Polypropylene film capacitor dielectric material molecular weight distribution prediction method and device based on machine learning
By constructing a fusion machine learning model, the problem of predicting the molecular weight distribution of polypropylene film capacitor dielectric materials in existing technologies has been solved, achieving rapid and accurate molecular weight distribution prediction and improving the performance optimization capability of film capacitors.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-29
- Publication Date
- 2026-03-27
AI Technical Summary
Existing technologies make it difficult to quickly and accurately predict the molecular weight distribution of polypropylene film capacitor dielectric materials, which affects the capacitor's voltage rating, reliability, and processing performance.
We employ three machine learning algorithms—gradient boosting regression, AdaBoost regression, and bagging regression—combined with Bayesian optimization to adjust hyperparameters and construct a fusion model to achieve rapid and accurate prediction of the molecular weight distribution of polypropylene film capacitor dielectric materials.
This method enables rapid and accurate prediction of the molecular weight distribution of polypropylene film capacitor dielectric materials, overcoming the limitations of traditional methods and providing theoretical support for the performance optimization of film capacitors.
Smart Images
Figure CN121747766A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of molecular weight distribution prediction technology, and in particular to a machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials and a machine learning-based device for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials. Background Technology
[0002] Film capacitors are widely used in power electronics, new energy vehicles, and photovoltaic power generation due to their excellent electrical properties and broad application prospects. Polypropylene (PP) film, as the core dielectric material of film capacitors, has a significant impact on its molecular weight distribution (MWD) regarding voltage withstand rating, reliability, and processing performance. Studies have shown that the molecular weight distribution of PP is closely related to its crystallization behavior, rheological properties, mechanical properties, and electrical properties. While an excessively wide molecular weight distribution is beneficial for processing, it leads to reduced film strength; conversely, an excessively narrow molecular weight distribution, while improving film strength, reduces processing performance. To balance mechanical and processing properties, a molecular weight distribution of PP between 4 and 6 is ideal. Therefore, precisely controlling the molecular weight distribution of PP is crucial for optimizing the performance of film capacitors.
[0003] However, traditional methods for determining molecular weight distribution (such as gel permeation chromatography) are time-consuming and cannot achieve real-time monitoring. With the development of machine learning technology, it has become possible to predict material properties using big data and algorithmic models. Currently, there is no efficient method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials. Summary of the Invention
[0004] The purpose of this invention is to provide a machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials to at least solve one of the aforementioned technical problems.
[0005] This invention provides the following solution:
[0006] According to one aspect of the present invention, a method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning is provided, the method comprising:
[0007] Obtain the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model, respectively.
[0008] Obtain the input variable information;
[0009] The input variable information is respectively input into one or more of the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model to obtain the molecular weight distribution information output by the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model.
[0010] Optionally, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes:
[0011] Obtain the training dataset;
[0012] Obtain a gradient boosting regression model;
[0013] The gradient boosting regression model is trained using a training dataset and its hyperparameters are adjusted using a Bayesian optimization method to obtain the trained gradient boosting regression model.
[0014] Optionally, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes:
[0015] Obtain the training dataset;
[0016] Obtain the AdaBoost regression model;
[0017] The AdaBoost regression model is trained using a training dataset and its hyperparameters are adjusted using a Bayesian optimization method to obtain the trained AdaBoost regression model.
[0018] Optionally, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes:
[0019] Obtain the training dataset;
[0020] Obtain the bagging regression model;
[0021] The bagging regression model is trained using a training dataset, and its hyperparameters are adjusted using a Bayesian optimization method to obtain the trained bagging regression model.
[0022] Optionally, the optimal hyperparameters of the trained gradient boosting regression model are a learning rate of 0.20, a maximum depth of 5, a number of base learners of 60, and a random seed of 10.
[0023] Optionally, the optimal hyperparameters of the trained AdaBoost regression model are a learning rate of 0.05, a base learner count of 77, and a random seed of 22.
[0024] Optionally, the optimal hyperparameters of the trained bagging regression model are 25 base learners and 26 random seeds.
[0025] Optionally, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes:
[0026] Meta-features are generated based on the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model.
[0027] Obtain the secondary model;
[0028] The secondary model is trained using meta-features, wherein the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model generate meta-features as base models, and the secondary model and the base models form a fusion model.
[0029] The input variable information is fed into the trained fusion model to obtain the molecular weight distribution information output by the fusion model.
[0030] Optionally, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes:
[0031] The model is evaluated using cross-validation to obtain the model with the best performance.
[0032] This application also provides a machine learning-based device for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials, the machine learning-based device for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials includes:
[0033] The model acquisition module is used to acquire the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model, respectively.
[0034] A variable information acquisition module, which is used to acquire input variable information;
[0035] A molecular weight distribution information acquisition module is used to input the input variable information into one or more of the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model, thereby acquiring the molecular weight distribution information output by the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model.
[0036] The machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials in this application achieves rapid and accurate prediction of the molecular weight distribution of polypropylene by constructing and optimizing a machine learning model, providing theoretical support for the performance optimization of film capacitors. Attached Figure Description
[0037] Figure 1 This is a flowchart illustrating a machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials in one embodiment of this application.
[0038] Figure 2 This is a fitting graph of the predicted and actual values of the GBR model in one embodiment of this application.
[0039] Figure 3 This is a fitting graph of the ABR model prediction and the actual value in one embodiment of this application.
[0040] Figure 4 This is a fitting graph of the predicted and actual values of the BAGR model in one embodiment of this application.
[0041] Figure 5 This is a heatmap of the correlation coefficients between input and output variables in one embodiment of this application.
[0042] Figure 6 This is a SHAP analysis diagram from one embodiment of this application.
[0043] Figure 7 This is a schematic diagram of the input and output features in one embodiment of this application. Detailed Implementation
[0044] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0045] like Figure 1 The machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials shown includes:
[0046] Obtain the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model, respectively.
[0047] Obtain the input variable information;
[0048] The input variable information is respectively input into one or more of the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model to obtain the molecular weight distribution information output by the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model.
[0049] In this embodiment, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes:
[0050] Obtain the training dataset;
[0051] Obtain a gradient boosting regression model;
[0052] The gradient boosting regression model is trained using a training dataset and its hyperparameters are adjusted using a Bayesian optimization method to obtain the trained gradient boosting regression model.
[0053] In this embodiment, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes:
[0054] Obtain the training dataset;
[0055] Obtain the AdaBoost regression model;
[0056] The AdaBoost regression model is trained using a training dataset and its hyperparameters are adjusted using a Bayesian optimization method to obtain the trained AdaBoost regression model.
[0057] In this embodiment, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes:
[0058] Obtain the training dataset;
[0059] Obtain the bagging regression model;
[0060] The bagging regression model is trained using a training dataset, and its hyperparameters are adjusted using a Bayesian optimization method to obtain the trained bagging regression model.
[0061] In this embodiment, the optimal hyperparameters of the trained gradient boosting regression model are a learning rate of 0.20, a maximum depth of 5, a number of base learners of 60, and a random seed of 10.
[0062] In this embodiment, the optimal hyperparameters of the trained AdaBoost regression model are a learning rate of 0.05, a base learner count of 77, and a random seed of 22.
[0063] In this embodiment, the optimal hyperparameters of the trained bagging regression model are 25 base learners and 26 random seeds.
[0064] In this embodiment, the data source can be obtained in the following ways:
[0065] Experimental data on the catalytic polymerization of polypropylene were collected from published literature both domestically and internationally, totaling 233 sets. The data cover different polymerization conditions, including reaction temperature, reaction pressure, and catalyst dosage.
[0066] In this embodiment, the data sources are divided into the training dataset and the test set as described above.
[0067] In this embodiment, the dataset is divided into a training dataset (80%) and a test dataset (20%), and the model is trained using the training dataset. The dataset is divided using the `train_test_split` function.
[0068] See Figure 7 In this embodiment, the input variable information includes: metal center (X1), polymerization temperature (X2, unit: °C), polymerization pressure (X3, unit: MPa), polymerization time (X4, unit: min), catalyst dosage (X5, unit: μmol), co-catalyst dosage (X6, unit: mmol), ratio of co-catalyst to catalyst (X7), catalytic activity (X8, unit: kg PE / mmolcat·h), number-average molecular weight (X9, unit: g / mol), and yield (X10, unit: g).
[0069] In this embodiment, Bayesian optimization is used to adjust the model's hyperparameters to determine the optimal parameter combination. Specific parameters include the learning rate, the number of base learners, and the maximum depth. Bayesian optimization is implemented using the scikit-learn library in Python, and the optimization process is as follows:
[0070] Gradient boosting regression (GBR): The optimal hyperparameters are learning rate 0.20, maximum depth 5, number of base learners 60, and random seed 10.
[0071] AdaBoost Regression (ABR): The optimal hyperparameters are learning rate 0.05, number of base learners 77, and random seed 22.
[0072] Bag Regression (BAGR): The optimal hyperparameters are 25 base learners and 26 random seeds.
[0073] In this embodiment, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes:
[0074] The model was evaluated using a 10th-order cross-validation method to ensure its robustness and generalization ability.
[0075] Evaluation metrics: Calculate the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). 2 Indicators such as ) are used to verify the accuracy and generalization ability of the model.
[0076] Gradient Boosting Regression (GBR): R 2 =0.9380, MAE=0.1545.
[0077] AdaBoost Regression (ABR): R 2 =0.8075, MAE=0.3079.
[0078] Bagging Regression (BAGR): R 2 =0.8076, MAE=0.2336.
[0079] See Figures 2 to 4 As can be seen from the results in the figure, the gradient boosting regression model exhibits the best predictive performance, with the highest good fit between the predicted and actual values and the smallest error.
[0080] In this embodiment, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes:
[0081] See Figure 5 We used the seaborn library in Python to draw a correlation heatmap between features and identify linear relationships between them. The results showed that the highest correlation coefficient, 0.68, was between catalyst dosage (X5) and catalytic activity (X8).
[0082] See Figure 6 By combining the SHapley Additive exPlanations (SHAP) method, the impact of each feature on the prediction results was analyzed in depth. The SHAP value can quantify the contribution of each feature to the model prediction results. The results show that the co-catalyst / catalyst ratio (X7), central metal (X1), catalytic activity (X8), and polymerization pressure (X3) have the greatest impact on the prediction results of molecular weight distribution.
[0083] Using the above technical solution, 1. a method for predicting the molecular weight distribution of polypropylene is proposed by combining machine learning technology with polymer materials science. This method overcomes the limitations of traditional experimental methods, enabling rapid and accurate prediction of the molecular weight distribution of polypropylene, and providing a new technical means for optimizing the performance of thin-film capacitor dielectric materials;
[0084] Simultaneously, three machine learning algorithms—Gradient Boosting Regression (GBR), AdaBoost Regression (ABR), and Bagging Regression (BAGR)—were used for modeling, and the hyperparameters were finely tuned using Bayesian optimization. This multi-algorithm comparison and optimization strategy ensured the model's high accuracy and generalization ability.
[0085] In another embodiment, the machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials of this application further includes:
[0086] Meta-features are generated based on the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model.
[0087] Obtain the secondary model;
[0088] The secondary model is trained using meta-features, wherein the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model generate meta-features as base models, and the secondary model and the base models form a fusion model.
[0089] The input variable information is fed into the trained fusion model to obtain the molecular weight distribution information output by the fusion model.
[0090] In this embodiment, the data in the training set is input into a trained gradient boosting regression model, a trained AdaBoost regression model, and a trained bagging regression model to obtain prediction results, and the prediction results are concatenated to form meta-features.
[0091] Train a secondary model (such as linear regression) using meta-features and true labels.
[0092] Validate the performance of the fusion model using a test set, and calculate MAE, RMSE, and R. 2 Indicators such as these.
[0093] By fusing GBR, ABR, and BAGR models using the Stacking method, prediction performance can be significantly improved. In practical applications, the model structure and parameters need to be adjusted according to specific data characteristics to achieve the best results.
[0094] This application also provides a machine learning-based device for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials. The machine learning-based device includes a model acquisition module, a variable information acquisition module, and a molecular weight distribution information acquisition module.
[0095] The model acquisition module is used to acquire the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model, respectively.
[0096] The variable information acquisition module is used to acquire the input variable information;
[0097] The molecular weight distribution information acquisition module is used to input the input variable information into one or more of the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model, respectively, so as to obtain the molecular weight distribution information output by the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model.
[0098] It should be noted that the foregoing explanation of the method embodiments also applies to the system of this embodiment, and will not be repeated here.
[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning, characterized in that, The machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials includes: Obtain the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model, respectively. Obtain the input variable information; The input variable information is respectively input into one or more of the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model to obtain the molecular weight distribution information output by the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model.
2. The method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning as described in claim 1, characterized in that, The machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes: Obtain the training dataset; Obtain a gradient boosting regression model; The gradient boosting regression model is trained using a training dataset and its hyperparameters are adjusted using a Bayesian optimization method to obtain the trained gradient boosting regression model.
3. The method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning as described in claim 1, characterized in that, The machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes: Obtain the training dataset; Obtain the AdaBoost regression model; The AdaBoost regression model is trained using a training dataset and its hyperparameters are adjusted using a Bayesian optimization method to obtain the trained AdaBoost regression model.
4. The method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning as described in claim 1, characterized in that, The machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes: Obtain the training dataset; Obtain the bagging regression model; The bagging regression model is trained using a training dataset, and its hyperparameters are adjusted using a Bayesian optimization method to obtain the trained bagging regression model.
5. The method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning as described in claim 2, characterized in that, The optimal hyperparameters of the trained gradient boosting regression model are a learning rate of 0.20, a maximum depth of 5, a number of base learners of 60, and a random seed of 10.
6. The method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning as described in claim 3, characterized in that, The optimal hyperparameter of the trained AdaBoost regression model is the learning rate. 0.05, number of base learners 77, random seed 22.
7. The method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning as described in claim 4, characterized in that, The optimal hyperparameters of the trained bagging regression model are 25 base learners and 26 random seeds.
8. The method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning as described in claim 1, characterized in that, The machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes: Meta-features are generated based on the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model. Obtain the secondary model; The secondary model is trained using meta-features, wherein the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model generate meta-features as base models, and the secondary model and the base models form a fusion model. The input variable information is fed into the trained fusion model to obtain the molecular weight distribution information output by the fusion model.
9. The method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning as described in any one of claims 1, 2, 3, 4, and 8, characterized in that, The machine learning-based method for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials further includes: The model is evaluated using cross-validation to obtain the model with the best performance.
10. A device for predicting the molecular weight distribution of polypropylene film capacitor dielectric materials based on machine learning, characterized in that, The machine learning-based polypropylene film capacitor dielectric material molecular weight distribution prediction device includes: The model acquisition module is used to acquire the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model, respectively. A variable information acquisition module, which is used to acquire input variable information; A molecular weight distribution information acquisition module is used to input the input variable information into one or more of the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model, thereby acquiring the molecular weight distribution information output by the trained gradient boosting regression model, the trained AdaBoost regression model, and the trained bagging regression model.