Method for regulating and controlling performance of filling type field-induced phase change material

Abstract

The invention belongs to the technical field of filling type field-induced phase-change materials, and discloses a method for regulating and controlling the performance of a filling type field-induced phase-change material. The method comprises the following steps: (1) data set acquisition: acquiring previous experimental data from publicly published literatures, tables and graphs and forming a data set; (2) data preprocessing: executing all preprocessing by using a Python language in the Juputer Notebook, wherein the Python language is used in the Juputer Notebook; standard Scaler is used to carry out standardization on the data set; (3) constructing a machine learning model: constructing a prediction model of the nonlinear conductivity of the filled polymer composite material by adopting support vector regression, kernel ridge regression, random forest regression, XGBoost regression and gradient enhanced regression; and (4) integrating prediction results by adopting a Stacking integration algorithm. According to the method, the SEL strategy is adopted, the performance of the Eb and alpha models is improved by integrating the five basic models, the prediction performance of the models can be improved, the generalization ability of the models can be enhanced, and the models can well express new and unseen data.

Description

Technical Field

[0001] The present invention relates to the technical field of filled-type field-induced phase change materials, and in particular to a prediction model for the performance of filled-type field-induced phase change materials. Background Art

[0002] With the continuous development of intelligent protection systems, especially in the development of intelligent electromagnetic shielding materials that can respond to changes in the external electromagnetic environment, filled polymer composites have shown great potential due to their tunable nonlinear conductivity. However, the nonlinear conductivity of such materials is affected by multiple factors, such as filler type, content, size, and temperature. Therefore, rationally designing the composition and optimizing the material properties are considered key steps in obtaining ideal intelligent electromagnetic shielding materials. In recent years, machine learning and data-driven methods have been widely applied in various fields of science and engineering, especially in material design and performance prediction. For example, machine learning algorithms can analyze complex relationships between composition and properties and help identify optimal material combinations. Furthermore, artificial intelligence technology provides the so-called "fourth paradigm" for scientific discovery by reducing the need for labor-intensive and costly trial-and-error experiments. Although machine learning techniques have demonstrated their superiority over traditional physical modeling methods in many materials science studies, research on filled nonlinear conductive composites for strong electromagnetic field intelligent shielding applications remains insufficient. To promote progress in this field, it is necessary to integrate machine learning techniques into materials engineering to more effectively design filled nonlinear conductive composites with specific nonlinear conductivity properties. Summary of the Invention

[0003] To solve the above technical problems, the present invention provides a prediction model for the performance of filled field-induced phase change materials. By collecting previous experimental data and forming a data set, a stable and effective prediction model for the performance of filled field-induced phase change materials is established, providing a feasible control method for the field-induced nonlinear conductive properties of filled composite materials.

[0004] The specific technical solution is as follows: A method for controlling the performance of a filled field-induced phase change material comprises the following steps:

[0005] (1) Dataset acquisition: Collect previous experimental data from publicly published literature, tables, and graphs to form a dataset, which includes classification features, numerical features, and target features;

[0006] (2) Data preprocessing: All preprocessing was performed in Python in Jupyter Notebook. The categorical features were converted into binary features represented by "0" and "1" using the one-hot encoding method. The mean strategy in SimpleImputer was used to handle missing data. The dataset was standardized using StandardScaler.

[0007] (3) Constructing machine learning models: using support vector regression (SVR), kernel ridge regression (KRR), random forest regression (RFR), XGBoost regression (XGB) and gradient boosting regression (GBT) to construct prediction models for the nonlinear conductive properties of filled polymer composites;

[0008] (4) The Stacking integration algorithm is used to integrate the prediction results.

[0009] Preferably, in step (1), a Digitizer drawing tool is used to convert graphic data that cannot be directly analyzed into numerical data.

[0010] Preferably, in step (1), a violin plot and a scatter plot are used to visualize the data set. Before training the machine learning model, the visualization analysis of the data set can help us understand the distribution and characteristics of the data and reveal the relationship between the characteristics and the target parameters (E b and α).

[0011] Preferably, in step (1), the classification features include filler, matrix and preparation process; the numerical features include filler size, mass fraction and sample thickness; the target features include the key parameter E for evaluating nonlinear conductivity characteristics. b and α. Switching field E b It represents the critical electric field strength at which the material begins to exhibit nonlinear characteristics. This parameter is of great significance for determining the application range of the material. Since it is difficult to accurately identify the exact point where the current-electric field (JE) curve changes from linear to nonlinear in experiments, the intersection of the linear region and the nonlinear region is usually taken as E b The nonlinear coefficient α is calculated by fitting the logarithmic relationship between current J and electric field E and is a key parameter for evaluating nonlinear conductivity characteristics:

[0012] α=(logJ2-logJ1) / (logE2-logE1)

[0013] Where E1 and E2 correspond to the electric fields of J1 and J2 respectively.

[0014] Preferably, in step (3), R 2 , MAE and RMSE three indicators for cross-validation, the data set is randomly divided into training set and test set, and this process is repeated 50 times. Each division will generate a new set of training set and test set. The model is trained and tested on these 50 different data divisions, and the evaluation index is calculated each time. The average value of these 50 evaluations is taken as the final evaluation result of the model performance.

[0015] Further preferably, in step (3), 80% of the samples in the data set are used as a training set, and 20% are used as a test set.

[0016] The beneficial effects of the present invention are:

[0017] The present invention adopts the SEL strategy and improves E by integrating five basic models (SVR, KRR, XGB, RFR and GBT). b The performance of the α model can not only improve the prediction performance of the model, but also enhance the generalization ability of the model, so that it can perform well on new and unseen data.

[0018] The present invention collects previous experimental data and forms a data set to establish a stable and effective performance prediction model for filled field-induced phase change materials, providing a feasible control method for the field-induced nonlinear conductive properties of filled composite materials.

[0019] Figures in the specification

[0020] Figure 1 A diagram of the machine learning workflow for the performance of filled field-induced phase change materials;

[0021] Figure 2a For packing type-E b Violin plot of ;

[0022] Figure 2b For base type -E b Violin plot of ;

[0023] Figure 2c Preparation process-E b Violin plot of ;

[0024] Figure 3a is the mass fraction -E b Scatter plot of

[0025] Figure 3b Packing size - E b Scatter plot of

[0026] Figure 3c Thickness-E b Scatter plot of

[0027] Figure 4a Violin plot for filler type-α;

[0028] Figure 4b Violin plot for matrix type-α;

[0029] Figure 4c is the violin plot of preparation process-α;

[0030] Figure 5a is the scatter plot of mass fraction-α;

[0031] Figure 5b is the scatter plot of packing size-α;

[0032] Figure 5c is the scatter plot of thickness-α;

[0033] Figure 6 For E b and statistical analysis graphs of α;

[0034] Figure 7 is the Pearson correlation coefficient graph;

[0035] Figure 8a For E b Comparison chart of experimental and predicted values;

[0036] Figure 8b A comparison chart of the experimental and predicted values ​​of α. DETAILED DESCRIPTION

[0037] The present invention will be further described in detail below with reference to specific embodiments.

[0038] A method for controlling the performance of a filled field-induced phase change material comprises the following steps (the machine learning workflow for the performance of a filled field-induced phase change material is shown in Figure 1 ):

[0039] (1) Obtaining a dataset

[0040] To ensure data accuracy and completeness, we collected data from multiple sources, including published literature, tables, and figures. Because some graphical data could not be directly analyzed, we used the Digitizer plotting tool to convert them into usable numerical data. Ultimately, we successfully extracted detailed feature information for 285 samples from 59 papers for use in model training.

[0041] Each sample in this dataset contains rich feature information and corresponding target values. The target features include the key parameter E for evaluating nonlinear conductivity characteristics. b and α; graphene, carbon nanotubes (CNT), zinc oxide (ZnO) and silicon carbide (SiC) were selected as filler variables for classification features, while epoxy resin (EP), silicone rubber (SiR), polydimethylsiloxane (PDMS) and polyethylene (PE) were selected as polymer matrices of the composite materials; melt blending and solution mixing were selected as the preparation process; numerical features included filler size, mass fraction and sample thickness;

[0042] The violin plot and scatter plot are used to visualize the E bThe distribution probability of and α on different preparation parameter characteristics. The violin plot is used to analyze the E b The relationship between the values ​​and classification characteristics (including filler type, matrix type and preparation process) is visualized in Figures 2a-2c ; The analysis showed that the E of samples containing SiC and ZnO b The value is significantly higher than that of other samples; among carbon-based materials, graphene is found to have a wider range of applications than carbon nanotubes, and carbon nanotubes E b The distribution of values ​​is relatively concentrated; the comparison of polymer matrices shows that the samples containing EP cover the widest range of E b Value range, for the most widely used matrix material, in comparison, PE's E b The value is relatively low, while the E b The value is relatively high. b The influence of the value is relatively weak, indicating that the intrinsic properties of the material (composition and microstructure) are the dominant regulatory factors of the conductive properties.

[0043] Use scatter plot to compare E b The relationship between the values ​​and numerical features (including mass fraction, filler size and thickness) is visualized in Figures 3a-3c The results showed that E b There is a negative correlation between the value and the mass fraction and filler size, that is, as the mass fraction and filler size increase, the E b The value shows a downward trend; in contrast, E b The values ​​do not show a significant dependence on the sample thickness. The analysis results show that mass fraction and filler size are key characteristic parameters and should be given priority when optimizing the nonlinear conductive properties of composites.

[0044] Violin plots were used to visualize the relationship between α values ​​and classification features (including filler type, matrix type, and preparation process). Figures 4a-4c The influence of different fillers and matrix materials on the α value can be observed intuitively. The research results show that the α value range of ZnO is relatively wide, ranging from 0.46 to 22.6, indicating that ZnO can maintain a stable nonlinear response in a wide electric field range. Its wide-range nonlinear response characteristics are derived from its piezoresistive effect, making ZnO a strong candidate material for designing composite materials with specific nonlinear characteristics. In contrast, the α value distributions of graphene, SiC, and CNT are relatively consistent. Among the four matrix materials, SiR exhibits a higher nonlinear conductivity coefficient α. As a typical nonlinear resistance field gradient material, this characteristic enables it to effectively suppress the local enhancement of the electric field.

[0045] The relationship between α value and numerical features (including mass fraction, filler size and thickness) was visualized using scatter plots. Figures 5a-5cThe results clearly demonstrate the significant influence of three numerical characteristics, namely filler size, mass fraction, and sample thickness, on the α value. Specifically, the α value is positively correlated with the filler size and mass fraction. As the filler size and mass fraction increase, the α value also increases, indicating that large-sized fillers and high filling ratios can enhance the nonlinear response characteristics through percolation network density. The α value is negatively correlated with sample thickness, meaning that the thinner the sample, the higher the α value. This may be because thicker samples face more significant restrictions on electron transport under the action of an electric field, resulting in a weakening of their nonlinear conductivity characteristics, while the thin layer structure has a focusing effect on the distribution of the interfacial field strength.

[0046] E of nonlinear conductive composite materials b and α for statistical analysis, such as Figure 6 As shown, E b The distribution range of the value is relatively wide, from 0.05kV / mm to 6.4kV / mm, which provides a control range for the adaptability design of the material in a wide field strength range. According to the specific application requirements and electric field conditions, the material with appropriate E b The material with the best nonlinear characteristics under the expected working conditions is selected. The value range of variable α is 0.43 to 22.6, covering the nonlinear characteristics from low to high. In practical applications, the ideal value of α should be at least greater than 10, because when the value of α exceeds 10, the effect of further increasing α on improving material performance will become less significant. b The negative correlation with α reveals the key design rules. It can be clearly seen from the figure that the high α value area is concentrated in the low E b interval, and E b The improvement of α is often accompanied by the attenuation of α, which requires material engineering to find a balance between nonlinear sensitivity and switching field threshold. b -The lower right area of ​​the α graph (low E b / high α), there may be an advantageous material system with both fast field-induced response and low switching field strength, which provides an optimal direction for the research and development of intelligent electromagnetic protection materials.

[0047] To further analyze the relationship between features, the Pearson correlation coefficient of each feature pair was calculated, as shown in Figure 7(Pearson correlation coefficient graph) is shown. The Pearson correlation coefficient is a statistical indicator that measures the degree of linear correlation between two variables, and its value ranges from -1 to 1. When the absolute value of the coefficient is ≥0.9, it is determined to be a strongly correlated feature pair. At this time, redundant features need to be eliminated to avoid multicollinearity problems in the model, confirming the independence and complementarity of the feature set and the absence of information redundancy. During the feature engineering stage, 14 input features were used to build the machine learning model. Analysis of the correlation between these features showed that the correlation between these 14 features was low, and there were no repeated or redundant features, indicating that each feature provides independent information to the model, thereby helping to improve the model's predictive accuracy and explanatory power.

[0048] (2) Data preprocessing

[0049] All preprocessing is performed in Python in Jupyter Notebook. The one-hot encoding method is used to convert categorical features into binary features represented by "0" and "1", and the original category feature labels are converted into a numerical form that is easier for machine learning models to handle. Specifically, if the sample belongs to a certain category, the feature value corresponding to the category is "1", otherwise it is "0".

[0050] The mean strategy in SimpleImputer is used to handle missing data. The mean strategy is a commonly used missing value filling method. It calculates the average of all non-missing values ​​and fills the missing values ​​with the average, thereby avoiding the information loss that may be caused by directly deleting samples containing missing values.

[0051] After processing the missing values, StandardScaler is used to standardize the dataset to ensure the scale consistency of different features and avoid some features from occupying too large a weight in model training due to their large numerical range.

[0052] Then 80% of the samples are used to train the model, and the remaining 20% ​​is used as a test set to evaluate the performance of the model.

[0053] (3) Build a machine learning model

[0054] A prediction model for the nonlinear conductive properties of filled polymer composites was constructed using support vector regression (SVR), kernel ridge regression (KRR), random forest regression (RFR), XGBoost regression (XGB), and gradient boosting regression (GBT). The dataset was randomly divided into training and test sets, and this process was repeated 50 times. Each partition generated a new training and test set. The model was trained and tested on these 50 different data partitions, and evaluation metrics were calculated for each iteration. The average of these 50 evaluations was used as the final evaluation result to reduce bias caused by random data partitioning.

[0055] (4) The Stacking integration algorithm is used to integrate the prediction results.

[0056] The present invention adopts five basic models, namely support vector regression (SVR), kernel ridge regression (KRR), random forest regression (RFR), XGBoost regression (XGB) and gradient boosting regression (GBT), to construct a prediction model for the nonlinear conductive properties of filled polymer composites. These models are all derived from popular machine learning libraries and have their own unique advantages and characteristics. The main advantage of the Stacking ensemble algorithm (SEL) is that it can integrate the predictive capabilities of different models, so that the final prediction results are more stable and accurate. Generally, the SEL strategy adopts a two-layer learning framework: a base learner layer and a meta-learner layer. The base learner is located in the first layer and is responsible for learning and making predictions from the original data; while the meta-learner is located in the second layer and is responsible for learning how to best combine the outputs of the first layer learners to generate the final predictions. The core idea of ​​this method is to first train the base learner on the original data, then build a new data set based on its prediction output, and finally train the meta-learner on the new data set to generate the final prediction results.

[0057] In order to more accurately evaluate the quality of different models, a comprehensive comparison of the prediction performance of five basic machine learning algorithms and SEL strategy on the test set was conducted, using RMSE, MRE and R 2 Three key indicators are used to evaluate the performance of the model in predicting the E b The accuracy and reliability of the α and α features are shown in Table 1.

[0058] Table 1 Evaluation of five basic machine learning models and SEL model

[0059]

[0060] From Table 1, we can see that in E b In the prediction of features, the R 2The value of 0.816 shows the highest accuracy, indicating that this model can best capture the data characteristics among all basic models. In contrast, the prediction accuracy of the KRR and SVR models is lower, which may indicate that these two models have certain limitations in dealing with the complex nonlinear relationships of this dataset. Similarly, in the prediction of the α feature, the random forest regression (RFR) model has an R value of 0.849. 2 The value is better than KRR and SVR models. This result further verifies that the decision tree ensemble learning algorithm is more suitable for the modeling needs of nonlinear relationships in this data set compared with the nonlinear models KRR and SVR. Among these basic models, E b and α obtained the best R on the test set respectively. 2 The values ​​are 0.816 and 0.849 respectively, which can fit the data well. It is worth noting that the SEL strategy can predict the results of E b The accuracy of E b and R on the test set 2 The values ​​reached 0.826 and 0.853, respectively, indicating that the SEL model has a stronger nonlinear correlation between components and properties through multi-model synergy, which improves the prediction accuracy of the model.

[0061] Figure 8a-Figure 8b The prediction results of the training and test data based on the SEL strategy are intuitively displayed. The dotted line represents the best fit line, which is used to evaluate the consistency between the predicted value and the actual value. The closer the point on the scatter plot is to the diagonal line, the closer the model's prediction result is to the actual value. b Prediction Model( Figure 8a ) and α prediction model ( Figure 8b ) are highly clustered near the diagonal, further demonstrating the feasibility and accuracy of the SEL model in predicting the properties of nonlinear conductive composites.

[0062] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for controlling the performance of a filled field-induced phase change material, characterized in that It includes the following steps: (1) Dataset acquisition: Collect previous experimental data from publicly published literature, tables, and graphs to form a dataset, which includes classification features, numerical features, and target features; (2) Data preprocessing: All preprocessing was performed in Python in Jupyter Notebook. One-hot encoding was used to convert categorical features into binary features represented by "0" and "1". The mean strategy in SimpleImputer was used to handle missing data. StandardScaler was used to standardize the dataset. (3) Constructing machine learning models: Using support vector regression, kernel ridge regression, random forest regression, XGBoost regression, and gradient boosting regression, a prediction model for the nonlinear conductive properties of filled polymer composites was constructed; (4) The Stacking integration algorithm is used to integrate the prediction results.

2. The method for controlling the performance of a filled field-induced phase change material according to claim 1, characterized in that: In the step (1), the Digitizer drawing tool is used to convert the graphic data that cannot be directly analyzed into numerical data.

3. The method for controlling the performance of a filled field-induced phase change material according to claim 1, wherein: In the step (1), a violin plot and a scatter plot are used to perform a visual analysis of the data set.

4. The method for controlling the performance of a filled field-induced phase change material according to claim 1, wherein: In step (1), the classification features include filler, matrix and preparation process; the numerical features include filler size, mass fraction and sample thickness; the target features include the key parameter E for evaluating nonlinear conductivity characteristics. b and α.

5. The method for controlling the performance of a filled field-induced phase change material according to claim 1, wherein: In step (3), the data set is randomly divided into a training set and a test set, and this process is repeated 50 times. Each division will generate a new set of training sets and test sets. The model is trained and tested on these 50 different data divisions, and the evaluation index is calculated each time. The average value of these 50 evaluations is taken as the final evaluation result of the model performance.

6. The method for controlling the performance of a filled field-induced phase change material according to claim 5, characterized in that: In step (3), 80% of the samples in the data set are used as training sets, and 20% are used as test sets.

Images