Shaping process parameter recommendation method and device, computer equipment and storage medium

Abstract

The invention discloses a setting process parameter recommendation method and device, computer equipment and a storage medium. The method comprises the steps of obtaining to-be-predicted production order data; preprocessing the production order data to be predicted; the preprocessing result is input into a prediction model for technological parameter prediction; outputting a prediction result; the prediction model is obtained by taking production order data as a sample set after the production order data is preprocessed and labeled and training a multi-weight main model by adopting an iterative training mode, and specifically, the quality score of each piece of preprocessed and labeled data is dynamically adjusted and is fused into a loss function for training; the multi-weight main model comprises a plurality of base models and meta-models, different weights are distributed according to the prediction performance of the base models, and results obtained through prediction of the weighted base models are integrated through the meta-models. By implementing the method disclosed by the invention, the effect of enhancing stacking by multiple models can be exerted, the dependence on artificial experience is reduced, the accuracy of process parameters and the one-time success rate of production are improved, and the product repair and production cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of machine learning, and more specifically to a method, device, computer equipment and storage medium for recommending forming process parameters. Background Art

[0002] Heat setting is a key step in textile dyeing and finishing. The process involves heating the fabric to a certain size, holding it at a high temperature for a specified period of time, and then cooling it at an appropriate rate. The heat setting process ensures that the fabric achieves the desired flatness, width, weight, and moisture content during the heating process.

[0003] However, the current selection and adjustment of heat-setting processes often relies on long-term experience and lacks precise process design. This approach suffers from low process design accuracy and poor adaptability, making it difficult to ensure consistent product quality. Variations in shrinkage, weight, and width within a single batch of products often occur, resulting in suboptimal heat-setting results and high scrap rates and production costs. When customer requirements for heat-setting fabric quality change or production conditions fluctuate, traditional empirical design methods are unable to meet these new production needs. Furthermore, even if specific process parameters are determined, it is difficult to accurately predict and estimate post-heat-setting quality indicators. Product quality often relies on manual sampling and inspection after the heat-setting process is complete. In other words, regardless of whether the final heat-setting quality is acceptable, the fabric must undergo a complete heat-setting process. This approach not only wastes considerable time and resources but also increases labor and material costs.

[0004] Currently, models are used to predict parameters and make recommendations. However, heat setting process parameters include target exiting machine seal, target exiting machine weight, target exiting machine WPI, exiting machine CPI, target temperature, target machine speed, target width baseline, target upper overfeed, first target width deviation, and second target width deviation. When using a single model for stacking prediction, the weighting of the base models is often overlooked, resulting in the over-amplification of the influence of certain base models and the weakening of the role of other models. This fails to fully utilize the strengths of each base model, affecting the prediction accuracy and generalization ability of the entire model.

[0005] Therefore, it is necessary to design a new method to give full play to the role of multi-model enhanced stacking, reduce dependence on manual experience, improve the accuracy of process parameters and the first-time success rate of production, and reduce product rework and production costs. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the prior art and provide a method, device, computer equipment and storage medium for recommending shaping process parameters.

[0007] To achieve the above-mentioned purpose, the present invention adopts the following technical solution: a method for recommending shaping process parameters, comprising:

[0008] Obtain production order data to be forecasted;

[0009] Preprocessing the production order data to be predicted to obtain a preprocessing result;

[0010] The preprocessing results are input into a prediction model to predict process parameters and obtain prediction results. The prediction model is obtained by preprocessing and labeling production order data as a sample set, and then training a multi-weighted master model using an iterative training method. Specifically, the quality score of each preprocessed and labeled data is dynamically adjusted and integrated into the loss function for training. The multi-weighted master model includes several base models and meta-models, and different weights are assigned according to the prediction performance of the base models. The results of the weighted base model predictions are integrated through the meta-model.

[0011] The prediction result is output.

[0012] The further technical solution is as follows: the prediction model is obtained by using the pre-processed and labeled production order data as a sample set and training the multi-weighted main model in an iterative training method. Specifically, the quality score of each pre-processed and labeled data is dynamically adjusted and integrated into the loss function for training, including:

[0013] Acquire production order data, and preprocess and label the production order data to obtain a sample set, wherein the sample set includes feature data and label data;

[0014] The label data is divided into two parts according to basic process parameters and non-basic process parameters to obtain label data corresponding to the basic process parameters and label data corresponding to the non-basic process parameters, wherein the basic process parameters include required outgoing machine cloth sealing, required outgoing machine grammage, required outgoing machine WPI, required outgoing machine CPI, temperature, machine speed, upper overfeed, and width reference, and the non-basic process parameters include a first width deviation and a second width deviation;

[0015] Construct a multi-weight master model;

[0016] Using the feature data and label data corresponding to the basic process parameters to train the multi-weighted main model to obtain a prediction process matrix;

[0017] Extracting data on a width reference from the label data corresponding to the basic process parameters and adding the data to the feature data to update the feature data;

[0018] Training the multi-weighted main model using the updated feature data and label data corresponding to the non-basic process parameters to obtain a prediction result;

[0019] Mapping the mean square error according to the prediction result and the mean square error corresponding to the label data to obtain a mapping result;

[0020] The data quality score is adjusted according to the mapping calculation, the parameters of the multi-weight main model are adjusted using the adjusted data quality score, and the training is repeated until the multi-weight main model converges to obtain the prediction model.

[0021] A further technical solution is as follows: constructing a multi-weighted main model includes:

[0022] Several models whose prediction performance meets the requirements are selected as base models, and a linear regression model is selected as a meta-model. The base models and the meta-model are stacked to form a multi-weight main model.

[0023] Its further technical solution is: the training process of the multi-weight main model includes:

[0024] Calculating a data quality score of the feature data;

[0025] Inputting the characteristic data into each base model to predict process parameters, and calculating the mean square error of each base model according to the prediction results and the corresponding label data;

[0026] Calculating the weight of the corresponding base model according to the mean square error of each base model;

[0027] Multiplying the weight of the base model by the result predicted by the base model to obtain a weighted prediction result;

[0028] Extracting corresponding process parameters from the weighted prediction results to form a matrix;

[0029] Determine the final prediction result according to the matrix combined with the linear regression model;

[0030] Using mean error as the loss function, and adding the data quality score of the feature data into the loss function;

[0031] By minimizing the loss function, adjusting the parameters of the meta-model, and repeating the training until the meta-model converges, a multi-weight main model is obtained.

[0032] A further technical solution is: calculating the data quality score of the feature data includes:

[0033] use Calculate the data quality score of the feature data, where s represents the quantity quality score of each feature data; S2 is the target deviation, which represents the degree of deviation between the actual test result and the target value of the parameter, and the value is between (0, 1]; E M is the relative error ratio of the cloth seal; E G is the relative error ratio of gram weight; M is the target deviation critical value when the difference between the actual test result and the target value is equal to the error value; actual is the actual value of the cloth seal; M target is the sealing target value; t M is the error range of the cloth seal; G actual G is the actual value of gram weight; target is the target weight value; t G is the error range of gram weight; the calculated data quality scores of each feature data are combined to form a data quality score set.

[0034] A further technical solution is: calculating the weight of the corresponding base model according to the mean square error of each base model includes:

[0035] use Calculate the weight of the corresponding base model, where W i is the weight of the i-th base model; MSE i is the mean square error of i basis models, m represents the number of base models; n is the amount of training data; h is the number of process parameters; y ij is the predicted value of the jth process parameter of the i-th row data; is the true value of the jth process parameter of the feature data in the i-th row; s i is the data quality score of the i-th feature data.

[0036] A further technical solution is: determining the final prediction result based on the matrix combined with the linear regression model includes:

[0037] Determining a coefficient matrix for the linear regression model;

[0038] Splitting and transposing the coefficient matrix row by row to obtain a new coefficient matrix;

[0039] Performing a point multiplication operation on the matrix and each of the new coefficient matrices to obtain a process parameter prediction value for each feature data;

[0040] All the process parameter prediction values ​​are combined to form a final prediction result.

[0041] The present invention also provides a shaping process parameter recommendation device, comprising:

[0042] A data acquisition unit, used to acquire production order data to be predicted;

[0043] A preprocessing unit, configured to preprocess the production order data to be predicted to obtain a preprocessing result;

[0044] A prediction unit, configured to input the preprocessing result into a prediction model to perform process parameter prediction to obtain a prediction result;

[0045] An output unit, configured to output the prediction result;

[0046] Among them, the prediction model is obtained by preprocessing and labeling the production order data as a sample set, and training the multi-weight main model using an iterative training method. Specifically, the quality score of each preprocessed and labeled data is dynamically adjusted and integrated into the loss function for training; the multi-weight main model includes several base models and meta-models, different weights are assigned according to the prediction performance of the base model, and the results obtained by weighted base model prediction are integrated through the meta-model.

[0047] The present invention further provides a computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the computer program.

[0048] The present invention also provides a storage medium, wherein the storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention automatically recommends process parameters by utilizing the trained prediction model, specifically by preprocessing the production order data and labeling it, thereby reducing manual intervention and ensuring data consistency and reliability; dynamically adjusting the quality score of each data and integrating it into the loss function, effectively reducing the impact of low-quality data and improving model accuracy; using a multi-weighted main model, by training multiple base models and assigning different weights according to prediction performance, the overall performance of the model is optimized; the results of the weighted base models are integrated through a meta-model to ensure that the advantages of each model are maximized and the stability and accuracy of the prediction are enhanced; the role of multi-model enhanced stacking is brought into play, reducing reliance on manual experience, improving the accuracy of process parameters and the one-time success rate of production, and reducing product rework and production costs.

[0050] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 A schematic diagram of an application scenario of the shaping process parameter recommendation method provided by an embodiment of the present invention;

[0053] Figure 2 A schematic flow chart of a method for recommending shaping process parameters according to an embodiment of the present invention;

[0054] Figure 3 A schematic diagram of a sub-flow diagram of a shaping process parameter recommendation method provided in an embodiment of the present invention;

[0055] Figure 4 A schematic diagram of a sub-flow diagram of a shaping process parameter recommendation method provided in an embodiment of the present invention;

[0056] Figure 5 A schematic diagram of a sub-flow diagram of a shaping process parameter recommendation method provided in an embodiment of the present invention;

[0057] Figure 6 A schematic diagram of a flow chart of a training process for a multi-weighted master model according to an embodiment of the present invention;

[0058] Figure 7 Schematic diagram of the interface for inputting basic batch information and target information provided by an embodiment of the present invention;

[0059] Figure 8 Schematic diagram of the output page provided by the embodiment of the present invention Figure 1 ;

[0060] Figure 9 Schematic diagram of the output page provided by the embodiment of the present invention Figure 2 ;

[0061] Figure 10 A schematic block diagram of a device for recommending shaping process parameters according to an embodiment of the present invention;

[0062] Figure 11 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0064] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0065] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0066] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0067] See also Figure 1 and Figure 2 , Figure 1 Schematic diagram of an application scenario of the shaping process parameter recommendation method provided in an embodiment of the present invention. Figure 2A schematic flow chart of a shaping process parameter recommendation method provided in an embodiment of the present invention. The shaping process parameter recommendation method is applied to a server. The server interacts with the terminal for data exchange, wherein the server can be a controller of a fabric heat-setting device, etc., and optimizes the selection of process parameters through advanced machine learning technology to meet specific production goals. This method not only improves production efficiency, but also ensures the consistency and stability of product quality. The method of this embodiment applies the Stacking algorithm technology to the recommendation of multiple process parameters in the textile dyeing and finishing industry, and improves the overall prediction performance by integrating multiple base models. The data quality score calculation is based on the label error and target deviation feedback of the initial model, and the quality of the training data is evaluated by the deviation value between the shaping test result and the target value. These quality scores are converted into weights and applied to model training, thereby improving the robustness and accuracy of the model. A base model weight distribution mechanism is introduced into the stacking model, and the influence weight of each base model is dynamically adjusted according to the loss value feedback of a single base model to optimize the performance of the stacking technology. The process parameters are divided into two stages for prediction using a chain iterative training method. Some of the process parameters predicted in the first stage are used as feature inputs of the second-stage model. This structure helps to gradually refine the prediction results and improve the accuracy and reliability of the final recommended process parameters.

[0068] Figure 2 FIG. 1 is a flow chart of a method for recommending shaping process parameters according to an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S140.

[0069] S110: Obtain production order data to be predicted.

[0070] In this embodiment, the production order data to be predicted includes data from the order data management system, specifically including cylinder number, process route, material composition, cloth seal, gram weight, density, temperature and other data, which can be obtained from Figure 7 The interface shown is filled in manually, by device scanning, or by automatic crawling.

[0071] S120: Preprocess the production order data to be predicted to obtain a preprocessing result.

[0072] In this embodiment, the preprocessing result refers to the preprocessed production order data. After receiving the production order data, preprocessing is required to improve the accuracy and efficiency of the prediction.

[0073] Specifically, missing data is a common problem in production order data, which may be due to omissions or errors in data collection. Missing data can affect the accuracy of model training and subsequent analysis, so it must be effectively handled. In the method of this embodiment, for missing pre- and post-cooking data, an attempt is first made to match it with the development office's data source by vat number. If the missing data still cannot be completed, the incomplete data is deleted. This method helps to avoid the adverse effects of missing data on the model, ensures data integrity and accuracy, and thus improves the reliability of subsequent analysis and model predictions.

[0074] Production order data may contain some duplicate records. If left unaddressed, these duplicates can lead to biased analysis results. To eliminate the impact of duplicate data on the model, feature data duplication analysis is used. Specifically, by comparing the feature data of each production order, if identical data is found, the most recent record is retained based on the order time and all other duplicates are deleted. This step ensures that the model does not introduce errors due to duplicate data, thereby improving analysis accuracy and data validity.

[0075] In the actual production process, due to manual input or other operational issues, there may be some outliers in the production order data. These outliers will seriously interfere with the learning and prediction of the model, so it is necessary to detect and process outliers in the data. Specifically, the following formula is used to determine outliers: K={k1,k2,…,k n}; where n is the amount of training data, and k is the ratio of target specifications to pre-cooking specifications; b i Practical seal for data target i; g i b′ is the target weight of the i-th data; i The first data is sealed before cooking; g′ i is the weight of the i-th data before cooking.

[0076] The method of this embodiment combines data characteristics and uses the interquartile range (IQR) method to detect outliers. The specific operation is to calculate the 25th percentile (q25) and 75th percentile (q75) of the data, and then determine the upper and lower limits of the outliers based on the IQR (i.e., q75-q25). Any data outside this range will be considered an outlier and deleted. This step can ensure that every data in the data set is within a reasonable range, avoid unnecessary interference of outliers on the model, and thus improve the accuracy and prediction effect of the model.

[0077] Specifically, the first quantile is also called the quarter quantile or the 25% quantile, and the third quantile is also called the quarter quantile or the 75% quantile. Let q 25is the first quantile, which means that 25% of the data points in data K are less than q 25 ,q 75 The third quantile means that 75% of the data points in the data K are greater than q 75 , then IQR=q 75 -q 25 ;

[0078] Thus, we can obtain the upper bound of the threshold and the lower bound of the threshold: lower = q 25 -1.5*IQR; upper=q 75 +1.5*IQR;

[0079] The data K is then filtered by the threshold [lower, upper], and values ​​that are not within the range are considered as outliers and deleted.

[0080] By processing missing data, duplicate data, and outliers, we ensure that input data is more complete, accurate, and consistent, providing a reliable foundation for subsequent analysis and model training. Data processing eliminates noise and bias, ensuring that models can better identify and learn real relationships between data, improving prediction reliability. By removing incomplete or outlier records, we avoid the negative impact of erroneous data on analysis results and help gain more accurate insights into production data.

[0081] These processing steps ensure the quality of the dataset, ultimately optimizing decision support and predictive capabilities of process parameters during production.

[0082] S130: Input the preprocessing result into a prediction model to predict process parameters to obtain a prediction result.

[0083] In this embodiment, the prediction result refers to the predicted values ​​corresponding to parameters such as the required cloth seal out of the machine, the required grammage out of the machine, the required WPI out of the machine, the required CPI out of the machine, temperature, machine speed, width reference, overfeed, first width deviation and second width deviation.

[0084] Among them, WPI and CPI represent the coil density index of the fabric, respectively. CPI (Courses Per Inch) refers to the number of horizontal rows of coils in the fabric within 1 inch, that is, the transverse density of the fabric, and WPI (Wales Per Inch) refers to the number of vertical rows of coils in the fabric within 1 inch, that is, the longitudinal density of the fabric. The prediction model is obtained by preprocessing and labeling the production order data as a sample set, and then training the multi-weight main model using an iterative training method. Specifically, the quality score of each preprocessed and labeled data is dynamically adjusted and integrated into the loss function for training. The multi-weight main model includes several base models and meta-models. Different weights are assigned according to the prediction performance of the base model, and the results predicted by the weighted base model are integrated through the meta-model.

[0085] In one embodiment, see Figure 3 The above-mentioned prediction model is obtained by preprocessing and labeling the production order data as a sample set, and training the multi-weight main model using an iterative training method. Specifically, the quality score of each preprocessed and labeled data is dynamically adjusted and integrated into the loss function for training, including steps S131 to S138.

[0086] S131. Obtain production order data, and pre-process and label the production order data to obtain a sample set, wherein the sample set includes feature data and label data.

[0087] In this embodiment, the characteristic data refers to the cylinder number (i.e., batch number), process route, material composition, cloth seal, gram weight, density, temperature and other data. The label data refers to the actual values ​​corresponding to the parameters such as the required cloth seal, required gram weight, required WPI, required CPI, temperature, machine speed, width reference, upper overfeed, first width deviation and second width deviation, which are called

[0088] The preprocessing in this step is consistent with the above-mentioned step S120 and will not be repeated here.

[0089] S132. Divide the label data into two parts according to basic process parameters and non-basic process parameters to obtain label data corresponding to the basic process parameters and label data corresponding to the non-basic process parameters, wherein the basic process parameters include required cloth sealing, required weight, required WPI, required CPI, temperature, machine speed, overfeed, and width reference; the non-basic process parameters include a first width deviation and a second width deviation.

[0090] In this embodiment, in the process parameter prediction, there is indeed a sequential dependency relationship between different parameters, especially process parameters such as the first width deviation, the second width deviation and the width reference, which are closely related to each other; since the three are process parameter values ​​at the same time, they are divided in the model training, and the basic process parameters including the width reference are predicted first, and then the predicted width reference is integrated into the original feature as a new feature to further predict other parameters.

[0091] Specifically, the label data Divided into two parts, namely and in The actual label value is composed of the required cloth seal, required weight, required WPI, required CPI, temperature, machine speed, overfeed, and width reference. is the true label value consisting of the first width deviation and the second width deviation. In the model used in this step, the feature data is Among them, n is the amount of training data, d is the number of feature fields; x ij ,i=1,2,…,n、j=1,2,…,d represents the jth feature field of the i-th row data. The label data is Use the multi-weight master model to train and obtain the prediction process matrix Y left . Among them, F is the main model; X is the feature data; n is the amount of training data; h is the number of process parameters; y is the predicted value of the process parameter.

[0092] S133. Construct a multi-weight main model.

[0093] In this embodiment, the multi-weight main model refers to a stacked model that incorporates weights.

[0094] Specifically, several models whose prediction performance meets the requirements are selected as base models, and a linear regression model is selected as a meta-model, and the base models and the meta-model are stacked to form a multi-weight main model.

[0095] The training process of a multi-weighted master model is essentially stacked ensemble learning, a method that combines the prediction results of multiple base models through a meta-model to improve overall prediction performance. This approach is particularly advantageous when dealing with complex regression problems. The stacked model architecture typically consists of two main components: a base model at the first level and a meta-model at the second level. To select base models, the original dataset is first divided into multiple subsets, typically using K-fold cross-validation to effectively reduce the risk of overfitting in the base models. Each base model (e.g., KNN, SVR, GBR, RFR, LR, DTR, etc.) is trained on each subset and generates predictions. The predictions of these base models are independent and may differ somewhat because each base model has different assumptions and preferences when processing the data. Base models with strong predictive power are selected to capture diverse patterns in the data. Diversity in base models can enhance ensemble performance. For example, models such as KNN, SVR, GBR, RFR, and DTR can be used as base models because they have different strengths when applied to different data characteristics. For the meta-model, the prediction results of all first-layer base models are combined with the original features to form a new dataset, which serves as the meta-model input. The meta-model is typically a relatively simple model (such as linear regression (LR)). It is responsible for further integrating and adjusting the outputs of the first-layer base models to generate the final prediction results. Meta-models generally use relatively simple structures and high computational efficiency, such as linear regression (LR). A simple meta-model can effectively avoid overfitting because it does not have many degrees of freedom to overfit the first-layer output.

[0096] In practical applications, selection is often made by evaluating the regression performance of different base models. Common evaluation metrics include the regression coefficient R² and mean squared error (MSE). Based on these metrics, the performance of the base models is ranked, and the best performing model is selected as the base model for the stacked model.

[0097] For example, the following base models were evaluated:

[0098] KNN (k-nearest neighbor regression); SVR (support vector machine regression); GBR (gradient boosting regression); RFR (random forest regression); LR (linear regression); DTR (decision tree regression);

[0099] By calculating the R 2 and MSE, and sorted by MSE from smallest to largest. Assume the top three best-performing models are selected: GBR, RFR, and KNN. These models will serve as the base models for the first layer.

[0100] Next, a model with a simple structure, such as LR (linear regression), is selected as the meta-model of the second layer. The output of the first layer base model is fused through it to finally obtain the prediction result of the stacked model.

[0101] Stacked ensemble learning improves overall prediction performance by passing the prediction results of multiple base models to a meta-model to fuse their predictions. The key steps include:

[0102] In the first layer, select base models that perform well and can capture different characteristics of the data.

[0103] A simple meta-model (such as linear regression) is used for fusion in the second layer to avoid overfitting.

[0104] By ranking base models based on evaluation metrics like MSE, selecting the optimal base model, and then combining it with a simple meta-model, the stability and accuracy of the entire model can be improved. This approach can effectively improve the prediction accuracy of regression problems and is particularly suitable for complex practical problems.

[0105] S134 , using the feature data and the label data corresponding to the basic process parameters to train the multi-weight main model to obtain a prediction process matrix.

[0106] In this embodiment, a multi-weighted master model is used for training to obtain the prediction process matrix Y left . Among them, F is the main model; X is the feature data; n is the amount of training data; h is the number of process parameters; y is the predicted value of the process parameter.

[0107] In one embodiment, see Figure 4 and Figure 6 The training process of the above-mentioned multi-weight main model includes steps S1341 to S1348.

[0108] S1341. Calculate the data quality score of the feature data.

[0109] In this embodiment,

[0110] use Calculate the data quality score of the feature data, where s represents the quantity quality score of each feature data; S2 is the target deviation, which represents the degree of deviation between the actual test result and the target value of the parameter, and the value is between (0, 1]; E M is the relative error ratio of the cloth seal; E G is the relative error ratio of gram weight; M is the target deviation critical value when the difference between the actual test result and the target value is equal to the error value; actual is the actual value of the cloth seal; Mtarget is the sealing target value; t M is the error range of the cloth seal; G actual G is the actual value of gram weight; target is the target weight value; t G is the error range of gram weight; the data quality scores of each feature data calculated are combined to form a data quality score set. S=[s1,s2,…s n ]; where n is the amount of training data, s n is the data quality score of the nth data.

[0111] Step S1341 is a quality scoring mechanism based on experimental data. It evaluates the deviation of test results for parameters such as fabric seal (M) and weight (G) from target values ​​during the production process, and then assigns a quality score to each piece of data. This approach combines the error margin, degree of deviation, and the difference from the target value to provide high-quality data support for the model while preserving data diversity.

[0112] Specifically, the test results for the two parameters of cloth seal and weight are analyzed to calculate the quality score for each data point. The key indicator of the evaluation is the deviation between the test result and the target value. The evaluation formula takes into account the following aspects:

[0113] Target Deviation (S): This indicator reflects the degree of deviation between the actual test results and the target value, and the score range is (0, 1). The closer the score is to 1, the higher the quality of the data is and the closer it is to the target value.

[0114] Error range (E M and E G ): The relative error between the actual value and the target value of the cloth seal and grammage is used to calculate the error ratio (E M and E G If both error ratios are within the allowable range (i.e. E M ≤1 and E G ≤1), the target deviation is calculated using the standard formula; if the error ratio exceeds the allowable range (i.e. E M >1 or E G >1), the target deviation decreases rapidly.

[0115] A quality score is calculated for each data point based on the error ratio between the fabric cover and the weight. This quality score calculation considers whether the test result meets the expected value and how quickly the quality score is reduced if the error exceeds the specified range. Test results outside the error range are not directly deleted, but are assigned a lower quality score to reflect their poor quality.

[0116] The quality scores of the training data form a dataset that represents the quality of all training data. It can reflect the quality level of the overall data and help screen out high-quality sample data for subsequent analysis or model training.

[0117] Even if some data test results fall outside the tolerance range, they are not directly deleted. Instead, they are retained by lowering their quality scores. This helps maintain dataset diversity while ensuring that quality issues with unqualified data are accurately reflected in subsequent analysis. By calculating the target deviation, the difference between the data and the target value can be quantified, and the quality score can be quickly adjusted based on the actual error. This precise assessment method makes the quality score of each data item more objective and scientific. If the test results fall outside the allowable error range, the quality score will drop rapidly. This design effectively prevents bias in the overall quality assessment caused by individual outliers. The closer the test results are to the target value, the higher the quality score, indicating that the data has a higher reference value for actual production and model training, thereby improving the quality of the final training set and enhancing the prediction results. By utilizing actual production historical data as the training set, this method can more realistically reflect the various possible situations in the production process, enhancing the adaptability and generalization of the model.

[0118] In summary, this quality scoring mechanism not only ensures data diversity but also ensures high-quality data used in model training through refined deviation processing, error range control, and dynamic adjustment of quality scores, thereby improving overall prediction accuracy and optimizing the production process.

[0119] S1342: Input the feature data into each base model to predict process parameters, and calculate the mean square error of each base model based on the prediction results and the corresponding label data.

[0120] In this embodiment, the feature data here refers to the input feature data, that is, it can be the feature data of step S134 or the feature data updated in step S136, and the label data refers to the label data of step S136 or step S134, specifically refers to the true value determined in the corresponding step.

[0121] S1343. Calculate the weight of the corresponding base model according to the mean square error of each base model.

[0122] In this embodiment, the Calculate the weight of the corresponding base model, where W i is the weight of the i-th base model; MSE i is the mean square error of i basis models, m represents the number of base models; n is the amount of training data; h is the number of process parameters; y ij is the predicted value of the jth process parameter of the i-th row data; is the true value of the jth process parameter of the feature data in the i-th row; s i is the data quality score of the i-th feature data.

[0123] In traditional stacking methods, the prediction results of multiple base models are typically synthesized through simple averaging. However, this embodiment adds a base model weighting mechanism, namely Weight Stacking, to the traditional stacking algorithm. Through this method, each base model is assigned a different weight based on its prediction performance and reliability, thereby achieving more accurate and stable results.

[0124] Specifically, the performance of each base model is evaluated by calculating the mean squared error (MSE) of the model and converting this error into a model weight. A smaller MSE indicates a more accurate model prediction and a larger corresponding weight; conversely, a larger error indicates a smaller weight. During the calculation process, the weights of all base models are normalized so that their sum is 1. In this way, weight stacking effectively improves the accuracy and stability of the overall prediction, assigning higher weights to more reliable base models.

[0125] Weight stacking introduces a weighting mechanism for base models, ensuring that the contribution of each model is proportional to its performance. Models with stronger performance automatically receive higher weights, thereby optimizing the accuracy of the final prediction results. This weighting mechanism assigns lower weights to models with larger errors, minimizing their impact on the final prediction results and effectively reducing the impact of model errors on the results. By considering the mean squared error (MSE) of each model, Weight stacking improves the stability of the combined prediction results. In the presence of data noise or instability in certain base models, more stable models receive higher weights to compensate, thus preventing fluctuations in the overall prediction results. MSE, as an evaluation metric, directly reflects the predictive performance of the base models. Using this metric, Weight stacking automatically selects high-performing and stable models, thereby enhancing the reliability of the overall model ensemble. When faced with complex data structures or multiple process parameters, Weight stacking effectively assigns appropriate weights to different base models, better adapting to diverse data characteristics and avoiding the risk that a single model may not be able to capture all data patterns. As training data and errors change, Weight stacking dynamically adjusts the weights of each model, enabling the model ensemble to continuously optimize and adapt to new data environments. This flexibility improves the long-term applicability of the algorithm.

[0126] In summary, Weight Stacking not only optimizes prediction accuracy but also improves model stability and reliability through a weighting mechanism based on model errors. It is an effective improvement to the traditional stacking method, especially suitable for multi-model combinations, and can show stronger adaptability on complex data sets.

[0127] S1344. Multiply the weight of the base model by the result predicted by the base model to obtain a weighted prediction result.

[0128] In this example, GBR, RFR, and KNN models are used as base models, and LR is used as a meta-model for stacking. In a production order, the characteristic data fields include target practical cloth seal, target grammage, D number, component ratio, pre-cooking cloth seal, pre-cooking grammage, pre-cooking WPI, pre-cooking CPI, post-cooking cloth seal, post-cooking grammage, post-cooking WPI, and post-cooking CPI. Parameters requiring recommendation include required cloth seal, required grammage, required WPI, required CPI, temperature, machine speed, width reference, and overfeed.

[0129] Assuming that the model input data is an n×d matrix, the feature data is represented as X: Among them, n is the amount of training data, d is the number of feature fields; x ij,i=1,2,…,n、j=1,2,…,d represents the jth feature field of the i-th row of data. Passing X to the base model, the predicted result is Y: Where n is the amount of training data; h is the number of process parameters; y ij ,i=1,2,…,n、j=1,2,…,h, which represents the predicted value of the jth process parameter of the i-th row data.

[0130] Each base model will generate its own prediction results, and the prediction results generated by m base models are integrated to form ′

[0131] Y: Where m represents the number of base models; Y i Represents the predicted data of the i-th base model; y ij , i = 1, 2, ..., n, j = 1, 2, ..., h represents the jth predicted value of the i-th row of data of the first base model. j = h + 1, h + 2, ..., h * 2 represents the jth predicted value of the i-th row of data of the second base model, and so on.

[0132] The weight of each base model is calculated in this way. A weighting mechanism is introduced on the basis of traditional stacking, which can assign different weights to each basic model, making the adjustment of model combination more flexible.

[0133] Bring the weights into the new dataset. In the subsequent meta-model training and prediction process, the weights of each basic model will play a role. The data after bringing in the weights is: Where m represents the number of base models; W m is the weight of the mth basis model; Y m is the process parameter matrix predicted by the mth base model; h is the number of process parameters; n is the amount of training data; y n(h*m) is the predicted value of the hth parameter of the nth data under the mth base model.

[0134] S1345. Extract corresponding process parameters from the weighted prediction results to form a matrix.

[0135] In this embodiment, the value of YY is extracted from the process parameters of each base model according to the process parameters, and the result is: A=[A1, A2, ..., A h ]; Where m represents the number of base models; W m is the weight of the mth base model; h is the number of process parameters; n is the amount of training data; y n(h*m) It is the predicted value of the hth parameter of the nth data under the mth base model.

[0136] S1346. Determine a final prediction result based on the matrix and the linear regression model.

[0137] In this embodiment, the linear regression model performs estimation analysis through the linear relationship between the independent variable and the dependent variable, and quantifies the influence of the independent variable on the dependent variable through the coefficient.

[0138] In one embodiment, see Figure 5 , the above-mentioned step S1346 may include S13461 to S13464.

[0139] S13461. Determine the coefficient matrix of the linear regression model;

[0140] S13462, splitting the coefficient matrix by rows and performing a transposition operation to obtain a new coefficient matrix;

[0141] S13463. Perform a dot multiplication operation on the matrix and each of the new coefficient matrices to obtain a process parameter prediction value for each feature data;

[0142] S13464. Combine all the process parameter prediction values ​​to form a final prediction result.

[0143] Specifically, each independent variable corresponds to a coefficient, which is multiplied by the independent variable in the model and then summed. The corresponding meta-model coefficient matrix of the present invention is P: Among them, h is the number of process parameters; m is the number of base models; p hm is the coefficient of the h-th process under the m-th base model.

[0144] Split P by rows and transpose it to get a new coefficient matrix: Among them, h is the number of process parameters; m is the number of base models; p hm is the coefficient of the h-th process under the m-th base model.

[0145] Finally, matrix multiplication is performed, A h The dimension is (n,m), The dimension is (m, 1), and the dot product is used to obtain the hth process parameter of each data, which has a dimension of (n, 1). After combining them, the final prediction result Z is formed, and the shape of Z is (n, h). The calculation formula is as follows: z nh =W1y n1 p h1 +W2y n(h*2) p h2 +…+W m y n(h*m) p hm ; Among them, zij , i=1,2,…,n、j=1,2,…,h represents the jth predicted value of the i-th row data; m represents the number of base models; h is the number of process parameters; n is the amount of training data; W m is the weight of the mth basis model; y n(h*m) is the predicted value of the hth parameter of the nth data under the mth base model; p hm is the coefficient of the h-th process under the m-th base model.

[0146] Compared to traditional stacking that directly combines each base model for prediction, this embodiment introduces a feature weighting mechanism to weight the prediction results of different base models. This enables more flexible adjustment of the contribution of each model, thereby optimizing the prediction performance of the model in a specific scenario. Weight stacking gives higher weights to models with better performance, which can more effectively utilize the strengths of each model, thereby improving the prediction accuracy of the integrated model. Lower weights are given to models with poor performance, thereby reducing their impact on the final prediction results and reducing the risk of the integrated model.

[0147] S1347. Use the mean error as a loss function, and add the data quality score of the feature data into the loss function.

[0148] In addition, during model training, the optimization direction of the model is controlled by calculating the loss value. The selected loss function is the mean square error, and the data quality score is applied in the calculation. Where J(θ) is the objective function, which is a function of the meta-model parameter θ; n is the amount of training data; h is the number of process parameters; y ij is the predicted value of the jth process parameter of the i-th row data; is the true value of the jth process parameter of the i-th row data; s i is the quality score of the i-th sample.

[0149] S1348. Adjust the parameters of the meta-model by minimizing the loss function, and repeat the training until the meta-model converges to obtain a multi-weight main model.

[0150] The loss function is calculated by the data quality score s i The influence of each sample is adjusted. Samples with higher quality scores contribute more to the loss function, thus affecting the model training process.

[0151] During model training, the goal is to minimize the loss function and optimize the prediction results by continuously adjusting the parameters θ of the meta-model. The training process is repeated until the model's loss function converges, and finally a multi-weight master model is obtained.

[0152] The introduction of a weighted data quality score allows model training to prioritize the impact of high-quality samples. Samples with high quality scores contribute more to the loss, helping the model more accurately learn and fit these important samples, thereby improving overall prediction accuracy. In real applications, data quality often varies. Using data quality scores, the model can dynamically adjust its focus on data of varying quality, reducing interference from low-quality data and improving its ability to learn from high-quality data. By minimizing the loss function weighted by the quality score, the training process can avoid the negative impact of outliers or low-quality data on model training. This improves the stability of model training, especially when working with noisy datasets. In traditional training, the loss function simply calculates the difference between the predicted and true values ​​without considering sample quality. By introducing the quality score, the loss function automatically adjusts the weights during calculation, allowing the model to more accurately prioritize high-quality samples during optimization, effectively improving optimization efficiency. Due to the weighted influence of high-quality samples, the model can more quickly identify important features, avoid interference from noise or low-quality data, and accelerate convergence. Compared to models without quality score weighting, this weighting mechanism makes training more efficient. By introducing data quality scores, the model can better adapt to diverse datasets, avoid overfitting to low-quality data, and thus improve its generalization ability to new data. This approach is particularly advantageous when data quality varies. During model training, the quality score can be flexibly adjusted based on the data source or application scenario. This flexibility enables the model to perform effectively in different environments, especially when working with real-time data, allowing for dynamic adjustment of training strategies.

[0153] By incorporating data quality scores into the loss function and dynamically adjusting weights during model training, the multi-weighted master model optimization method proposed in this paper can effectively improve the accuracy, stability, and efficiency of model training. This method not only reduces the negative impact of low-quality data but also accelerates training convergence and enhances the model's generalization ability. It is particularly suitable for applications with uneven data quality or high noise.

[0154] S135 , extracting data on a width reference from the label data corresponding to the basic process parameters, and adding the data to the feature data to update the feature data.

[0155] In this embodiment, The width reference data is extracted from and added to the training feature data X as the training data for the second model training. Among them, n is the amount of training data; d is the number of feature fields; x nd The dth characteristic field of the nth row of data; y nfis the true value of the width reference in the nth row of data.

[0156] S136 , using the updated feature data and label data corresponding to the non-basic process parameters to train the multi-weighted main model to obtain a prediction result.

[0157] In this embodiment, the updated feature data and the label data corresponding to the non-basic process parameters are used to train the multi-weighted main model to obtain the predicted process matrix Y right : Among them, F is the main model; X ′ is the feature data after adding the width reference data; n is the data volume of the training data; h is the number of process parameters; y is the predicted value of the process parameters.

[0158] Combine the results of the two models to obtain the complete process parameter output results: Where n is the amount of training data; h is the number of process parameters; y nh The predicted value of the hth process parameter for the nth row of data

[0159] In practical applications, the results of the first model are calculated first, and then the predicted width benchmark is placed in the input feature. Then the second model calculation is performed, and the results of the two are combined to obtain the final model result.

[0160] S137. Map the mean square error according to the prediction result and the mean square error corresponding to the label data to obtain a mapping result.

[0161] In this embodiment, the prediction matrix of the process parameters is obtained, and the predicted value and the actual value are calculated to obtain the mean square error of each training data: L = [l1, l2, ..., l i ,…,l n ]; Where L is the set of average errors between the predicted values ​​and actual values ​​of the label data of the training data; n is the amount of training data; h is the number of process parameters; y ij is the predicted value of the jth process parameter of the i-th data; is the true value of the jth process parameter of the i-th data.

[0162] L is mapped to between 1 and 0 through an exponential function. The smaller the mean square error l is, the smaller the deviation is, and the higher the score is, the highest being 1. in, is the set of mean square errors after mapping; n is the amount of training data; l i is the average error between the predicted value and the actual value of the label data of the i-th data; e is a natural constant.

[0163] S138. Adjust the data quality score according to the mapping calculation, adjust the parameters of the multi-weight main model using the adjusted data quality score, and repeat the training until the multi-weight main model converges to obtain the prediction model.

[0164] In this embodiment, during the training process of the multi-weighted main model, different weights are formulated according to the influence of the target deviation and the model feedback error, and a new data quality score is obtained through fusion calculation: S ′ =[s1 ′ ,s2 ′ ,…,s i ′ ,…,s n ′ ]; Among them, S ′ is the set of adjusted data quality scores; n is the amount of training data; w1 is the weight of target deviation; s i is the target deviation of the i-th data; w2 is the weight of the model feedback error; The average error between the predicted value and the actual value of the label data after mapping the i-th data.

[0165] After obtaining the new data quality score, the entire prediction model training process, namely steps S133 to S138, needs to be re-executed. The iterative chain training method solves the correlation between the many output process parameters, thereby improving the effect. First, some parameters are predicted, and then the predicted values ​​are used as input to predict the remaining parameters.

[0166] S140: Output the prediction result.

[0167] In this embodiment, the prediction results include the results of the two outputs of the multi-weight main model, specifically including the predicted values ​​of the required cloth seal, required weight, required WPI, required CPI, temperature, machine speed, overfeed, width reference, first width deviation and second width deviation, and are finally presented on the specified interface, such as Figures 8 and 9 shown.

[0168] The above-mentioned method for recommending shaping process parameters automatically recommends process parameters by utilizing the trained prediction model. Specifically, it pre-processes the production order data and labels it to reduce manual intervention and ensure data consistency and reliability; dynamically adjusts the quality score of each data and integrates it into the loss function to effectively reduce the impact of low-quality data and improve model accuracy; uses a multi-weighted main model to train multiple base models and assign different weights according to prediction performance to optimize the overall performance of the model; integrates the results of the weighted base models through the meta-model to ensure that the advantages of each model are maximized and enhance the stability and accuracy of the prediction; plays the role of multi-model enhanced stacking, reduces reliance on manual experience, improves the accuracy of process parameters and the first-time success rate of production, and reduces product rework and production costs.

[0169] Figure 10 FIG. 3 is a schematic block diagram of a shaping process parameter recommendation device 300 provided by an embodiment of the present invention. Figure 10 As shown, corresponding to the above shaping process parameter recommendation method, the present invention also provides a shaping process parameter recommendation device 300. The shaping process parameter recommendation device 300 includes a unit for executing the above shaping process parameter recommendation method, and the device can be configured in a server. Figure 10 The shaping process parameter recommendation device 300 includes a data acquisition unit 301 , a preprocessing unit 302 , a prediction unit 303 and an output unit 304 .

[0170] The data acquisition unit 301 is used to acquire production order data to be predicted; the preprocessing unit 302 is used to preprocess the production order data to be predicted to obtain a preprocessing result; the prediction unit 303 is used to input the preprocessing result into the prediction model to predict process parameters to obtain a prediction result; the output unit 304 is used to output the prediction result;

[0171] Among them, the prediction model is obtained by preprocessing and labeling the production order data as a sample set, and training the multi-weight main model using an iterative training method. Specifically, the quality score of each preprocessed and labeled data is dynamically adjusted and integrated into the loss function for training; the multi-weight main model includes several base models and meta-models, different weights are assigned according to the prediction performance of the base model, and the results obtained by weighted base model prediction are integrated through the meta-model.

[0172] In one embodiment, the shaping process parameter recommendation system further includes a model training unit, and the model training unit includes:

[0173] The data acquisition unit 301 is used to acquire production order data, and pre-process and label the production order data to obtain a sample set, wherein the sample set includes feature data and label data;

[0174] a dividing unit, configured to divide the label data into two parts according to basic process parameters and non-basic process parameters, so as to obtain label data corresponding to the basic process parameters and label data corresponding to the non-basic process parameters, wherein the basic process parameters include required outgoing machine cloth sealing, required outgoing machine grammage, required outgoing machine WPI, required outgoing machine CPI, temperature, machine speed, upper overfeed, and width reference, and the non-basic process parameters include a first width deviation and a second width deviation;

[0175] A construction unit for constructing a multi-weighted main model;

[0176] A first training unit is configured to train the multi-weighted main model using the feature data and label data corresponding to the basic process parameters to obtain a prediction process matrix;

[0177] an extraction unit, configured to extract data on a width reference from the label data corresponding to the basic process parameters, and add the data to the feature data to update the feature data;

[0178] A second training unit is used to train the multi-weighted main model using the updated feature data and label data corresponding to the non-basic process parameters to obtain a prediction result;

[0179] A mapping unit, configured to map the mean square error according to the prediction result and the mean square error corresponding to the label data to obtain a mapping result;

[0180] A score adjustment unit is used to adjust the data quality score according to the mapping calculation, adjust the parameters of the multi-weight main model using the adjusted data quality score, and repeat the training until the multi-weight main model converges to obtain the prediction model.

[0181] In one embodiment, the construction unit is used to select several models whose prediction performance meets the requirements as base models, and select a linear regression model as a meta-model, and stack the base models and the meta-model to form a multi-weight main model.

[0182] In one embodiment, the first training unit and the second training unit respectively include:

[0183] The score calculation subunit is used to calculate the data quality score of the feature data; specifically,

[0184] Calculate the data quality score of the feature data, where s represents the quantity quality score of each feature data; S2 is the target deviation, which represents the degree of deviation between the actual test result and the target value of the parameter, and the value is between (0, 1]; E M is the relative error ratio of the cloth seal; E G is the relative error ratio of gram weight; M is the target deviation critical value when the difference between the actual test result and the target value is equal to the error value; actual is the actual value of the cloth seal; M target is the sealing target value; t M is the error range of the cloth seal; G actual G is the actual value of gram weight; target is the target weight value; t G is the error range of gram weight; the calculated data quality scores of each feature data are combined to form a data quality score set.

[0185] A base model prediction subunit, configured to input the feature data into each base model to perform process parameter prediction, and calculate the mean square error of each base model based on the prediction result and the corresponding label data;

[0186] The weight calculation subunit is used to calculate the weight of the corresponding base model according to the mean square error of each base model; specifically, Calculate the weight of the corresponding base model, where W i is the weight of the i-th base model; MSE i is the mean square error of i basis models, m represents the number of base models; n is the amount of training data; h is the number of process parameters; y ij is the predicted value of the jth process parameter of the i-th row data; is the true value of the jth process parameter of the feature data in the i-th row; s i is the data quality score of the i-th feature data.

[0187] a multiplication subunit, configured to multiply the weight of the base model by the result predicted by the base model to obtain a weighted prediction result;

[0188] an extraction subunit, configured to extract corresponding process parameters from the weighted prediction results to form a matrix;

[0189] A regression processing subunit, configured to determine a final prediction result based on the matrix in combination with the linear regression model;

[0190] a loss function construction subunit, configured to use the mean error as a loss function and add the data quality score of the feature data into the loss function;

[0191] The adjustment subunit is used to adjust the parameters of the meta-model by minimizing the loss function, and repeat the training until the meta-model converges to obtain a multi-weight main model.

[0192] In one embodiment, the regression processing subunit includes:

[0193] A coefficient matrix determination module, used to determine the coefficient matrix of the linear regression model;

[0194] A matrix processing module is used to split the coefficient matrix by rows and perform a transposition operation to obtain a new coefficient matrix;

[0195] a dot product module, configured to perform a dot product operation on the matrix and each of the new coefficient matrices to obtain a process parameter prediction value for each feature data;

[0196] The combination module is used to combine all the process parameter prediction values ​​to form a final prediction result.

[0197] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned shaping process parameter recommendation device 300 and each unit can refer to the corresponding description in the aforementioned method embodiment, and for the convenience and brevity of description, it will not be repeated here.

[0198] The above-mentioned shaping process parameter recommendation device 300 can be implemented in the form of a computer program. The computer program can be used in Figure 11 Runs on the computer equipment shown.

[0199] See also Figure 11 , Figure 11 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0200] See Figure 11 The computer device 500 includes a processor 502 , a memory, and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0201] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, which, when executed, can enable the processor 502 to execute a shaping process parameter recommendation method.

[0202] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.

[0203] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a method for recommending forming process parameters.

[0204] The network interface 505 is used to communicate with other devices through the network. Figure 11 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0205] The processor 502 is configured to execute a computer program 5032 stored in the memory to implement the following steps:

[0206] Acquire production order data to be predicted; preprocess the production order data to be predicted to obtain a preprocessing result; input the preprocessing result into a prediction model to predict process parameters to obtain a prediction result; wherein, the prediction model is obtained by preprocessing and labeling the production order data as a sample set, and adopting an iterative training method to train a multi-weight main model, specifically by dynamically adjusting the quality score of each preprocessed and labeled data, and integrating it into the loss function for training; the multi-weight main model includes several base models and meta-models, different weights are assigned according to the prediction performance of the base model, and the results obtained by weighted base model prediction are integrated through the meta-model; output the prediction result.

[0207] In one embodiment, the processor 502 implements the prediction model by using preprocessed and labeled production order data as a sample set and training the multi-weighted main model using an iterative training method. Specifically, the processor 502 dynamically adjusts the quality score of each preprocessed and labeled data and integrates it into the loss function to perform the training step, and specifically implements the following steps:

[0208] Acquire production order data, and pre-process and label the production order data to obtain a sample set, wherein the sample set includes feature data and label data; divide the label data into two parts according to basic process parameters and non-basic process parameters to obtain label data corresponding to the basic process parameters and label data corresponding to the non-basic process parameters, wherein the basic process parameters include required cloth sealing, required weight, required WPI, required CPI, temperature, machine speed, overfeed, and width reference, and the non-basic process parameters include the first width deviation and the second width deviation; construct a multi-weight master model; use the feature data and the basic process parameters to obtain the label data corresponding to the basic process parameters. The multi-weight main model is trained with the label data to obtain a prediction process matrix; data on the width reference in the label data corresponding to the basic process parameters are extracted and added to the feature data to update the feature data; the multi-weight main model is trained with the updated feature data and the label data corresponding to the non-basic process parameters to obtain a prediction result; the mean square error is mapped according to the prediction result and the mean square error corresponding to the label data to obtain a mapping result; the data quality score is adjusted according to the mapping calculation, the parameters of the multi-weight main model are adjusted with the adjusted data quality score, and the training is repeated until the multi-weight main model converges to obtain the prediction model.

[0209] In one embodiment, when implementing the step of constructing the multi-weighted main model, the processor 502 specifically implements the following steps:

[0210] Several models whose prediction performance meets the requirements are selected as base models, and a linear regression model is selected as a meta-model. The base models and the meta-model are stacked to form a multi-weight main model.

[0211] In one embodiment, the processor 502 implements the following steps when implementing the training step of the multi-weight main model:

[0212] Calculate the data quality score of the feature data; input the feature data into each of the base models to predict the process parameters, and calculate the mean square error of each base model based on the predicted results and the corresponding label data; calculate the weight of the corresponding base model based on the mean square error of each base model; multiply the weight of the base model by the result obtained by the base model prediction to obtain a weighted prediction result; extract the corresponding process parameters from the weighted prediction result to form a matrix; determine the final prediction result based on the matrix and the linear regression model; use the mean error as the loss function, and add the data quality score of the feature data to the loss function; adjust the parameters of the meta-model by minimizing the loss function, and repeat the training until the meta-model converges to obtain a multi-weight main model.

[0213] In one embodiment, when the processor 502 implements the step of calculating the data quality score of the feature data, it specifically implements the following steps:

[0214] use Calculate the data quality score of the feature data, where s represents the quantity quality score of each feature data; S2 is the target deviation, which represents the degree of deviation between the actual test result and the target value of the parameter, and the value is between (0, 1]; E M is the relative error ratio of the cloth seal; E G is the relative error ratio of gram weight; M is the target deviation critical value when the difference between the actual test result and the target value is equal to the error value; actual is the actual value of the cloth seal; M target is the sealing target value; t M is the error range of the cloth seal; G actual G is the actual value of gram weight; target is the target weight value; t G is the error range of gram weight; the calculated data quality scores of each feature data are combined to form a data quality score set.

[0215] In one embodiment, when the processor 502 implements the step of calculating the weight of the corresponding base model according to the mean square error of each base model, it specifically implements the following steps:

[0216] use Calculate the weight of the corresponding base model, where W i is the weight of the i-th base model; MSE i is the mean square error of i basis models, m represents the number of base models; n is the amount of training data; h is the number of process parameters; y ij is the predicted value of the jth process parameter of the i-th row data; is the true value of the jth process parameter of the feature data in the i-th row; s i is the data quality score of the i-th feature data.

[0217] In one embodiment, when implementing the step of determining the final prediction result based on the matrix combined with the linear regression model, the processor 502 specifically implements the following steps:

[0218] Determine the coefficient matrix of the linear regression model; split the coefficient matrix by rows and transpose it to obtain a new coefficient matrix; perform a dot multiplication operation on the matrix and each of the new coefficient matrices to obtain a process parameter prediction value for each feature data; combine all the process parameter prediction values ​​to form a final prediction result.

[0219] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0220] Those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program includes program instructions, which can be stored in a storage medium that is computer-readable. The program instructions are executed by at least one processor in the computer system to implement the steps in the method of the above-described embodiment.

[0221] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor performs the following steps:

[0222] Acquire production order data to be predicted; preprocess the production order data to be predicted to obtain a preprocessing result; input the preprocessing result into a prediction model to predict process parameters to obtain a prediction result; wherein, the prediction model is obtained by preprocessing and labeling the production order data as a sample set, and adopting an iterative training method to train a multi-weight main model, specifically by dynamically adjusting the quality score of each preprocessed and labeled data, and integrating it into the loss function for training; the multi-weight main model includes several base models and meta-models, different weights are assigned according to the prediction performance of the base model, and the results obtained by weighted base model prediction are integrated through the meta-model; output the prediction result.

[0223] In one embodiment, the processor, when executing the computer program, implements the prediction model by preprocessing and labeling the production order data as a sample set and training the multi-weighted main model using an iterative training method. Specifically, the processor dynamically adjusts the quality score of each preprocessed and labeled data and incorporates it into the loss function to perform the training step, specifically implementing the following steps:

[0224] Acquire production order data, and pre-process and label the production order data to obtain a sample set, wherein the sample set includes feature data and label data; divide the label data into two parts according to basic process parameters and non-basic process parameters to obtain label data corresponding to the basic process parameters and label data corresponding to the non-basic process parameters, wherein the basic process parameters include required cloth sealing, required weight, required WPI, required CPI, temperature, machine speed, overfeed, and width reference, and the non-basic process parameters include the first width deviation and the second width deviation; construct a multi-weight master model; use the feature data and the basic process parameters to obtain the label data corresponding to the basic process parameters. The multi-weight main model is trained with the label data to obtain a prediction process matrix; data on the width reference in the label data corresponding to the basic process parameters are extracted and added to the feature data to update the feature data; the multi-weight main model is trained with the updated feature data and the label data corresponding to the non-basic process parameters to obtain a prediction result; the mean square error is mapped according to the prediction result and the mean square error corresponding to the label data to obtain a mapping result; the data quality score is adjusted according to the mapping calculation, the parameters of the multi-weight main model are adjusted with the adjusted data quality score, and the training is repeated until the multi-weight main model converges to obtain the prediction model.

[0225] In one embodiment, when the processor executes the computer program to implement the step of constructing a multi-weighted master model, the processor specifically implements the following steps:

[0226] Several models whose prediction performance meets the requirements are selected as base models, and a linear regression model is selected as a meta-model. The base models and the meta-model are stacked to form a multi-weight main model.

[0227] In one embodiment, when the processor executes the computer program to implement the training step of the multi-weight main model, the processor specifically implements the following steps:

[0228] Calculate the data quality score of the feature data; input the feature data into each of the base models to predict the process parameters, and calculate the mean square error of each base model based on the predicted results and the corresponding label data; calculate the weight of the corresponding base model based on the mean square error of each base model; multiply the weight of the base model by the result obtained by the base model prediction to obtain a weighted prediction result; extract the corresponding process parameters from the weighted prediction result to form a matrix; determine the final prediction result based on the matrix and the linear regression model; use the mean error as the loss function, and add the data quality score of the feature data to the loss function; adjust the parameters of the meta-model by minimizing the loss function, and repeat the training until the meta-model converges to obtain a multi-weight main model.

[0229] In one embodiment, when the processor executes the computer program to implement the step of calculating the data quality score of the feature data, the processor specifically implements the following steps:

[0230] use Calculate the data quality score of the feature data, where s represents the quantity quality score of each feature data; S2 is the target deviation, which represents the degree of deviation between the actual test result and the target value of the parameter, and the value is between (0, 1]; E M is the relative error ratio of the cloth seal; E G is the relative error ratio of gram weight; M is the target deviation critical value when the difference between the actual test result and the target value is equal to the error value; actual is the actual value of the cloth seal; M target is the sealing target value; t M is the error range of the cloth seal; G actual G is the actual value of gram weight; target is the target weight value; t G is the error range of gram weight; the calculated data quality scores of each feature data are combined to form a data quality score set.

[0231] In one embodiment, when the processor executes the computer program to implement the step of calculating the weight of the corresponding base model according to the mean square error of each base model, the processor specifically implements the following steps:

[0232] use Calculate the weight of the corresponding base model, where W i is the weight of the i-th base model; MSE i is the mean square error of i basis models, m represents the number of base models; n is the amount of training data; h is the number of process parameters; y ijis the predicted value of the jth process parameter of the i-th row data; is the true value of the jth process parameter of the feature data in the i-th row; s i is the data quality score of the i-th feature data.

[0233] In one embodiment, when the processor executes the computer program to implement the step of determining the final prediction result based on the matrix combined with the linear regression model, the processor specifically implements the following steps:

[0234] Determine the coefficient matrix of the linear regression model; split the coefficient matrix by rows and transpose it to obtain a new coefficient matrix; perform a dot multiplication operation on the matrix and each of the new coefficient matrices to obtain a process parameter prediction value for each feature data; combine all the process parameter prediction values ​​to form a final prediction result.

[0235] The storage medium may be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0236] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0237] In the several embodiments provided herein, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the various units is merely a logical functional division, and actual implementation may employ other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0238] The steps in the methods of the embodiments of the present invention may be adjusted in order, combined, or deleted as needed. The units in the devices of the embodiments of the present invention may be combined, divided, or deleted as needed. Furthermore, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit.

[0239] If this integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, terminal, or network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present invention.

[0240] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. The method for recommending shaping process parameters is characterized in that: include: Obtain production order data to be predicted; Preprocessing the production order data to be predicted to obtain a preprocessing result; The preprocessing result is input into the prediction model to predict the process parameters to obtain the prediction result; wherein, the prediction model is obtained by preprocessing the production order data and labeling it as a sample set, and training the multi-weight main model by an iterative training method, specifically by dynamically adjusting the quality score of each preprocessed and labeled data, and integrating it into the loss function for training; the multi-weight main model includes several base models and meta-models, different weights are assigned according to the prediction performance of the base model, and the results obtained by weighted base model prediction are integrated through the meta-model; The prediction result is outputted.

2. The method for recommending shaping process parameters according to claim 1, characterized in that: The prediction model is obtained by preprocessing and labeling the production order data as a sample set, and training the multi-weight main model in an iterative training manner. Specifically, the quality score of each preprocessed and labeled data is dynamically adjusted and integrated into the loss function for training, including: Acquire production order data, and preprocess and label the production order data to obtain a sample set, wherein the sample set includes feature data and label data; The label data is divided into two parts according to basic process parameters and non-basic process parameters, so as to obtain label data corresponding to the basic process parameters and label data corresponding to the non-basic process parameters, wherein the basic process parameters include required outgoing cloth sealing, required outgoing gram weight, required outgoing WPI, required outgoing CPI, temperature, machine speed, upper overfeed, and width reference, and the non-basic process parameters include a first width deviation and a second width deviation; Construct a multi-weighted master model; Using the feature data and label data corresponding to the basic process parameters to train the multi-weighted main model to obtain a prediction process matrix; Extracting data on a width reference from the label data corresponding to the basic process parameters, and adding the data to the feature data to update the feature data; Training the multi-weighted main model using the updated feature data and label data corresponding to the non-basic process parameters to obtain a prediction result; Mapping the mean square error according to the prediction result and the mean square error corresponding to the label data to obtain a mapping result; The data quality score is adjusted according to the mapping calculation, the parameters of the multi-weight main model are adjusted using the adjusted data quality score, and the training is repeated until the multi-weight main model converges to obtain the prediction model.

3. The method for recommending shaping process parameters according to claim 2, characterized in that: The construction of the multi-weight main model comprises: Several models whose prediction performance meets the requirements are selected as base models, and a linear regression model is selected as a meta-model. The base models and the meta-model are stacked to form a multi-weight main model.

4. The method for recommending shaping process parameters according to claim 3, characterized in that: The training process of the multi-weight main model includes: Calculating a data quality score of the feature data; Input the characteristic data into each base model to predict process parameters, and calculate the mean square error of each base model according to the prediction result and the corresponding label data; Calculate the weight of the corresponding base model according to the mean square error of each base model; Multiplying the weight of the base model by the result predicted by the base model to obtain a weighted prediction result; Extracting corresponding process parameters from the weighted prediction results to form a matrix; Determine the final prediction result according to the matrix combined with the linear regression model; Using mean error as a loss function, and adding a data quality score of the feature data into the loss function; By minimizing the loss function, adjusting the parameters of the meta-model, and repeating the training until the meta-model converges, a multi-weight main model is obtained.

5. The method for recommending shaping process parameters according to claim 4, characterized in that: The calculating the data quality score of the feature data comprises: use Calculate the data quality score of the feature data, where s represents the quantity quality score of each feature data; S2 is the target deviation, which represents the degree of deviation between the actual test result and the target value of the parameter, and takes a value between (0, 1]; E M E is the relative error ratio of the cloth seal; G is the relative error ratio of gram weight; M is the target deviation critical value when the difference between the actual test result and the target value is equal to the error value; actual is the actual value of the cloth seal; M target is the sealing target value; t M is the error range of the cloth seal; G actual G is the actual value of gram weight; target is the target weight value; t G is the error range of gram weight; the calculated data quality score of each characteristic data is combined to form a data quality score set.

6. The method for recommending shaping process parameters according to claim 4, characterized in that: The step of calculating the weight of the corresponding base model according to the mean square error of each base model comprises: use Calculate the weight of the corresponding base model, where W i is the weight of the i-th base model; MSE i is the mean square error of the i-base models, m represents the number of base models; n is the amount of training data; h is the number of process parameters; y is the number of training data; ij is the predicted value of the jth process parameter of the i-th row of data; is the true value of the jth process parameter of the feature data in the i-th row; s i is the data quality score of the i-th feature data.

7. The method for recommending shaping process parameters according to claim 4, characterized in that: Determining the final prediction result according to the matrix in combination with the linear regression model includes: Determining a coefficient matrix of the linear regression model; The coefficient matrix is ​​split and transposed row by row to obtain a new coefficient matrix; Performing a point multiplication operation on the matrix and each of the new coefficient matrices to obtain a process parameter prediction value for each feature data; All the process parameter prediction values ​​are combined to form a final prediction result.

8. The device for recommending shaping process parameters is characterized in that: include: A data acquisition unit, used for acquiring production order data to be predicted; A preprocessing unit, used for preprocessing the production order data to be predicted to obtain a preprocessing result; A prediction unit, used for inputting the preprocessing result into a prediction model to perform process parameter prediction to obtain a prediction result; An output unit, used for outputting the prediction result; Among them, the prediction model is obtained by preprocessing and labeling the production order data as a sample set, and training the multi-weight main model by iterative training. Specifically, the quality score of each preprocessed and labeled data is dynamically adjusted and integrated into the loss function for training; the multi-weight main model includes several base models and meta-models, different weights are assigned according to the prediction performance of the base model, and the results obtained by weighted base model prediction are integrated through the meta-model.

9. A computer device, characterized in that: The computer device comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images