CF4 / N2 separation-oriented molecular sieve performance cross-temperature zone rapid prediction and optimization method

Abstract

The invention discloses a CF4 / N2 separation-oriented molecular sieve performance cross-temperature zone rapid prediction and optimization method. The method mainly comprises the following steps: calculating single-component and mixed adsorption amounts of CF4 and N2 in a certain temperature range of a zeolite database based on giant regular Monte Carlo simulation high throughput; fitting a plurality of adsorption isotherm models, and screening the optimal adsorption isotherm model to construct an adsorption capacity-temperature-pressure data matrix; data enhancement is carried out in combination with CCR, and rapid and continuous generation of the adsorption isotherm at any temperature within a wide temperature range is realized; predicting the CF4 / N2 adsorption selectivity based on an IAST interpolation method, and comparing the CF4 / N2 adsorption selectivity with a GCMC calculation result for verification; finally, within the applicable temperature range, the adsorption capacity and the selectivity performance target of the mixed components are integrated and optimized, and intelligent screening of the high-performance molecular sieve is achieved. The method can significantly improve the acquisition efficiency and accuracy of adsorption data at different temperatures, and is suitable for rational design and intelligent optimization of the adsorbent under a wide temperature condition.

Description

Technical Field

[0001] This invention relates to the field of predictive adsorption separation technology, and more particularly to a smart method that combines mechanistic models and data science for rapidly and accurately screening molecular sieve adsorbents for efficient separation of CF4 / N2 over a wide temperature range. Background Technology

[0002] Carbon tetrafluoride (CF4) is a widely used plasma etching gas in the semiconductor manufacturing industry. The exhaust gases produced during its production and use often contain large amounts of nitrogen (N2). Due to CF4's extremely high global warming potential, its efficient capture and recovery are crucial. Molecular sieve-based adsorption separation technology is considered a promising solution due to its relatively low energy consumption and ease of operation.

[0003] However, the temperature of industrial exhaust gases often fluctuates, making the development of adsorbents that maintain high adsorption capacity and selectivity over a wide temperature range a current challenge. Traditional experimental screening methods are time-consuming, labor-intensive, and costly. While high-throughput computational screening based on molecular simulations can accelerate this process, it still has the following limitations: (1) Using giant canonical Monte Carlo simulation to directly calculate the adsorption equilibrium of a large number of molecular sieve materials at different temperatures, especially under mixed component conditions, is computationally intensive and time-consuming, making it difficult to achieve a truly wide temperature range rapid scan.

[0004] (2) Limited simulation or experimental data points may lead to large deviations in the fitting of adsorption isotherms, affecting the accuracy of subsequent predictions of the adsorption behavior of mixed components based on thermodynamic models (such as IAST).

[0005] (3) The traditional IAST model method is highly dependent on the fitting quality of adsorption isotherms and fails to predict when there are insufficient data points or the model is not well-fitted. On the other hand, the pure data-driven machine learning method is based on a "black box" model, which has limited interpretability and limited prediction accuracy and generalization ability when there is insufficient training data or imperfect feature engineering.

[0006] Therefore, there is an urgent need in this field for a new method that can balance computational efficiency and prediction accuracy, reliably predict the adsorption and separation performance of molecular sieves over a wide temperature range, and achieve intelligent screening. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for rapid prediction and optimization of molecular sieve performance across temperature ranges for CF4 / N2 separation.

[0008] The objective of this invention is achieved through the following technical solution: A rapid prediction and optimization method for molecular sieve performance across temperature range for CF4 / N2 separation is proposed, which mainly includes the following specific steps: Step S1: Calculate the single-component adsorption capacity and binary mixed adsorption capacity of CF4 and N2 at several discrete temperatures and several discrete pressure points in the molecular sieve database based on high-throughput molecular simulation. Step S2: Based on the single-component adsorption capacity of CF4 and N2, fit several typical adsorption isotherm models, select the optimal model, and calculate the adsorption capacity-temperature-pressure data matrix under a specific adsorption capacity. Step S3: Based on the Clausius-Clapeyron equation, the least squares regression method is used to fit the isotropic adsorption heat (Q) corresponding to a specific adsorption amount for a specific material. st ) qi and equation constant C qi ; Step S4: Using the structural descriptor, chemical descriptor, temperature, and pressure of the molecular sieve material as input features, predict the isotropic adsorption heats (Q) of CF4 and N2 fitted in Step S3 based on the transformer model. st ) qi and equation constant C qi ; Step S5: Using the adsorption amount corresponding to different pressures at one temperature as a benchmark, data interpolation is performed by fitting the Clausius-Clapeyron equation to obtain the pressure values ​​corresponding to these adsorption amounts at different temperatures, thereby quickly generating a complete single-component adsorption isotherm at any new temperature. Step S6: Predict the multi-component adsorption selectivity based on the IAST algorithm (Ideal Adsorbed Solution Theory), compare it with the actual mixed component selectivity calculated by GCMC (Grand Canonical Monte Carlo Simulation), and also compare it with the prediction results of the IAST Model_Isotherm model (isotherm model) algorithm and the "black box" model of machine learning to verify the efficiency, accuracy and applicable temperature range of the method in obtaining selectivity data; based on this, quickly extend the multi-component adsorption selectivity at any temperature within the applicable temperature range; Step S7: Within the applicable temperature range, considering both CF4 adsorption capacity and CF4 / N2 selectivity, select high-performance molecular sieves and their optimal operating temperatures.

[0009] Furthermore, the GCMC calculation in step S1 is implemented using RASPA 2.0 software, and the adsorption simulation conditions include temperature points and pressure points; the temperature points include 273 K, 298 K, 330 K, and 373 K; the pressure points include 10000 Pa, 30000 Pa, 50000 Pa, 70000 Pa, 90000 Pa, and 100000 Pa; the selection and distribution of temperature and pressure are set based on the specificity of the subsequent prediction task, and are not arbitrary or conventional sampling.

[0010] Furthermore, the candidate isotherm models used in step S2 include Radke-Prausnitz, Redlich-Peterson, Freundlich, Quadratic, DSL, Langmuir, Temkin, BET, and quaKK. These models require cross-validation, while also considering mathematical compatibility and extrapolation stability within the subsequent thermodynamic equation framework, selecting the best model from each set of fits. The cross-validation includes R... 2 RMSE and MAE are fitting performance indicators.

[0011] Furthermore, the Clausius–Clapeyron equation in step S3 is:

[0012] In the formula, P(qi,T) is the equilibrium pressure corresponding to the adsorption amount q at temperature T, R is the gas constant, and (Q st ) qi With C qi The adsorption heat and equation constants were obtained by least-squares linear fitting of at least three sets of temperature-pressure data (1 / T, lnP).

[0013] Furthermore, in step S4, the structural descriptor of the molecular sieve material includes the maximum cavity diameter (LCD), the confined pore size (PLD), the accessible specific surface area, the accessible volume, the pore-occupied accessible volume, and the material density; the chemical descriptor includes the adsorption capacity of CF4 single component, the adsorption capacity of N2 single component, the Henry's constant / adsorption heat of CF4, and the Henry's constant / adsorption heat of N2; using the descriptor as input, the trained Transformer model directly outputs the isoproton adsorption heat (Q) at the specified adsorption capacity. st ) qi and the Clausius-Clapeyron equation constant C qi .

[0014] Furthermore, in step S5, the pressure data of different adsorption amounts at different temperatures calculated based on the Clausius-Clapeyron equation are used to attempt to fit all the adsorption isotherm models in step S2. For different materials, the model with the highest fitting degree is selected for interpolation to back-calculate the adsorption amount corresponding to the actual pressure point (i.e., the pressure point selected by GCMC calculation), thereby forming an enhanced single-component isotherm data matrix under multiple temperature conditions.

[0015] As a preferred embodiment of the present invention, in step S6, the IAST calculation of adsorption selectivity preferentially adopts the Interpolator_Isotherm interpolation algorithm, and the degree of agreement between the predicted selectivity and the actual selectivity calculated by GCMC is evaluated under different pressure and temperature conditions to determine the applicable temperature / pressure range of the invention; the degree of agreement includes the coefficient of determination R. 2 RMSE and MAE indicators.

[0016] As a preferred embodiment of the present invention, the selectivity prediction of the IAST Interpolator_Isotherm interpolation algorithm and the IAST Model_Isotherm model algorithm in step S6 is compared, and the adsorption model includes Langmuir, Quadratic, Brunauer-Emmett-Teller (BET), Dual-site Langmuir (DSLangmuir), and Temkin model.

[0017] As a preferred embodiment of the present invention, the selectivity predicted by the IAST Interpolator_Isotherm interpolation algorithm and the machine learning prediction model in step S6 is compared. The machine learning algorithm includes transformer, SVM, XGB, RF, HGB, ExtraTrees, and Bagging, and its prediction performance is evaluated using k-fold cross-validation.

[0018] Furthermore, in step S7, the CF4 adsorption capacity and CF4 / N2 selectivity of the candidate molecular sieves within the applicable temperature range are obtained; the product or weighted score of the adsorption capacity and selectivity is used as a comprehensive performance index to screen out the molecular sieve with the best comprehensive performance and its corresponding working pressure / temperature; the screened high-performance molecular sieve adsorbents include molecular sieves with AEI and MFS topologies.

[0019] The working process and principle of this invention are as follows: Addressing the drawbacks of time-consuming and resource-intensive GCMC calculations, and the potential for large isotherm fitting errors due to limited adsorption data, this method proposes a rapid data generation method combining an adsorption isotherm model with the Clausius-Clapeyron (CCR) thermodynamic equation. This method utilizes GCMC data at a limited number of temperature anchor points (selected to cover the industrial operating range) and combines this data with the CCR equation to fit the isothermal heat of adsorption (Q) corresponding to a specific adsorption amount. st ) qi and equation constant C qi Then, a Transformer model is constructed to find the intrinsic mathematical relationship between the material's structural features and the aforementioned thermodynamic parameters. This process can be repeated to cover all (Q) values ​​across the entire adsorption capacity range. st ) qi and C qi This invention utilizes the CCR equation to quickly and accurately reconstruct adsorption data at any temperature over a wide temperature range, reconstructing a complete adsorption isotherm at any new temperature. This avoids time-consuming global GCMC calculations for each temperature point, significantly reducing computation time and resource consumption. The invention creatively discovers that the adsorption data calculated by GCMC can be transformed into (Q...) using the CCR equation. st ) qi and C qi This allows for a more effective capture of the intrinsic relationship between molecular sieve structure and temperature-dependent adsorption performance, resulting in higher learning efficiency and more accurate predictions for the Transformer model.

[0020] Furthermore, this method also utilizes AI models (especially advanced Transformer architectures) to predict specific parameters of the CCR equation from the descriptor of the molecular sieve (Q). st ) qi and C qi This approach bypasses the traditional requirement of a complete isotherm for calculation (Q). st ) qi This streamlined the previously cumbersome process, enabling rapid end-to-end prediction from structural descriptors to core thermodynamic parameters. It is particularly important to note that (Q... st ) qi and C qi These are complex thermodynamic functions that vary with the amount of adsorption. Directly and accurately predicting them places extremely high demands on the model architecture and training. Therefore, the Transformer model (whose attention mechanism can better handle the complex nonlinear relationships and long-range dependencies between molecular sieve descriptors) was chosen instead of a regular fully connected neural network.

[0021] Furthermore, addressing the drawback of traditional IAST methods, which suffer from decreased selective prediction accuracy due to data sparsity or inappropriate model selection, this invention proposes an IAST interpolation algorithm based on data augmentation using the Clausius-Clapeyron equation. This method utilizes highly reliable single-component adsorption data augmented with data as input and directly employs isotherm integration (i.e., interpolation) for selective prediction. Practical application demonstrates that, compared to pure "black box" machine learning models, this method, incorporating physical laws (CCR+IAST), exhibits better extrapolation and interpretability; and compared to traditional IAST algorithms relying on adsorption isotherm models, this method significantly improves data acquisition efficiency and cross-temperature-range prediction accuracy due to the introduction of an AI architecture. Therefore, the method described in this invention achieves a deep integration and unification of "rigorous physical mechanisms" and "efficient AI prediction."

[0022] Compared with the prior art, the present invention also has the following advantages: (1) The method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation provided by the present invention is highly efficient: by introducing the Clausius–Clapeyron equation for thermodynamic interpolation, only a small amount of GCMC calculation data at temperature anchor points is needed to quickly generate adsorption data at any temperature in a wide temperature range, avoiding massive direct simulation calculations, significantly improving screening efficiency and saving computing resources.

[0023] (2) The method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation provided by this invention is reliable: by screening the optimal adsorption model for precise parameterization and combining it with the Clausius–Clapeyron equation with clear physical meaning for data enhancement, the physical rationality and accuracy of the generated data are ensured. The selectivity is predicted using the validated IAST interpolation method, and the prediction results are in high agreement with the GCMC results in the range of 298 K-373 K, demonstrating high reliability.

[0024] (3) The method for rapid prediction and optimization of molecular sieve performance across temperature ranges for CF4 / N2 separation provided by this invention is intelligent and universal: the method forms a complete closed-loop process of "mechanism simulation - model parameterization - thermodynamic extension - performance prediction - intelligent screening", realizing the dual-drive of mechanism and data. This method is not only applicable to the CF4 / N2 system, but its technical ideas can also be extended to the adsorbent screening of other gas separation systems, and has wide applicability. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the method for rapid prediction and optimization of molecular sieve performance across temperature ranges for CF4 / N2 separation provided by this invention.

[0026] Figure 2 This is a schematic diagram of the distribution of CF4 / N2 selectivity and molecular sieve physical structure characteristics provided by the present invention.

[0027] Figure 3 This is a schematic diagram of the optimal adsorption isotherm model provided by the present invention.

[0028] Figure 4 This is a schematic diagram provided by the present invention for verifying the accuracy of thermodynamic data enhancement methods.

[0029] Figure 5 This is a schematic diagram illustrating the predictive selectivity effect of the IAST interpolation method provided by the present invention.

[0030] Figure 6 This is a schematic diagram illustrating the further analysis of the performance of the IAST interpolation method as a function of pressure, provided by the present invention.

[0031] Figure 7 This is a comparative schematic diagram of the selectivity calculation effect of the IAST model method provided by the present invention.

[0032] Figure 8 This is a schematic diagram comparing the selectivity performance of the seven machine learning models provided by this invention in directly predicting the entire temperature range.

[0033] Figure 9 This is a schematic diagram comparing the overlap between the IAST predicted selectivity and the GCMC calculated selectivity of the three molecular sieves (AEI, MFS, RWR) provided by this invention at different temperatures. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the present invention clearer and more explicit, the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0035] Example 1: like Figures 1 to 9 As shown in the figure, this embodiment discloses a method for rapid prediction and optimization of molecular sieve performance across temperature ranges for CF4 / N2 separation. The method mainly includes the following specific steps: Step S1: Calculate the single-component adsorption capacity and binary mixed adsorption capacity of CF4 and N2 at several discrete temperatures and several discrete pressure points in the molecular sieve database based on high-throughput molecular simulation. Step S2: Based on the single-component adsorption capacity of CF4 and N2, fit several typical adsorption isotherm models, select the optimal model, and calculate the adsorption capacity-temperature-pressure data matrix under a specific adsorption capacity. Step S3: Based on the Clausius-Clapeyron equation, the least squares regression method is used to fit the isotropic adsorption heat (Q) corresponding to a specific adsorption amount for a specific material. st ) qi and equation constant C qi ; Step S4: Using the structural descriptor, chemical descriptor, temperature, and pressure of the molecular sieve material as input features, predict the isotropic adsorption heats (Q) of CF4 and N2 fitted in Step S3 based on the transformer model. st ) qi and equation constant C qi ; Step S5: Using the adsorption amount corresponding to different pressures at one temperature as a benchmark, data interpolation is performed by fitting the Clausius-Clapeyron equation to obtain the pressure values ​​corresponding to these adsorption amounts at different temperatures, thereby quickly generating a complete single-component adsorption isotherm at any new temperature. Step S6: Predict the multi-component adsorption selectivity based on the IAST algorithm (Ideal Adsorbed Solution Theory), compare it with the actual mixed component selectivity calculated by GCMC (Grand Canonical Monte Carlo Simulation), and also compare it with the prediction results of the IAST Model_Isotherm model (isotherm model) algorithm and the "black box" model of machine learning to verify the efficiency, accuracy and applicable temperature range of the method in obtaining selectivity data; based on this, quickly extend the multi-component adsorption selectivity at any temperature within the applicable temperature range; Step S7: Within the applicable temperature range, considering both CF4 adsorption capacity and CF4 / N2 selectivity, select high-performance molecular sieves and their optimal operating temperatures.

[0036] Furthermore, the GCMC calculation in step S1 is implemented using RASPA 2.0 software, and the adsorption simulation conditions include temperature points and pressure points; the temperature points include 273 K, 298 K, 330 K, and 373 K; the pressure points include 10000 Pa, 30000 Pa, 50000 Pa, 70000 Pa, 90000 Pa, and 100000 Pa; the selection and distribution of temperature and pressure are set based on the specificity of the subsequent prediction task, and are not arbitrary or conventional sampling.

[0037] Furthermore, the candidate isotherm models used in step S2 include Radke-Prausnitz, Redlich-Peterson, Freundlich, Quadratic, DSL, Langmuir, Temkin, BET, and quaKK. These models require cross-validation, while also considering mathematical compatibility and extrapolation stability within the subsequent thermodynamic equation framework, selecting the best model from each set of fits. The cross-validation includes R... 2 RMSE and MAE are fitting performance indicators.

[0038] Furthermore, the Clausius–Clapeyron equation in step S3 is:

[0039] In the formula, P(qi,T) is the equilibrium pressure corresponding to the adsorption amount q at temperature T, R is the gas constant, and (Q st ) qi With C qi The adsorption heat and equation constants were obtained by least-squares linear fitting of at least three sets of temperature-pressure data (1 / T, lnP).

[0040] Furthermore, in step S4, the structural descriptor of the molecular sieve material includes the maximum cavity diameter (LCD), the confined pore size (PLD), the accessible specific surface area, the accessible volume, the pore-occupied accessible volume, and the material density; the chemical descriptor includes the adsorption capacity of CF4 single component, the adsorption capacity of N2 single component, the Henry's constant / adsorption heat of CF4, and the Henry's constant / adsorption heat of N2; using the descriptor as input, the trained Transformer model directly outputs the isoproton adsorption heat (Q) at the specified adsorption capacity. st ) qi and the Clausius-Clapeyron equation constant C qi .

[0041] Furthermore, in step S5, the pressure data of different adsorption amounts at different temperatures calculated based on the Clausius-Clapeyron equation are used to attempt to fit all the adsorption isotherm models in step S2. For different materials, the model with the highest fitting degree is selected for interpolation to back-calculate the adsorption amount corresponding to the actual pressure point (i.e., the pressure point selected by GCMC calculation), thereby forming an enhanced single-component isotherm data matrix under multiple temperature conditions.

[0042] As a preferred embodiment of the present invention, in step S6, the IAST calculation of adsorption selectivity preferentially adopts the Interpolator_Isotherm interpolation algorithm, and the degree of agreement between the predicted selectivity and the actual selectivity calculated by GCMC is evaluated under different pressure and temperature conditions to determine the applicable temperature / pressure range of the invention; the degree of agreement includes the coefficient of determination R. 2 RMSE and MAE indicators.

[0043] As a preferred embodiment of the present invention, the selectivity prediction of the IAST Interpolator_Isotherm interpolation algorithm and the IAST Model_Isotherm model algorithm in step S6 is compared, and the adsorption model includes Langmuir, Quadratic, Brunauer-Emmett-Teller (BET), Dual-site Langmuir (DSLangmuir), and Temkin model.

[0044] As a preferred embodiment of the present invention, the selectivity predicted by the IAST Interpolator_Isotherm interpolation algorithm and the machine learning prediction model in step S6 is compared. The machine learning algorithm includes transformer, SVM, XGB, RF, HGB, ExtraTrees, and Bagging, and its prediction performance is evaluated using k-fold cross-validation.

[0045] Furthermore, in step S7, the CF4 adsorption capacity and CF4 / N2 selectivity of the candidate molecular sieves within the applicable temperature range are obtained; the product or weighted score of the adsorption capacity and selectivity is used as a comprehensive performance index to screen out the molecular sieve with the best comprehensive performance and its corresponding working pressure / temperature; the screened high-performance molecular sieve adsorbents include molecular sieves with AEI and MFS topologies.

[0046] Example 2: This embodiment details the complete process from basic data generation to data augmentation and verification. For example... Figure 1 As shown, this method includes the following steps: S1. Using the IZA zeolite database as the material library, and employing RASPA 2.0 software, the single-component adsorption isotherms of CF4 and N2 and the mixed adsorption capacity at a molar ratio of CF4:N2=1:9 were calculated by GCMC simulation at four temperatures: 273 K, 298 K, 330 K, and 373 K. The pressure points were set at 10000, 30000, 50000, 70000, 90000, and 100000 Pa.

[0047] The adsorption selectivity of GCMC can be calculated from the mixed adsorption amount calculated by GCMC:

[0048] Plotting the CF4 / N2 selectivity calculated using GCMC against the physical structural characteristics of the molecular sieve (e.g.) Figure 2 As shown in the figure, this graph consists of four subplots, corresponding to four temperatures: 273 K, 298 K, 330 K, and 373 K. Each subplot is a honeycomb scatter plot, with the horizontal axis representing the normalized values ​​of CF4 / N2 selectivity and various physical descriptors (including maximum cavity diameter, confined pore size, accessible specific surface area, accessible volume, pore occupied accessible volume, and density). Black dots in the graph represent materials with outliers in selectivity, while gray dots represent materials without outliers. These materials with outliers in selectivity are also marked in black on the physical structure feature honeycomb plot. It can be seen that the lower the temperature, the more materials with outliers there are, and these outliers have relatively similar structural characteristics, such as the normalized values ​​of the maximum cavity diameter (PLD) concentrated between 0.6 and 0.8.

[0049] S2. Multiple adsorption isotherm models (Radke-Prausnitz, Redlich-Peterson, Freundlich, Quadratic, DSL, Langmuir, Temkin, BET, quaKK) were used to fit the single-component adsorption data. For example... Figure 3 As shown, each subplot in this figure represents a temperature (273 K, 298 K, 330 K, 373 K). The figure is presented in violin plot format, with the horizontal axis representing different adsorption isotherm models and the vertical axis representing the goodness-of-fit index log1p_RMSE. In each subplot, the gray half of the violin represents the fitting error distribution for CF4, and the black half represents the fitting error distribution for N2. By systematically comparing the logarithm of the root mean square error (log1p_RMSE), the Radke-Prausnitz model was determined to be the optimal model, exhibiting the smallest and most stable fitting error. The model parameters for all gas-material-temperature combinations were saved to complete the precise parameterization calibration of the initial adsorption behavior.

[0050] The error in step S2 is the fitting error between the GCMC data and the adsorption model, denoted by RMSE, and the formula is as follows:

[0051] S3. Based on the above calibration results, perform thermodynamically driven data augmentation: (1) Equivalent adsorption heat fitting: The single-component adsorption amount at 373 K was selected as the specific adsorption amount qᵢ. Using the saved Radke-Prausnitz model parameters, the equilibrium pressure of each qᵢ at four temperature points was calculated in reverse, forming (T, P) data pairs. According to the Clausius-Clapeyron equation lnP(qi,T) = -(Q st ) qi / (R·T) + C(qi), the equivalent adsorption heat (Q) corresponding to each qᵢ is obtained by fitting using the least squares method. st ) qi And the constant C(qi).

[0052] (2) Data interpolation and validation: using the fitted (Q) st ) qi Adsorption isotherms for any temperature range (including the original four temperature points) from 273 K to 373 K were re-interpolated using the Clausius-Clapeyron equation (CCR). To verify the reliability of this method, the interpolated adsorption amount was compared with the adsorption amount directly calculated by the original GCMC. Figure 4 As shown, the first row of four subplots represents the data for CF4 at 273 K, 298 K, 330 K, and 373 K, respectively, and the second row represents the corresponding data for N2. Each subplot shows a scatter plot of the correlation between the single-component adsorption capacity calculated using the Clausius-Clapeyron equation and the single-component adsorption capacity calculated using the original GCMC. The coefficient of determination R is marked on the plot. 2 Within the range of 298 K to 373 K, the coefficient of determination R between the estimated and calculated values ​​of CF4 and N2 is... 2 All are above 0.98 (e.g., R of CF4 at 298 K). 2 =0.9824, R of N2 2 =0.9970), demonstrating the extremely high reliability of this data enhancement method, which can replace time-consuming direct GCMC calculations for rapidly generating adsorption data over a wide temperature range. However, it is noted that the adsorption amount calculated by the CCR equation at lower temperatures (273 K) deviates significantly from the actual GCMC calculation results. This may be because molecular diffusion is slow at low temperatures, making it difficult for the CCR equation to accurately describe the specific interactions between molecules and materials, and between molecules and molecules, at low temperatures. Therefore, the applicable temperature range of this invention should be at least above 298 K.

[0053] Here, the indicator for evaluating the accuracy of step S3 is the coefficient of determination R. 2 The formula is as follows:

[0054] Based on the enhanced high-density single-component adsorption data, three selectivity prediction schemes were compared: (1) IAST interpolation method (preferred method in this invention): The augmented data is used as the input of the IAST interpolation method to predict the CF4 / N2 selectivity. For example... Figure 5 As shown, the four subplots in this figure correspond to temperatures of 273 K, 298 K, 330 K, and 373 K, respectively. The vertical axis of each subplot represents the CF4 / N2 selectivity predicted using IAST interpolation, and the horizontal axis represents the selectivity calculated using GCMC. The coefficient of determination R is labeled in the figure for each temperature. 2 And the number of valid data points N. Within the range of 298 K to 373 K, the R-value of the 298 K predicted value versus the GCMC calculated value. 2 Better than 0.75, R0.75 for 330 K to 373 K predicted values ​​compared to GCMC calculated values. 2 The accuracy exceeds 0.91 and can be applied to over 200 materials, achieving a good balance between accuracy and universality. However, below 298 K, such as at 273 K, the selectivity predicted by the IAST interpolation method deviates significantly from the actual GCMC calculation results. This may be because the CCR equation struggles to describe the actual interactions between molecules and materials at low temperatures (e.g., ...). Figure 4 (As shown).

[0055] like Figure 6 The diagram shown is a further analysis of the performance of the IAST interpolation method as a function of pressure in an embodiment of the IAST interpolation method of the present invention. Figure 6 The top chart is a scatter plot, with the vertical axis representing the ratio of the selectivity predicted by IAST to the actual selectivity calculated by GCMC, and the horizontal axis representing different pressure points (10, 30, 50, 70, 90, 100 kPa), with different shades of gray dots distinguishing temperature. The bottom chart is a bar chart, with the vertical axis representing R... 2 Different filling patterns were used to distinguish temperatures, and each column was labeled with a specific R value. 2 As can be seen from the values, the accuracy of the IAST interpolation method for predicting selectivity is affected not only by temperature but also by pressure. Higher pressure results in higher accuracy. This may be because at low pressures, the molecular dispersion within the material cavity is relatively sparse, making it difficult to reach adsorption potential equilibrium for the mixed components. Since the IAST theory relies on adsorption equilibrium, the prediction effect is poor at low pressures. Therefore, when using the method of this invention to predict the selectivity of mixed components, attention should be paid to the applicable temperature and pressure ranges. Special care should be taken with excessively low temperatures (below 298 K) or pressures (below 3000 Pa).

[0056] (2) IAST Model Method (Comparative Example): Augmented data is used as input to the IAST model method to predict CF4 / N2 selectivity. For example... Figure 7 As shown, the subplots correspond to four temperatures. The vertical axis of each subplot represents the selectivity calculated using the IAST model, and the horizontal axis represents the selectivity calculated using GCMC. The R value is marked on the graph. 2 And the amount of material that can be successfully calculated. This method completely fails at 273 K (R0). 2 =-1.9588), and the number of materials that can be successfully calculated at high temperatures decreases sharply (only 28 at 373 K), making it impractical.

[0057] (3) Machine learning prediction method (comparative example): Seven mainstream models (TF, SVM, XGB, RF, HGB, ET, Bag) are used to directly predict selectivity across the entire temperature range (273 K, 298 K, 330 K, and 373 K). For example... Figure 8 As shown, the upper subplot box plot has the ordinate representing the ratio of predicted selectivity to true selectivity, and the abbreviation of the seven models (TF, SVM, XGB, RF, HGB, ET, Bag) on ​​the x-axis; the lower subplot is a dual-axis histogram, with the left axis representing R... 2 (Dark gray bars), right vertical axis represents MAE (light gray grid bars). Test set R for the best model. 2 The accuracy is only about 0.792, which is less than that of the IAST interpolation method.

[0058] Therefore, through comparison of the three methods, it can be verified that the "IAST interpolation method based on Clausius-Clapeyron equation-enhanced data" of this invention exhibits excellent accuracy and versatility within its applicable temperature and pressure range. Based on this, further screening and verification of high-performance molecular sieves are conducted. Within the applicable temperature range of 273 K-373 K, the adsorption capacity and selectivity data at more temperatures are expanded using the CCR equation. Then, considering the CF4 adsorption capacity and selectivity indicators comprehensively, high-performance molecular sieves are screened. Figure 9 The figures show the CF4 / N2 selectivity of three topological molecular sieves: AEI, MFS, and RWR, with subplots corresponding to four temperatures. In each subplot, the horizontal axis represents pressure (Pa), and the vertical axis represents the selectivity calculated by GCMC or IAST interpolation. Different scatter plot shapes represent the GCMC calculation results. The IAST predicted selectivity of these three materials shows a high degree of agreement with the actual GCMC calculation results at different temperatures and pressures, verifying the accuracy and efficiency of the method of this invention in predicting the selectivity of mixed components. Furthermore, the CF4 / N2 selectivity of AEI and MFS topological molecular sieves can reach over 10 at 273 K and 10000 Pa, demonstrating the high efficiency of the method of this invention in screening high-performance separation materials.

[0059] The above method can be programmed into a computer, stored on a computer-readable storage medium (such as a USB flash drive, hard disk, or memory), and executed by a computing device (such as a computer or server) containing a processor and memory, thereby achieving automated intelligent screening.

[0060] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation, characterized in that, Includes the following steps: Step S1: Calculate the single-component adsorption capacity and binary mixed adsorption capacity of CF4 and N2 at several discrete temperatures and several discrete pressure points in the molecular sieve database based on high-throughput molecular simulation. Step S2: Based on the single-component adsorption capacity of CF4 and N2, fit several typical adsorption isotherm models, select the optimal model, and calculate the adsorption capacity-temperature-pressure data matrix under a specific adsorption capacity. Step S3: Based on the Clausius-Clapeyron equation, least squares regression fitting is used to fit the isotropic adsorption heat (Q) corresponding to a specific adsorption amount for a specific material. st ) qi and equation constant C qi ; Step S4: Using the structural descriptor, chemical descriptor, temperature, and pressure of the molecular sieve material as input features, predict the isotropic adsorption heats (Q) of CF4 and N2 fitted in Step S3 based on the transformer model. st ) qi and equation constant C qi ; Step S5: Using the adsorption amount corresponding to different pressures at one temperature as a benchmark, data interpolation is performed by fitting the Clausius-Clapeyron equation to obtain the pressure values ​​corresponding to these adsorption amounts at different temperatures, thereby quickly generating a complete single-component adsorption isotherm at any new temperature. Step S6: Predict the multi-component adsorption selectivity based on the IAST algorithm, compare it with the actual mixed component selectivity calculated by GCMC, and compare it with the prediction results of the IAST Model_Isotherm model algorithm and the "black box" model of machine learning to verify the efficiency, accuracy and applicable temperature range of the method in obtaining selectivity data; based on this, quickly extend the multi-component adsorption selectivity at any temperature within the applicable temperature range. Step S7: Within the applicable temperature range, considering both CF4 adsorption capacity and CF4 / N2 selectivity, select high-performance molecular sieves and their optimal operating temperatures.

2. The method for rapid prediction and optimization of molecular sieve performance across temperature ranges for CF4 / N2 separation according to claim 1, characterized in that the GCMC calculation in step S1 adopts RASPA. The adsorption simulation is implemented using software 2.0, and the adsorption simulation conditions include temperature points and pressure points. The temperature points include 273 K, 298 K, 330 K, and 373 K; the pressure points include 10000 Pa, 30000 Pa, 50000 Pa, 70000 Pa, 90000 Pa, and 100000 Pa. The selection and distribution of temperature and pressure are set based on the specific characteristics of the subsequent prediction task and are not arbitrary or conventional sampling.

3. The method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation according to claim 1, characterized in that, The candidate isotherm models used in step S2 include Radke-Prausnitz, Redlich-Peterson, Freundlich, Quadratic, DSL, Langmuir, Temkin, BET, and quaKK. These models require cross-validation, and the best model from each fitting group is selected, considering both mathematical compatibility and extrapolation stability within the subsequent thermodynamic equation framework. The cross-validation includes R... 2 RMSE and MAE are fitting performance indicators.

4. The method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation according to claim 1, characterized in that, The Clausius–Clapeyron equation in step S3 is: In the formula, P(qi,T) is the equilibrium pressure corresponding to the adsorption amount q at temperature T, R is the gas constant, and (Q st ) qi With C qi The adsorption heat and equation constants were obtained by least-squares linear fitting of at least three sets of temperature-pressure data (1 / T, lnP).

5. The method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation according to claim 1, characterized in that, The structural descriptor of the molecular sieve material in step S4 includes the maximum cavity diameter, the confined pore size, the accessible specific surface area, the accessible volume, the pore-occupied accessible volume, and the material density; the chemical descriptor includes the adsorption capacity of CF4 single component, the adsorption capacity of N2 single component, the Henry's constant / heat of adsorption of CF4, and the Henry's constant / heat of adsorption of N2. Taking a descriptor as input, the trained Transformer model directly outputs the equivalent heat of adsorption (Q) at a specified adsorption amount. st ) qi and the Clausius-Clapeyron equation constant C qi .

6. The method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation according to claim 1, characterized in that, In step S5, the pressure data of different adsorption amounts at different temperatures calculated based on the Clausius-Clapeyron equation are used to attempt to fit the adsorption isotherm models in step S2. For different materials, the model with the highest fitting degree is selected for interpolation to inversely calculate the adsorption amount corresponding to the actual pressure point, thereby forming an enhanced single-component isotherm data matrix under multiple temperature conditions.

7. The method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation according to claim 1, characterized in that, In step S6, the IAST calculation of adsorption selectivity preferentially uses the Interpolator_Isotherm interpolation algorithm, and evaluates the degree of agreement between the predicted selectivity and the actual selectivity calculated by GCMC under different pressure and temperature conditions to determine the applicable temperature / pressure range of the invention; the degree of agreement includes the coefficient of determination R. 2 RMSE and MAE indicators.

8. The method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation according to claim 1, characterized in that, The selectivity predictions of the IAST Interpolator_Isotherm interpolation algorithm and the IAST Model_Isotherm model algorithm in step S6 are compared. The adsorption models include Langmuir, Quadratic, Brunauer-Emmett-Teller, Dual-site Langmuir, and Temkin models.

9. The method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation according to claim 1, characterized in that, In step S6, the selectivity of the IAST Interpolator_Isotherm interpolation algorithm is compared with that predicted by the machine learning prediction model. The machine learning algorithms include transformer, SVM, XGB, RF, HGB, ExtraTrees, and Bagging, and their prediction performance is evaluated using k-fold cross-validation.

10. The method for rapid prediction and optimization of molecular sieve performance across temperature range for CF4 / N2 separation according to any one of claims 1 to 9, characterized in that, In step S7, the CF4 adsorption capacity and CF4 / N2 selectivity of the candidate molecular sieves within the applicable temperature range are obtained; the product or weighted score of the adsorption capacity and selectivity is used as a comprehensive performance index to screen out the molecular sieve with the best comprehensive performance and its corresponding working pressure / temperature; the screened high-performance molecular sieve adsorbents include molecular sieves with AEI and MFS topologies.

Images