Air-jet vortex spun yarn quality prediction model based on M estimation optimization width learning

Abstract

The invention relates to the technical field of vortex spinning yarn, in particular to an air-jet vortex spinning yarn quality prediction model based on M estimation optimization width learning, which comprises the following steps: data acquisition and preprocessing: acquiring process parameter data and corresponding yarn quality index data in an air-jet vortex spinning production process; performing normalization processing on the collected data and dividing the data into a training set and a test set, constructing a robust width learning model R-M-BLS based on M estimation optimization, initializing network structure parameters, generating feature layer nodes and enhancement layer nodes, and performing calculation through an optimal output weight matrix to obtain a predicted value of a resultant yarn quality index. A width learning system is optimized through an M estimator, a Tukey function is adopted as a loss function to replace traditional L2 norm square loss, outlier interference is effectively inhibited through dynamic distribution of sample weights, the problem of insufficient model robustness in a small sample and sample unbalanced data environment in the textile industry is solved, and the robustness of the model is improved. And the stability and the reliability of resultant yarn quality prediction are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vortex spinning yarn, in particular to a jet vortex spinning yarn quality prediction model based on M-estimate optimization of width learning. BACKGROUND

[0002] The jet vortex spinning process is a complex process of interaction of fiber assembly, rigid body, fluid and elastomer, and the fiber needs to go through multiple actions such as stretching, deformation and instability, and is easily affected by extreme environment, so that it is difficult to identify the key influencing factors of the yarn quality, and there are three core problems in the current yarn quality prediction: first, there are too many process parameters, the feature selection algorithm has insufficient applicability, and it is difficult to provide effective data support, second, the existing model has weak anti-interference ability to outliers, poor generalization ability and insufficient small sample adaptability, and third, the prediction accuracy and stability are low, which cannot meet the actual needs of the factory.

[0003] The existing yarn quality prediction methods at home and abroad mainly include mathematical statistics, grey theory and machine learning, the mathematical statistics method is difficult to handle complex nonlinear relationship, and the prediction effect is poor, the grey theory method has slightly improved accuracy, but still does not reach the practical standard, the traditional machine learning method such as BP neural network has the problems of large parameter uncertainty and strong prediction randomness, and the width learning system BLS has general approximation ability, but is sensitive to outliers, in the data environment of small sample and sample imbalance in the textile industry, it is easy to lead to the decrease of prediction accuracy and the weakening of stability. SUMMARY

[0004] In view of the problems in the prior art, the present application provides a jet vortex spinning yarn quality prediction model based on M-estimate optimization of width learning.

[0005] The technical scheme adopted by the present application to solve its technical problems is: the jet vortex spinning yarn quality prediction model based on M-estimate optimization of width learning, comprising the following steps: S1, data acquisition and preprocessing: collecting process parameter data and corresponding yarn quality index data in the jet vortex spinning production process, normalizing the collected data and dividing the training set and the test set; S2, constructing a robust width learning model R-M-BLS based on M-estimate optimization: initializing network structure parameters, generating feature layer nodes and enhanced layer nodes, constructing a hidden layer output matrix and initializing an output weight matrix; S3, weight iterative updating based on M-estimate: calculating the prediction error vector of the training set sample, determining the threshold and calculating the sample weight matrix, and iteratively updating the output weight matrix until the termination condition is met; S4. Yarn quality prediction: Input the preprocessed test set process parameter data into the trained RM-BLS model, and calculate the predicted value of the yarn quality index through the optimal output weight matrix.

[0006] Specifically, the process parameters in step S1 include spinning speed, nozzle air pressure, and spinning machine component parameters, and the yarn quality indicators include yarn breaking strength and hairiness H value.

[0007] Specifically, the network structure parameters mentioned in step S2 include the number of feature layer windows. The number of feature nodes contained in each window of the feature layer Number of enhancement layer windows The number of enhancement nodes contained in each window of the enhancement layer Regularization parameters in the objective function Termination of tolerance Maximum number of loops .

[0008] Specifically, in step S2, when generating feature layer nodes, the weights of the feature layer nodes are randomly generated. and bias , through Calculated And then through formula Merging to obtain the feature layer Randomly generate the weights of the enhancement layer nodes. and bias , through Calculated ,in, Indicates the first The covariance matrix is ​​estimated by the forward propagation, and then expressed as follows: Merging to obtain an enhancement layer At the same time, the corresponding uncertainty estimate is obtained. .

[0009] Specifically, in step S3, when calculating the prediction error vector, , For the first The true label value of each sample Output matrix for hidden layer The Row data, based on the training sample error vector Median calculation threshold The calculation formula is: ,in The median of the absolute values ​​of the error vector. The parameter is set to 4.6851.

[0010] Specifically, in step S3, the Tukey function is selected as the M estimator, based on the error vector. and threshold Calculate the weight of each sample The weighting function is: Construct a diagonal sample weight matrix from the weights of each sample. ,in This represents the number of training samples.

[0011] Specifically, the update formula for the output weight matrix in step S3 is as follows: ,in Given the identity matrix; determine if the iteration termination condition is met: or number of iterations Reaching the maximum number of loops If the condition is met, the iteration stops, and the optimal output weight matrix is ​​obtained. .

[0012] The beneficial effects of this invention are: (1) The jet vortex spinning yarn quality prediction model based on M estimation and width learning described in this invention optimizes the width learning system through M estimator, uses Tukey function as loss function to replace traditional L2 norm squared loss, effectively suppresses outlier interference by dynamically allocating sample weights, solves the problem of insufficient model robustness in the textile industry under small sample and unbalanced data environment, and significantly improves the stability and reliability of yarn quality prediction.

[0013] (2) The jet vortex spinning yarn quality prediction model based on M-estimation optimization width learning described in this invention, through hierarchical feature extraction of feature layer and enhancement layer, combined with weight iterative update strategy, not only retains the advantages of simple structure and efficient training of width learning system, but also enhances the fitting ability of nonlinear relationship between complex process parameters and yarn quality index through robust optimization. The prediction accuracy is greatly improved compared with traditional mathematical statistics, grey theory and BP neural network methods, which can meet the actual production needs of textile factories.

[0014] (3) The jet vortex spinning yarn quality prediction model based on M estimation optimization width learning described in this invention clarifies the mapping relationship between key process parameters of jet vortex spinning such as spinning speed and nozzle air pressure and core quality indicators such as yarn breaking strength and hairiness H value. It provides data support for the precise control of process parameters in the production process, helps to reduce production energy consumption and reduce defect rate, and is in line with the development direction of specialization, refinement and innovation and green and low carbon in the textile industry. Attached Figure Description

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0016] Figure 1 The flowchart shows the method for predicting the quality of jet vortex spinning yarn based on M-estimation optimization width learning provided by this invention. Detailed Implementation

[0017] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0018] like Figure 1 As shown, the present invention provides the following technical solution: The jet vortex spinning yarn quality prediction model based on M-estimation optimization width learning includes the following steps: S1. Data Acquisition and Preprocessing: Collect process parameter data and corresponding yarn quality index data during the air jet vortex spinning production process. The process parameters include spinning speed, nozzle air pressure, spinning machine component parameters, etc., and the yarn quality indexes include yarn breaking strength, hairiness H value, etc.; normalize the collected data to eliminate dimensional differences, and divide the data into training set and test set.

[0019] S2. Construct a robust width learning model RM-BLS based on M-estimation optimization. 2.1 Initialize network structure parameters: Set the network structure parameters, i.e., the number of feature layer windows. The number of feature nodes contained in each window of the feature layer Number of enhancement layer windows The number of enhancement nodes contained in each window of the enhancement layer Regularization parameters in the objective function Termination of tolerance Maximum number of loops .

[0020] 2.2 Generating Feature Layer Nodes: Randomly generating weights for feature layer nodes. and bias , through Calculated And then through formula Merging to obtain the feature layer .

[0021] 2.3 Generate Enhancement Layer Nodes: Randomly generate the weights of enhancement layer nodes. and bias , through Calculated ,in, Indicates the first The covariance matrix is ​​estimated by the forward propagation, and then expressed as follows: Merging to obtain an enhancement layer At the same time, the corresponding uncertainty estimate is obtained. .

[0022] 2.4 Constructing the hidden layer output matrix: The feature layer output... and enhancement layer output By concatenating the components, we obtain the final output of the hidden layer. .

[0023] 2.5 Initialize the output weight matrix: Initialize the output weight matrix through pseudo-inverse operation. .

[0024] S3. Iterative weight update based on M-estimation 3.1 Calculate the error vector: based on the current output weight matrix Calculate the prediction error vector of the training set samples. ,in , For the first The true label value of each sample Output matrix for hidden layer The Row data.

[0025] 3.2 Determining the threshold Based on the training sample error vector Median calculation threshold The calculation formula is: ,in The median of the absolute values ​​of the error vector. The parameter is set to 4.6851.

[0026] 3.3 Calculate the sample weight matrix: Choose the Tukey function as the M-estimater, based on the error vector... and threshold Calculate the weight of each sample The weighting function is: Construct a diagonal sample weight matrix from the weights of each sample. ,in This represents the number of training samples.

[0027] 3.4 Iteratively update the output weight matrix: based on the sample weight matrix Hidden layer output matrix A and true label vector The output weight matrix is ​​updated using the following formula: ,in Given the identity matrix; determine if the iteration termination condition is met: or number of iterations Reaching the maximum number of loops If the condition is met, the iteration stops, and the optimal output weight matrix is ​​obtained. Otherwise, return to step 3.1 and continue iterating.

[0028] S4. Yarn Quality Prediction: Input the preprocessed test set process parameter data into the trained RM-BLS model, and then use the optimal output weight matrix... The predicted values ​​of the yarn quality index were calculated.

[0029] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A jet vortex spinning yarn quality prediction model based on M-estimation optimization width learning, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing: Collect process parameter data and corresponding yarn quality index data during the jet vortex spinning production process, normalize the collected data and divide it into training set and test set; S2. Construct a robust width learning model RM-BLS based on M estimation optimization: initialize network structure parameters, generate feature layer nodes and enhancement layer nodes, construct the hidden layer output matrix and initialize the output weight matrix; S3. Weight Iterative Update Based on M-Estimation: Calculate the prediction error vector of the training set samples, determine the threshold and calculate the sample weight matrix, and iteratively update the output weight matrix until the termination condition is met. S4. Yarn quality prediction: Input the preprocessed test set process parameter data into the trained RM-BLS model, and calculate the predicted value of the yarn quality index through the optimal output weight matrix.

2. The jet vortex spinning yarn quality prediction model based on M-estimation optimization width learning according to claim 1, characterized in that: The process parameters mentioned in step S1 include spinning speed, nozzle air pressure, and spinning machine component parameters. The yarn quality indicators include yarn breaking strength and hairiness H value.

3. The jet vortex spinning yarn quality prediction model based on M-estimation optimization width learning according to claim 1, characterized in that: The network structure parameters mentioned in step S2 include the number of feature layer windows. The number of feature nodes contained in each window of the feature layer Number of enhancement layer windows The number of enhancement nodes contained in each window of the enhancement layer Regularization parameters in the objective function Termination of tolerance Maximum number of loops .

4. The jet vortex spinning yarn quality prediction model based on M-estimation optimization width learning according to claim 1, characterized in that: In step S2, when generating feature layer nodes, the weights of the feature layer nodes are randomly generated. and bias , through Calculated And then through formula Merging to obtain the feature layer Randomly generate the weights of the enhancement layer nodes. and bias , through Calculated ,in, Indicates the first The covariance matrix is ​​estimated by the forward propagation, and then expressed as follows: Merging to obtain an enhancement layer At the same time, the corresponding uncertainty estimate is obtained. .

5. The jet vortex spinning yarn quality prediction model based on M-estimation optimization width learning according to claim 1, characterized in that: When calculating the prediction error vector in step S3, , For the first The true label value of each sample Output matrix for hidden layer The Row data, based on the training sample error vector Median calculation threshold The calculation formula is: ,in The median of the absolute values ​​of the error vector. The parameter is set to 4.6851.

6. The jet vortex spinning yarn quality prediction model based on M-estimation optimization width learning according to claim 1, characterized in that: In step S3, the Tukey function is selected as the M estimator, based on the error vector. and threshold Calculate the weight of each sample The weighting function is: Construct a diagonal sample weight matrix from the weights of each sample. ,in This represents the number of training samples.

7. The jet vortex spinning yarn quality prediction model based on M-estimation optimization width learning according to claim 1, characterized in that: The update formula for the output weight matrix in step S3 is as follows: ,in Given the identity matrix; determine if the iteration termination condition is met: or number of iterations Reaching the maximum number of loops If the condition is met, the iteration stops, and the optimal output weight matrix is ​​obtained. .

Images