeny performance data, and pedigree records, a high-quality raw dataset was constructed, specifically including: (1) Based on the multi-generational genomic data, a clustering algorithm is used to divide the data into multiple genomic data groups according to individuals. Specifically, the acquired multi-generational genomic data is divided according to individuals using the K-means clustering algorithm. For example, for the SNP data of 2000 cattle accumulated over the past 5 generations, the number of clusters k=5 is set. After iterative calculation, genetically similar individuals are grouped into the same genomic data group, resulting in 5 groups. The average genetic similarity within each group can reach 85%.

[0025] (2) Determine whether the kinship records between different individuals match the genetic consistency of their genomic groups, and determine the kinship chain when a match is found. For example, the system reads the record "Cow A (Group 1) and Bull B (Group 2) are the parents of calf C". Then, it checks the group to which the calf C's genomic data belongs. If C belongs to Group 1 or 2, or is close to the centroid of both groups, it is considered a match, and the kinship chain from A, B to C is determined. If there is no match, such as C belonging to a completely different Group 5, an anomaly marker is triggered.

[0026] (3) Based on the bloodline, identify recurring gene inheritance patterns in the original dataset; based on the recurring gene inheritance patterns, use regression analysis to predict genomic change trends. First, based on all confirmed bloodline, the system scans the entire dataset in time series to identify recurring gene inheritance patterns. For example, it finds a combination of SNP loci associated with high lean meat percentage; in three consecutive generations, as long as the ancestor G0 carries it, about 70% of the grandchildren G2 will also express high lean meat percentage. Based on this recurring pattern, the system uses regression analysis, such as linear regression, to predict genomic change trends and establish mathematical models, such as predicting that the frequency of this favorable SNP locus will increase at a rate of 0.5% per year in the next generation.

[0027] (4) Based on the predicted genomic change trend, determine whether the newly acquired genomic data has deviated, and mark it as an abnormal gene point if it deviates. During continuous operation of the system, when newly acquired genomic data, such as the detection data of newborn cattle, are entered, the system determines whether it has deviated based on the predicted genomic change trend. For example, if a newborn cattle has a homozygous deletion at a key SNP site, but the model predicts that it should be heterozygous, and the deviation exceeds 3 standard deviations, it is marked as an abnormal gene point.

[0028] (5) The abnormal gene points are compared with the relevant pedigree records. Based on the comparison results, the pedigree records, genomic data, and historical offspring performance data are integrated and corrected to construct the original dataset. In this embodiment, the abnormal gene points are compared with the relevant pedigree records in depth. If a recording error is found, such as an error in the ear tag scan, the pedigree record is corrected; if it is confirmed to be a rare real mutation, the data is retained but a high-confidence label is added. Based on the comparison results, the system performs a final integration and correction of the pedigree records, genomic data, and historical offspring performance data, removing contradictions and filling in gaps to construct an internally consistent, high-quality original dataset.

[0029] Step 2: Based on the original dataset, construct a kinship matrix and use a random forest algorithm to classify multi-generational genetic pathways, identifying key genetic pathways representing the gene transmission chain from ancestor to offspring. This process is executed by the genetic pathway analysis module, specifically as follows: First, kinship data is extracted from the original dataset, and kinship coefficients between individuals are calculated to construct the kinship matrix. Specifically, kinship data is extracted from the original dataset, and kinship coefficients between individuals are calculated. For example, for a population of N cattle, an N×N kinship matrix K is generated using pedigree data and the common ancestor coefficient formula, where elements... The genetic correlation between individuals i and j was quantified. Then, a random forest classification algorithm was used to classify multiple genetic paths derived from the kinship matrix, identifying key genetic path features. Each path was defined as a probabilistic chain of gene transmission from an ancestor to its descendants, with features including the genotype combination of individuals along the path, path length, and average phenotypic value of individuals along the path. For example, the random forest model was set... A decision tree, through learning these features, identifies key genetic pathway features, namely those pathway attributes that are crucial for distinguishing between offspring with excellent segregation performance and ordinary offspring.

[0030] Then, based on the key genetic pathway features, a path graph structure is constructed with individuals as nodes and directed edges representing parent-child relationships with significant genetic contributions. In this graph, the edge weights are assigned by the importance score of the connection to the transmission of superior traits, calculated by a random forest model. Next, the connection strength of each node in the path graph structure is calculated, and paths with connection strength exceeding a preset strength threshold are identified as primary transmission paths. Specifically, the system calculates the connection strength of each node in the path graph structure, i.e., the edge weights, and identifies paths with connection strength exceeding a preset strength threshold, such as 0.7, as primary transmission paths, focusing on the chains with the most significant genetic contributions. Finally, based on the primary transmission paths and their associated individual attributes, the layout of the key genetic pathway graph is completed. These individual attributes include individual ID and generation. After completing the layout of the key genetic pathway graph, a visual map is generated. This map represents the gene transmission chain from ancestor to offspring, intuitively revealing the core channels and bottlenecks in the transmission of superior genes in the population. The above process transforms a static pedigree into a dynamic and quantifiable analytical object.

[0031] Step 3: Extract node data where the genetic similarity between nodes in the key genetic path graph exceeds a preset similarity threshold, and extract their hidden genomic features through a deep learning network to form an input feature set.

[0032] The feature extraction and model building module performs this step, which includes: calculating the whole-genome genetic similarity between individuals corresponding to any two nodes in the key genetic pathway graph; if the genetic similarity between the two nodes exceeds the preset similarity threshold, then extracting the genomic data of the individuals corresponding to the two nodes from the original dataset and inputting it into a convolutional neural network to extract high-dimensional latent genomic features; grouping all the extracted latent genomic features using a clustering method to form a grouped feature set; and normalizing the grouped feature set to obtain the input feature set.

[0033] In a specific embodiment, the system calculates the whole-genome genetic similarity between individuals corresponding to any two nodes in the critical genetic pathway graph. For example, using the genomic relation matrix method, the genotypic correlation of all SNP loci between node A (individual 001) and node B (individual 005) is calculated, yielding a similarity of 0.82. If the genetic similarity between the two nodes exceeds the preset similarity threshold, such as 0.8, the genomic data of the individuals corresponding to the two nodes are extracted from the original dataset. These two high-dimensional SNP vectors, each 500,000-dimensional, are input into a convolutional neural network (CNN). This CNN network contains three convolutional layers and two fully connected layers, using the ReLU activation function, and can extract complex multi-gene interaction latent genomic features that cannot be represented by a single SNP locus, outputting a 128-dimensional feature vector. All extracted latent genomic features are grouped using clustering methods such as hierarchical clustering, aggregating samples with similar features to form grouped feature sets. Subsequently, the grouped feature sets are normalized, such as by Min-Max normalization, scaling them to the [0,1] interval to obtain the input feature set.

[0034] Step 4: Based on the input feature set and the historical offspring performance data, construct an offspring potential prediction model. The offspring potential prediction model employs a deep neural network or a gradient boosting tree model trained using the gradient descent algorithm.

[0035] Specifically, this process is executed by the feature extraction and model building module. The model can be a deep neural network trained using gradient descent, such as a multilayer perceptron (MLP), or a gradient boosting tree model like XGBoost. When using the XGBoost model, the system takes normalized 128-dimensional features as input, the corresponding historical offspring performance values ​​(e.g., weight) as labels, sets the learning rate to 0.1, and the maximum tree depth to 6, training an ensemble model containing 100 decision trees. This model is the initial offspring potential prediction model. Alternatively, the offspring potential prediction model can employ a deep neural network, such as a multilayer perceptron. This network can contain an input layer, one or more hidden layers, and an output layer. The number of neurons in the input layer is the same as the dimension of the input feature set; the hidden layers can be one to three layers, with each layer containing 64 to 256 neurons, and the activation function can be the ReLU function; the output layer consists of one neuron, used to output the potential assessment value. The model's loss function can be mean squared error, the optimizer can be Adam, and the learning rate can be set between 0.001 and 0.01. The network depth, width, and hyperparameters can be adjusted according to the actual data scale and complexity.

[0036] Step 5: Preprocess the input feature set and integrate environmental variables from breeding records to adjust feature weights, thereby obtaining an optimized genetic prediction vector.

[0037] The process of obtaining the optimized genetic prediction vector specifically includes: processing the input feature set using the Z-score normalization method to obtain a normalized feature set; identifying a subset of environmental features related to environmental factors from the normalized feature set; if the subset of environmental features is not empty, calculating the mutual information value between each environmental feature and the remaining features; wherein the environmental features are taken from the subset of environmental features, and the remaining features are features in the normalized feature set other than the subset of environmental features; adjusting the initial weights of the remaining features corresponding to each calculated mutual information value, and fusing the adjusted weights with the normalized feature set using a linear weighting method to generate the optimized genetic prediction vector.

[0038] In this embodiment, the feature optimization module performs the operation; see details below. Figure 2 First, the input feature set is preprocessed, specifically using Z-score normalization. For each feature column, its mean is calculated. and standard deviation For each original value in this column ,calculate A standardized feature set is obtained. From this feature set, a subset of environmental features related to environmental factors is identified. For example, features such as "average daily temperature before weaning" and "crude protein content in feed during the fattening period" are matched from feature names to form an environmental feature subset. If the environmental feature subset is not empty, the mutual information value between each environmental feature and the remaining features is calculated. Here, the environmental features are taken from the environmental feature subset, and the remaining features are features in the standardized feature set other than the environmental feature subset. The mutual information value quantifies the association strength between environmental factors and genetic characteristics. Based on the calculated mutual information values, the initial weights of the corresponding remaining features are adjusted, for example, set to 1. The adjustment strategy can be: New weight = Initial weight × (1 + ... × mutual information value), where The adjustment coefficient can be determined through grid search based on actual data, for example, within the range of [0.1, 1.0]. Finally, a linear weighting method is used to fuse the adjusted weights and the standardized feature set to generate the optimized genetic prediction vector. This process innovatively incorporates environmental variables, enabling the prediction model to adapt to fluctuations in actual aquaculture conditions, improving the model's robustness, and solving the problem of neglecting the intertwined influence of environmental factors.

[0039] Step 6: Extract multi-generational path dynamic parameters based on the optimized genetic prediction vector, analyze the impact of these parameters on offspring performance using a random forest regression algorithm, and determine the bias correction coefficient accordingly. The system's bias correction analysis module executes this step, specifically including: From the optimized genetic prediction vector, a set of dynamic parameters for multiple generations of paths is extracted. These dynamic parameters include path strength, path stability, and path length, calculated based on the key genetic path graph. Specifically, path strength is the average weight of each edge on the path, path stability is calculated as the coefficient of variation of the connection strength between generations, and path length is the number of generations from ancestor to offspring. Using this set of dynamic parameters as input features and historical offspring performance data as the target variable, a random forest regression model is constructed. After training, the model output features are ranked by importance, and parameters with importance higher than a preset importance threshold (e.g., 0.1) are marked as key path parameters. For example, path stability might be found to have the greatest impact on offspring performance. On the validation set, the average difference between the model predictions and actual values ​​of the key path parameters is calculated. For example, the mean absolute error, based on which the deviation correction factor is determined. Specifically, the settings ,when When the prediction error is large, If the value is less than 1, the original predicted value of the model will be significantly revised downwards; when... Smaller, meaning smaller prediction bias. The value is close to 1, indicating a small correction margin. (Coefficient) As a scaling factor, it is used to correct systematic biases caused by ignoring path dynamics.

[0040] Step 7: Update the offspring potential prediction model using the deviation correction coefficient, and optimize iteratively to make the matching degree between the simulated offspring performance data output by the model and the historical offspring performance data reach a preset matching degree threshold, thereby obtaining an accurate offspring potential assessment value.

[0041] This step includes: initially correcting the output of the offspring potential prediction model using the deviation correction coefficient to obtain preliminary simulated data; calculating the cosine similarity between the preliminary simulated data and historical offspring performance data as the matching degree; if the matching degree is lower than the matching degree threshold, adjusting the parameters of the offspring potential prediction model using the gradient descent algorithm and regenerating simulated data for calculation until the matching degree reaches or exceeds the matching degree threshold. The above process is executed by the model optimization and evaluation module to achieve closed-loop optimization of the model.

[0042] Specifically, the offspring potential prediction model is updated using a bias correction factor. The original prediction output of the model is multiplied by the bias correction factor. Preliminary calibration is performed to obtain preliminary simulated data. The cosine similarity between the preliminary simulated data and historical offspring performance data is calculated as the matching degree; in this embodiment, the cosine similarity is 0.88. If the matching degree is lower than the matching degree threshold (e.g., a preset value of 0.9), the gradient descent algorithm is used to adjust the parameters of the offspring potential prediction model, such as the output values ​​of leaf nodes in the XGBoost model or the weights of the MLP, and simulated data is regenerated for calculation. This process is repeated until the matching degree reaches or exceeds the matching degree threshold. Finally, an accurate offspring potential assessment value is obtained. This iterative process ensures that the model can continuously self-optimize and approach the optimal prediction performance.

[0043] Step 8: Based on the precise offspring potential assessment values, generate a recommended sequence for population expansion and determine the breeding priority of high-potential cattle. This step includes: obtaining the precise offspring potential assessment values ​​for all candidate individuals in the population, and using these assessment values ​​as input, performing secondary correction using a regression model to eliminate population bias in the assessment values; sorting all individuals in descending order according to the corrected potential values ​​to generate a preliminary candidate sequence; selecting the top-ranked individuals from the preliminary candidate sequence as high-potential cattle; calculating a breeding priority score by combining the health index of the high-potential cattle with the estimated breeding cost; ranking the high-potential cattle according to the breeding priority score to generate the final recommended sequence for population expansion and determine the breeding priority of the high-potential cattle.

[0044] In one specific embodiment, the decision output module performs the following steps: It obtains accurate progeny potential assessment values ​​for all candidate individuals in the population. Using these assessment values ​​as input, a regression model is employed for secondary correction to eliminate any potential minor population biases. Based on the corrected potential values, all individuals are sorted in descending order to generate a preliminary candidate sequence. The top-ranked individuals (e.g., the top 20%) from this preliminary candidate sequence are selected as high-potential cattle. A breeding priority score is calculated by combining the health index of these high-potential cattle with the estimated breeding cost. In this embodiment, The high-potential cattle are then ranked according to their breeding priority scores, generating a final recommended breeding sequence and determining the breeding priority of these high-potential cattle. For example, the output report indicates that cattle numbered 101, 205, and 078 should be prioritized for breeding. This process integrates genetic potential and economic feasibility, providing farmers with scientific and intuitive decision support, ultimately achieving the invention's objective of resolving prediction bias and optimizing breeding decisions.

[0045] This embodiment also provides a Pingliang Red Cattle population expansion data acquisition and analysis system, used to execute the aforementioned Pingliang Red Cattle population expansion data acquisition and analysis method, including: The data acquisition and preprocessing module is used to acquire multi-generational genomic data, historical offspring performance data, and bloodline records of the Pingliang Red Cattle population to form the original dataset; The genetic path analysis module is used to construct a kinship matrix based on the original dataset and to classify multi-generational genetic paths using a random forest algorithm to determine the key genetic path graph. The feature extraction and model building module is used to extract hidden genomic features from the key genetic pathway map to form an input feature set, and to build a progeny potential prediction model based on this set. The feature optimization module is used to preprocess the input feature set and perform environmental variable fusion processing to obtain the optimized genetic prediction vector. The deviation correction analysis module is used to determine the deviation correction coefficient based on the analysis of dynamic parameters of multiple generations of paths. The model optimization and evaluation module is used to update and iteratively optimize the offspring potential prediction model using the bias correction coefficient until an accurate offspring potential assessment value is obtained. The decision output module is used to generate a recommended sequence for population expansion based on the accurate offspring potential assessment value.

[0046] The modules described above can be deployed on a cloud server cluster, receiving data and returning analysis results through API interfaces, together realizing the aforementioned complete methodological process.

[0047] Embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0048] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0049] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0050] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0051] Contents not described in detail in this specification are prior art known to those skilled in the art. It is hereby indicated that the above description is intended to help those skilled in the art understand this invention, but does not limit the scope of protection of this invention. Any equivalent substitutions, modifications, improvements, or simplifications of the above descriptions that do not depart from the essential content of this invention fall within the scope of protection of this invention.

Claims

1. A method for collecting and analyzing data on the population expansion of Pingliang Red Cattle, characterized in that, include: Obtain multi-generational genomic data, historical offspring performance data, and pedigree records of the Pingliang Red Cattle population to construct an original dataset; The bloodline records are obtained by automatically linking the identification devices installed on the cattle with the data acquisition devices. Based on the original dataset, a blood relationship matrix was constructed, and a random forest algorithm was used to classify multi-generational genetic paths to determine the key genetic path graphs representing the gene transmission chain from ancestor to offspring. Extract node data from the key genetic path graph where the genetic similarity between nodes exceeds a preset similarity threshold, and extract their hidden genomic features through a deep learning network to form an input feature set; Based on the input feature set and the historical offspring performance data, a prediction model for offspring potential is constructed. The input feature set is preprocessed and environmental variables from breeding records are fused to adjust the feature weights, resulting in an optimized genetic prediction vector. Based on the optimized genetic prediction vector, multi-generational path dynamic parameters are extracted. The random forest regression algorithm is used to analyze the impact of the path dynamic parameters on the performance of offspring, and the bias correction coefficient is determined accordingly. The offspring potential prediction model is updated using the deviation correction coefficient, and the model is iteratively optimized to make the matching degree between the simulated offspring performance data output by the model and the historical offspring performance data reach a preset matching degree threshold, thereby obtaining an accurate offspring potential assessment value. Based on the precise offspring potential assessment values, a recommended sequence for population expansion is generated, and the breeding priority of high-potential cattle is determined.

2. The method for collecting and analyzing data on the population expansion of Pingliang Red Cattle according to claim 1, characterized in that, The construction of the original dataset specifically includes: Based on the multi-generational genomic data, a clustering algorithm is used to divide the data into multiple genomic data groups according to individuals; Determine whether the kinship records between different individuals match the genetic consistency of their genomic groups, and determine the kinship chain when matching; Based on the bloodline, identify recurring genetic patterns in the original dataset; Based on the recurring gene inheritance pattern, regression analysis is used to predict genomic change trends. Based on the predicted genomic change trend, determine whether the newly acquired genomic data has deviated, and mark any deviations as abnormal gene points; The abnormal gene points are compared with the relevant kinship records. Based on the comparison results, the kinship records, genomic data and historical offspring performance data are integrated and corrected to construct the original dataset.

3. The method for collecting and analyzing data on the population expansion of Pingliang Red Cattle according to claim 1, characterized in that, The determination of the key genetic pathway map representing the gene transfer chain from ancestor to offspring specifically includes: Kinship data is extracted from the original dataset, and kinship coefficients between individuals are calculated to construct the blood relationship matrix. The random forest classification algorithm is used to classify multiple genetic paths derived from the kinship matrix and identify key genetic path features; Based on the key genetic pathway features, a path graph structure is constructed with individuals as nodes and parent-child relationships with significant genetic contributions as directed edges. Calculate the connection strength of each node in the path graph structure, and determine the path with connection strength exceeding a preset strength threshold as the main transmission path; Based on the main transmission pathways and their associated individual attributes, the layout of the key genetic pathway map is completed.

4. The method for collecting and analyzing data on the population expansion of Pingliang Red Cattle according to claim 1, characterized in that, The formation of the input feature set specifically includes: Calculate the genome-wide genetic similarity between individuals corresponding to any two nodes in the critical genetic pathway graph; If the genetic similarity between the two nodes exceeds the preset similarity threshold, the genomic data of the individuals corresponding to the two nodes are extracted from the original dataset and input into the convolutional neural network to extract high-dimensional hidden genomic features. All extracted latent genomic features were grouped using clustering methods to form grouped feature sets; The grouped feature set is normalized to obtain the input feature set.

5. The method for collecting and analyzing data on the population expansion of Pingliang Red Cattle according to claim 1, characterized in that, The offspring potential prediction model employs a deep neural network or gradient boosting tree model trained using a gradient descent algorithm.

6. The method for collecting and analyzing data on the population expansion of Pingliang Red Cattle according to claim 1, characterized in that, The optimized genetic prediction vector obtained specifically includes: The input feature set is processed using the Z-score normalization method to obtain a normalized feature set; Identify a subset of environmental features related to environmental factors from the standardized feature set; If the environmental feature subset is not empty, then the mutual information value between each environmental feature and the remaining features is calculated; wherein, the environmental features are taken from the environmental feature subset, and the remaining features are features in the standardized feature set other than the environmental feature subset; Based on the calculated mutual information values, the initial weights of the remaining features corresponding to each value are adjusted, and the adjusted weights are fused with the standardized feature set using a linear weighting method to generate the optimized genetic prediction vector.

7. The method for collecting and analyzing data on the population expansion of Pingliang Red Cattle according to claim 1, characterized in that, The determination of the deviation correction coefficient specifically includes: From the optimized genetic prediction vector, a set of dynamic parameters for multiple generations of paths is parsed out. The dynamic parameters include path strength, path stability, and path length calculated based on the key genetic path graph. Using the dynamic parameter set as input features and historical offspring performance data as target variables, a random forest regression model is constructed. Based on the importance ranking of the features output by the random forest regression model, parameters with importance higher than a preset importance threshold are marked as critical path parameters; Calculate the average difference between the model predictions and actual values ​​of the critical path parameters, and determine the deviation correction coefficient based on this difference.

8. The method for collecting and analyzing data on the population expansion of Pingliang Red Cattle according to claim 1, characterized in that, The step of iteratively optimizing the model output to achieve a preset matching degree threshold between simulated offspring performance data and historical offspring performance data specifically includes: The output of the offspring potential prediction model is initially corrected using the deviation correction coefficient to obtain preliminary simulation data; The cosine similarity between the preliminary simulation data and the historical performance data of future generations is calculated as the matching degree. If the matching degree is lower than the matching degree threshold, the gradient descent algorithm is used to adjust the parameters of the offspring potential prediction model, and simulated data is regenerated for calculation until the matching degree reaches or exceeds the matching degree threshold.

9. The method for collecting and analyzing data on the population expansion of Pingliang Red Cattle according to claim 1, characterized in that, The process of generating recommended sequences for population expansion and determining breeding priorities for high-potential cattle specifically includes: The precise offspring potential assessment value of all candidate individuals in the population is obtained, and the assessment value is used as input to perform secondary correction using a regression model to eliminate the population bias of the assessment value; All individuals are sorted in descending order based on the corrected potential values ​​to generate a preliminary candidate sequence; Individuals ranked higher in the preliminary candidate sequence were selected as high-potential cattle. The breeding priority score is calculated by combining the health index of the high-potential cattle with the estimated breeding cost. High-potential cattle are ranked according to the breeding priority score, a final population propagation recommendation sequence is generated, and the breeding priority of the high-potential cattle is determined.

10. A data collection and analysis system for the population expansion of Pingliang Red Cattle, characterized in that, A method for collecting and analyzing data on the population expansion of Pingliang Red Cattle, as described in any one of claims 1-9, includes: The data acquisition and preprocessing module is used to acquire multi-generational genomic data, historical offspring performance data, and bloodline records of the Pingliang Red Cattle population to form the original dataset; The genetic path analysis module is used to construct a kinship matrix based on the original dataset and to classify multi-generational genetic paths using a random forest algorithm to determine the key genetic path graph. The feature extraction and model building module is used to extract hidden genomic features from the key genetic pathway map to form an input feature set, and to build a progeny potential prediction model based on this set. The feature optimization module is used to preprocess the input feature set and perform environmental variable fusion processing to obtain the optimized genetic prediction vector. The deviation correction analysis module is used to determine the deviation correction coefficient based on the analysis of dynamic parameters of multiple generations of paths. The model optimization and evaluation module is used to update and iteratively optimize the offspring potential prediction model using the bias correction coefficient until an accurate offspring potential assessment value is obtained. The decision output module is used to generate a recommended sequence for population expansion based on the accurate offspring potential assessment value.

Images