Weaned Hu sheep lamb seed selection method based on machine learning

Abstract

The invention relates to a weaned Hu sheep lamb seed selection method based on machine learning. The method comprises the following steps: collecting, sorting and storing long-term production practice data of Hu sheep; preprocessing the collected data, and screening out the most effective features; and selecting a plurality of machine learning models to train and comprehensively analyze the data to obtain a model with an optimal prediction effect. The prediction result is evaluated through the model, a threshold value is set, in combination with production requirements, a seed selection decision is obtained, the selected breeding sheep is marked, hybridized and tracked for a long time, the production and reproduction conditions of the breeding sheep are recorded, and real-time improvement and optimization of the model are ensured. According to the method, the growth performance of the lambs can be accurately predicted, and the selection decision is optimized according to the prediction result.

Description

Technical Field

[0001] The present invention relates to the field of Hu sheep breeding, and in particular to a method for breeding Hu sheep weaned lambs based on machine learning. Background Art

[0002] Hu sheep is a local sheep breed unique to my country, especially represented by the Hu sheep breeding industry in Huzhou City. It has excellent meat performance and high fertility, and occupies an important position in animal husbandry. Traditional Hu sheep selection methods mostly rely on artificial experience or indicators based on reproductive performance, such as the reproductive capacity and maternal performance of the flock. However, these methods have certain limitations when facing complex multi-dimensional influencing factors.

[0003] The selection of weaned lambs is particularly faced with the problem of how to accurately assess the growth potential of lambs. Traditional selection methods often rely only on individual weight, genetic background and environmental factors, but lack effective tools for deeper predictions of growth performance. With the development of modern genetics and data science, precision selection methods based on bioinformatics and big data analysis have gradually been applied. Many studies have shown that a single genetic evaluation method has limitations in optimizing breeding results, especially for the evaluation of diversified indicators such as growth rate and weight. Machine learning models can integrate multidimensional data and mine complex nonlinear relationships from them, thereby providing more accurate and comprehensive decision-making basis for precision breeding.

[0004] Especially in the field of sheep breeding, growth performance prediction methods based on machine learning have made some progress. For example, using support vector machines (SVM), neural networks (ANN) and other methods to predict the growth performance of sheep can provide more accurate selection strategies. However, in the selection of weaned Hu sheep lambs, multi-dimensional data analysis on their growth characteristics, feeding management and environmental changes is still relatively scarce. Summary of the invention

[0005] The technical problem to be solved by the present invention is: in order to effectively make up for the shortcomings of the traditional method in prediction accuracy, a more scientific and efficient solution is provided for Hu sheep breeding. A machine learning method for selecting weaned Hu sheep lambs is designed. The details are as follows:

[0006] Step S1: Collecting data, including family information data, growth performance data, and reproductive performance data;

[0007] Step S2: preprocess the data, including using the mean filling method and interpolation method to supplement the data set; based on the standard deviation 3σ criterion and IQR method, identify the data that deviates significantly from the normal range and correct or eliminate them; standardize and normalize the data dimensions of different characteristics; calculate the average lambing rate of each ewe as the prediction target variable of the model, and the calculation formula of the average lambing rate is:

[0008]

[0009] Among them, R avg is the average lambing rate of ewes, N total is the total number of lambs produced by the breeding ewe during the reproductive cycle, T births The lambing order of the ewe;

[0010] Pearson's correlation coefficient was used to analyze the correlation of continuous traits and to screen out traits that were highly correlated with reproductive performance;

[0011] Step S3: Perform model optimization and training, wherein the residual network and Swish activation function are introduced to optimize the deep neural network model, and the data is divided into training and validation classes at a ratio of 8:2; grid search and random search are used to perform hyperparameter tuning; during training, the model is optimized by minimizing the loss function, and the model parameters are updated by the gradient descent method;

[0012] Step S4: To perform performance prediction, first input data: input new lamb sample data (including genetic, growth performance and other characteristic data) into the trained model; then predict output: finally, evaluate and verify accuracy by calculating error indicators;

[0013] Step S5: Based on the prediction results of the trained model and the actual production needs of the sheep farm, reasonable breeding sheep screening and decision-making are carried out, including:

[0014] Set a threshold value, calculate the lambing rate of the top 20% of the breeding sheep in the sheep farm's historical data, and set it as the baseline value: Let L = {L1, L2, ..., L n} is the lambing rate of all ewes in the test set, where L i The lambing rate of the i-th ewe is obtained by calculating the lowest lambing rate of the top 20% of the breeding sheep in the data set. baseline , the formula is:

[0015] L baseline =min{L i |L i ∈L 20％}

[0016] In practical application, according to the baseline lambing rate L baseline , the sheep farm can further select breeding sheep that meet this standard and cultivate them as reserve breeding sheep;

[0017] Comprehensive evaluation of multiple indicators: comprehensive evaluation based on lamb skin grade, genetic genes, pedigree and other factors;

[0018] Step S6: Implementation of seed selection, marking and registration of breeding sheep, data tracking and adjustment and update model

[0019] As a preferred embodiment: the standardization method in step S2 is to center the data of each feature according to its mean and scale it according to the standard deviation; normalization is to ensure that the contribution of each feature to the model training is balanced by scaling the data to the range of [0,1].

[0020] As a preferred embodiment, the data preprocessing in step S2 also includes adopting a feature importance evaluation method of a random forest to further identify the features that contribute most to the prediction.

[0021] As a preferred embodiment, the data preprocessing in step S2 also includes data enhancement processing, using random noise injection and resampling technology based on statistical distribution to generate new data samples.

[0022] First, by using machine learning algorithms to analyze multi-dimensional data, we can accurately predict the growth performance of lambs and optimize the selection decision based on the prediction results. Compared with traditional manual experience and single indicator evaluation, this invention effectively avoids human bias and improves the accuracy and reliability of selection.

[0023] Secondly, the growth prediction model based on machine learning can accelerate the genetic improvement process of the Hu sheep population. Through the comprehensive analysis of multiple influencing factors, weaned lambs with excellent genetic potential can be quickly identified, thereby improving the efficiency and genetic quality of breeding sheep selection. Compared with traditional methods, the present invention can more efficiently optimize the production performance of the Hu sheep population.

[0024] Finally, the present invention can significantly improve the production efficiency of Huyang farms. Through accurate seed selection, the reproduction of undesirable individuals can be reduced, thereby improving the overall production efficiency. In addition, the machine learning method can adjust the seed selection strategy in real time, enhancing the ability to adapt to different environmental conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Attached Figure 1 :A flow chart of a method for selecting weaned Hu sheep lambs based on machine learning DETAILED DESCRIPTION

[0026] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] The specific implementation of the present invention includes the steps of data collection, data preprocessing, machine learning model training, performance prediction, seed selection decision and seed selection implementation, which are described in detail below:

[0028] Step S1: Data collection. The data collection of the present invention mainly comes from the long-term production practice data of Huyang farms. These data are collected by multiple Huyang farms in Huzhou through standardized recording methods, and are systematically organized and stored to ensure the integrity and reliability of the data. Specific data categories include the following aspects: First, in terms of genetic data, detailed pedigree information of each sheep is collected, including parent information, grandparent information, and grandparent information. Through the construction of a pedigree map, the genetic origin of the sheep can be fully traced and the transmission of excellent genes in the family can be evaluated. These genetic data are crucial for a deep understanding of the growth potential, fertility and other aspects of Huyang individuals, and can provide basic support for breeding decisions.

[0029] Secondly, in terms of growth performance data, multiple growth indicators of Hu sheep at different growth stages are recorded in detail. Specifically, key indicators such as weight, body length, body height and chest circumference at each growth stage, such as birth, weaning, six months of age, and adulthood, are recorded. In addition, important reproductive performances such as lamb fur grade and number of lambs per birth are also recorded. These growth performance data not only reflect the performance of individuals at specific growth stages, but also provide a quantitative basis for evaluating the overall growth trend of Hu sheep.

[0030] These multi-dimensional data will serve as input for machine learning models to help accurately predict the future growth potential of weaned lambs and achieve scientific breeding goals for breeding sheep.

[0031] Step S2: Data preprocessing is a key step to ensure that machine learning models can be trained accurately and efficiently. Raw data usually contains missing values, outliers, and noise, so a series of preprocessing operations are required to ensure data quality and model training results. Specifically, the following steps are included:

[0032] Step S21: During the data collection process, due to complex environment or operational errors, some data are often missing. The missing value processing mainly adopts a strategy combining mean filling method and interpolation method. Among them, for missing data of quantitative variables such as weight, height, and length, the mean filling method can be used to estimate the missing values ​​to ensure that the overall distribution characteristics of the data are maintained; for severely missing data samples, the present invention processes them through the threshold deletion method, that is, if the missing ratio of the data sample exceeds a predetermined threshold (such as 20%), the sample is directly eliminated. This method ensures data quality through strict threshold setting, thereby avoiding the negative impact of low-quality samples on model performance.

[0033] Step S22: The presence of data outliers will seriously affect the training effect of the model, so outlier detection is an important step in data cleaning. First, a statistical analysis method (based on the standard deviation 3σ criterion and IQR method) is used to analyze the data distribution and identify values ​​that significantly deviate from the normal range. For the detected outliers, they are corrected or eliminated as appropriate to ensure the accuracy of the data and the stability of the model training.

[0034] Step S23: Since different features in the data have different dimensions, such as weight in kilograms and height in centimeters, directly using these data for model training may cause the impact of certain features on the model to be over-amplified or ignored. For this reason, data standardization and normalization are essential.

[0035] For data standardization, the Z-score standardization method is used to center the data of each feature by its mean and scale it by the standard deviation so that the data has a standard normal distribution with a mean of 0 and a standard deviation of 1. For each feature X, the data is standardized by the following formula:

[0036]

[0037] Among them, μ is the mean of feature X, and σ is the standard deviation of feature X.

[0038] Normalization ensures that each feature contributes evenly to model training by scaling the data to the range of [0,1]. The formula is as follows:

[0039]

[0040] Among them, x min and x max are the minimum and maximum values ​​of feature X. The above steps greatly improve the convergence speed and stability of model training.

[0041] Step S24: In data preprocessing, the average lambing rate of each ewe needs to be calculated as the prediction target variable of the model. The calculation formula of the average lambing rate is:

[0042]

[0043] Among them, R avg is the average lambing rate of ewes, N total is the total number of lambs produced by the breeding ewe during the reproductive cycle, T births The average lambing rate is the number of lambings of the ewe. By calculating the average lambing rate, the reproductive capacity of each ewe can be evaluated, which is of great reference value for the selection of breeding sheep.

[0044] Step S25: Feature selection is an important part of data preprocessing, which aims to improve the training efficiency and generalization ability of the model by screening out features that have a significant impact on the target variable and reducing the interference of redundant features on the model. The present invention adopts a dual feature selection strategy based on correlation and model-driven: First, the Pearson correlation coefficient is used to perform correlation analysis on continuous features to screen out features that are highly correlated with reproductive performance. The formula is as follows:

[0045]

[0046] Among them, r xy represents the correlation coefficient between feature x and target variable y, and are the means of the features and target variables, respectively. Features with higher feature importance will be retained for model training.

[0047] Furthermore, the feature importance evaluation method of random forest is used to further identify the features that contribute most to the prediction. This feature selection strategy that combines correlation and model-driven ensures that the feature screening process has both theoretical support and is verified by the actual model, so that the final selected features can effectively improve the performance of the model.

[0048] Step S26: Data enhancement. In order to improve the generalization performance of the model when the training data samples are insufficient, the present invention also performs data enhancement processing. For the production performance data of Hu sheep, random noise injection and resampling technology based on statistical distribution are used to generate new data samples to enrich the training set of the model.

[0049] x augmental =x+∈,∈~N(0,σ 2 )

[0050] Among them, ∈ is random noise that obeys the normal distribution, and σ is the standard deviation of the noise. By generating multiple enhanced samples, the robustness of the model can be effectively improved and overfitting can be avoided.

[0051] Step S3: After data preprocessing, the selection and training process of the machine learning model is crucial. To ensure that the model can effectively learn the inherent laws of the data, a suitable machine learning algorithm is usually selected based on the characteristics of the data, task objectives, and performance requirements. The training process of the present invention includes multiple key steps, such as model selection, data partitioning, hyperparameter tuning, training process, etc. The specific steps are as follows:

[0052] Step S31: After the data preprocessing is completed, select a suitable machine learning model for training. Traditional machine learning methods, such as support vector machine (SVM), random forest (RF), and XGBoost and LightGBM based on gradient boosting trees, have been widely used in various regression and classification tasks. These traditional methods perform well in many tasks. In the present invention, we design a deep neural network (DNN) model and combine it with traditional machine learning methods to improve the accuracy of the prediction model.

[0053] Considering the characteristics of the data in this study - small data sets, complex nonlinear relationships, and the need for high-order features, DNN was designed as the main model architecture. By introducing the residual network (ResNet) and the Swish activation function, the model can effectively avoid the gradient vanishing problem, improve learning efficiency, and thus improve prediction performance.

[0054] Swish(x)=x·σ(x)

[0055] Where σ(x) is the Sigmoid function, defined as The Swish function has been proven in experiments to be superior to the traditional ReLU activation function in many deep learning tasks, especially in deeper networks, where it can provide better gradient flow and alleviate the gradient vanishing problem. Therefore, using Swish as the activation function of this model can help the network update weights more efficiently during training, avoiding the "neuron death" problem that may occur with ReLU in deep networks.

[0056] The core structure of this model is the residual block. Each residual block includes two fully connected layers, batch normalization, Dropout, and residual connections. Residual connections help the network skip some layers, thereby preventing the gradient vanishing and information loss problems that occur when the network is too deep. The form of the residual block is:

[0057]

[0058] Among them, W1 and W2 are the weight matrices of the linear layer, b1 is the bias term, x is the input, and the residual connection is implemented by adding the input x directly to the output. In this way, the network can more easily learn the residual (that is, the difference between the input and the output) instead of learning the entire mapping. In this way, the network can effectively learn deep representations while avoiding the gradient disappearance problem common in deep networks.

[0059] The input layer maps the raw data from the features to a higher dimensional space with a size of 512 nodes. The four residual blocks gradually reduce the number of nodes from 512 nodes to 64 nodes. This layer-by-layer reduction method can help the model gradually extract features from coarse to fine. The size of the output layer is 1, and it outputs the predicted continuous value.

[0060] Step S32: After the model selection is determined, the next step is to divide the data reasonably. We use 80% of the data to train the model and 20% of the data to verify the effect of the model. This division ratio can ensure that the model has enough data to learn during training, while retaining a portion of the data for evaluating the generalization ability of the model.

[0061] Step S33: Hyperparameter tuning aims to improve the prediction accuracy of the model by optimizing hyperparameters (learning rate, regularization coefficient, number of network layers, dropout ratio, etc.). In this study, grid search was used to select hyperparameters.

[0062] Grid search discretizes the hyperparameter space and tries each possible parameter combination one by one to find the best combination. For this study, the selection of hyperparameters mainly includes the following aspects: Learning rate: The value range is [10 -5 , 10 -4 , 10 -3 , 10 -2 , 10 -1 ], the grid search tries these discrete values ​​to find the best learning coefficient that can avoid model oscillation while ensuring convergence speed:

[0063] Regularization coefficient: The value range is [10 -6 , 10 -4 , 10 -3 , 10 -2 ], the choice of regularization parameters is based on preventing the model from overfitting the training data while not affecting the effective learning of the growth performance of Hu sheep.

[0064] Number of network layers and number of neurons per layer: To determine the optimal model depth, grid search tested network structures with 3 to 6 layers.

[0065] Dropout ratio: The value range is [0.3, 0.5, 0.7]. Its main function is to reduce overfitting. The impact on the generalization ability of the model is observed through different ratio combinations.

[0066] Step S34: training process. During the model training process, the model is optimized by minimizing the loss function (mean square error), and the model parameters are updated by gradient descent or other optimization algorithms.

[0067] The specific process is as follows:

[0068] First, in order to measure the difference between the model prediction value and the true value, the mean square error (MSE) is used as the loss function. MSE can effectively amplify the impact of the prediction error, and its formula is:

[0069]

[0070] Among them, y i represents the true value, y i Represents the predicted value, and N is the number of samples. MSE can ensure that the model prioritizes reducing data points with large errors during training, thereby converging to a better solution faster.

[0071] Secondly, in order to optimize the loss function, we use the gradient descent method, which is an iterative optimization algorithm that updates the parameters along the negative gradient direction of the loss function to minimize the loss function value. The basic formula of the gradient descent method is:

[0072]

[0073] Where: θ t represents the parameter vector in the tth iteration, η is the learning rate, which controls the step size of parameter update. is the gradient of the loss function with respect to the parameter θ. The core idea of ​​the gradient descent method is to use the gradient of the loss function to guide the direction and magnitude of parameter updates, so that each iteration is in the direction of reducing the loss. Through multiple iterations, the global optimal solution is gradually approached.

[0074] Step S4: Performance prediction is a crucial part of this study, which aims to use the trained model to predict the newly input weaned lamb samples, especially the average lambing rate of each lamb. This process can help screen out breeding sheep with low lambing numbers, optimize breeding decisions, and thus improve the production efficiency of the entire population.

[0075] Step S41: Data input: Input new lamb sample data (including genetic, growth performance and other characteristic data) into the trained model. When inputting data, firstly perform necessary preprocessing on it to ensure that the format of the data matches the model requirements.

[0076] Step S42: prediction output: the model outputs prediction results. Through these prediction results, the growth potential of lambs can be quantitatively evaluated, thereby providing a basis for the selection of breeding sheep.

[0077] Step S43: Evaluation and verification: Use the verification set and real sample data to evaluate the prediction results, calculate error indicators (such as mean square error, mean absolute error, etc.) to ensure the prediction accuracy of the model.

[0078] Step S5: The breeding decision is to make reasonable breeding sheep screening and decision-making based on the prediction results of the trained model and the actual production needs of the sheep farm. The goal of this step is to select high-quality reserve breeding sheep in a scientific and data-driven way. The steps are as follows:

[0079] Step S51: Setting the threshold. In the seed selection decision of the present invention, setting the seed selection standard is a crucial link. According to the fact that sheep farms in actual production update about 20% of the breeding sheep every year, we introduced a threshold setting method based on lambing rate. The specific threshold setting process is as follows: First, according to the lambing rate of the breeding sheep ranked in the top 20% in the sheep farm's historical data, calculate its lambing rate and set it as the baseline value: Let L = {L1, L2, ..., L n} is the lambing rate of all ewes in the test set, where L i The lambing rate of the i-th ewe is obtained by calculating the lowest lambing rate of the top 20% of the breeding sheep in the data set. baseline , the formula is:

[0080] L baseline =min{L i |L i ∈L 20％}

[0081] In practical application, according to the baseline lambing rate L baseline , the sheep farm can further select the breeding sheep that meet this standard and cultivate them as reserve breeding sheep. Through this standard, the sheep farm can select the breeding sheep according to the lambing rate, eliminate the inefficient breeding sheep that do not reach the baseline lambing rate, and improve the reproductive efficiency of the entire population.

[0082] Step S52: Comprehensive evaluation of multiple indicators. In the breeding decision-making process, in addition to the main factor of lambing rate, multiple other important indicators should be considered comprehensively, such as lamb skin grade, genetic potential (such as genotype), pedigree and other factors, to select the most promising weaned lambs.

[0083] Step S53: weight assignment. Different breeding objectives may lead to different importance of different indicators. Different weights can be assigned according to their importance, and the final selection is carried out through weighted scoring.

[0084] In the selection decision-making process, different weights can be assigned to various indicators according to the actual situation of the sheep farm. For example, with the goal of improving reproductive capacity, lambing rate may be given a higher weight to ensure that the selected sheep have excellent reproductive characteristics. For sheep farms that use fur as an important source of income, the weight of lamb grade will be increased accordingly to ensure that the selected sheep have better fur characteristics. Sheep farms that introduce excellent pedigree breeds from outside: the weight of pedigree background will be increased to ensure the transmission and retention of excellent genes.

[0085] S final =ω1×S 产羔率 +ω2×S 羔皮等级 +ω3×S 血统

[0086] Among them, ω1, ω2, and ω3 represent the weights of each indicator, which can be flexibly adjusted by the sheep farm according to the actual breeding needs. final Scoring is used to select the best breeding sheep.

[0087] Step S6: Seed selection implementation: after completing the seed selection decision, implement the seed selection process

[0088] Step S61: Marking and registration, marking the selected breeding sheep and recording their relevant information so that they can be clearly distinguished and tracked in the subsequent management and breeding process;

[0089] Step S62: Breeding of breeding sheep directly affects the production performance and genetic structure of the offspring. A scientific breeding plan is adopted according to the genetic background, production goals and actual needs of the sheep farm. The design of the breeding strategy not only considers the genetic potential of a single breeding sheep, but also includes the overall genetic diversity of the group to avoid inbreeding and genetic bottlenecks.

[0090] Step S63: Tracking and adjustment: Track the selected breeding sheep for a long time, record their production and reproduction, ensure that their performance in actual production meets expectations, and adjust the breeding strategy when necessary. As the sheep farm production data is continuously updated, the model will also be updated with the continuous input of new data to ensure continuous improvement and optimization.

[0091] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for selecting weaned Hu sheep lambs based on machine learning, characterized in that: The following steps are involved: Step S1: Collecting data, including family information data, growth performance data, and reproductive performance data; Step S2: preprocessing the data, including using a combination of mean filling method and interpolation method to supplement the data set; based on the standard deviation 3σ criterion and IQR method, identifying data that significantly deviates from the normal range, and correcting or eliminating them; The data dimensions of different characteristics are standardized and normalized; the average lambing rate of each ewe is calculated as the prediction target variable of the model. The calculation formula of the average lambing rate is: Among them, R avg is the average lambing rate of ewes, N total is the total number of lambs produced by the breeding ewe during the reproductive cycle, T birth s The lambing order of the ewe; Pearson's correlation coefficient was used to analyze the correlation of continuous traits and to screen out traits that were highly correlated with reproductive performance; Step S3: Perform model optimization and training, wherein the residual network and Swish activation function are introduced to optimize the deep neural network model, and the data is divided into training and validation classes at a ratio of 8:2; grid search and random search are used to perform hyperparameter tuning; during training, the model is optimized by minimizing the loss function, and the model parameters are updated by the gradient descent method; Step S4: To perform performance prediction, first input data: input new lamb sample data (including genetic, growth performance and other characteristic data) into the trained model; then predict the output; finally, evaluate and verify the accuracy by calculating the error index; Step S5: Based on the prediction results of the trained model and the actual production needs of the sheep farm, reasonable breeding sheep screening and decision-making are carried out, including: Set a threshold value, calculate the lambing rate of the top 20% of the breeding sheep in the sheep farm's historical data, and set it as the baseline value: Let L = {L1, L2, ..., L n } is the lambing rate of all ewes in the test set, where L i The lambing rate of the i-th ewe is obtained by calculating the lowest lambing rate of the top 20% of the breeding sheep in the data set. baseline , the formula is: L baseline =min{L i |L i ∈L 20％ } In practical application, according to the baseline lambing rate L baseline , the sheep farm can further select breeding sheep that meet this standard and cultivate them as reserve breeding sheep; Comprehensive evaluation of multiple indicators: comprehensive evaluation based on lamb skin grade, genetic genes, pedigree and other factors; Step S6: Implement seed selection, mark and register the breeding sheep, track data and adjust and update the model.

2. The method according to claim 1, characterized in that The standardization method in step S2 is to center the data of each feature according to its mean and scale it according to the standard deviation; normalization is to ensure that each feature contributes evenly to model training by scaling the data to the range of [0,1].

3. The method according to claim 1, characterized in that: Step S2 of preprocessing the data also includes using a random forest feature importance evaluation method to further identify the features that contribute most to the prediction.

4. The method according to claim 1, 2 or 3, characterized in that: The data preprocessing in step S2 also includes data enhancement processing, using random noise injection and resampling technology based on statistical distribution to generate new data samples.

Images