A high-performance nanofiltration membrane preparation method based on machine learning aided design

Abstract

The application discloses a kind of high-performance nanofiltration membrane preparation method based on machine learning aided design, belong to nanofiltration membrane technical field.The method includes: the construction includes target screening ion physicochemical property, functionalized nanomaterial physicochemical property, composite nanofiltration membrane preparation condition, composite membrane physicochemical structure feature, separation process environmental condition and separation performance parameter fine parameter data set;Data set is divided into training set and test set, input machine learning model builds composite membrane performance prediction model, establishes the nonlinear mapping relationship between multiple parameters and separation performance;Model optimization is realized by hyperparameter tuning and evaluation index;Key parameters are identified using interpretable methods such as feature importance ranking, and the optimal parameter combination is obtained through global optimization, to guide the directional preparation of high-performance composite nanofiltration membranes for different ion screening systems.The application decouples the coupling effect of multiple factors, reduces the experimental trial and error cost, and improves the ion separation performance of nanofiltration membrane.

Description

Technical Field

[0001] This invention relates to the field of nanofiltration membrane technology, and more specifically to a method for preparing high-performance nanofiltration membranes based on machine learning-aided design. Background Technology

[0002] The separation and purification of substances play a vital role in human life and industrial development. Selective ion separation is a common requirement in fields such as energy storage and conversion, pollution monitoring and control, and clean industrial processes. Ion separation processes are becoming increasingly important in exploring new resources and enhancing the recovery and reuse of existing resources. Membrane separation technology, as a highly efficient, energy-saving, and environmentally friendly separation method, can achieve selective ion sieving at the microscopic scale through the design and control of membrane structures. However, the hydrated diameter of ions is typically less than 1 nm, and the differences in valence and size between different ions are extremely small. Compared to the separation of larger molecules or particles, achieving precise ion sieving is extremely difficult. Furthermore, to achieve excellent ion screening performance, the separation membrane must simultaneously achieve precise ion selectivity and high permeability. This means that the membrane must allow free permeation of target ions while possessing high selectivity to ensure that unwanted ions do not permeate. However, limited by the mass transfer mechanism of membrane separation processes, conventional ion separation membranes suffer from a "trade-off" effect, where permeability and selectivity are mutually constrained. Membrane-based selective ion mass transfer requires a trade-off between selectivity and permeability. Therefore, nanofiltration separation of ion systems is a complex process involving "multi-mechanism synergy" and "multi-factor coupling". How to decouple the interaction between multiple factors and the ion separation performance of nanofiltration membranes, and identify and extract the key characteristic parameters affecting membrane performance from the complex and numerous factors, has become one of the most challenging and hot scientific issues in the field of membrane ion separation.

[0003] Existing research often employs conventional non-equilibrium thermodynamic models and pore structure-based mechanical models to guide the design and fabrication of ion-separation nanofiltration membranes. This typically requires a massive number of single-factor experiments to obtain optimal experimental schemes, incurring significant costs while the ion separation performance of the fabricated membranes remains insignificant. Therefore, there is an urgent need to develop a membrane-assisted fabrication and design method that addresses the complex multi-factor processes involved in ion-separation nanofiltration membrane preparation. This method should be able to more quickly and comprehensively elucidate the relationships between multiple factors and variables, such as ion properties, guest material properties, membrane structure, and operating parameters, and membrane ion separation performance, based on different ion separation systems. This would guide the design, fabrication, and modification of membranes, significantly reducing experimental trial-and-error costs while further enhancing the ion separation performance of the fabricated nanofiltration membranes, possessing immense scientific value and application potential. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a high-performance nanofiltration membrane preparation method based on machine learning-aided design. It establishes an extremely complex and detailed dataset of basic parameters based on large-scale data collection, and further combines machine learning algorithms to construct a highly predictive neural network model. For different ion separation systems, it decouples multiple factors affecting membrane ion separation performance and proposes key characteristic parameters. It establishes a multi-scale mapping relationship between ion properties, nanomaterial characteristics, membrane physicochemical structure, membrane preparation and separation operation conditions, and membrane separation performance, ultimately obtaining the optimal combination of membrane preparation condition parameters and realizing the targeted preparation of high-performance ion separation composite nanofiltration membranes.

[0005] This invention discloses a method for preparing high-performance nanofiltration membranes based on machine learning-aided design, comprising: Step 1: Based on the data collected from experimental tests and literature reviews, construct a refined parameter dataset; wherein, the refined parameter dataset includes the physicochemical property parameters of the target sieve ions, the physicochemical property parameters of the functionalized nanomaterials, the preparation condition parameters of the composite nanofiltration membrane modified by the functionalized nanomaterials, the physicochemical structural characteristic parameters of the composite membrane, the external environmental condition parameters of the composite membrane separation process, and the separation performance parameters of the composite membrane. Step 2: Divide the refined parameter dataset into training and testing sets according to the proportion. Input the training and testing sets into a machine learning or deep learning model to construct a composite membrane performance prediction model. Perform data training and prediction to establish a nonlinear mapping relationship between the target sieve molecule physicochemical property parameters, functional nanomaterial physicochemical property parameters, composite nanofiltration membrane preparation condition parameters, composite membrane physicochemical structural characteristic parameters, composite membrane separation process external environmental condition parameters, and composite membrane separation performance parameters. Step 3: Combining mathematical algorithms and evaluation indicators, perform hyperparameter tuning and performance evaluation on the composite membrane performance prediction model to achieve model optimization; Step 4: Based on the optimized composite membrane performance prediction model, the key parameters affecting the separation performance of the composite membrane are identified and extracted using an interpretable method. The optimal parameter combination is obtained before membrane preparation, during membrane preparation, and during membrane separation. This guides the targeted preparation of high-performance composite nanofiltration membranes for different ion sieving systems.

[0006] As a further improvement of the present invention, in step 1, the physicochemical properties of the target sieve ions include the basic physicochemical properties of ions with different valence states under different separation systems, wherein the ions with different valence states include Cl. - SO4 2- Li + Na + K + / Mg 2+ and Ca 2+ / Al 3+The basic physicochemical properties include ionic valence state, hydrated ionic radius, dehydrated ionic radius, mass, and hydration free energy.

[0007] As a further improvement of the present invention, in step 1, the physicochemical properties parameters of the functionalized nanomaterials include the basic material properties of the functionalized two-dimensional nanomaterials. The functionalized two-dimensional nanomaterials include one or more of the following: aminated molybdenum disulfide, carboxylated molybdenum disulfide, hydroxylated molybdenum disulfide, aminated carbon nitride, carboxylated carbon nitride, hydroxylated carbon nitride, aminated graphene, carboxylated graphene, and hydroxylated graphene. The basic material properties include the size and interlayer spacing of the nanosheets, the dispersion characteristics of the nanosheets, the hydrophilicity and hydrophobicity of the wall surface, the wall surface roughness, the near-wall liquid viscosity, the wall surface wettability, the wall surface curvature, the functional group loading rate, and the density.

[0008] As a further improvement of the present invention, in step 1, the preparation conditions and parameters of the composite nanofiltration membrane modified with functionalized nanomaterials include: substrate membrane material and molecular weight, substrate membrane pretreatment time and temperature, aqueous phase monomer type and concentration, organic phase monomer type and concentration, type of functionalized nanomaterials and concentration of their dispersion, soaking time and temperature of the membrane in the functionalized nanomaterial dispersion-aqueous phase solution-organic phase solution for interfacial polymerization reaction, operating parameters of the air knife used to dry the residual solution on the membrane surface, time and temperature required for air drying the membrane after interfacial polymerization, and temperature and time of thermosetting treatment; wherein... The substrate material is one or more of polyethersulfone, polyacrylonitrile, polyvinylidene fluoride, polysulfone, and polyvinyl chloride, and the molecular weight is 50~100 kDa; Aqueous monomers for anion separation systems include piperazine, piperazine-2-carboxylic acid, and m-phenylenediamine; aqueous monomers for cation separation systems include polyethyleneimine, branched polyethyleneimine, and polyallylamine hydrochloride. The operating parameters of the air knife used to dry the residual solution on the membrane surface include air supply pressure, distance between the air knife nozzle and the membrane surface, angle between the air knife and the membrane surface, and membrane flow speed.

[0009] As a further improvement of the present invention, in step 1, the physicochemical structural characteristic parameters of the composite membrane are obtained by one or more characterization methods, including Fourier transform infrared spectroscopy, Raman spectroscopy, photoelectron spectroscopy, scanning electron microscopy, transmission electron microscopy, atomic force microscopy, X-ray diffraction, water contact angle and Zeta potential analysis, including membrane thickness, membrane surface roughness, elemental composition of membrane surface and vertical gradient, membrane surface loading, water contact angle, Zeta potential, average pore size and mathematical distribution of membrane, surface and vertical gradient charge properties / intensity / density, Donnan distribution coefficient and size sieving factor.

[0010] As a further improvement of the present invention, in step 1, the external environmental condition parameters of the composite membrane separation process refer to the operating conditions involved in the ion separation process of using the prepared functionalized two-dimensional nanomaterial composite nanofiltration membrane in different separation systems, including feed concentration, pre-pressurization time and pressure, separation operation pressure and temperature, and effluent time and pressure. The composite membrane separation performance parameters refer to the ion separation performance of different ion separation systems using functionalized two-dimensional nanomaterial composite nanofiltration membranes, including the membrane's pure water flux, single salt rejection rate, mixed salt selective separation factor, and long-term operational stability.

[0011] As a further improvement of the present invention, in step 2, the refined parameter dataset is divided into a training set and a test set according to a certain ratio, including: organizing the data according to the anion and cation separation system through experimental testing and literature data collection, ensuring that there are more than 500 effective data sets and more than 15,000 effective data points to form an initial dataset; dividing the initial dataset into multiple types of feature datasets: selecting the physicochemical properties parameters of the target sieve ions as background features, selecting the physicochemical properties parameters of functionalized nanomaterials and the physicochemical structural parameters of composite membranes as structural features, selecting the preparation condition parameters of composite nanofiltration membranes and the external environmental condition parameters of the composite membrane separation process as operational features, and selecting the separation performance parameters of composite membranes as separation result features; preprocessing each feature data sequence: using the box plot method to remove outliers, using cubic spline interpolation or linear interpolation to process blank values ​​and missing values, and using the Min-Max normalization method to map the original data to the [0,1] interval, the normalization formula is: In the formula, for The corresponding value after normalization For the m-th valid data point in the N-th type of feature data sequence, The minimum eigenvalue in the Nth class of feature data sequence. The largest eigenvalue in the Nth class of feature data sequence.

[0012] After normalizing the data, each feature dataset is divided into training and testing datasets in a ratio of 7:3 or 8:2.

[0013] As a further improvement of the present invention, in step 2, the machine learning or deep learning model includes one or more of the following: convolutional neural network, support vector machine, extreme gradient boosting algorithm, adaptive boosting algorithm, and random forest; Based on the selected learning model, the established and partitioned training dataset is input into the model, and the parameters are adjusted for data training to establish a complex mapping relationship between the separation performance parameters of the composite membrane for different ions and other input feature parameters. The specific feature relationship is shown in the following formula: In the formula, is the feature data matrix of the separation performance parameters of the composite membrane, and is the output variable; The matrix represents the physicochemical property parameters of the target sieve ions, and the background feature values ​​are the input variables. , , , The feature data matrices of functionalized nanomaterial physicochemical properties, composite nanofiltration membrane preparation conditions, composite membrane physicochemical structure, and composite membrane separation process external environmental conditions are all used as target input variables of the model. For residual variables.

[0014] As a further improvement of the present invention, in step 3, one or more of the following algorithms are used to perform hyperparameter tuning on the model: feature discretization algorithm, Bayesian optimization algorithm, and K-fold cross-validation, and the coefficient of determination R is used. 2 The performance of the model parameter scheme is evaluated using five evaluation indicators: mean error (MB), mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE), and the optimal parameter scheme is selected.

[0015] As a further improvement of the present invention, in step 4, based on the optimized model, interpretability methods such as feature importance ranking are used to identify and extract key parameters affecting the separation performance of the composite membrane. Specifically, on the basis of establishing the optimal prediction model, local interpretability methods such as feature importance ranking, NNKX, Deep Taylor, ICE, and SHAP are used to measure the contribution score and weight of different input parameter features in the model prediction results, quantify the importance of the feature parameter, and thus identify and extract key parameter features affecting the separation performance of the composite membrane.

[0016] As a further improvement of the present invention, in step 4, obtaining the optimal parameter combination before, during, and after membrane preparation and separation processes specifically means that, based on the feature importance assessment, key features that have a significant impact on the performance of the composite membrane are identified and extracted. Iterative optimization using a global optimization algorithm and a machine learning model is employed, and the fitness value is calculated in combination with the objective function. Based on the optimal ion separation performance of the composite nanofiltration membrane under different sieving systems, the global optimal parameter values ​​of the corresponding key features are predicted in reverse, thereby obtaining the optimal parameter combination for operation before, during, and after membrane preparation and separation processes.

[0017] As a further improvement of the present invention, in step 4, the final step is to guide the directional design and preparation of high-performance composite nanofiltration membranes for different ion sieving systems. Specifically, for different ion sieving systems, the optimal combination of parameters for the preparation and separation process of composite nanofiltration membranes, predicted by machine learning models, is used. This includes the optimal combination of key parameters that have a significant impact on membrane performance, such as the type of functional groups of nanomaterials, substrate material, type and concentration of aqueous monomers, concentration of functional nanomaterial dispersion, soaking time and temperature of the membrane in functional nanomaterial dispersion-aqueous solution-organic solution (interfacial polymerization reaction), temperature and time of thermosetting treatment, water contact angle, Zeta potential, average pore size and mathematical distribution of the membrane, properties / intensity / density of surface and vertical gradient charge, Donnan partition coefficient and size sieving factor, pre-compression time and pressure of separation operation, and pressure and temperature of separation process. This allows for the directional preparation of composite nanofiltration membranes, ultimately resulting in high-performance composite nanofiltration membranes.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention differs from traditional nanofiltration membrane fabrication methods. It creatively employs a combination of big data analysis and machine learning models to provide targeted guidance for the design and fabrication of nanofiltration membranes. This effectively reduces the significant costs associated with conventional membrane fabrication and modification methods that rely on extensive trial-and-error experiments based on single factors. At the same time, it further enhances the ion separation performance of the membrane. This invention addresses different ion separation systems by decoupling multiple factors affecting membrane ion separation performance and proposing key characteristic parameters. It establishes a multi-scale mapping relationship between ion properties, nanomaterial characteristics, membrane physicochemical structure, membrane preparation and separation operation conditions, and membrane separation performance. This allows for a faster and more comprehensive explanation of the influence of multiple factors and variables, such as ion properties, guest material properties, membrane structure, and operating parameters, on membrane ion separation performance. Consequently, it guides the design, preparation, and modification of membranes, ultimately obtaining the optimal combination of membrane preparation condition parameters and achieving the targeted preparation of high-performance ion separation composite nanofiltration membranes. Attached Figure Description

[0019] Figure 1 This is a flowchart of the high-performance nanofiltration membrane preparation method based on machine learning-aided design disclosed in this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. 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.

[0021] The present invention will now be described in further detail with reference to the accompanying drawings: like Figure 1 As shown, this invention provides a method for preparing high-performance nanofiltration membranes based on machine learning-aided design. Unless otherwise specified, all steps below are performed at room temperature, including: S1. Based on experimental tests and extensive literature review, data were collected and organized on the following: physicochemical properties of anions / cations with various valence states, physicochemical properties of functionalized two-dimensional nanomaterials (taking functionalized molybdenum disulfide (MoS2-OH, MoS2-COOH, MoS2-NH2) as an example, but not limited to functionalized molybdenum disulfide), preparation conditions of composite nanofiltration membranes modified with functionalized nanomaterials, physicochemical structural characteristics of composite membranes, external environmental conditions of composite membrane separation process, and separation performance parameters of composite membranes. A total of 1200 sets of valid data were collected, containing 18200 valid data points. S2. The entire dataset is divided into multiple types of feature datasets. Based on different sieving systems, the physicochemical properties of the target ion are selected as background features (B), the physicochemical properties of functionalized nanomaterials (F) and the physicochemical structural parameters of composite membranes (M) are selected as structural features, the preparation conditions of composite nanofiltration membranes (C) and the external environmental conditions of the composite membrane separation process (S) are selected as operational features, and the separation performance parameters of the composite membrane for different ions (R) are selected as separation result features. Furthermore, effective data points of different feature data sequences are preprocessed. Box plots are used to remove outliers in the same type of feature data sequences, and cubic spline interpolation is used to process blank and missing values ​​in the same type of feature data sequences. After preprocessing different types of feature data sequences, Min-Max normalization is used to normalize all original data within the same type of data sequence, and each type of feature dataset is divided into training and testing datasets in an 80:20 ratio. S3. Select the Adaptive Boosting Algorithm (AdaBoost) and Random Forest (RF) machine learning models as the base models (not limited to AdaBoost and RF models), and establish the objective function as follows: In the formula, is the feature data matrix of the separation performance parameters of the composite membrane, and is the output variable; The matrix represents the physicochemical property parameters of the target sieve ions, and the background feature values ​​are the input variables. , , , The feature data matrices of functionalized nanomaterial physicochemical properties, composite nanofiltration membrane preparation conditions, composite membrane physicochemical structure, and composite membrane separation process external environmental conditions are all used as target input variables of the model. For residual variables.

[0022] The pre-defined training and test sets are sequentially input into the AdaBoost-RF composite model. Five-fold cross-validation and Bayesian optimization algorithms are used to further partition and validate the training set to enhance the model's predictive performance, robustness, and generalization ability. Parameter tuning is then performed based on 5-8 rounds of AdaBoost adaptive iterative optimization, using the coefficient of determination (R²). 2 The performance of the model parameter scheme was evaluated using five typical evaluation metrics: mean error (MB), mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). The optimization focused on parameters such as learning_rate, n_estimators, and max_depth of the RF module in the AdaBoost-RF composite model. After 7 iterations, the optimal parameter scheme was obtained, resulting in the final enhanced AdaBoost-RF composite model.

[0023] S4. SHAP (Shapley Additive Explanation) analysis was used to identify and extract key parameter features affecting the separation performance of the composite membrane, and the specific contribution of each input variable to the membrane performance prediction results was quantified. The results are shown in Table 1 below: Table 1. Key parameters affecting membrane performance under different sieving systems (sum of contribution equals 1.00) Based on the identified key parameters affecting the separation performance of the composite membrane, a global optimization algorithm is further employed, combined with the objective function, to calculate the fitness value and obtain the optimal parameter combination corresponding to the key parameter features. According to the global optimization calculation results, the optimal guiding operating parameter combination for functionalized nanomaterial (taking functionalized MoS2 as an example) composite nanofiltration membranes, provided by the AdaBoost-RF composite model for different ion separation systems, is shown in Table 2 below: Table 2 Optimal Parameter Combinations and Parameter Assignments Based on the optimal conditions given above, the composite nanofiltration membrane was prepared. After completion, the performance of pure water flux, single salt rejection rate, and mixed salt selective separation factor was tested. The specific results are shown in Table 3 below.

[0024] Comparative Example 1 The nanofiltration membrane was not prepared according to the method of Example 1. The difference from Example 1 is that the membrane was not prepared in the order of steps (1) to (4). Instead, conventional membrane preparation was carried out based on single-factor experiments. The key characteristic parameters in Table 2 are different from those in Example 1, while the other parameters are the same as those in Example 1. After the membrane was prepared, the performance of pure water flux, single salt rejection rate, and mixed salt selective separation factor was tested. The specific results are shown in Table 3 below. Table 3. Performance Reference Table for Examples and Comparative Examples In summary, the high-performance nanofiltration membrane designed and prepared using machine learning-assisted methods in this invention exhibits significantly higher permeability and separation selectivity than the nanofiltration membrane prepared using single-factor experimental experience in the comparative example, regardless of whether it is anion or cation sieving systems. This demonstrates that the method employed in this invention addresses different ion separation systems, decouples multiple factors affecting membrane ion separation performance, proposes key characteristic parameters, and establishes a multi-scale mapping relationship between ion properties, nanomaterial characteristics, membrane physicochemical structure, membrane preparation and separation operation conditions, and membrane separation performance. Ultimately, it obtains the optimal combination of membrane preparation condition parameters, achieving the targeted preparation of a high-performance ion separation composite nanofiltration membrane. Therefore, the high-performance nanofiltration membrane and its preparation method based on machine learning-assisted design presented in this invention demonstrate excellent performance.

[0025] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for preparing high-performance nanofiltration membranes based on machine learning-aided design, characterized in that, include: Step 1: Based on the data collected from experimental tests and literature reviews, construct a refined parameter dataset; wherein, the refined parameter dataset includes the physicochemical property parameters of the target sieve ions, the physicochemical property parameters of the functionalized nanomaterials, the preparation condition parameters of the composite nanofiltration membrane modified by the functionalized nanomaterials, the physicochemical structural characteristic parameters of the composite membrane, the external environmental condition parameters of the composite membrane separation process, and the separation performance parameters of the composite membrane. Step 2: Divide the refined parameter dataset into training and testing sets according to the proportion. Input the training and testing sets into a machine learning or deep learning model to construct a composite membrane performance prediction model. Perform data training and prediction to establish a nonlinear mapping relationship between the target sieve molecule physicochemical property parameters, functional nanomaterial physicochemical property parameters, composite nanofiltration membrane preparation condition parameters, composite membrane physicochemical structural characteristic parameters, composite membrane separation process external environmental condition parameters, and composite membrane separation performance parameters. Step 3: Combining mathematical algorithms and evaluation indicators, perform hyperparameter tuning and performance evaluation on the composite membrane performance prediction model to achieve model optimization; Step 4: Based on the optimized composite membrane performance prediction model, the key parameters affecting the separation performance of the composite membrane are identified and extracted using an interpretable method. The optimal parameter combination is obtained before membrane preparation, during membrane preparation, and during membrane separation. This guides the targeted preparation of high-performance composite nanofiltration membranes for different ion sieving systems.

2. The preparation method according to claim 1, characterized in that, In step 1, the physicochemical properties of the target sieve ions include the basic physicochemical properties of ions with different valence states under different separation systems, and the ions with different valence states include Cl. - SO4 2- Li + Na + K + / Mg 2+ and Ca 2+ / Al 3+ The basic physicochemical properties include ionic valence state, hydrated ionic radius, dehydrated ionic radius, mass, and hydration free energy.

3. The preparation method according to claim 1, characterized in that, In step 1, the physicochemical properties of the functionalized nanomaterials include the basic material properties of the functionalized two-dimensional nanomaterials. The functionalized two-dimensional nanomaterials include one or more of the following: aminated molybdenum disulfide, carboxylated molybdenum disulfide, hydroxylated molybdenum disulfide, aminated carbon nitride, carboxylated carbon nitride, hydroxylated carbon nitride, aminated graphene, carboxylated graphene, and hydroxylated graphene. The basic material properties include the size and interlayer spacing of the nanosheets, the dispersion characteristics of the nanosheets, the hydrophilicity and hydrophobicity of the wall surface, the wall surface roughness, the near-wall liquid viscosity, the wall surface wettability, the wall surface curvature, the functional group loading rate, and the density.

4. The preparation method according to claim 1, characterized in that, In step 1, the preparation conditions and parameters for the composite nanofiltration membrane modified with functionalized nanomaterials include: substrate material and molecular weight, substrate pretreatment time and temperature, type and concentration of aqueous monomers, type and concentration of organic monomers, type of functionalized nanomaterials and concentration of their dispersion, soaking time and temperature of the membrane in the functionalized nanomaterial dispersion-aqueous solution-organic solution for interfacial polymerization reaction, operating parameters of the air knife used to dry the residual solution on the membrane surface, time and temperature required for air drying the membrane after interfacial polymerization, and temperature and time for thermosetting treatment; among which... The substrate material is one or more of polyethersulfone, polyacrylonitrile, polyvinylidene fluoride, polysulfone and polyvinyl chloride, and the molecular weight is 50~100 kDa; Aqueous monomers for anion separation systems include piperazine, piperazine-2-carboxylic acid, and m-phenylenediamine; aqueous monomers for cation separation systems include polyethyleneimine, branched polyethyleneimine, and polyallylamine hydrochloride. The operating parameters of the air knife used to dry the residual solution on the membrane surface include air supply pressure, distance between the air knife nozzle and the membrane surface, angle between the air knife and the membrane surface, and membrane flow speed.

5. The preparation method according to claim 1, characterized in that, In step 1, the physicochemical structural characteristic parameters of the composite membrane are obtained through one or more characterization methods, including Fourier transform infrared spectroscopy, Raman spectroscopy, photoelectron spectroscopy, scanning electron microscopy, transmission electron microscopy, atomic force microscopy, X-ray diffraction, water contact angle and Zeta potential analysis. These parameters include membrane thickness, membrane surface roughness, elemental composition of membrane surface and vertical gradient, membrane surface loading, water contact angle, Zeta potential, average pore size and mathematical distribution of membrane, surface and vertical gradient charge properties / intensity / density, Donan partition coefficient and size sieve factor.

6. The preparation method according to claim 1, characterized in that, In step 1, the external environmental condition parameters of the composite membrane separation process refer to the operating conditions involved in the ion separation process of using the prepared functionalized two-dimensional nanomaterial composite nanofiltration membrane in different separation systems, including feed concentration, pre-pressurization time and pressure, separation operation pressure and temperature, and effluent time and pressure. The composite membrane separation performance parameters refer to the ion separation performance of different ion separation systems using functionalized two-dimensional nanomaterial composite nanofiltration membranes, including the membrane's pure water flux, single salt rejection rate, mixed salt selective separation factor, and long-term operational stability.

7. The preparation method according to claim 1, characterized in that, In step 2, the refined parameter dataset is divided into training and testing sets according to a ratio. This includes: organizing data according to anion and cation separation systems through experimental testing and literature data collection, ensuring that there are more than 500 effective data sets and more than 15,000 effective data points to form an initial dataset; dividing the initial dataset into multiple types of feature datasets: selecting the physicochemical properties of the target sieve ions as background features, selecting the physicochemical properties of functionalized nanomaterials and the physicochemical structure of composite membranes as structural features, selecting the preparation conditions of composite nanofiltration membranes and the external environmental conditions of the composite membrane separation process as operational features, and selecting the separation performance parameters of composite membranes as separation result features; preprocessing each feature data sequence: using box plot method to remove outliers, using cubic spline interpolation or linear interpolation to process blank and missing values, and using Min-Max normalization method to map the original data to the [0,1] interval. The normalization formula is: In the formula, for The corresponding value after normalization For the m-th valid data point in the N-th type of feature data sequence, The minimum eigenvalue in the Nth class of feature data sequence. The largest eigenvalue in the Nth class of feature data sequence.

8. The preparation method according to claim 1, characterized in that, In step 2, the machine learning or deep learning model includes one or more of the following: convolutional neural network, support vector machine, extreme gradient boosting algorithm, adaptive boosting algorithm, and random forest. Based on the selected learning model, the established and partitioned training dataset is input into the model, and the parameters are adjusted for data training to establish a complex mapping relationship between the separation performance parameters of the composite membrane for different ions and other input feature parameters. The specific feature relationship is shown in the following formula: In the formula, is the feature data matrix of the separation performance parameters of the composite membrane, and is the output variable; The matrix represents the physicochemical property parameters of the target sieve ions, and the background feature values ​​are the input variables. , , , The feature data matrices of functionalized nanomaterial physicochemical properties, composite nanofiltration membrane preparation conditions, composite membrane physicochemical structure, and composite membrane separation process external environmental conditions are all used as target input variables of the model. For residual variables.

9. The preparation method according to claim 1, characterized in that, In step 3, one or more of the following algorithms are used to tune the hyperparameters of the model: feature discretization algorithm, Bayesian optimization algorithm, and K-fold cross-validation. The coefficient of determination R0 is then used. 2 The performance of the model parameter scheme is evaluated using five evaluation indicators: mean error (MB), mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE), and the optimal parameter scheme is selected.

10. The preparation method according to claim 1, characterized in that, In step 4 One or more of the following methods—feature importance ranking, NNKX, Deep Taylor, ICE, and SHAP interpretability—are used to measure the contribution score and weight of each input parameter feature in the model prediction results, and to identify and extract key parameters that affect the separation performance of composite membranes. Iterative optimization using a global optimization algorithm and machine learning model, combined with the objective function to calculate the fitness value, is employed to predict the global optimal parameter values ​​of corresponding key features based on the best ion separation performance of the composite nanofiltration membrane under different sieving systems. This yields the optimal parameter combination, including the type of functionalized groups of nanomaterials, substrate material, type and concentration of aqueous monomers, concentration of functionalized nanomaterial dispersion, immersion time and temperature for interfacial polymerization reaction, temperature and time for thermosetting treatment, water contact angle, Zeta potential, average pore size and mathematical distribution of the membrane, properties / intensity / density of surface and vertical gradient charge on the membrane, Donald distribution coefficient and size sieving factor, pre-compression time and pressure for separation operation, and pressure and temperature during separation process. The composite nanofiltration membrane is then prepared in a directional manner based on the optimal parameter combination.

Images