Anode aluminum foil performance prediction method and system based on stacking model

Abstract

The invention relates to an anode aluminum foil performance prediction method and system based on a stacking model. The method comprises the steps of obtaining target parameters in a to-be-detected anode aluminum foil production process; the target parameters in the production process of the to-be-detected anode aluminum foil are input into an anode aluminum foil performance prediction model, the performance prediction result of the to-be-detected anode aluminum foil is obtained, the anode aluminum foil performance prediction model is constructed based on machine learning model stacking and obtained based on training of a training set, and the performance prediction result of the to-be-detected anode aluminum foil is obtained. The training set comprises target parameters in the production process of the anode aluminum foils of different specifications. According to the invention, the performance of the AAF can be predicted stably and accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of anode aluminum foil production processes, and particularly to a method and system for predicting the performance of anode aluminum foil based on a stacking model. Background Art

[0002] Anode aluminum foil (AAF) is widely used in fields such as capacitors, aluminum anode technology, and lithium-ion batteries due to its excellent electrochemical and physical properties. Anode aluminum foil is produced by subjecting high-purity aluminum foil to corrosion pore formation and anodization treatments. The electrochemical formation and acid etching stages of anode aluminum foil involve numerous non-linear process parameters, which pose a great challenge to accurately predicting its performance and studying corrosion and formation parameters. In addition, with the continuous emergence of new technologies and processes, the production process requirements and performance requirements of AAF are also constantly changing, further exacerbating the instability of AAF performance control. Therefore, accurately and reliably predicting the performance of anode aluminum foil is crucial for optimizing processing technologies.

[0003] According to existing research, the models used to predict the performance of anode aluminum foil include theoretical models, machine learning models, and stacking models. Theoretical models include electrochemical reactions, mass, and momentum transfer. It is difficult to find a convergent solution in the steady-state equation when predicting performance. Using machine learning models to predict the performance of anode aluminum foil is a common practice in this field. However, the prediction accuracy of these basic machine learning and neural networks is still insufficient.

[0004] In recent years, the emergence of stacking models has prompted some scholars to apply them to the prediction of the performance of anodic aluminum foils, and very good results have been achieved. However, there are still deficiencies in stacking models, including: (1) Insufficient data processing: The method of removing outliers is simple and easy to implement, but it may also lead to the loss of data information. Whether to adopt this method requires weighing the impact degree of missing values on the overall analysis of the data set or the performance of the model, as well as the effectiveness and representativeness of the remaining data after data cleaning. (2) Insufficient feature selection: Currently, the number of features used in existing prediction models is insufficient, and the influencing factors are not fully considered. The production of anodic aluminum foil is a multi-process operation, and its performance is affected by various environmental factors, including current, voltage, and conductivity. However, some features may contain irrelevant or duplicate information, and may even hinder the performance of the model. Excluding these features can improve the accuracy and generalization ability of the model. (3) Deficiencies in prediction models: Determining the optimal fusion method is a challenging task because the choice of fusion methods and strategies for different-level models has a significant impact on the results. The increase in the complexity of stacking models usually leads to overfitting of the training data, resulting in poor performance on new data. In addition, due to the limited number of anodic aluminum foil samples, it is difficult for the model to fully capture the potential patterns and rules in the data, resulting in poor generalization ability. (4) Deficiencies in optimization algorithms: The hyperparameter space of stacking models is extensive. Therefore, hyperparameter optimization may fall into local optimal solutions, especially when the complexity of the stacking model is large.

[0005] To address the above deficiencies of the stacking model, the present invention provides a method and system for predicting the performance of anodic aluminum foil based on the stacking model. Summary of the Invention

[0006] The objective of the present invention is to provide a method and system for predicting the performance of anodic aluminum foil based on the stacking model, which can stably and accurately predict the performance of anodic aluminum foil AAF.

[0007] To achieve the above objective, the present invention provides the following solutions:

[0008] A method for predicting the performance of anodic aluminum foil based on the stacking model includes:

[0009] Obtain target parameters in the production process of the anodic aluminum foil to be detected;

[0010] Input the target parameters in the production process of the anodic aluminum foil to be detected into the anodic aluminum foil performance prediction model to obtain the performance prediction result of the anodic aluminum foil to be detected, where the anodic aluminum foil performance prediction model is constructed based on the stacking of machine learning models and trained based on a training set, and the training set includes target parameters in the production processes of different specifications of anodic aluminum foils.

[0011] Optionally, obtaining the training set includes:

[0012] Obtain the process parameters and performance parameters of anode aluminum foils with different specifications on a normal production line;

[0013] Combine the Pearson correlation coefficient PCC and the multicollinearity test MT to perform data screening on the process parameters and the performance parameters, and obtain the screened data;

[0014] Perform Box-Cox transformation on the screened data to obtain the target parameters in the production process of anode aluminum foils with different specifications.

[0015] Optionally, combining the Pearson correlation coefficient PCC and the multicollinearity test MT to perform data screening on the process parameters and the performance parameters includes:

[0016] Use the Pearson correlation coefficient PCC to obtain the linear relationship between the process parameters and the performance parameters;

[0017] Through the multicollinearity test MT, obtain the variance inflation factor of the process parameters;

[0018] According to the linear relationship and the variance inflation factor, perform data screening on the process parameters and the performance parameters to obtain the screened data.

[0019] Optionally, the anode aluminum foil performance prediction model includes a base learner and a meta-learner;

[0020] The base learner includes Random Forest RF, Gradient Boosting Decision Tree GB, Extra Trees ET, Adaptive Boosting AB;

[0021] The meta-learner includes Extreme Gradient Boosting XGB, LightGBM.

[0022] Optionally, training the anode aluminum foil performance prediction model based on a training set includes:

[0023] Input the training set into the base learner to obtain the prediction results of the Random Forest RF, Gradient Boosting Decision Tree GB, Extra Trees ET, Adaptive Boosting AB;

[0024] Input the prediction results into the meta-learner to obtain the final prediction results.

[0025] Optionally, the hyperparameters of the anode aluminum foil performance prediction model are optimized using the Optuna optimization algorithm.

[0026] The present invention also provides a system for the anode aluminum foil performance prediction method based on a stacking model, including: a parameter acquisition module and a performance prediction module;

[0027] The parameter acquisition module is used to acquire the target parameters in the production process of the anodic aluminum foil to be detected;

[0028] The performance prediction model is used to input the target parameters in the production process of the anodic aluminum foil to be detected into the anodic aluminum foil performance prediction model, and obtain the performance prediction result of the anodic aluminum foil to be detected. Among them, the anodic aluminum foil performance prediction model is constructed based on the stacking of machine learning models and trained based on a training set, and the training set includes the target parameters in the production processes of anodic aluminum foils of different specifications.

[0029] Optionally, acquiring the training set includes:

[0030] Acquire the process parameters and performance parameters of anodic aluminum foils of different specifications on a normal production line;

[0031] Combine the Pearson correlation coefficient PCC and the multicollinearity test MT to perform data screening on the process parameters and the performance parameters, and obtain the screened data;

[0032] Perform Box-Cox transformation on the screened data to obtain the target parameters in the production processes of anodic aluminum foils of different specifications.

[0033] Optionally, the anodic aluminum foil performance prediction model includes a base learner and a meta-learner;

[0034] The base learner includes Random Forest RF, Gradient Boosting Decision Tree GB, Extra Trees ET, and Adaptive Boosting AB;

[0035] The meta-learner includes Extreme Gradient Boosting XGB and LightGBM.

[0036] Optionally, training the anodic aluminum foil performance prediction model based on the training set includes:

[0037] Input the training set into the base learner to obtain the prediction results of Random Forest RF, Gradient Boosting Decision Tree GB, Extra Trees ET, and Adaptive Boosting AB;

[0038] Input the prediction results into the meta-learner to obtain the final prediction result.

[0039] The beneficial effects of the present invention are as follows: First, the present invention uses the Pearson correlation coefficient (PCC) and the multicollinearity test (MT) to screen out the most relevant and representative features from the original dataset, thereby reducing the computational burden. The data is processed through the Box-Cox transformation (BCT) to improve the normality of the data. Then, a stacked ensemble learning model is constructed, and the Optuna algorithm is used to determine the parameters of each model to enhance the performance of the stacked model. Finally, the actual production data is used to predict the multiple performances of AAF. The present invention can stably and accurately predict the performance of AAF. Description of the Drawings

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0041] Figure 1 It is a flowchart of the method for predicting the performance of anodic aluminum foil based on a stacked model according to an embodiment of the present invention;

[0042] Figure 2 It is the correlation analysis result of an embodiment of the present invention. Among them, (a) is the correlation heat map between process parameters and breakdown voltage, and (b) is the correlation heat map between process parameters and specific capacitance;

[0043] Figure 3 It is the multicollinearity test result of an embodiment of the present invention. Among them, (a) is the multicollinearity test result of breakdown voltage, and (b) is the multicollinearity test result of specific capacitance;

[0044] Figure 4Results of the Cox-Box transformation for the embodiments of the present invention. Among them, (a) is the data distribution and normal fitting curve of feature X25 (transformed into conductivity in three segments) before Cox-Box data transformation, (b) is the data distribution and normal fitting curve of feature X25 (transformed into conductivity in three segments) after Cox-Box data transformation, (c) is the data distribution and normal fitting curve of feature X77 (curing 3 into time) before Cox-Box data transformation, (d) is the data distribution and normal fitting curve of feature X77 (curing 3 into time) after Cox-Box data transformation, (e) is the data distribution and normal fitting curve of feature X24 (transformed into voltage in three segments) before Cox-Box data transformation, (f) is the data distribution and normal fitting curve of feature X24 (transformed into voltage in three segments) after Cox-Box data transformation, (g) is the data distribution and normal fitting curve of feature X43 (transformed into conductivity in five segments) before Cox-Box data transformation, (h) is the data distribution and normal fitting curve of feature X43 (transformed into conductivity in five segments) after Cox-Box data transformation;

[0045] Figure 5 Prediction performance results of different models for the embodiments of the present invention. Among them, (a) is the bar chart of the predicted breakdown voltage values of the base model, (b) is the bar chart of the predicted breakdown voltage values of the base model after hyperparameter optimization, (c) is the bar chart of the predicted breakdown voltage values of each stacked model after hyperparameter optimization, (d) is the bar chart of the predicted specific capacitance values of the base model, (e) is the bar chart of the predicted specific capacitance values of the base model after hyperparameter optimization, (f) is the bar chart of the predicted specific capacitance values of each stacked model after hyperparameter optimization;

[0046] Figure 6 Box plots of the prediction errors of different models for the embodiments of the present invention. Among them, (a) is the box plot of the prediction error of the breakdown voltage value of the base model, (b) is the box plot of the prediction error of the breakdown voltage value of the base model after hyperparameter optimization, (c) is the box plot of the prediction error of the breakdown voltage value of each stacked model after hyperparameter optimization, (d) is the box plot of the prediction error of the specific capacitance value of the base model, (e) is the box plot of the prediction error of the specific capacitance value of the base model after hyperparameter optimization, (f) is the box plot of the prediction error of the specific capacitance value of each stacked model after hyperparameter optimization;

[0047] Figure 7Prediction error distributions of different models in the embodiments of the present invention. Among them, (a) is the bar chart of the breakdown voltage prediction error of the proposed model, (b) is the bar chart of the breakdown voltage prediction error of the random forest model after hyperparameter optimization, (c) is the bar chart of the breakdown voltage prediction error of the random forest model, (d) is the bar chart of the specific capacitance prediction error of the proposed model, (e) is the bar chart of the specific capacitance prediction error of the random forest model after hyperparameter optimization, and (f) is the bar chart of the specific capacitance prediction error of the random forest model;

[0048] Figure 8 Scatter plots of the actual values and predicted values of the proposed model in the embodiments of the present invention. Among them, (a) is the scatter plot of the breakdown voltage prediction result of the proposed anode aluminum foil prediction model, and (b) is the scatter plot of the specific capacitance prediction result of the proposed anode aluminum foil prediction model;

[0049] Figure 9 Comparison of microscopic morphologies in the embodiments of the present invention. Among them, (a) is the cross-section of the etched foil, (b) is the surface of the etched foil, (c) is the cross-section of the AAF, and (d) is the surface of the AAF. Detailed implementation manners

[0050] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0051] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0052] Embodiment 1:

[0053] This embodiment provides a method for predicting the performance of anode aluminum foil based on a stacked model, including:

[0054] Obtain the target parameters in the production process of the anode aluminum foil to be detected;

[0055] Input the target parameters in the production process of the anode aluminum foil to be detected into the anode aluminum foil performance prediction model to obtain the performance prediction result of the anode aluminum foil to be detected. Among them, the anode aluminum foil performance prediction model is constructed based on a stacked machine learning model and trained based on a training set, and the training set includes the target parameters in the production processes of different specifications of anode aluminum foil.

[0056] Further, obtaining the training set includes:

[0057] Obtain the process parameters and performance parameters of anode aluminum foils with different specifications on a normal production line;

[0058] Combine the Pearson correlation coefficient PCC and the multicollinearity test MT to perform data screening on the process parameters and performance parameters, and obtain the screened data;

[0059] Perform Box-Cox transformation on the screened data to obtain the target parameters in the production process of anode aluminum foils with different specifications.

[0060] Specifically, in order to improve the accuracy and generalization ability of the model, this embodiment eliminates redundant and noisy features through the PCC-MT method. This method can reduce the impact of irrelevant information on the model, enabling the model to give priority to important and valuable features.

[0061] The Box-Cox transformation is mainly used to make the data more conform to the normal distribution, so that the data more conforms to the statistical hypothesis. In practice, it is often used to solve the problem that the data does not conform to the normal distribution in regression analysis and variance analysis, thereby improving the accuracy and reliability of the model. The mathematical formula of the Box-Cox transformation is:

[0062]

[0063] Among them, y is the original data, and λ is the parameter of BCT. When λ = 0, logarithmic transformation is used; otherwise, the BCT formula is used for transformation.

[0064] Furthermore, combining the Pearson correlation coefficient PCC and the multicollinearity test MT to perform data screening on the process parameters and performance parameters includes:

[0065] Adopt the Pearson correlation coefficient PCC to obtain the linear relationship between the process parameters and the performance parameters;

[0066] Through the multicollinearity test MT, obtain the variance inflation factor of the process parameters;

[0067] According to the linear relationship and the variance inflation factor, perform data screening on the process parameters and performance parameters to obtain the screened data.

[0068] Specifically, the Pearson correlation coefficient is a statistical tool used to represent the change trend and degree between two variables. In this embodiment, PCC is used to determine the linear relationship between the process parameters and the performance parameters. The expression of PCC is as follows:

[0069]

[0070] Among them, x i and y i are respectively the i-th parameters of the feature and the target; and They are the means of the features and the target respectively; n is the number of samples.

[0071] The variance inflation factor (VIF) is an important indicator for detecting multicollinearity. In MT, if there is a high correlation between two or more independent variables, the variances of these independent variables will be overestimated, resulting in inaccurate estimation of regression coefficients. VIF is used to quantify the degree of this variance inflation. The calculation formula of VIF is:

[0072]

[0073] where VIF i is the VIF of the i-th independent variable; is the coefficient of determination obtained when the i-th independent variable is used as the dependent variable and other independent variables are used as independent variables for regression.

[0074] Furthermore, the anode aluminum foil performance prediction model includes a base learner and a meta-learner;

[0075] The base learner includes Random Forest (RF), Gradient Boosting Decision Tree (GB), Extra Trees (ET), and Adaptive Boosting (AB);

[0076] The meta-learner includes Extreme Gradient Boosting (XGB) and LightGBM.

[0077] Even further, training the anode aluminum foil performance prediction model based on the training set includes:

[0078] Input the training set into the base learner to obtain the prediction results of Random Forest (RF), Gradient Boosting Decision Tree (GB), Extra Trees (ET), and Adaptive Boosting (AB);

[0079] Input the prediction results into the meta-learner to obtain the final prediction results.

[0080] Specifically, the stacking ensemble framework, i.e., the anode aluminum foil performance prediction model, consists of a base learner and a meta-learner. First, divide the original dataset into multiple sub-datasets and input each sub-dataset into the base learners (RF, GB, ET, and AB) in the first layer. The base learners output their respective prediction results. Then, use the output of the first layer as input to the learners in the second layer to train the meta-learner. Finally, the meta-learner (XGB and LGBM) outputs the final prediction results. In this embodiment, the prediction effects of XGB and LGBM are compared, and it is found that LGBM as the meta-learner has a better prediction effect.

[0081] Furthermore, the hyperparameters of the anode aluminum foil performance prediction model are optimized using the Optuna optimization algorithm.

[0082] Specifically, the core components of the Optuna algorithm are the hyperparameter space and the objective function. Optuna searches for the optimal combination of hyperparameters within the hyperparameter space and optimizes based on the results of the objective function.

[0083] In this embodiment, the mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ) are used as the evaluation metrics for the prediction model.

[0084]

[0085] where n is the number of samples, y i is the actual value of the i-th sample, is the predicted value of the model, is the average of all actual values.

[0086] As Figure 1 shown, the prediction model proposed in this embodiment is based on PCC-MT feature screening, BCT data transformation, stacked ensemble model, and Optuna parameter optimization algorithm. The following are the specific steps of its design:

[0087] Step 1: PCC-MT feature screening:

[0088] Use the PCC-MT method to screen out the most valuable features from the process parameters of the anodic aluminum foil (AAF). Evaluate the linear relationship between each process parameter and the target performance through PCC, and detect and eliminate the multicollinearity problem through MT, thereby reducing the feature dimension and improving the computational efficiency and stability of the model. In this embodiment, feature selection can identify the key parameters in production, that is, the target parameters.

[0089] Step 2: BCT data transformation:

[0090] Perform Box-Cox transformation (BCT) on the screened feature data to transform the non-normally distributed data into approximately normally distributed data. BCT can improve the normality of the data, thereby improving the prediction accuracy and reliability of the model.

[0091] Step 3: Basic model prediction:

[0092] Use the processed data to train multiple basic models (KNN, random forest (RF), gradient boosting decision tree (GB), extra trees (ET), adaptive boosting (AB), extreme gradient boosting (XGB), and light gradient boosting machine (LGBM)) respectively to predict the performance of the anodic aluminum foil. At the same time, compare the prediction results of these basic models with the prediction results of the stacked model to evaluate the performance improvement of the stacked model.

[0093] Step 4: Optuna Parameter Optimization:

[0094] Use the Optuna optimization algorithm to optimize the hyperparameters of the stacking model. Optuna searches for the optimal parameter combination within the hyperparameter space to improve the prediction performance of the stacking model.

[0095] Step 5: Stacking Strategy Comparison and Model Selection:

[0096] Conduct a performance comparison analysis of the stacking models with different stacking strategies to determine the optimal stacking strategy. Further optimize the prediction performance of the stacking model by adjusting the combination method of the base models and the meta - learner.

[0097] Through the above steps, the stacking model can make full use of the advantages of each base model. At the same time, by means of feature screening, data transformation, and parameter optimization, etc., it can improve the prediction accuracy and generalization ability of the model, so as to achieve accurate prediction of multiple properties of the anodic aluminum foil.

[0098] The following tools were used: Intel(R)Pentium(R)CPU G4560@3.50GHz, Windows 10, Jupyter Notebook, and Python 3.9. Analyze the method of this embodiment:

[0099] 1. Dataset Description:

[0100] In this embodiment, the production data of a certain factory's actual anodic aluminum foil (AAF) production line in 2023 was used, including the production process parameters and performance parameters of AAF. The total amount of data was 2000 groups. Detailed information such as code, name, unit, minimum value, and maximum value mainly included data such as temperature, current, voltage, conductivity, pH value, GOE (etching solution), and time at each stage. The performance parameters included withstand voltage (WV) and specific capacitance (SC). There were 80 production process parameters and 2 performance parameters. The ratio of the training set to the prediction set was 8:2.

[0101] 2. Correlation Analysis:

[0102] To improve the quality of data features and identify key process parameters, it is crucial to conduct a Pearson correlation coefficient (PCC) analysis. Determine whether the differences between features are real or just the result of random factors through the significance level (probability value, P - value). As Figure 2As shown in (a)-(b), the PCCs of X61 (voltage at the second repair stage), X42 (voltage at the fifth formation stage), X73 (voltage at the third repair stage), and X51 (voltage at the first repair stage) with WV are 0.80, 0.72, 0.72, and 0.71 respectively, and all the corresponding P-values are less than 0.001. This indicates that there is a strong correlation between the voltages at the formation and repair stages and WV, and it is minimally affected by other factors. The PCCs of X61, X62 (conductivity at the second repair stage), X73, X42 with SC are -0.71, 0.65, -0.63, and -0.63 respectively, and all the corresponding P-values are less than 0.001. This indicates that there is a strong correlation between the voltages at the formation and repair stages and SC, and it is minimally affected by other factors. The results of previous studies have shown that the thickness of the dense layer of the oxide film is positively correlated with the voltages at the formation and repair stages. This indicates that it is reasonable to select the main process parameters that affect WV and SC.

[0103] 3. Multicollinearity analysis:

[0104] To improve the stability and interpretability of the regression model, a multicollinearity test (MT) was performed on the input features. The complete analysis results are as Figure 3 shown in (a)-(b). Generally, when the VIF value exceeds 10, there is a serious multicollinearity problem. For example, the input features X51 and X61 used to predict WV and SC exhibit significant multicollinearity problems, which may have a negative impact on the prediction accuracy of the model. Therefore, features with VIF values exceeding 10 must be removed.

[0105] 4. Box-Cox transformation:

[0106] To improve the normality of the data and the accuracy of the model, the Box-Cox transformation (BCT) was applied to the dataset. The detailed analysis results are as Figure 4 shown in (a)-(h). Taking X25 (conductivity at the third formation stage) as an example, the skewness of the data before transformation is 1.5498, and after the Box-Cox transformation, the skewness is -0.0391. The skewness is close to 0, indicating that the data distribution is relatively symmetric and the normality of the data is significantly improved. In addition, 5 indicators were collected to describe the statistical characteristics of the input variables. The model input features for WV include X25, X77, X24, X43, X31, X52, X74, X62, X51, X73, X42, and X61. The model input features for SC include X25, X77, X31, X24, X74, X43, X52, X51, X73, X42, X62, and X61.

[0107] 5. Analysis of multi-performance prediction results:

[0108] This embodiment comprehensively analyzes the rationality of each component of the stacked model, as Figure 5 shown.

[0109] 5.1. Comparison with the base model:

[0110] To verify the effectiveness of the stacked model in performance prediction, it was compared with seven base models (KNN, RF, ET, GB, AB, XGB, and LGBM models). The comparison results are as Figure 5 (a)-(f) shown. As shown in Table 1 of the performance of the stacked model and the base model, the stacked model shows better performance in predicting WV and SC, with lower error metrics (MAE, MSE, RMSE) and higher R 2 value, having an obvious advantage compared with the base model.

[0111] Table 1

[0112]

[0113] 5.2. Comparison of data processing and model optimization methods based on PCC-MT-BCT-Optuna:

[0114] To verify the ability of the PCC-MT-BCT-Optuna method in improving the performance of the stacked model, relevant analyses were carried out. As Figure 5 (b) and (c) shown, the performance of the stacked models based on different PCC-MT-BCT-Optuna models is shown in the data of Table 2. The stacked model based on PCC-MT-BCT-Optuna is superior to the seven base models in terms of prediction stability and accuracy. Compared with the base models, the model based on PCC-MT-BCT-Optuna shows a significant performance improvement in predicting WV and SC. The reasons are as follows: the best input variables of the prediction model are determined by the PCC-MT method; the normality of the data is improved by BCT; finally, Optuna is used to determine the best parameters of the model. In contrast, the comparison model based on PCC-MT-BCT-Optuna can fully demonstrate the advantages of feature selection, the effectiveness of data processing, and the necessity of multi-parameter optimization. The PCC-MT-BCT-Optuna method effectively improves the accuracy and stability of the model.

[0115] Table 2

[0116]

[0117] 5.3. Comparison results with other stacking strategy models:

[0118] Eight stacking models were selected for the evaluation of accuracy and stability. The calculation results are as Figure 5As shown in (c) and (f), the performance of the stacking model and different stacking models in this embodiment is shown in Table 3. It can be seen that the proposed model is superior to the comparative stacking model in predicting the performance of anodic aluminum foil (WV, SC). Therefore, the stacking strategy (RF-ADA-GB-ET-LGB) improves the accuracy and stability of the stacking model.

[0119] Table 3

[0120]

[0121]

[0122] 6. Model error analysis:

[0123] In this embodiment, the prediction error of the stacking model is analyzed to evaluate the performance of the model. As Figure 6 shown, the error box plot of the stacking model shows that the error distribution is more concentrated and there are fewer outliers, indicating that the model has good performance. By comparing Figure 6 (a), (b), (d), (e), it can be seen that the PCC-MT-BCT-Optuna optimization algorithm significantly reduces the occurrence of error outliers. By comparing Figure 6 (b), (c), (e), (f), it can be seen that the proposed stacking model significantly improves the overall prediction performance of the stacking model. The comparison results of the error box plots of each group show that the optimization algorithm and the stacking strategy are crucial for the effectiveness of the stacking model.

[0124] 7. Prediction error distribution analysis:

[0125] As Figure 7 (a)-(f) shown, by comparing each subplot, it can be clearly seen that as the model is improved, the error becomes more concentrated around 0 and the dispersion degree becomes lower, indicating that the prediction accuracy of the model is continuously improving. Generally speaking, the performance of the stacking model is better than that of the comparative model. Specifically:

[0126] The error distribution of the stacking model is the most concentrated and has the fewest outliers, indicating that its prediction results are the most stable and accurate.

[0127] The error distribution of the basic model optimized based on PCC-MT-BCT-Optuna is also relatively concentrated, but there are still a small number of outliers, indicating that the optimization algorithm has a significant effect on improving the performance of the basic model.

[0128] The error distribution of the unoptimized basic model is relatively dispersed and there are many outliers, indicating that its prediction performance is poor.

[0129] 8. Analysis of the scatter plot of prediction results:

[0130] The scatter plot of the prediction results of the stacked model is as follows Figure 8 (a)-(b). It can be seen that when the WV is at a relatively low level, the prediction error is relatively large; when the SC is at a relatively high level, the prediction error is also relatively large. As shown in Figure 9 , the level of WV depends on the thickness of the oxide film dielectric layer. The value of SC depends on the relative dielectric constant of the dielectric, the plate area, and the plate spacing. As WV decreases, the required thickness of the oxide film also decreases accordingly. However, the decrease in the thickness of the oxide film is accompanied by a decrease in the plate spacing, which leads to an increase in SC. The results are consistent with the actual situation, and the prediction error is within the allowable range. Therefore, the stacked model can be used to evaluate WV and SC, thereby promoting the optimization of process parameters.

[0131] 9. Mechanistic analysis:

[0132] As shown in Figure 9 (a)-(d), the etched foil of the anodic aluminum foil undergoes a multi-stage formation process, and a layer of oxide film dielectric layer that can withstand a specific voltage is formed on its surface. During the chemical synthesis process, hydroxide ions (OH - ) in water enter the film layer under the action of an electric field and combine with aluminum ions (Al 3+ ) to form a porous hydroxide layer. At high temperatures, the porous hydroxide layer dehydrates to form an amorphous alumina layer, and then the amorphous Al2O3 layer is transformed into a crystalline layer through heat treatment. During the transformation process, the density of the oxide film dielectric layer of the anodic aluminum foil changes, and the thickness decreases. The concentration of oxygen ions at the interface between the oxide film and the electrolyte increases, especially in the defect areas of the film. This causes electrons to enter the film layer, and the electric field exerts a force on the ions, accelerating them and hitting the film structure, resulting in local dielectric breakdown and flashover phenomena. The flashover voltage is closely related to the concentration of the electrolyte, so it is necessary to ensure that the electric field voltage is lower than the flashover voltage. The withstand voltage (WV) of the anodic aluminum foil is proportional to the electric field voltage, indicating that the features selected by the model meet the theoretical criteria.

[0133] Example 2:

[0134] This example provides a system for implementing the method for predicting the performance of anodic aluminum foil based on a stacked model, including: a parameter acquisition module and a performance prediction module;

[0135] The parameter acquisition module is used to acquire the target parameters in the production process of the anodic aluminum foil to be detected;

[0136] The performance prediction model is used to input the target parameters in the production process of the anodic aluminum foil to be detected into the anodic aluminum foil performance prediction model to obtain the performance prediction results of the anodic aluminum foil to be detected. Among them, the anodic aluminum foil performance prediction model is constructed based on the stacking of machine learning models and trained based on a training set, and the training set includes the target parameters in the production processes of different specifications of anodic aluminum foils.

[0137] Further, obtaining the training set includes:

[0138] Obtaining the process parameters and performance parameters of anode aluminum foils with different specifications on a normal production line;

[0139] Combining the Pearson correlation coefficient PCC and the multicollinearity test MT to perform data screening on the process parameters and performance parameters, and obtaining the screened data;

[0140] Performing a Box-Cox transformation on the screened data to obtain the target parameters in the production processes of anode aluminum foils with different specifications.

[0141] Further, the anode aluminum foil performance prediction model includes a base learner and a meta-learner;

[0142] The base learner includes Random Forest RF, Gradient Boosting Decision Tree GB, Extra Trees ET, and Adaptive Boosting AB;

[0143] The meta-learner includes Extreme Gradient Boosting XGB and LightGBM.

[0144] Furthermore, training the anode aluminum foil performance prediction model based on the training set includes:

[0145] Inputting the training set into the base learner to obtain the prediction results of Random Forest RF, Gradient Boosting Decision Tree GB, Extra Trees ET, and Adaptive Boosting AB;

[0146] Inputting the prediction results into the meta-learner to obtain the final prediction results.

[0147] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for predicting the performance of anodic aluminum foil based on a stacked model, characterized in that, Including: Obtain the target parameters in the production process of the anodic aluminum foil to be detected; Input the target parameters in the production process of the anodic aluminum foil to be detected into the anodic aluminum foil performance prediction model, and obtain the performance prediction result of the anodic aluminum foil to be detected. Among them, the anodic aluminum foil performance prediction model is constructed based on the stacking of machine learning models and obtained by training based on a training set, and the training set includes the target parameters in the production processes of different specifications of anodic aluminum foils.

2. The method for predicting the performance of an anodic aluminum foil based on a stacked model according to claim 1, wherein Obtaining the training set includes: Obtain the process parameters and performance parameters of different specifications of anodic aluminum foils on a normal production line; Combined with the Pearson correlation coefficient PCC and the multicollinearity test MT, perform data screening on the process parameters and the performance parameters to obtain the screened data; Perform Box-Cox transformation on the screened data to obtain the target parameters in the production processes of different specifications of anodic aluminum foils.

3. The method for predicting the performance of an anodic aluminum foil based on a stacked model according to claim 2, wherein Combined with the Pearson correlation coefficient PCC and the multicollinearity test MT, performing data screening on the process parameters and the performance parameters includes: Using the Pearson correlation coefficient PCC to obtain the linear relationship between the process parameters and the performance parameters; Through the multicollinearity test MT, obtain the variance inflation factor of the process parameters; According to the linear relationship and the variance inflation factor, perform data screening on the process parameters and the performance parameters to obtain the screened data.

4. The method for predicting the performance of an anodic aluminum foil based on a stacked model according to claim 1, wherein The anodic aluminum foil performance prediction model includes a base learner and a meta-learner; The base learner includes Random Forest RF, Gradient Boosting Decision Tree GB, Extra Trees ET, and Adaptive Boosting AB; The meta-learner includes Extreme Gradient Boosting XGB and LightGBM.

5. The method for predicting the performance of an anodic aluminum foil based on a stacked model according to claim 4, wherein Training the anodic aluminum foil performance prediction model based on the training set includes: Input the training set into the base learner to obtain the prediction results of the Random Forest RF, Gradient Boosting Decision Tree GB, Extra Trees ET, and Adaptive Boosting AB; Input the prediction results into the meta-learner to obtain the final prediction results.

6. The method for predicting the performance of an anodic aluminum foil based on a stacked model according to claim 1, characterized in that, The hyperparameters of the anodic aluminum foil performance prediction model are optimized using the Optuna optimization algorithm.

7. The system of the method for predicting the performance of an anodic aluminum foil based on a stacked model according to any one of claims 1-6, characterized in that, A parameter acquisition module and a performance prediction module; The parameter acquisition module is used to obtain the target parameters in the production process of the anodic aluminum foil to be detected; The performance prediction model is used to input the target parameters in the production process of the anodic aluminum foil to be detected into the anodic aluminum foil performance prediction model, and obtain the performance prediction result of the anodic aluminum foil to be detected. Among them, the anodic aluminum foil performance prediction model is constructed based on the stacking of machine learning models and obtained by training based on a training set, and the training set includes the target parameters in the production processes of different specifications of anodic aluminum foils.

8. The system according to claim 7, characterized in that, Obtaining the training set includes: Obtain the process parameters and performance parameters of different specifications of anodic aluminum foils on a normal production line; Combined with the Pearson correlation coefficient PCC and the multicollinearity test MT, perform data screening on the process parameters and the performance parameters to obtain the screened data; Perform Box-Cox transformation on the screened data to obtain the target parameters in the production processes of different specifications of anodic aluminum foils.

9. The system according to claim 7, characterized in that, The anodic aluminum foil performance prediction model includes a base learner and a meta-learner; The base learners include Random Forest (RF), Gradient Boosting (GB), Extra Trees (ET), and Adaptive Boosting (AB); The meta - learner includes Extreme Gradient Boosting (XGB) and Light Gradient Boosting Machine (LGBM).

10. The system according to claim 9, characterized in that, Training the anodic aluminum foil performance prediction model based on the training set includes: Inputting the training set into the base learners to obtain the prediction results of the Random Forest (RF), Gradient Boosting (GB), Extra Trees (ET), and Adaptive Boosting (AB); Inputting the prediction results into the meta - learner to obtain the final prediction result.

Images