Carbon nanotube film performance prediction method based on WGAN-GP and Super Learner ensemble learning

Abstract

The invention discloses a method for predicting the performance of a carbon nano tube film based on WGAN-GP and Super Learner ensemble learning. According to the method, the high-quality synthetic data is generated through the WGAN-GP, so that the training data set is expanded. And then, a Super Learner ensemble learning model is utilized to combine a plurality of regression models, so that the precision and generalization ability of performance prediction are improved. According to the model, through a multi-level learning structure, the problems of data scarcity and noise are effectively solved while efficient training is kept. The method not only can accurately predict the performance of the carbon nanotube film, but also has high expandability, and is suitable for predicting and analyzing the performance of other materials. According to the invention, the limitation of processing small sample and high noise data in the prior art is overcome, and an efficient and reliable performance prediction tool is provided for the fields of material science and engineering.

Description

Technical Field

[0001] The present invention relates to the field of carbon nanotube film performance prediction, in particular to a method for predicting the water permeability of carbon nanotube films based on generative adversarial networks (WGAN-GP) and Super Learner ensemble learning. The method can effectively predict the water permeability of carbon nanotube films under different conditions by expanding and optimizing the experimental data of water permeation of carbon nanotube films and combining multiple machine learning algorithms, and belongs to the intersection of materials science, nanotechnology and machine learning. Background Art

[0002] Carbon nanotube membranes have broad application prospects in water treatment, filtration and membrane separation technology, especially in terms of water permeability. Carbon nanotube membranes have gradually become a hot spot for research and application due to their excellent nanoscale channel structure and high flux characteristics. However, since carbon nanotube membranes exhibit different permeability under different preparation conditions, operating environments and usage conditions, how to accurately predict their water permeability has become a key issue. At present, the prediction of water permeability mainly relies on experimental data and physical models. However, due to the high cost of experimental data acquisition, long experimental cycle, and the fact that only small-scale samples can be obtained in experiments, existing methods have problems such as insufficient accuracy and poor applicability when facing complex working conditions and small sample data. Summary of the invention

[0003] In order to solve the problems existing in the background technology, the present invention combines limited experimental data and multiple machine learning algorithms to perform performance prediction in order to improve the prediction accuracy of the model. This method effectively improves the prediction ability under small sample data by mining and modeling data. The technical solution of the present invention is implemented as follows:

[0004] The present invention provides a method for predicting the water permeability of carbon nanotube membranes based on WGAN-GP and Super Learner integrated learning, aiming to solve the problem of inaccurate performance prediction due to insufficient sample size in the prior art. The original experimental data is expanded by a generative adversarial network (WGAN-GP), and the water permeability of carbon nanotube membranes is predicted with high precision in combination with a Super Learner model. The present invention has a significant improvement in the field of water permeability prediction of carbon nanotube membranes, especially under the condition of a small sample data set, by combining data expansion with an integrated learning model, the accuracy and stability of the prediction are significantly improved.

[0005] In order to achieve the above object, the technical solution of the present invention is achieved as follows:

[0006] A carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning includes the following steps:

[0007] Step 1: Build a data set based on the carbon nanotube film experimental data and perform feature extraction; build a WGAN-GP generative adversarial network model, optimize the network structure of the generator and discriminator through gradient penalty, and ensure the effectiveness of the model.

[0008] Step 2: Use the trained WGAN-GP model to generate a new data set, preprocess the newly generated data set, merge the processed data with the original data set, and then divide it into training set and test set.

[0009] Step 3: Use the Super Learner ensemble learning model to build a two-layer regression architecture containing multiple base learners and meta-learners, and perform model training and parameter optimization for the extended data set.

[0010] Step 4: According to the model evaluation index, the performance of the Super Learner model is evaluated, and the performance data of the carbon nanotube film to be tested is predicted using the trained model, and the prediction results are output.

[0011] Preferably, based on the experimental data of carbon nanotube films, a generative adversarial network model WGAN-GP is constructed, wherein the generator is used to generate new synthetic data, the discriminator is used to distinguish the generated data from the real data, and the Wasserstein distance is optimized by introducing a gradient penalty to ensure that the discriminator satisfies Lipschitz continuity and avoids gradient vanishing or exploding phenomena; WGAN-GP is used to generate additional synthetic data for expanding the data set, and the generated data is combined with the original experimental data to form a new set of data sets.

[0012] Preferably, the generator of the WGAN-GP generative adversarial network is a multi-layer fully connected neural network, and the input is a standard normal distribution. The noise vector sampled in the network, each hidden layer in the network uses the ReLU activation function, expressed as h i =max(0,W i h i-1 +b i ), where W i is the weight matrix, b i is the bias vector, h i is the output of each layer; the output layer uses the Softplus activation function Softplus(x)=ln(1+e x ), x is the intermediate result of the generator network output to ensure that the generated data is positive and consistent with the physical properties of the carbon nanotube film; the generator minimizes the loss function in To generate data, is the evaluation value of the discriminator on the generated data, E is the expected value of the noise distribution z, and the network parameters are updated to make the generated data gradually approach the real data distribution.

[0013] Preferably, during the training process of the generative adversarial network, a gradient penalty term is used to ensure the Lipschitz continuity of the discriminator. The discriminator structure is a multi-layer neural network, and each layer uses the LeakyReLU activation function LeakyReLU(x)=max(0.01x,x), which effectively prevents the gradient vanishing problem. In addition, the discriminator minimizes the loss function at each weight update: Among them, λ is the weight hyperparameter of the gradient penalty term, is the interpolation point between the real data and the generated data, α~U(0,1), and is the discriminator interpolation point The gradient of , in this way, ensures the Lipschitz constraint and improves the generation quality of the generator.

[0014] Preferably, the generated data and the original experimental data are first standardized to have the same scale, and then the quantile cutoff method is applied to remove outliers. Specifically, by calculating the upper and lower boundary quantiles of the data (such as 1% and 99%), the extreme data points outside the range are removed to ensure the rationality of the data distribution. The cleaned data set will be used for subsequent model training and performance prediction to improve the generalization ability and prediction accuracy of the model.

[0015] Preferably, the performance of the carbon nanotube film is predicted using an integrated learning model based on Super Learner, wherein the Super Learner model includes multiple base learners and meta learners. The first layer of base learners includes a random forest regression model, which constructs multiple decision trees and integrates the prediction results, and the final prediction value is the average value of all trees. The formula is Where N is the number of decision trees, h i (x) is the prediction result of the i-th tree; the XGBoost regression model minimizes the objective function by iteratively updating the residual The objective function consists of two parts: loss function and regularization term. The first part is to calculate the prediction error of the model, where n represents the number of samples in the training data set, and y i is the true value of the i-th sample; is the model's predicted value for the i-th sample; is the loss function of the tth round. The second part is the penalty for model complexity, where t represents the number of rounds of the current iteration, Ω(f k) is the regularization term of the kth tree; the LightGBM regression model optimizes data distribution and training efficiency through the histogram-based gradient boosting tree algorithm;

[0016] The second layer of base learners includes: Gaussian process regression model, using kernel function The data covariance is calculated to model the nonlinear relationship of carbon nanotube film performance, where l is the length scale hyperparameter, x and x′ are the characteristic variables m support vector regression model SVR, whose optimization goal is Where w is the weight vector, representing the parameter vector used to define the model in the support vector regression model, C is the penalty coefficient, which controls the balance between the model's tolerance to errors and regularization, n is the number of training samples, and y is the weight vector used to define the model in the support vector regression model. i is the true value of the i-th sample, f(x i ) is the prediction function of the model, which represents the support vector regression model for the input x i The predicted value of ∈ is the accuracy threshold, which defines an error range that does not require penalty, max(0,|y i -f(x i )|-∈) represents the loss of each sample i. The loss is when the prediction error |y i -f(x i )| will be calculated only when it exceeds ∈. If the predicted value f(x i ) is far from the true value y i If the error is less than ∈, the item is zero and no penalty is imposed. If the error is greater than ∈, the portion exceeding ∈ will be added to the total loss. The meta-learner uses a ridge regression model to generate the final prediction result by weighted integration of the outputs of the first and second layer base learners.

[0017] Preferably, the Super Learner ensemble learning model is trained by cross-validation to avoid overfitting. The performance of the model is measured by the mean square error (MSE) and the coefficient of determination (R 2 ) is used for evaluation, and the formula for the mean square error is: in is the predicted value, y i is the true value. 2 The calculation formula is in is the mean of the true value. These two evaluation indicators ensure the prediction accuracy and fitting effect of the model.

[0018] Preferably, the target variable of the carbon nanotube film performance is subjected to power transformation, i.e., Yeo-Johnson transformation, to transform the target variable into data close to normal distribution. The specific formula is: Where y is the original data of the target variable, y′ is the transformed data, and λ is the transformation parameter, which is automatically determined by maximum likelihood estimation. This method can effectively reduce the deviation in the performance data of carbon nanotube membranes, improve the symmetry of their distribution, ensure the continuity and stability of the model input, and thus improve the accuracy of the regression model in predicting the physical water permeability of carbon nanotube membranes. Combined with the data after power conversion, the Super Learner ensemble learning model is used to train the combined data, and finally an accurate prediction result of the carbon nanotube membrane performance is obtained.

[0019] Beneficial Effects of the Invention

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] (1) The invention generates high-quality synthetic data through the WGAN-GP generative adversarial network, which solves the problem of insufficient training caused by data scarcity in traditional methods and significantly improves the data volume and prediction accuracy of the model.

[0022] (2) The Super Learner ensemble learning model was used in combination with multiple base learners such as random forest, XGBoost, and LightGBM. The two-layer regression architecture was used to improve the accuracy of predicting the water permeability of carbon nanotube membranes, especially when dealing with complex multivariate relationships.

[0023] (3) Cross-validation is used to avoid model overfitting and ensure the stability of the prediction results. The use of power conversion further reduces data bias and makes the model more robust to different types of carbon nanotube film data.

[0024] (4) The present invention integrates steps such as data expansion, preprocessing, model training and evaluation, simplifies the prediction process, and reduces manual intervention. Compared with traditional manual analysis or empirical speculation, the present invention provides an automated prediction method suitable for large-scale carbon nanotube film performance prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the basic implementation framework of the present invention.

[0026] Figure 2 This is a schematic diagram of the adversarial generation network structure of the present invention.

[0027] Figure 3 The figure is a flow chart for constructing the Super Learner integrated learning model of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] like Figure 1 As shown, this embodiment is a carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning, comprising the following steps:

[0030] Step 1: Construct a data set based on the experimental data of carbon nanotube films and perform feature extraction. Figure 2 As shown in the figure, a WGAN-GP generative adversarial network model is constructed, and the network structure of the generator and discriminator is optimized through gradient penalty to ensure the effectiveness of the model.

[0031] Step 2: Use the trained WGAN-GP model to generate a new data set, and preprocess the newly generated data set, including using the quantile truncation method to remove outliers, standardization, and feature engineering, and merge the processed data with the original data set, and then divide it into training set and test set.

[0032] Step 3: Use the Super Learner ensemble learning model to build a two-layer regression architecture containing multiple base learners (such as random forest, XGBoost, LightGBM, etc.) and a meta-learner, and perform model training and parameter optimization for the extended data set.

[0033] Step 4: According to the model evaluation indicators, the performance of the Super Learner model is evaluated, and the trained model is used to predict the performance data of 3450 sets of synthesized carbon nanotube films, and the prediction results are output.

[0034] In step 1, the original carbon nanotube membrane experimental data obtained from the laboratory are first processed and cleaned. Since the experimental data contain a small number of missing values, 584 sets of complete experimental data are finally obtained by eliminating incomplete data sets. Table 1 shows the range of experimental parameters of carbon nanotube membrane materials, among which pressure (Pressure), tube diameter (Sigma) and potential well depth (Epsilon) are three key material parameters. The pressure parameter varies in the MPa order of magnitude, increasing from 300MPa to 500MPa. This parameter simulates the force applied to the membrane material during the filtration process. Sigma (tube diameter) ranges from arrive The influence of surface tension on membrane porosity and filtration efficiency is investigated by changing the surface tension, which affects the retention capacity of the membrane and the fluidity of water molecules. Finally, Epsilon represents the interaction energy, which ranges from 0.0936Kcal / mole to 5.3333Kcal / mole, and involves the intermolecular interaction force and its influence on the transport performance of water molecules.

[0035] Table 1 Experimental parameter range

[0036]

[0037] Then, the cleaned 584 sets of experimental data were input into the constructed WGAN-GP (Wasserstein Generative Adversarial Network-Gradient Penalty) model for data expansion. Tables 2 and 3 are the parameter settings of the generator and discriminator in WGAN-GP, which are designed to ensure that the model can efficiently generate and discriminate data to meet the needs of predicting the water permeability performance of carbon nanotube membranes.

[0038] The generator first maps the 100-dimensional noise vector to 256 dimensions through a linear layer, gradually generating higher-dimensional features. The hidden layer dimension is fixed to 256, and the ReLU activation function is used to capture complex features and avoid gradient disappearance. The output layer generates 6-dimensional data that is the same as the real data, and the Softplus activation function is used to ensure that the output is positive, which meets the requirements of physical properties.

[0039] The discriminator extracts the key features of the data by gradually reducing the dimensions. The input layer maps the 6-dimensional data to 256 dimensions, and then each hidden layer uses the LeakyReLU activation function to prevent the "dead neuron" problem. The output layer uses the Sigmoid activation function to generate a scalar value to determine the authenticity of the data, effectively distinguishing between generated and real data.

[0040] Table 2 Generator parameter settings

[0041]

[0042] Table 3 Discriminator parameter settings

[0043]

[0044] In step 2, the generator of the fully trained WGAN-GP model is used to generate random noise. 3000 sets of new carbon nanotube membrane water permeability performance data are generated in . The generated data are processed by a multi-layer fully connected network and nonlinear activation functions (ReLU and Softplus), and 3000 sets of data related to the performance of 6-dimensional carbon nanotube membranes are output. Some of the generated data are shown in Table 4. First, for the k-water, k-Na and k-Cl data in the target variables, the power transformation, i.e., Yeo-Johnson transformation, is used to process the target variables containing negative values ​​and zero values ​​to ensure that the data is more in line with the normal distribution. Secondly, the generated data is processed by the quantile truncation method to eliminate outliers. Specifically, the upper quantile (99%) and lower quantile (1%) of the generated data are used as truncation thresholds, and the data above the 99% quantile or below the 1% quantile are cleaned. This method eliminated 134 sets of abnormal data and retained 2866 sets of more reasonable generated data.

[0045] After data preprocessing, the processed 2866 sets of generated data were merged with the original 584 sets of experimental data to form a data set with a total of 3450 sets of samples. The data set was then divided into a training set and a test set, usually with 80% of the data used for training and 20% for testing.

[0046] Table 4 Partial data generated by the WGAN-GP model

[0047]

[0048]

[0049] In step 3, the water permeability of the carbon nanotube membrane is predicted by constructing a Super Learner ensemble learning model. Figure 3 As shown, first, the key characteristic variables and target variables of the carbon nanotube membrane are extracted from the merged data set. In this embodiment, the characteristic variables include three physical parameters of the carbon nanotube membrane, and the target variables are water permeability (k-water), sodium ion permeability (k-Na) and chloride ion permeability (k-Cl). These three properties are used as the core indicators of model prediction. Next, the standardized formula is used The generated data is normalized, where μ is the mean of the generated data and σ is the standard deviation of the generated data. The normalization process adjusts the data to zero mean and unit variance, ensuring that the generated data is in the same scale range as the original experimental data. For the model training phase, the first layer of the SuperLearner model is constructed, and the feature variables X and the target variables y are input in sequence. Through multiple training and cross-validation, the optimal parameter configuration of each base learner is obtained. As shown in Table 5, the hyperparameters of the Random Forest Regressor are set to a maximum depth of 12 and the number of trees is 200; the XGBoost Regressor improves the prediction accuracy by adjusting the learning rate, objective function and the number of trees; the LightGBM Regressor sets 30 leaf nodes and adjusts the learning rate.

[0050] Table 5 Parameter settings of the three models

[0051]

[0052] In the second layer of the model, Gaussian Process Regressor and Support Vector Regressor (SVR) are selected as base learners, and the covariance matrix of the data is constructed through the kernel function to predict the target variable. Among them, l is the length scale hyperparameter, and x and x′ are feature variables.

[0053] In the final prediction of the model, the meta-learner Ridge Regression is used to weight the prediction results of the base learners to improve the generalization ability of the model. The loss function of Ridge Regression is expressed as:

[0054] L(w)=||y-Xw|| 2 +α||w|| 2 , where L(w) is the loss function, y is the target value vector, X is the feature matrix, w is the regression coefficient vector, and α is the regularization coefficient.

[0055] The model was validated by the cross-validation method, selecting 80% of the data set as the training set and 20% as the test set to ensure the stability and generalization ability of the model on different data sets. The Super Learner model trained with the generated data set finally outputs the prediction results of the water permeability performance of the carbon nanotube membrane.

[0056] After completing the training of the Super Learner ensemble learning model in step 4, the performance of the model needs to be evaluated to verify its accuracy and robustness in predicting the water permeability of carbon nanotube membranes. The evaluation indicators mainly include mean square error (MSE) and determination coefficient (R 2 ), these indicators can effectively measure the prediction accuracy and fitting effect of the model.

[0057] First, use the mean square error formula The deviation between the predicted value of the quantitative model and the true value, where n is the number of samples and y is the i is the actual value, is the predicted value of the model. The smaller the mean square error, the higher the prediction accuracy of the model.

[0058] Secondly, using the coefficient of determination formula Evaluate how well the model fits the data, where is the mean of the true values. 2 The closer the value is to 1, the better the model fit is.

[0059] In the performance evaluation stage, the cross-validation method was used to verify the training set and the test set to ensure the robustness of the model on different data sets. The evaluation results are shown in Table 6. The model has high accuracy and stability in predicting the water permeability (k-water), sodium ion permeability (k-Na) and chloride ion permeability (k-Cl) of carbon nanotube membranes.

[0060] Table 6 Model performance evaluation results

[0061]

[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A carbon nanotube film performance prediction method based on WGAN-GP and Super Learner ensemble learning, characterized in that: The steps include: Step 1: Build a data set based on the experimental data of carbon nanotube film and perform feature extraction; build a WGAN-GP generative adversarial network model, optimize the network structure of the generator and discriminator through gradient penalty to ensure the effectiveness of the model; Step 2: Use the trained WGAN-GP model to generate a new data set, preprocess the newly generated data set, merge the processed data with the original data set, and then divide it into training set and test set; Step 3: Use the Super Learner ensemble learning model to build a two-layer regression architecture containing multiple base learners and meta-learners, and perform model training and parameter optimization for the extended data set; Step 4: According to the model evaluation index, the performance of the Super Learner model is evaluated, and the performance data of the carbon nanotube film to be tested is predicted using the trained model, and the prediction results are output.

2. The carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning according to claim 1, characterized in that: The generator of the WGAN-GP generative adversarial network is a multi-layer fully connected neural network, and the input is a standard normal distribution. The noise vector sampled in the network, each hidden layer in the network uses the ReLU activation function, expressed as h i =max(0,W i h i-1 +b i ), where W i is the weight matrix, b i is the bias vector, h i is the output of each layer; The output layer uses the Softplus activation function Softplus(x)=ln(1+e x ), x is the intermediate result of the generator network output to ensure that the generated data is positive and consistent with the physical properties of the carbon nanotube film; the generator minimizes the loss function Where x is the generated data, is the evaluation value of the discriminator on the generated data, E is the expected value of the noise distribution z, and the network parameters are updated to make the generated data gradually approach the real data distribution.

3. The carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning according to claim 1, characterized in that: During the training process of the generative adversarial network, a gradient penalty term is used to ensure the Lipschitz continuity of the discriminator; the discriminator structure is a multi-layer neural network, and each layer uses the LeakyReLU activation function LeakyReLU(x)=max(0.01x,x), which effectively prevents the gradient vanishing problem; The discriminator minimizes the loss function at each weight update: Among them, λ is the weight hyperparameter of the gradient penalty term, is the interpolation point between the real data and the generated data, α~U(0,1), and is the discriminator interpolation point The gradient of , in this way, ensures the Lipschitz constraint and improves the generation quality of the generator.

4. The carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning according to claim 1, characterized in that: The preprocessing described in step 2 includes using the quantile truncation method to eliminate outliers, standardization processing, and feature engineering.

5. The carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning according to claim 4 is characterized in that: The generated data and the original experimental data are first standardized to have the same scale, and then the quantile truncation method is applied to remove outliers; specifically, the upper and lower boundary quantiles of the data are calculated and the extreme data points outside the range are removed to ensure the rationality of the data distribution.

6. The carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning according to claim 1, characterized in that: The performance of the carbon nanotube film is predicted using an integrated learning model based on Super Learner, wherein the Super Learner model includes multiple base learners and meta learners, wherein: The first layer of base learners includes: The random forest regression model constructs multiple decision trees and integrates the prediction results. The final prediction value is the average of all trees. The formula is Where N is the number of decision trees, h i (x) is the prediction result of the i-th tree; XGBoost regression model, by iteratively updating the residual, minimizes the objective function The objective function consists of two parts: loss function and regularization term. The first part is to calculate the prediction error of the model, where n represents the number of samples in the training data set and y i is the true value of the i-th sample; is the model's predicted value for the i-th sample; is the loss function of the tth round; the second part is the penalty for model complexity, where t represents the number of rounds of the current iteration, Ω(f k ) is the regularization term of the k-th tree; LightGBM regression model, which optimizes data distribution and training efficiency through histogram-based gradient boosting tree algorithm; The second layer of base learners includes: Gaussian process regression model, using kernel function The data covariance was calculated to model the nonlinear relationship of carbon nanotube film performance, where l is the length scale hyperparameter and x and x′ are characteristic variables; Support vector regression model SVR, its optimization goal is Where w is the weight vector, representing the parameter vector used to define the model in the support vector regression model, C is the penalty coefficient, which controls the balance between the model's tolerance to errors and regularization, n is the number of training samples, and y is the weight vector used to define the model in the support vector regression model. i is the true value of the i-th sample, f(x i ) is the prediction function of the model, which represents the support vector regression model for the input x i The predicted value of ∈ is the accuracy threshold, which defines an error range that does not require penalty, max(0,|y i f(x i )|-∈) represents the loss of each sample i. The loss is when the prediction error |y i f(x i )| will be calculated only when it exceeds ∈. If the predicted value f(x i ) is far from the true value y i If the error is less than ∈, the term is zero and no penalty is imposed. If the error is greater than ∈, the portion exceeding ∈ will be added to the total loss. The meta-learner uses a ridge regression model to generate the final prediction result by weighted integration of the outputs of the first and second layer base learners.

7. The carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning according to claim 1, characterized in that: The Super Learner ensemble learning model is trained by cross-validation to avoid overfitting.

8. The carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning according to claim 7, characterized in that: The performance of the model is measured by the mean square error (MSE) and the coefficient of determination (R). 2 Conduct an assessment where: The formula for mean square error is in is the predicted value, y i is the true value; Coefficient of determination R 2 The calculation formula is in is the mean of the true values.

9. The carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning according to claim 8, characterized in that: Power conversion is performed on the carbon nanotube film performance data to be predicted to convert it into data close to normal distribution. The specific formula is: y<0, where y is the original data, y′ is the transformed data, and λ is the transformation parameter, which is automatically determined by maximum likelihood estimation.

10. The carbon nanotube film performance prediction method based on WGAN-GP and Super Learner integrated learning according to claim 1, characterized in that: The feature extraction in step 1 includes: pressure, tube diameter, potential well depth, water permeability k-water, sodium ion permeability k-Na and chloride ion permeability k-Cl; the performance prediction includes carbon nanotube membrane water permeability k-water, sodium ion permeability k-Na and chloride ion permeability k-Cl.

Images