Laying hen lossless data processing method based on improved SVR

Abstract

The invention discloses a laying hen lossless data processing method based on improved SVR, relates to the technical field of data processing, and adopts an adaptive cap cleaning data based on quartile distance to dynamically adjust an abnormal value elimination boundary, avoid mistaken deletion of effective physiological signals, improve the robustness of data cleaning and improve the accuracy of data cleaning. The defect that data distribution is damaged by a traditional fixed threshold value and a simple filling method is overcome. Through a preprocessing assembly line of KNN topology filling and Yeo-Johnson manifold transformation, missing data are effectively repaired, data skewed distribution is corrected, and the problem of feature space measurement failure is solved. Based on self-adaptive feature screening of F statistics, redundant features can be accurately removed, data dimension reduction is realized, and dimensionality disaster and model overfitting are avoided. According to the established TTR-SVR model and parameter optimization strategy, optimization precision and calculation efficiency are both considered, the problem that traditional SVR hyper-parameter selection is unreasonable is solved, and the accuracy and efficiency of laying hen nondestructive monitoring data processing are integrally improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a lossless data processing method for laying hens based on an improved SVR. Background Technology

[0002] With the rapid development of modern animal husbandry towards intensification, large-scale operation, and intelligentization, precise data monitoring and analysis in the egg-laying hen farming process has become a crucial link in improving farming efficiency, ensuring food safety, and achieving refined management. In actual production scenarios, farms typically deploy a large number of sensors to collect real-time growth and physiological indicators of laying hens. Efficient processing and accurate modeling of this multi-dimensional time-series data have significant engineering application value for achieving non-destructive testing of laying hen health status and prediction of production performance.

[0003] However, existing data processing and prediction technologies for laying hen farming still face many challenges in practical applications. First, due to the complexity of the farming environment and the stability of the sensors themselves, the raw data collected is often of low quality, containing a large number of missing values ​​and outlier noise. Traditional data cleaning methods often use simple mean or median imputation, which easily destroys the original distribution structure of the data, leading to the accumulation of biases. Meanwhile, anomaly removal methods based on fixed thresholds lack adaptive dynamic boundary adjustment capabilities, easily erroneously deleting extreme physiological signals with important biological significance, and have poor robustness to non-normally distributed data.

[0004] Secondly, laying hen growth data typically exhibits highly nonlinear and non-normal distribution characteristics, and existing modeling schemes often neglect the impact of feature distribution on model convergence. Although some studies have employed standardization, they have failed to effectively correct the skewed distribution of the data, causing machine learning algorithms based on Euclidean distance or dot product operations to become ineffective in terms of metric performance in the feature space. Furthermore, the original feature set often contains a large number of redundant or irrelevant variables, lacking an adaptive feature selection mechanism based on statistical significance testing, which can easily lead to the curse of dimensionality and model overfitting, thereby reducing prediction accuracy.

[0005] While Support Vector Regression (SVR) is widely used due to its excellent generalization ability, its performance is highly dependent on the selection of hyperparameters. Traditional grid search methods are computationally expensive, while random search methods, although faster, struggle to lock the global optimum. Existing single search strategies fail to achieve a balance between computational efficiency and optimization accuracy. Therefore, developing a lossless data processing method that integrates robust data cleaning, adaptive feature transformation, and efficient parameter optimization has become an urgent need in the field of smart aquaculture. Summary of the Invention

[0006] In order to solve the above-mentioned technical problems, this application proposes the following technical solution: In a first aspect, embodiments of this application provide a method for lossless processing of laying hen data based on an improved SVR, including: Multidimensional monitoring data of laying hens is acquired, and the multidimensional monitoring data is cleaned based on the adaptive capping of the interquartile range to eliminate extreme outliers. Construct multidimensional monitoring data after preprocessing pipeline repair and cleaning based on KNN topology filling and Yeo-Johnson manifold transformation; The adaptive feature percentage screening based on the F statistic is used to remove redundant features and reduce the dimensionality of the repaired multidimensional monitoring data to obtain multidimensional monitoring sample data. A target transformation support vector regression (TTR-SVR) model is constructed, and the model parameters are optimized using the multidimensional monitoring sample data to determine the final TTR-SVR model.

[0007] In one possible implementation, the acquisition of multidimensional monitoring data of laying hens, and the cleaning of the multidimensional monitoring data based on an adaptive capping method using interquartile range to eliminate extreme outliers, includes: The original observation space was constructed based on the multidimensional monitoring data of the laying hens: in: Indicates the first The input feature vector of each sample, Indicates the first The target observation value for each sample Indicates the total number of samples; express 3D feature vectors This represents the total feature dimension, i.e., the number of features contained in each sample. It is the Cartesian product symbol, representing the combination of the input space and the output space; Represents a one-dimensional real space, corresponding to the target observation in scalar form. ; For any feature dimension Define its empirical cumulative distribution function: in: Indicates the first The empirical cumulative distribution function of the dimensional feature, Index representing the feature dimension Represents an arbitrary real-valued variable. Indicates all Iterate through and accumulate each sample. This represents an indicator function; if true, the function value is... If this is not true, then the function value is ; Indicates the first The sample at the th The original feature values ​​on the dimensional feature; Based on the inverse cumulative distribution function, a quantile operator is defined to accurately extract the distribution feature points of the data: in: Indicates the first quantile function of each feature This represents the quantile value, which ranges from 0 to 1. The infimum symbol represents all conditions that satisfy the condition. The minimum value among the values; Constructing a robust discreteness metric matrix , its first Each component The interquartile range is defined as the robustness of data distribution after excluding extreme values ​​at both ends. Define the set of anomaly truncation thresholds ,in: This is the upper bound of the set of abnormal truncation thresholds. This is the lower bound of the set of abnormal truncation thresholds; Finally, a nonlinear capping mapping operator is constructed. The initial multidimensional monitoring data was coarsely cleaned to remove obvious outliers. The component mapping rules are as follows: in: This represents a two-sided truncation operator that forces outlier data that exceeds the normal range to be mapped onto the boundary manifold, thereby eliminating extreme noise interference while preserving the amount of data. This represents a minimum value function that takes the smaller of the two values ​​within the parentheses. This represents the maximum value function, which takes the larger of the two values ​​within the parentheses.

[0008] In one possible implementation, the construction of multidimensional monitoring data after preprocessing pipeline repair and cleaning based on KNN topology filling and Yeo-Johnson manifold transformation includes: To address the missing values ​​in the cleaned multidimensional monitoring data, for each missing sample vector, a Euclidean metric-based search is conducted within the complete feature subspace. Topological filling is performed on the topological neighborhood formed by the nearest neighbor samples; For the filled data matrix Perform maximum likelihood estimation transformation to improve the normality of the feature distribution. For each column of features... To find the optimal parameters.

[0009] In one possible implementation, for a missing sample vector, the search for its Euclidean metric-based value in the complete feature subspace is performed. Topological filling is performed on the topological neighborhood formed by the nearest neighbor samples, including: Define the weighted fill operator: in: Indicates the first The feature vectors of the samples to be filled Indicates the first The feature vectors of the neighboring samples Indicates the first The sample to be filled in the first Estimates of missing values ​​in each feature dimension; Indicates the sample to be filled The missing part; This represents the observed feature subspace, i.e., the sample. and A set of all existing feature dimensions without any missing features; Indicates sample of nearest neighbor set This represents a hyperparameter, indicating how many nearest neighbors to select; This represents the kernel function, used to convert distance into weights; Indicates sample and exist Euclidean distance on; Representing neighbor samples In the The true value of each feature; Kernel function Using the reciprocal distance kernel form: in, This represents the distance kernel function, with distance as input. The output is the weights. Represents the distance between two samples; It is the standard Euclidean distance calculation formula. Indicates traversal Each feature dimension in, It is a very small positive number that prevents the denominator from being zero.

[0010] In one possible implementation, the maximum likelihood estimation transformation is performed on the imputed data matrix to improve the normality of the feature distribution. For each column of features... Finding the optimal parameters includes: in: Indicates the optimal parameters. These are the core parameters of the Yeo-Johnson transform, which determine the specific form of the transform. This refers to finding the parameters in the real number field that maximize the objective function. ; This represents the transformed data value. This represents the mean of the transformed data. It is the variance of the transformed data; The sign function is used to preserve the sign of the data before and after the transformation. Transformation basis functions Defined as a piecewise power function: .

[0011] In one possible implementation, the adaptive feature percentage screening based on the F-statistic performs redundant feature removal and data dimensionality reduction on the repaired multidimensional monitoring data to obtain multidimensional monitoring sample data, including: Constructing a feature importance evaluation operator ,in, Indicates the percentage to be retained. The output is 3D feature vectors This indicates that the mapping relationship maps the high-dimensional feature space to the low-dimensional feature space; Calculate each feature vector With the target vector Univariate regression -Statistics ; Then based on Define feature selection indicator function When this value is 1, it indicates that the feature is preserved. A value of 0 indicates that the value should be removed. in: express -Quantile function of the statistic This is the formula for calculating the threshold of quantiles; if If the value is positive, the feature is considered important and should be retained; otherwise, the feature is considered less important and should be removed. The final output is the filtered feature matrix. Achieving data dimensionality reduction: in, It is a set construction symbol, representing a set consisting of all column vectors that satisfy the condition; Represents the first element in the original data matrix. Column feature vectors; This is a filter condition; it only keeps columns where the indicator function equals 1. arrive This means that by traversing all the original feature columns, the number of columns in the final matrix will be reduced to [number missing]. List.

[0012] In one possible implementation, the computation of each feature vector With the target vector Univariate regression -Statistics ,include: Define the sum of squared residuals Indicates the first The explanatory variance of each feature for the target is defined, and the regression sum of squares is also defined. Indicates unexplained variation: in: Represents the feature index, indicating the feature index for the first... Column feature calculation; Indicates only the first A univariate linear regression model is established for the first feature. The predicted values ​​obtained from each sample; Represent the target variable The sample mean; This indicates the degree to which the calculated predicted value deviates from the mean; the larger the value, the stronger the model's explanatory power. Indicates the first The true target value of each sample; This represents the sum of squares of the prediction errors; the smaller the value, the better the model fit. The construction follows the degree of freedom of The statistical test of the F-distribution Used to examine features With target variable Is there a significant linear relationship between them? in: The sum of squares divided by its degrees of freedom represents the average sum of squares of regression. The sum of squared residuals divided by its degrees of freedom represents the average sum of squared residuals. : indicates that the statistic follows a set of degrees of freedom. of -distributed.

[0013] In one possible implementation, constructing the target transformation support vector regression (TTR-SVR) model includes: In the regenerating nucleus Hilbert space Constructing a regression hyperplane: in: This represents the objective function that needs to be minimized. The normal vector of the regression hyperplane determines the direction of the hyperplane; The intercept of the regression hyperplane determines the position of the hyperplane; Represents the slack variable vector. Measure the first The degree of error of a sample exceeding the upper bound. Measure the degree of error exceeding the lower bound; Represents the regularization term, and represents the weight vector. In the regenerating nucleus Hilbert space The square of the norm; This represents the penalty coefficient, which controls the tolerance for error. The larger the value, the heavier the penalty for error, and the more the model tends to fit each sample. The smaller the value, the higher the tolerance, and the smoother the model. This represents the sum of all sample slack variables, i.e., the total empirical error; The constraints are set as follows: in: Indicates the first The true target value of each sample; This represents a nonlinear mapping function that maps samples in a low-dimensional input space. Mapped to a high-dimensional feature space; This represents the inner product operation in a high-dimensional space, i.e. Representative model for samples The predicted value; This represents the bandwidth of the insensitive loss band; if the difference between the predicted and actual values ​​is within... If the prediction is within the acceptable range, it is considered accurate and no loss is accounted for. Using the Lagrange multiplier method to construct the dual function, the final regression function expression is: in: This represents a Lagrange multiplier, with each sample corresponding to a pair of multipliers; This represents a weighted summation over all support vectors, with coefficients... This reflects the magnitude of the pull of the support vector on the final decision surface; This represents the Gaussian radial basis function kernel, replacing the inner product operation in high-dimensional space. It directly calculates sample similarity in a low-dimensional space: in: This represents the natural exponential function. These represent the kernel function coefficients, which control the influence range of a single training sample. The larger the value, the narrower the Gaussian distribution and the more complex the model. The smaller the value, the wider the distribution, and the smoother the model. This represents the squared Euclidean distance between two sample vectors.

[0014] In one possible implementation, optimizing the model parameters of the TTR-SVR model using the multidimensional monitoring sample data includes: Define hyperparameter space Construct a two-level optimization model to find the global optimal solution. The objective function is Minimizing the expected risk under cross-validation: in: This indicates the search for the independent variable that minimizes the objective function. Operation; Indicates the number of folds in cross-validation. Indicates to The results of the two verifications are summed and then divided by . To obtain average performance, This represents the mean squared error loss function, used to measure the difference between the model's predicted values ​​and the actual values; This indicates that after removing the first... The model trained on the training set after folding the data, index This indicates that the model uses a specific combination of hyperparameters, indicated by the superscript. Indicates that it does not include the first Partial data, Indicates the first Feature data of the validation set, Indicates the first The true label data of the validation set; Logarithmic hypercube space for random coarse search phase Monte Carlo sampling was conducted inside: in: Indicates the first The parameter combination obtained from the random sampling. Indicates a logarithmic uniform distribution. The probability measure is a log-uniform distribution, ensuring that the search probability of the parameter is equal on different orders of magnitude; In the coarse search for the optimal solution of - Build a dense mesh within the neighborhood A deterministic traversal search is performed to pinpoint the exact optimal combination of parameters, where: This represents the initial optimal solution found during the coarse search phase. Indicates the search radius or neighborhood size. This indicates that the optimal solution is the center of the coarse search. A locally subdivided mesh constructed within the specified range.

[0015] In one possible implementation, the optimal parameter combination is input into the TTR-SVR model to obtain the final TTR-SVR model, which is used to directly receive various types of lossless data from laying hens and to process the lossless data.

[0016] In this embodiment, an adaptive capping cleaning method based on interquartile range is employed to dynamically adjust outlier removal boundaries, avoiding the accidental deletion of valid physiological signals, improving the robustness of data cleaning, and overcoming the shortcomings of traditional fixed threshold and simple imputation methods that disrupt data distribution. A preprocessing pipeline using KNN topology imputation and Yeo-Johnson manifold transformation effectively repairs missing data and corrects skewed data distribution, resolving the problem of feature space metric failure. Adaptive feature selection based on the F-statistic accurately removes redundant features, achieves data dimensionality reduction, and avoids the curse of dimensionality and model overfitting. The constructed TTR-SVR model and parameter optimization strategy balance optimization accuracy and computational efficiency, solving the problem of unreasonable hyperparameter selection in traditional SVR, and comprehensively improving the accuracy and efficiency of non-destructive monitoring data processing for laying hens, providing reliable technical support for non-destructive testing of laying hen health and prediction of production performance. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a method for lossless processing of laying hens based on an improved SVR, provided in an embodiment of this application; Figure 2 Comparison of pectoral muscle tenderness prediction results of the TTR-SVR model provided in the embodiments of this application; Figure 3 A comparison chart showing the prediction results of the overall sensory preference of the pectoral muscles using the TTR-SVR model provided in the embodiments of this application. Detailed Implementation

[0018] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.

[0019] See Figure 1 The lossless data processing method for laying hens based on improved SVR provided in this embodiment includes: S101, acquire multidimensional monitoring data of laying hens, and clean the multidimensional monitoring data based on the adaptive capping of interquartile range to eliminate extreme outliers.

[0020] First, acquire multidimensional monitoring data of laying hens to construct the original observation space: in: Indicates the first The input feature vector of the nth sample; in a chicken farming scenario, it represents the nth... The set of multidimensional environmental parameters at the time of sampling. Indicates the first The target observation value for each sample Indicates the index subscript of the sample, with a value range from arrive . This represents the total number of samples, i.e., the number of rows of data contained in the dataset. It is a subset notation, representing a dataset. It is contained within the mathematical space defined on the right. express 3D real Euclidean space; This represents the total feature dimension, which is the number of features contained in each sample. It is the Cartesian product symbol, representing the combination of the input space and the output space; Represents a one-dimensional real space, corresponding to the objective value in scalar form. .

[0021] For any feature dimension Define its empirical cumulative distribution function: in, Indicates the first The empirical cumulative distribution function of a feature. It is a step function used to describe the probability distribution of the feature data on the numerical axis. Indicates the index of the feature dimension, with a range of values. Specifically refers to a column of features that is currently being processed; Let represent an arbitrary real-valued variable, which serves as the independent variable of the function and represents the possible values ​​of the characteristic. The summation symbol represents summation over all... Each sample is iterated and accumulated. This indicates an indicator function, whose logical definition is: the condition within the parentheses (i.e., If true, then the function value is If this is not true, then the function value is... ; Indicates the first The sample at the th The specific numerical value of the feature.

[0022] A quantile operator is defined based on the inverse cumulative distribution function to accurately extract the distribution feature points of the data. in: Indicates the first quantile function of each feature This represents the quantile level, with values ​​ranging from 0 to 1. The infimum symbol represents all conditions that satisfy the condition. The minimum value among the values.

[0023] Then construct the robust discreteness metric matrix. , its first Each component Defined as interquartile range, this index effectively characterizes the robustness of data distribution after excluding extreme values: in: This represents a vector or matrix composed of the dispersion indices of all features. Indicates the first The interquartile differences of each feature dimension. Indicates the first The third quartile (upper quartile) of a feature is the value that is located at the 75th percentile after being arranged from smallest to largest. Indicates the first The first quartile (lower quartile) of a feature is the value that is located at the 25th percentile after being arranged from smallest to largest.

[0024] Define the set of anomaly truncation thresholds The lower bound is defined as: The upper bound is defined as: in: The anomaly discrimination coefficient, as a hyperparameter for adjusting the cleaning intensity, determines the model's tolerance range for outliers.

[0025] Finally, a nonlinear capping mapping operator is constructed. Its component mapping rule is: in: This represents a two-sided truncation operator that forces outlier data that exceeds the normal range to be mapped onto the boundary manifold, thereby eliminating extreme noise interference while preserving the amount of data. This represents the input variable, i.e., the original feature values ​​before processing. This represents a minimum value function that takes the smaller of the two values ​​within the parentheses. This represents the maximum value function, which takes the larger of the two values ​​within the parentheses.

[0026] The above steps complete the initial dataset coarse cleaning, removing obvious outliers to obtain coarsely cleaned data.

[0027] S102, construct multidimensional monitoring data after preprocessing pipeline repair and cleaning based on KNN topology filling and Yeo-Johnson manifold transformation.

[0028] After obtaining the coarsely cleaned data, although outliers have been removed, there may still be missing values ​​within the data, and the data distribution may not follow a normal distribution. Since the SVR model is very sensitive to the integrity and distribution shape of the data, more in-depth data preprocessing is required.

[0029] For missing values ​​in the data, for missing sample vectors In a complete feature subspace Searching for its Euclidean metric The topological neighborhood formed by the nearest neighbor samples Define the weighted fill operator: in: Indicates the first The feature vectors of the samples to be filled Indicates the first The feature vectors of the neighboring samples Indicates the first The sample at the th Estimates of missing values ​​in each feature dimension; superscript : Refers to a specific feature dimension, subscript This refers to a sample. The missing part. This represents the observed feature subspace, i.e., the sample. and The set of all existing feature dimensions without any missing features is used to calculate distance based solely on these common features. Indicates sample of nearest neighbor set This represents a hyperparameter, indicating how many nearest neighbors to select (e.g., 5). This represents the kernel function, used to convert distance into weights; Indicates sample and In the characteristic subspace Euclidean distance on, Representing neighbor samples In the The authenticity of each characteristic.

[0030] Kernel function Using the reciprocal distance kernel form (Formula 9), this distance decay kernel function reflects the spatial correlation principle that "the closer the distance, the greater the influence": in: This represents the distance kernel function, with distance as input. The output is the weights. This represents the distance between two samples. It is the standard Euclidean distance calculation formula. Indicates traversing the characteristic subspace Each feature dimension in the dataset. It is a very small positive number, called the smoothing term, which is used to prevent calculation errors caused by the denominator being 0 when the distance between two samples is 0.

[0031] For the filled data matrix Maximum likelihood estimation is performed to improve the normality of the feature distribution. For each column of features... Finding the optimal parameters To maximize the log-likelihood function : Our goal is to find the function that maximizes this value. , These are the core parameters of the Yeo-Johnson transform, which determine the specific form of the transform (such as logarithmic transform, square root transform, reciprocal transform, etc.). This represents the optimal parameter, which is the best value obtained by maximizing the likelihood function. This refers to finding the parameters in the real number field that maximize the objective function. , Represents the total number of samples. This represents the transformed data value. This represents the mean of the transformed data. It is the variance of the transformed data, the first term of the formula. To make the variance of the transformed data as stable as possible. This represents the sign function, used to preserve the sign of the data before and after the transformation. When When it is 1, when When it is -1, when The time is 0.

[0032] Transformation basis functions Defined as a piecewise power function: In this embodiment, the obtained coarsely cleaned data is filled with holes, and then the Yeo-Johnson transform is used to transform the data into a standard normal distribution. High-quality standard data is output, thus laying the foundation for the smooth progress of subsequent steps.

[0033] S103, based on the adaptive feature percentage screening of the F statistic, redundant features are removed and data dimensionality is reduced to obtain multidimensional monitoring sample data.

[0034] After the data processing in step S102, the data quality is now very high, but it may still contain many useless redundant features. If all features are fed into the model, the calculation will be slow and it is easy to overfit. Therefore, it is necessary to remove redundant features.

[0035] In this embodiment, a feature importance evaluation operator is constructed to eliminate redundant features. .in, It is a key hyperparameter that represents the "percentage retained". Indicates that the input is 3D feature vectors The output is dimensional feature vectors, where ; The operator represents a mapping relationship, mapping a high-dimensional feature space to a low-dimensional feature space.

[0036] Calculate each feature vector With the target vector Univariate regression -Statistics Define the sum of squared residuals. Indicates the first The explanatory variance of each feature for the target is defined, and the regression sum of squares is also defined. Indicates unexplained variation: in: Represents the feature index, indicating the feature index for the first... Column feature calculation. Indicates only the first A univariate linear regression model is established for the first feature. The predicted value obtained from each sample. Represent the target variable The sample mean, This indicates the degree to which the calculated predicted value deviates from the mean; the larger the value, the stronger the model's explanatory power. Indicates the first The true target value of each sample This represents the sum of squares of the prediction errors; the smaller the value, the better the model fit.

[0037] The construction follows the degree of freedom of The statistical test of the F-distribution Used to examine features With target variable Is there a significant linear relationship between them? in: This represents the regression sum of squares divided by its degrees of freedom. Since the regression degrees of freedom are 1, it represents the average regression sum of squares because it is a univariate regression. This represents the sum of squared residuals divided by its degrees of freedom, where the degrees of freedom of the residuals are... ,in That is the sample size, which represents the average sum of squared residuals. This indicates that the statistic follows a set of degrees of freedom. of -Distribution. If the calculated The value is very large, indicating that the variance explained by the regression is much greater than the variance of the random error, and the feature... Significantly effective.

[0038] Then, based on the percentage hyperparameter of the preserved features Define feature selection indicator function When this value is 1, it indicates that the feature is preserved. A value of 0 indicates that the value should be removed. in: express - The quantile function of the statistic gives the value in - The critical value corresponding to a specific cumulative probability under a distribution. This is the formula for calculating the threshold of quantiles. If... This indicates that the importance of this feature is ranked first. ,reserve( Otherwise, it indicates that the feature is of low importance and should be removed. ).

[0039] The final output is the filtered feature matrix. This achieves effective dimensionality reduction.

[0040] in: It is a set construction symbol, representing a set consisting of all column vectors that satisfy the condition. Represents the first element in the original data matrix. Column feature vectors; This is a filter condition; it only keeps columns where the indicator function equals 1. arrive This indicates that all original feature columns will be traversed. The final matrix will have fewer columns. List.

[0041] Finally, after passing the F test in step S103, the most critical features have been selected by percentage, thereby achieving noise removal and outputting a high signal-to-noise ratio feature set.

[0042] S104, construct the target transformation support vector regression (TTR-SVR) model, and use the multidimensional monitoring sample data to optimize the model parameters of the TTR-SVR model to determine the final TTR-SVR model.

[0043] After obtaining the high signal-to-noise ratio feature set, the TTR-SVR model is established. Specifically, in the reproducing kernel Hilbert space... By constructing a regression hyperplane, the original problem is formulated as a convex quadratic programming problem with slack variables: in: This represents the objective function that needs to be minimized, which consists of two parts: structural risk (model complexity) and empirical risk (training error). The normal vector of the regression hyperplane determines the direction of the hyperplane. The intercept of the regression hyperplane determines the position of the hyperplane; Represents the slack variable vector. Measure the first The degree of error of a sample exceeding the upper bound. It measures the degree of error that exceeds the lower bound. Represents the regularization term, and represents the weight vector. In the regenerating nucleus Hilbert space The squared norm of the model. Minimizing this term makes the model smoother and prevents overfitting (minimizing structural risk). The penalty coefficient is a hyperparameter that controls the tolerance for error. The larger the value, the heavier the penalty for error, and the more the model tends to fit each sample. The smaller the value, the higher the tolerance and the smoother the model. The sum of all sample slack variables, i.e., the total empirical error.

[0044] The constraints are set as follows: in: Indicates the first The true target value of each sample This represents a nonlinear mapping function that maps samples in a low-dimensional input space. Mapped to a high-dimensional feature space. This represents the inner product operation in a high-dimensional space, i.e. , representing the model's performance on samples The predicted value. This represents the bandwidth of the insensitive loss band; if the difference between the predicted and actual values ​​is within... If the prediction is within the acceptable range, it is considered accurate and no loss is incurred. This means that the difference between the actual value and the predicted value should not exceed [a certain value]. Add relaxation (upper bound constraint). This means that the predicted value minus the actual value should not exceed [a certain value]. Add relaxation (lower bound constraint). This indicates that slack variables must be non-negative.

[0045] Using the Lagrange multiplier method to construct the dual function, the final regression function expression is: in: Represents a Lagrange multiplier, where each sample corresponds to a pair of multipliers, and the pair is valid only if the sample falls within the range of Lagrange multipliers. - When outside the band or on the boundary, the corresponding or Talent is not zero; This means that a sample cannot violate both the upper and lower bound constraints at the same time. This represents a weighted summation over all support vectors, with coefficients... This reflects the magnitude of the pull of the support vector on the final decision surface. This represents the Gaussian radial basis function kernel, replacing the inner product operation in high-dimensional space. It directly calculates sample similarity in a low-dimensional space: in: This represents the natural exponential function. These represent the kernel function coefficients, which control the influence range of a single training sample. The larger the value, the narrower the Gaussian distribution and the more complex the model. The smaller the value, the wider the distribution, and the smoother the model. This represents the squared Euclidean distance between two sample vectors. The closer the distance, the closer the kernel function value is to 1; the farther the distance, the closer the kernel function value is to 0.

[0046] The TTR-SVR model was successfully constructed, and the kernel function was used to successfully map the data to a high-dimensional space to find the optimal regression hyperplane. Although the TTR-SVR model has been established, the existing hyperparameter combination is not optimal and is insufficient to bring the model to its best state. It may even prevent the model from running. Furthermore, the feature selection ratio ρ itself is a parameter that needs to be dynamically adjusted.

[0047] To obtain the optimal combination of hyperparameters, this embodiment defines a hyperparameter space. Construct a two-level optimization model to find the global optimal solution. The objective function is Minimizing the expected risk under -fold cross-validation aims to find a set of parameters This makes the model in The average prediction error is minimized using folded cross-validation. in: This indicates the search for the independent variable that minimizes the objective function. The operation, Indicates the number of folds in cross-validation. Indicates to The results of the two verifications are summed and then divided by . The average performance was obtained. This represents the mean squared error loss function, used to measure the difference between the model's predicted values ​​and the actual values. This indicates that after removing the first... The model trained on the training set after folding the data, index This indicates that the model uses a specific combination of hyperparameters, indicated by the superscript. Indicates that it does not include the first Partial data. Indicates the first Feature data of the validation set; Indicates the first The true label data of the validation set.

[0048] Random coarse search phase: in the logarithmic hypercube space Monte Carlo sampling was conducted inside: in: Indicates the first The parameter combination obtained from the random sampling. Indicates a logarithmic uniform distribution. It is a log-uniform distribution probability measure, which ensures that the search probability of the parameter is equal on different orders of magnitude.

[0049] Grid search phase: In the initial coarse search for the optimal solution of - Build a dense mesh within the neighborhood A deterministic traversal search is performed to pinpoint the precise optimal combination of parameters. This represents the initial optimal solution found during the coarse search phase. Indicates the search radius or neighborhood size. This indicates that the optimal solution is the center of the coarse search. A locally subdivided mesh constructed within the specified range.

[0050] By employing a coarse-search + fine-search strategy, the optimal parameter combination that minimizes prediction error is automatically found, thus laying the foundation for the model to reach its optimal state. Finally, the obtained hyperparameters are input into the model output to obtain the final TTR-SVR model. The TTR-SVR model obtained based on the optimal parameter combination can directly accept various types of non-destructive data from laying hens, realizing non-destructive data processing and used for tasks such as non-destructive target detection, attribute discrimination, and state classification.

[0051] To verify the non-destructive data processing method for laying hens based on the improved SVR provided in the above embodiments, this embodiment further adopts the above method to predict the tenderness of the breast muscles and the overall sensory preference of the breast muscles of laying hens using B-mode ultrasound information.

[0052] Ultrasonic data collection for live dual-purpose (laying and meat) chickens during their laying period: A total of 284 dual-purpose chickens (Jishan Black Chicken II strain) at different stages of their laying period (26, 29, 33, 37, 38, 40, 41, 43, 45, 46, and 48 weeks of age) from the same commercial farm were collected. Feeding of the chickens was stopped 12 hours prior to measurement. Before measurement, the breast feathers of the chickens were removed using an animal defecation device.

[0053] Turn on the portable B-mode ultrasound scanner, connect the scanner to the 5.0 cm linear array probe, apply ultrasound coupling agent to the surface of the probe's sensing part, and ensure the probe's sensing part is in full contact with the chicken's body at the following location: two-fifths of the length of the fossilized bone, and 1.3 cm lateral to the right edge of the fossilized bone. Position the probe tip towards the chicken's head, perpendicular to the surface of the chicken's breast, and then slowly adjust the probe until a clear bright white line appears in the captured ultrasound image. Lock the image, measure the thickness of the flocculent tissue layer below the top white line in the image, and record this as the breast fat thickness; measure the distance from the bottom of the flocculent tissue layer to the next bright white line, and record this as the breast muscle thickness.

[0054] Information collection on meat quality characteristics of dual-purpose laying hens: After ultrasound scanning, the chickens rested for 30 minutes before slaughter. Slaughter was performed using the bleeding method, severing the carotid artery and jugular vein in the neck with a sharp knife for approximately 5 minutes. The chickens were then placed in an automatic defecation machine containing 60°C hot water for defecation. Next, the chickens were immersed in 0°C ice water for 1 hour. Subsequently, the pectoral muscles were dissected by cutting along the bony ridges, clavicle, and shoulder bones, as well as the attachment lines of these muscles to the ribs. The hip and tarsal joints were separated, and the pistol leg was dissected. The pectoral muscles and pistol leg were each packaged separately in vacuum-sealed bags and stored in a 4°C cold storage for 24 hours before proceeding with subsequent steps.

[0055] A whole piece of pectoral muscle was placed in a vacuum-sealed bag and heated in an 80°C water bath until the center temperature of the pectoral muscle reached 72°C. The sample was then removed, cooled to room temperature, and stored in a 4°C cold storage for 24 hours. After the cold storage period, the sample was cut into uniform sizes (2cm×1cm×1cm) along the direction parallel to the muscle fibers and placed in a texture analyzer to measure the shear force.

[0056] The evaluation panel consisted of 15 trained meat researchers (8 men and 7 women) with extensive experience in meat research. Their ages ranged from 20 to 35. During the sensory analysis, evaluators were required to remain neither hungry nor mentally fatigued. Smoking, chewing gum, and consuming strongly scented foods were prohibited within one hour prior to sample analysis. All sample evaluations were conducted in a standard sensory evaluation room, ensuring adequate lighting, good ventilation, and a noise-free environment. Evaluators were seated at appropriate intervals within the evaluation room, maintaining silence and refraining from discussion during the evaluation process.

[0057] The pectoral muscles were placed in vacuum-sealed bags and heated in an 80°C water bath until the center temperature reached 72°C. The samples were then removed and placed in a 60°C water bath for 30 minutes. Immediately afterward, the pectoral muscles were divided into eight equal parts and presented to the panel for evaluation. All samples were coded with a three-digit random number and presented on white porcelain plates. Evaluators assigned an overall sensory preference score to each sample using a 15-point scale (1 - extremely unacceptable, 3 - very unacceptable, 5 - moderately unacceptable, 7 - slightly unacceptable, 9 - slightly acceptable, 11 - moderately acceptable, 13 - very acceptable, 15 - extremely acceptable). After evaluating each sample, evaluators used water and unseasoned soda crackers to eliminate any interfering tastes, rested for 20 seconds, and then evaluated the next sample. Finally, the average score for each sample was recorded as the overall sensory preference score.

[0058] First, construct the original observation space. Set the total number of samples. For each laying hen, its pectoral muscle thickness (in mm) and subcutaneous fat thickness (in mm) are used as input feature vectors. Simultaneously, the corresponding pectoral muscle tenderness (unit: N) or overall sensory preference score of the pectoral muscles will be used as the target observation value. Therefore, a dataset is constructed. ,in This indicates that the feature dimension is 2.

[0059] For the characteristic dimension of pectoral muscle thickness (let's call it...) First, its empirical cumulative distribution function (ECDF) is calculated to describe the overall distribution of the data. Specifically, for any given thickness value... ,calculate When the first The thickness of the pectoral muscles of a chicken is less than or equal to The value is 1 if the cumulative distribution function is active, and 0 otherwise. Based on this cumulative distribution function, a quantile operator is defined using inverse operations. This allows for the precise extraction of key distribution nodes in the pectoral muscle thickness data.

[0060] Subsequently, the robust dispersion measure matrix of the pectoral muscle thickness data was calculated using the quantile operator. The first component The specific calculation process is as follows: First, determine the lower quartile of the data. (i.e., the 25th percentile) and the upper quartile (i.e., the 75th percentile), then calculate the difference between the two to obtain the interquartile range. .

[0061] After obtaining the interquartile range, in order to define the fluctuation range of normal data, a set of abnormal cutoff thresholds for pectoral muscle thickness is defined. Set the anomaly detection coefficient. This hyperparameter determines the tolerance for outliers. The lower bound threshold is calculated as follows: The upper bound threshold is calculated as follows: Similarly, regarding the characteristics of subcutaneous fat thickness in the chest ( Repeat the above process to calculate the corresponding lower bound. and the Upper Realm .

[0062] Finally, the nonlinear capping mapping operator is applied. The feature data of all 284 samples were cleaned. For the first... Observational value of pectoral muscle thickness of laying hens Its value after cleaning According to the formula Calculations are performed. Specifically, if the measured thickness of the pectoral muscle of a laying hen is below the lower limit... Then force it to be corrected to If it is higher than the upper limit Then it is corrected to If it falls between the two, then the original value remains unchanged.

[0063] The system randomly divides the cleaned dataset into training and testing sets. In this embodiment, to ensure the reproducibility of experimental results, a fixed random seed of 42 is used for the division process, and the division ratio is set as follows: 80% of the data is used for model training, and 20% of the data is used for model performance evaluation.

[0064] A preprocessing pipeline based on KNN topology imputation and Yeo-Johnson manifold transform is constructed. First, KNN topology imputation is performed. For some laying hen samples where data on pectoral muscle thickness or subcutaneous fat thickness is missing due to sensor malfunction or human error, interpolation is performed to repair the missing data by finding similar samples. Let the... Feature vector of a laying hen If missing values ​​exist (e.g., missing pectoral muscle thickness), the feature subspace is first constructed from the remaining complete samples without missing values. In this study, the Euclidean distance metric was used to find the chicken most similar to the characteristics of the laying hen. Each laying hen constitutes its topological neighborhood set. .

[0065] After determining the neighborhood samples, a distance decay mechanism is introduced to more accurately estimate the missing values, and the distance decay mechanism is calculated. Only one laying hen and the first in the neighboring area Only one laying hen ( Kernel function weights between ( ). Kernel function Using the reciprocal distance kernel form, i.e. The terms within the summation symbol represent the sum of squared differences between the two laying hens along their shared feature dimensions. To prevent tiny positive numbers with a denominator of zero (take...) ).

[0066] Subsequently, the weighted average method was used to calculate the first... The first laying hen Filled values ​​for dimensional features (such as pectoral muscle thickness) The calculation formula is: Through this process, by utilizing the local topological information of a group of 284 laying hens, accurate and unbiased repair of missing data was achieved.

[0067] Then, the Yeo-Johnson manifold transformation was applied to improve the normality of the data distributions for pectoral muscle thickness, subcutaneous fat thickness, and the target variable (pectoral muscle tenderness or sensory preference). This transformation used maximum likelihood estimation (MLE) to find the optimal transformation parameters, making the transformed data distribution as close as possible to a Gaussian distribution. For the imputed data matrix... Each column of features (e.g., the column vector of pectoral muscle thickness), constructing a transformation parameter log-likelihood function Its expression is The first term of the likelihood function measures the variance stability of the transformed data, while the second term corrects for the impact of the transformation on the probability density using the Jacobian determinant.

[0068] The optimal transformation parameters for this feature dimension are obtained by finding the maximum value of the log-likelihood function using an optimization algorithm. Based on the calculation Value, using piecewise power functions A non-linear mapping is applied to the feature values ​​of each sample. The specific transformation rule is as follows: when the feature value... and hour, ;when and hour, ;when and hour, ;when hour, .

[0069] Adaptive Feature Percentage Screening Based on F-Statistic: Constructing a Feature Importance Evaluation Operator From the thickness of the pectoral muscles, the thickness of subcutaneous fat, etc. From the candidate features, a subset of key features that best explains the target variable (such as the shear force value of pectoral muscle tenderness) is selected. First, for each feature vector... (For example, the first) One feature is pectoral muscle thickness), which is then compared with the target vector. A univariate linear regression model is used to calculate the sum of squared residuals of the regression model, relating factors such as (e.g., pectoral muscle tenderness). With regression sum of squares ,in This section explains the variation in pectoral muscle tenderness due to the characteristics of pectoral muscle thickness. For the first The predicted tenderness value for a single laying hen based on this feature. The mean of all tenderness observations; and This indicates the remaining variation that could not be explained by this feature.

[0070] Subsequently, a statistical test metric was constructed based on the above sum of squares. This is used to quantify the significance of the linear correlation between the feature and the target variable. The calculation formula is as follows: The numerator has 1 degree of freedom (corresponding to univariate regression), and the denominator has 284 - 2 = 282 degrees of freedom. This statistic... Obeying the degree of freedom The F-distribution, i.e. If a certain feature is calculated to obtain The larger the value, the stronger its ability to explain the variation of the target variable, meaning the more important this feature is for predicting pectoral muscle tenderness; conversely, if... A smaller value means that the feature may be noise or an irrelevant variable.

[0071] To achieve adaptive filtering, a hyperparameter is set to retain a percentage of features. Based on this hyperparameter, calculate the quantile threshold of the set of all feature F statistics. Define the feature selection indicator function. When the first Each feature satisfies the condition hour, , indicates that the feature is retained; otherwise This indicates elimination. Through this filtering mechanism, the final output is a feature matrix optimized by dimensionality reduction. .

[0072] In Regenerative Nuclear Hilbert Space (RKHS) A TTR-SVR model is constructed. First, the original optimization problem is defined as a convex quadratic programming problem with slack variables, and its objective function is: The first item The first term represents the squared norm of the weight vector in the Hilbert space, used to control the complexity of the model to prevent overfitting, reflecting the principle of minimizing structural risk; the second term is a penalty term, where... This is the penalty coefficient, used to balance the relationship between model complexity and training error; and For the introduced slack variables, respectively allow the first The predicted value for the tenderness of the breast muscle of a single laying hen deviated from the actual value. - Insensitive band upper and lower bounds, thereby enhancing the model's robustness to noisy data.

[0073] To solve the above optimization problem, a series of constraints must be satisfied: for each sample of laying hens ( Its prediction error must be limited to an allowable range, i.e. and At the same time, ensure that the slack variables are non-negative. By introducing Lagrange multipliers and The original constrained optimization problem is transformed into a dual problem for solution. Using the KKT conditions, the final analytical expression of the regression function is: .

[0074] To effectively handle the complex nonlinear relationship between pectoral muscle thickness and tenderness, the kernel function... The Gaussian radial basis function kernel is selected, and its expression is: This kernel function uses parameters... By controlling the influence range of a single sample, the nonlinear relationship in the original feature space can be mapped to a linear relationship in a high-dimensional space, thereby achieving high-precision prediction of the tenderness of the breast muscle and sensory preference of laying hens. Finally, through the inverse transformation of the target transformer, the standardized predicted values ​​output by the model are restored to the real physical dimensions (such as shear force N), completing the entire prediction process.

[0075] By regularization coefficient Radial basis kernel function parameters Insensitive loss band width and feature filtering retention percentage Establish hyperparameter space To find the global optimal solution First, define the objective function as follows: - Minimize the expected risk under cross-validation, i.e. In this embodiment, The value is set to 5, indicating the use of 5-fold cross-validation. Specifically, the 284 laying hen samples are randomly divided into 5 non-overlapping subsets. Each time, 4 subsets are selected as the training set, and the remaining subset is used as the validation set. For each set of candidate hyperparameters... Train the model on the training set. Then in the validation set Calculate the mean square error above Finally, the average of the five verification errors was taken as the performance evaluation index for this set of parameters.

[0076] Setting the search space for parameters Penalty coefficient The sampling range is kernel function coefficients The sampling range is Insensitivity tolerance The sampling range is For discrete parameters Then in the set Uniform sampling is used. The result is a logarithmic hypercube, and for each hyperparameter... Monte Carlo sampling is performed within a preset range, i.e. ,in This is a logarithmic uniform distribution probability measure. In this embodiment, a total of... After several random samplings, the cross-validation error corresponding to each set of sampled parameters is calculated. Finally, the best-performing set of parameters is selected as the optimal solution for the coarse search. .

[0077] Perform a grid search to find the optimal solution obtained from the coarse search. Centered on, in its - Build a dense mesh within the neighborhood The specific operation is as follows: for each continuous hyperparameter ( ), taking 0.5 times, 1.0 times, and 2.0 times the optimal value of the coarse search as new candidate points, and constructing a high-resolution search grid. For discrete hyperparameters If the feature retention percentage is calculated, the optimal percentage obtained from the coarse search is directly determined. Subsequently, a deterministic traversal search is performed on each node in the grid, and the model performance is evaluated again using 5-fold cross-validation. Finally, the validation errors of all grid points are compared, and the set of parameters with the smallest error is selected as the globally optimal parameter combination. = .

[0078] Finally, the system selects the set of parameters with the smallest verification error as the globally optimal hyperparameter configuration (Table 1) and embeds it into the model configuration for subsequent full training.

[0079] Table 1. Optimal hyperparameter configurations for each prediction model determined by the hierarchical optimization strategy. After reconstructing the model using globally optimal parameters, the system calculates the following metrics on the test set to quantify model accuracy: Coefficient of determination: Root mean square error: Mean absolute error: Mean absolute percentage error: In addition, the system performs a generalization stability test procedure: on the full training data. Cross-validation (taking) ), calculate on each fold subset Standard deviation of scores : like If the value is less than the preset threshold, the system determines that the model has been successfully built and saves the model file; otherwise, it prompts a risk of generalization.

[0080] The final prediction results are as follows Figure 2 and Figure 3 This visually demonstrates the improvement of the SVR model in the present invention regarding the tenderness of the breast muscle in laying hens. Figure 2 ) and overall sensory preference ( Figure 3 The model's predictive performance on two key quality indicators is shown in the graph. The solid black line represents the actual values ​​measured in the laboratory, while the dashed red line represents the model's predicted output. The horizontal and vertical axes correspond to the test sample number and quality value, respectively. As can be seen from the curve trends in both graphs, the red predicted curve closely follows the peaks and troughs of the black actual curve, maintaining a high degree of consistency in both overall trend and local fluctuations, without significant lag or deviation. Quantitative indicators show that the model's determination coefficient for predicting breast muscle tenderness is as high as 0.864, and its prediction coefficient for the more subjective sensory preference is also 0.813. This fully verifies the model's superior fitting accuracy and robustness in handling complex nonlinear regression problems, demonstrating its ability to effectively utilize ultrasound image features to achieve accurate and non-destructive inversion of egg-laying chicken meat quality.

[0081] To fully verify the effectiveness of the improved support vector regression (SVR) prediction model proposed in this embodiment in the detection of chicken meat quality in live dual-purpose egg-laying chickens, this experiment first quantitatively evaluated the prediction accuracy of the model on the independent test set, and the relevant indicators are summarized in Table 2.

[0082] Table 2. Evaluation results of the predictive performance of the improved SVR model for different meat quality indicators. As can be seen from the statistical data in Table 2, the model demonstrated excellent fitting ability for both key meat quality indicators. Specifically, for the physical indicator of "breast muscle tenderness," the model's coefficient of determination (COP) was [missing information]. The model achieved a mean square error (RMSE) of 0.8634, indicating that it can explain over 86% of the tenderness variability. Simultaneously, its root mean square error (RMSE) was 2.7496, and its mean absolute percentage error (MAPE) was only 6.28%, demonstrating a high degree of agreement between the predicted and actual values, with errors kept within a very small range. For the more complex subjective indicator of "overall sensory preference for pectoral muscles," the model also maintained a high level of predictive ability. The model achieved a mean value of 0.8125 and a mean amplitude of MAPE as low as 3.85%, demonstrating that it can not only accurately capture objective physical characteristics but also effectively establish a nonlinear mapping relationship between ultrasonic signals and subjective sensory evaluations.

[0083] After confirming the model's prediction accuracy, in order to further test the model's robustness and generalization ability when facing different data distributions, this experiment conducted a 5-fold cross-validation test on the full dataset, and the results are shown in Table 3.

[0084] Table 3. Generalization stability test results of the model based on 5-fold cross-validation Statistical analysis of the results from five independent experiments revealed a high degree of consistency in the model's performance across different data subsets. For the pectoral muscle tenderness prediction task, the average value of its 5-fold cross-validation was [data missing]. The average value was 0.8504, very close to the result of the independent test set (0.8634), and the standard deviation was only 0.0453; for the prediction of overall sensory preference, the average... The standard deviation was 0.8067 and the value was 0.0529. The extremely low standard deviation strongly demonstrates that the model has not experienced overfitting or underfitting, exhibiting excellent generalization stability. It can maintain the reliability and credibility of prediction results in live testing of different batches or individuals, fully meeting the technical requirements for rapid, non-destructive, and continuous monitoring of meat quality in actual production.

[0085] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0086] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A lossless data processing method for laying hens based on improved SVR, characterized in that, include: Multidimensional monitoring data of laying hens is acquired, and the multidimensional monitoring data is cleaned based on the adaptive capping of the interquartile range to eliminate extreme outliers. Construct multidimensional monitoring data after preprocessing pipeline repair and cleaning based on KNN topology filling and Yeo-Johnson manifold transformation; The adaptive feature percentage screening based on the F statistic is used to remove redundant features and reduce the dimensionality of the repaired multidimensional monitoring data to obtain multidimensional monitoring sample data. A target transformation support vector regression (TTR-SVR) model is constructed, and the model parameters are optimized using the multidimensional monitoring sample data to determine the final TTR-SVR model.

2. The method for lossless data processing of laying hens based on improved SVR according to claim 1, characterized in that, The process of acquiring multidimensional monitoring data of laying hens and cleaning the multidimensional monitoring data based on an adaptive capping method using interquartile range to eliminate extreme outliers includes: The original observation space was constructed based on the multidimensional monitoring data of the laying hens: in: Indicates the first The input feature vector of each sample, Indicates the first The target observation value for each sample Indicates the total number of samples; express 3D feature vectors This represents the total feature dimension, i.e., the number of features contained in each sample. It is the Cartesian product symbol, representing the combination of the input space and the output space; Represents a one-dimensional real space, corresponding to the target observation in scalar form. ; For any feature dimension Define its empirical cumulative distribution function: in: Indicates the first The empirical cumulative distribution function of the dimensional feature, Index representing the feature dimension Represents an arbitrary real-valued variable. Indicates all Iterate through and accumulate each sample. This represents an indicator function; if true, the function value is... If this is not true, then the function value is ; Indicates the first The sample at the th The original feature values ​​on the dimensional feature; Based on the inverse cumulative distribution function, a quantile operator is defined to accurately extract the distribution feature points of the data: in: Indicates the first quantile function of each feature This represents the quantile value, which ranges from 0 to 1. The infimum symbol represents all conditions that satisfy the condition. The minimum value among the values; Constructing a robust discreteness metric matrix , its first Each component Defined as interquartile range, it is used to characterize the robustness of data distribution after excluding extreme values ​​at both ends; Define the set of anomaly truncation thresholds ,in: This is the upper bound of the set of abnormal truncation thresholds. This is the lower bound of the set of abnormal truncation thresholds; Finally, a nonlinear capping mapping operator is constructed. The initial multidimensional monitoring data was coarsely cleaned to remove obvious outliers. The component mapping rules are as follows: in: This represents a two-sided truncation operator that forces outlier data that exceeds the normal range to be mapped onto the boundary manifold, thereby eliminating extreme noise interference while preserving the amount of data. This represents a minimum value function that takes the smaller of the two values ​​within the parentheses. This represents the maximum value function, which takes the larger of the two values ​​within the parentheses.

3. The method for lossless data processing of laying hens based on improved SVR according to claim 1 or 2, characterized in that, The construction of multidimensional monitoring data after preprocessing pipeline repair and cleaning based on KNN topology filling and Yeo-Johnson manifold transformation includes: To address the missing values ​​in the cleaned multidimensional monitoring data, for each missing sample vector, a Euclidean metric-based search is conducted within the complete feature subspace. Topological filling is performed on the topological neighborhood formed by the nearest neighbor samples; For the filled data matrix Perform maximum likelihood estimation transformation to improve the normality of the feature distribution. For each column of features... To find the optimal parameters.

4. The method for lossless data processing of laying hens based on improved SVR according to claim 3, characterized in that, For missing sample vectors, the method involves searching for their Euclidean metric-based values ​​in the complete feature subspace. Topological filling is performed on the topological neighborhood formed by the nearest neighbor samples, including: Define the weighted fill operator: in: Indicates the first The feature vectors of the samples to be filled Indicates the first The feature vectors of the neighboring samples Indicates the first The sample to be filled in the first Estimates of missing values ​​in each feature dimension; Indicates the sample to be filled The missing part; This represents the observed feature subspace, i.e., the sample. and A set of all existing feature dimensions without any missing features; Indicates sample of Nearest neighbor set This represents a hyperparameter, indicating how many nearest neighbors to select; This represents the kernel function, used to convert distance into weights; Indicates sample and exist Euclidean distance on; Representing neighbor samples In the The true value of each feature; Kernel function Using the reciprocal distance kernel form: in, This represents the distance kernel function, with distance as input. The output is the weights. Represents the distance between two samples; It is the standard Euclidean distance calculation formula. Indicates traversal Each feature dimension in, It is a very small positive number that prevents the denominator from being zero.

5. The method for lossless data processing of laying hens based on improved SVR according to claim 4, characterized in that, The maximum likelihood estimation transformation performed on the imputed data matrix improves the normality of the feature distribution. For each column of features... Finding the optimal parameters includes: in: Indicates the optimal parameters. These are the core parameters of the Yeo-Johnson transform, which determine the specific form of the transform. This refers to finding the parameters in the real number field that maximize the objective function. ; This represents the transformed data value. This represents the mean of the transformed data. It is the variance of the transformed data; The sign function is used to preserve the sign of the data before and after the transformation. Transformation basis functions Defined as a piecewise power function: 。 6. The method for lossless data processing of laying hens based on improved SVR according to claim 2, characterized in that, The adaptive feature percentage screening based on the F-statistic removes redundant features and reduces the dimensionality of the repaired multidimensional monitoring data to obtain multidimensional monitoring sample data. include: Constructing a feature importance evaluation operator ,in, Indicates the percentage to be retained. The output is 3D feature vectors This indicates that the mapping relationship maps the high-dimensional feature space to the low-dimensional feature space; Calculate each feature vector With the target vector Univariate regression -Statistics ; Then based on Define feature selection indicator function When this value is 1, it indicates that the feature is preserved. A value of 0 indicates that the value should be removed. in: express -Quantile function of the statistic This is the formula for calculating the threshold of quantiles; if If the value is positive, the feature is considered important and should be retained; otherwise, the feature is considered less important and should be removed. The final output is the filtered feature matrix. Achieving data dimensionality reduction: in, It is a set construction symbol, representing a set consisting of all column vectors that satisfy the condition; Represents the first element in the original data matrix. Column feature vectors; This is a filter condition; it only keeps columns where the indicator function equals 1. arrive This means that by traversing all the original feature columns, the number of columns in the final matrix will be reduced to [number missing]. List.

7. The method for lossless data processing of laying hens based on improved SVR according to claim 6, characterized in that, The calculation of each feature vector With the target vector Univariate regression -Statistics ,include: Define the sum of squared residuals Indicates the first The explanatory variance of each feature for the target is defined, and the regression sum of squares is also defined. Indicates unexplained variation: in: Represents the feature index, indicating the feature index for the first... Column feature calculation; Indicates only the first A univariate linear regression model is established for the first feature. The predicted values ​​obtained from each sample; Represent the target variable The sample mean; This indicates the degree to which the calculated predicted value deviates from the mean; the larger the value, the stronger the model's explanatory power. Indicates the first The true target value of each sample; This represents the sum of squares of the prediction errors; the smaller the value, the better the model fit. The construction follows the degree of freedom of The statistical test of the F-distribution Used to examine features With target variable Is there a significant linear relationship between them? in: The sum of squares divided by its degrees of freedom represents the average sum of squares of regression. The sum of squared residuals divided by its degrees of freedom represents the average sum of squared residuals. : indicates that the statistic follows a set of degrees of freedom. of -distributed.

8. The method for lossless data processing of laying hens based on improved SVR according to claim 1, characterized in that, The construction of the target transformation support vector regression (TTR-SVR) model includes: In the regenerating nucleus Hilbert space Constructing a regression hyperplane: in: This represents the objective function that needs to be minimized. The normal vector of the regression hyperplane determines the direction of the hyperplane; The intercept of the regression hyperplane determines the position of the hyperplane; Represents the slack variable vector. Measure the first The degree of error of a sample exceeding the upper bound. Measure the degree of error exceeding the lower bound; Represents the regularization term, and represents the weight vector. In the regenerating nucleus Hilbert space The square of the norm; This represents the penalty coefficient, which controls the tolerance for error. The larger the value, the heavier the penalty for error, and the more the model tends to fit each sample. The smaller the value, the higher the tolerance, and the smoother the model. This represents the sum of all sample slack variables, i.e., the total empirical error; The constraints are set as follows: in: Indicates the first The true target value of each sample; This represents a nonlinear mapping function that maps samples in a low-dimensional input space. Mapped to a high-dimensional feature space; This represents the inner product operation in a high-dimensional space, i.e. Representative model for samples The predicted value; This represents the bandwidth of the insensitive loss band; if the difference between the predicted and actual values ​​is within... If the prediction is within the acceptable range, it is considered accurate and no loss is accounted for. Using the Lagrange multiplier method to construct the dual function, the final regression function expression is: in: This represents a Lagrange multiplier, with each sample corresponding to a pair of multipliers; This represents a weighted summation over all support vectors, with coefficients... This reflects the magnitude of the pull of the support vector on the final decision surface; This represents the Gaussian radial basis function kernel, replacing the inner product operation in high-dimensional space. It directly calculates sample similarity in a low-dimensional space: in: This represents the natural exponential function. These represent the kernel function coefficients, which control the influence range of a single training sample. The larger the value, the narrower the Gaussian distribution and the more complex the model. The smaller the value, the wider the distribution, and the smoother the model. This represents the squared Euclidean distance between two sample vectors.

9. The method for lossless data processing of laying hens based on improved SVR according to claim 8, characterized in that, The optimization of model parameters for the TTR-SVR model using the multidimensional monitoring sample data includes: Define hyperparameter space Construct a two-level optimization model to find the global optimal solution. The objective function is Minimizing the expected risk under cross-validation: in: This indicates the search for the independent variable that minimizes the objective function. Operation; Indicates the number of folds in cross-validation. Indicates to The results of the two verifications are summed and then divided by . To obtain average performance, This represents the mean squared error loss function, used to measure the difference between the model's predicted values ​​and the actual values; This indicates that after removing the first... The model trained on the training set after folding the data, index This indicates that the model uses a specific combination of hyperparameters, indicated by the superscript. Indicates that it does not include the first Partial data, Indicates the first Feature data of the validation set, Indicates the first The true label data of the validation set; Logarithmic hypercube space during random coarse search phase Monte Carlo sampling was conducted inside: in: Indicates the first The parameter combination obtained from the random sampling. Indicates a logarithmic uniform distribution. The probability measure is a log-uniform distribution, ensuring that the search probability of the parameter is equal on different orders of magnitude; In the coarse search for the optimal solution of - Build a dense mesh within the neighborhood A deterministic traversal search is performed to pinpoint the exact optimal combination of parameters, where: This represents the initial optimal solution found during the coarse search phase. Indicates the search radius or neighborhood size. This indicates that the optimal solution is the center of the coarse search. A locally subdivided mesh constructed within the specified range.

10. The method for lossless data processing of laying hens based on improved SVR according to claim 9, characterized in that, The optimal parameter combination is input into the TTR-SVR model to obtain the final TTR-SVR model, which is used to directly receive various types of lossless data from laying hens and to process the lossless data.

Images