Screening method of zeolite molecular sieve for adsorbing electronic gas
Through high-throughput computational simulation and machine learning algorithms, the most suitable zeolite molecular sieve is screened for electronic gas adsorption, solving the problems of high energy consumption and equipment requirements in the existing technology, and achieving rapid and efficient material screening.
Patent Information
- Application Number
- CN202311152038.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-07
- Publication Date
- 2025-03-07
AI Technical Summary
The existing electronic special gas separation and purification technology consumes high energy and has strict equipment requirements. Due to the wide variety of porous materials, it is difficult to quickly screen out the most suitable materials for specific gas adsorption.
High-throughput calculation simulation and machine learning algorithms are used to obtain the adsorption amount of electron gas by zeolite molecular sieves, and the most important structural features are screened using random forest model and linear model to screen zeolite molecular sieves.
It has achieved rapid screening of zeolite molecular sieves that efficiently adsorb electron gases, overcome the problems of high energy consumption and equipment requirements in the original technology, and significantly accelerated the speed of material screening.
Smart Images
Figure CN119580882A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the fields of computational chemistry and electronic gas adsorption and separation technology, and in particular to a method for screening zeolite molecular sieves for adsorbing electronic gases. Background Art
[0002] Electronic specialty gases refer to a group of highly purified gases specifically used in electronic manufacturing processes, designed to meet stringent purity and quality requirements to ensure optimal performance and reliability of electronic equipment and components. The production of electronic specialty gases involves rigorous manufacturing and purification processes to achieve the required purity levels. The importance of electronic specialty gas separation and purification technology is becoming increasingly prominent, and the existing main separation method, such as distillation, is not only energy-intensive, but also has very demanding equipment requirements.
[0003] In contrast, adsorption separation technology has many advantages such as low cost, low energy consumption, low equipment requirements and high selectivity. As more and more porous materials are synthesized, zeolite molecular sieve materials have been widely used in chemical production. Among many porous materials, zeolite molecular sieves not only have the high stability of activated carbon, but also have the structural adjustability of MOF, so it is one of the most promising materials in electronic gas adsorption separation technology.
[0004] Porous materials are rich in variety, and a variety of pore structures, pore microenvironments, and active sites can be designed, which makes it possible to specifically select and adsorb a certain gas. However, due to the large number of materials available, selecting the most suitable porous material for a specific gas purification scenario has become a research hotspot. The key to solving this problem is to be able to quickly obtain the performance of candidate materials and evaluate whether the materials meet the requirements of the current adsorption scenario based on the performance of each material.
[0005] As a rapidly developing technology, machine learning can learn rules from large amounts of data, so it can be used to model structure-activity relationships. Compared with computational simulation, machine learning can greatly shorten the calculation time and reduce the computing resources used. It can provide the properties of materials with relatively accurate accuracy and extremely fast time, thus facilitating large-scale material screening.
[0006] Therefore, it is necessary to design a screening method for zeolite molecular sieves that efficiently adsorb electron gases to solve the above problems. Summary of the invention
[0007] In order to solve the above technical problems, the present invention provides a method for screening zeolite molecular sieves that adsorb electron gases, the method comprising the following steps:
[0008] (1) Using high-throughput computational simulation to obtain the adsorption amount of electron gas by zeolite molecular sieves;
[0009] (2) Use machine learning algorithms to model the structural characteristics of zeolite molecular sieves and the adsorption of electron gases;
[0010] (3) Based on the random forest model and / or linear model, the most important structural features that affect the adsorption of different electron gases are obtained to screen zeolite molecular sieves.
[0011] In step (1) of the present invention, the high throughput computing (HTC) is to split the large-scale intensive computing into subtasks and assign them to cluster computers for computing. The high throughput computing is a well-known technology in the art.
[0012] According to an embodiment of the present invention, in step (1), the electronic gas is selected from at least one of AsH3, B2H6, BCl3, C2F6, ClF3, HBr, NF3, PF3, SF6, PH3, SiH4 or GeH4.
[0013] According to the embodiment of the present invention, step (1) is specifically: calculating the geometric parameters of the electron gas by density functional theory, and calculating the RESP charge of the electron gas and the DDEC6 charge of the atoms on the zeolite molecular sieve, and then using the above calculated parameters and UFF force field parameters to simulate the adsorption process of the electron gas on the zeolite molecular sieve by grand canonical Monte Carlo, so as to obtain the adsorption amount of the electron gas by the molecular sieve on the zeolite. Wherein, the calculated parameters include the geometric configuration of the electron gas, the RESP charge of the electron gas and the DDEC6 charge of the atoms on the zeolite molecular sieve.
[0014] According to an embodiment of the present invention, in step (1), a grand canonical Monte Carlo simulation is used to simulate the equilibrium adsorption process of the electron gas on the zeolite molecular sieve to obtain the adsorption amount of the electron gas on the zeolite molecular sieve at normal temperature and pressure, i.e., sample data. The sample data is used to obtain the relationship between the structural characteristics of the electron gas and the adsorption amount of the zeolite molecular sieve.
[0015] Preferably, the geometric parameters of the electron gas molecules are calculated as follows: first, the initial configuration of the electron gas is constructed using GaussView 6.0, and then DFT calculations are performed using Gaussian 16. During the calculations, all electron gases are electrically neutral and the spin multiplicities are set to 1. The specific calculation parameters use the B3LYP hybrid functional, select the def2TZVP basis set, and consider DFT-D3 dispersion correction.
[0016] Preferably, the calculation process of the RESP charge of the electron gas is: firstly, the electrostatic potential of the electron gas is calculated using Gaussian 16, and the result is output to a chk or fchk file, and then the file is read into the Multiwfn program, and then the RESP module is entered for fitting.
[0017] Preferably, the calculation process of the DDEC6 charge of the zeolite molecular sieve is: firstly calculate the total electron density of the electron gas using the quantum chemistry program VASP and output it to a CHGCAR or AECCAR2 file, then read the file using the chargemol program and decompose it.
[0018] According to an embodiment of the present invention, in step (2), the structural characteristics of the zeolite molecular sieve are calculated using the Zeo++ program to obtain a structural characteristic data set.
[0019] According to an embodiment of the present invention, in step (2), the structural characteristics include pore size, density (Density), accessible surface area (ASA), accessible pore volume (AV), accessible porosity (AV_VF) and molecular pore volume that can be occupied by probe molecules.
[0020] Preferably, the aperture includes a global maximum pore diameter (GCD), a pore limit diameter (PLD) and a maximum pore diameter (LCD) as well as PLDs and LCDs in three crystal axis directions (PLD_a, PLD_b, PLD_c, LCD_a, LCD_b and LCD_c).
[0021] Preferably, the probe molecule-occupiable molecular pore volume includes probe molecule accessible occupiable pore volume (POAV), probe molecule inaccessible occupiable pore volume (PONAV), probe molecule accessible porosity (POAV_AV) and probe molecule inaccessible accessible porosity (PONAV_AV).
[0022] According to an embodiment of the present invention, in step (2), a structure-activity relationship model between the structural characteristics of the zeolite molecular sieve and the adsorption amount of the zeolite molecular sieve to the electron gas in step (1) is established.
[0023] Preferably, the structural features include pore size (9 in total, including GCD, PLD, LCD, PLD_a, PLD_b, PLD_c, LCD_a, LCD_b, LCD_c), density, accessible surface area (ASA) of zeolite molecular sieve materials, accessible pore volume (AV), accessible porosity (AV VF), molecular pore volume occupyable by probe molecules (occupiable pore volume accessible to probe molecules, i.e. POAV, occupiable pore volume inaccessible to probe molecules, i.e. PONAV, porosity accessible to probe molecules (POAV_AV) and accessible porosity inaccessible to probe molecules (PONAV_AV) in total 4) for a total of 17.
[0024] According to an embodiment of the present invention, in step (2), the machine learning algorithm includes a multivariate linear model or a random forest model for modeling, and the index for evaluating the model performance is R 2 value.
[0025] In the present invention, the multivariate linear model and random forest model are known models in the art.
[0026] According to an embodiment of the present invention, step (2) also includes establishing a data set, wherein the data set includes sample data, structural feature data and label data; wherein the sample data is zeolite molecular sieve data (for example, pure silica zeolite molecular sieve data, and for example, data of 240 pure silica zeolite molecular sieves), the structural feature data set contains zeolite molecular sieve material data having the structural features, and the label data is the adsorption amount data of the zeolite molecular sieve for electron gas in step (1).
[0027] According to an embodiment of the present invention, in step (2), Z-score is used in the training process, and the structural feature data and label data are standardized at the same time, and then all the data are divided into training set and test set according to the ratio, and the experiment is repeated for multiple times to take the average value. The said all data refers to the structural feature data and label data. Preferably, the said ratio is, for example, a ratio of 4:1. Preferably, each experiment can be repeated at least 5 times, preferably 10 times.
[0028] According to an embodiment of the present invention, in step (3), the most important structural features affecting the adsorption amount of different molecules are obtained using the random forest Gini importance index.
[0029] In a preferred embodiment of the present invention, the method of the present invention comprises the steps of:
[0030] (1) Calculating the geometric parameters of the electron gas by using the B3LYP hybrid functional, and calculating the RESP charge of the electron gas and the DDEC6 charge of the atoms on the zeolite molecular sieve, and then using the above-calculated parameters and the UFF force field parameters to simulate the adsorption process of the electron gas on the zeolite molecular sieve by using the grand canonical Monte Carlo, so as to obtain the adsorption amount of the electron gas by the zeolite molecular sieve; wherein the calculated parameters include the geometric configuration of the electron gas, the RESP charge of the electron gas and the DDEC6 charge of the atoms on the zeolite molecular sieve; wherein the adsorption amount of the electron gas by the zeolite molecular sieve is the label data;
[0031] (2) Using grand canonical Monte Carlo simulation to simulate the equilibrium adsorption of electron gas on zeolite molecular sieves, the adsorption of electron gas on 240 pure silica zeolite molecular sieves at room temperature and pressure was obtained, i.e., 240 sample data;
[0032] The Zeo++ program was used to calculate the structural characteristics of zeolite molecular sieves and obtain a structural characteristic data set;
[0033] Establishing a data set, wherein the data set includes 240 sample data, structural feature data, and label data;
[0034] During the training process, Z-score was used to standardize the structural feature data and label data. The structural feature data and label data were then divided into training sets and test sets according to the proportion. Each experiment was repeated 10 times to take the average value.
[0035] (3) The random forest Gini importance index is used to measure the relative importance of each feature, and the most important structural features that affect the adsorption amount of different molecules are obtained for the screening of zeolite molecular sieves.
[0036] Beneficial effects of the present invention:
[0037] The method for screening zeolite molecular sieves with high efficiency in adsorbing electron gas of the present invention utilizes computational chemistry to perform high-throughput screening of zeolite molecular sieves, thus overcoming the drawbacks of the original high-throughput calculations and high-throughput experiments and greatly accelerating the screening speed. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 This is a flow chart of the method for screening zeolite molecular sieves for efficient adsorption of electronic gases in Example 1 of the present invention;
[0039] Figure 2 is the performance result of the linear model obtained in Example 1 of the present invention on the data set;
[0040] Figure 3 is the performance result of the random forest model obtained in Example 2 of the present invention on the data set;
[0041] Figure 4 It is the importance index of each structural feature of the random forest model taking C2F6 as an example in Example 2 of the present invention.
[0042] Figure 5 It is the importance index of each structural feature of the random forest model taking B2H6 as an example in Example 2 of the present invention. DETAILED DESCRIPTION
[0043] The technical scheme of the present invention will be further described in detail below in conjunction with specific embodiments. It should be understood that the following embodiments are only exemplary descriptions and explanations of the present invention and should not be construed as limiting the scope of protection of the present invention. All technologies implemented based on the above content of the present invention are included in the scope that the present invention is intended to protect.
[0044] Unless otherwise specified, the raw materials and reagents used in the following examples are commercially available or can be prepared by known methods.
[0045] Example 1
[0046] See also Figure 1-5 The present invention provides a method for screening zeolite molecular sieves with high efficiency in adsorbing electron gas, which is based on a machine learning system and comprises the following steps:
[0047] The electronic gases are AsH3, B2H6, BCl3, C2F6, ClF3, HBr, NF3, PF3, SF6, PH3, SiH4 and GeH4 respectively.
[0048] Step 1 includes: selecting the UFF force field to describe the Lennard-Jones interaction between the zeolite molecular sieve and the electron gas and the Lennard-Jones interaction between the electron gases.
[0049] To calculate the geometric parameters, for the selected electron gas molecules, GaussView 6.0 was first used to construct the initial configuration of the electron gas, and then Gaussian 16 was used for DFT calculation. During the calculation, all electron gas molecules were electrically neutral and the spin multiplicity was set to 1. The specific calculation parameters used the B3LYP hybrid functional, the def2TZVP basis set was selected, and the DFT-D3 dispersion correction was considered.
[0050] The RESP charges of the electron gas molecules and the DDEC6 charges of the zeolite molecular sieves are calculated to describe the Coulomb interactions.
[0051] To calculate the RESP charge of the electron gas molecules, first use Gaussian 16 to calculate the electrostatic potential of the electron gas, output it to the chk or fchk file, then use the Multiwfn program to read the file, and then enter the RESP module for fitting;
[0052] To calculate the DDEC6 charge of zeolite molecular sieves, first use the quantum chemistry program VASP to calculate the total electron density of the electron gas and output it to the CHGCAR or AECCAR2 file. Then use the chargemol program to read the file and decompose it.
[0053] Monte Carlo simulation was performed using RASPA2 software, and GCMC simulation was performed using RASPA2 before the formal adsorption simulation to calculate the helium porosity of the zeolite molecular sieve, which was used to convert the absolute adsorption amount into excess adsorption amount. The calculation blueprint came from the RASPA2 document. The specific parameters of the GCMC simulation are as follows: the initial MC step number and equilibrium step number of the GCMC simulation were both set to 8000, and the cutoff radius was set to The simulated temperature is 298.0K, and the simulated pressure is 101000Pa. The Ewald summation method is used to calculate the electrostatic interaction. The equilibrium adsorption of electron gas on zeolite molecular sieves is simulated using grand canonical Monte Carlo simulation, and the adsorption of electron gas on 240 pure silica zeolite molecular sieves at room temperature and pressure is obtained, that is, 240 sample data.
[0054] Step 2: The structural characteristics of the zeolite molecular sieve are calculated using the Zeo++ program to obtain a structural characteristic data set. The structural characteristics include the global maximum pore diameter (GCD), the pore limit diameter (PLD) and the maximum pore diameter (LCD), as well as the PLD and LCD in the three crystal axis directions (PLD_a, PLD_b, PLD_c, LCD_a, LCD_b and LCD_c); density (Density) and accessible surface area (ASA); accessible pore volume (AV) and accessible void fraction (AV_VF); accessible pore volume (POAV) of the probe molecule, occupied pore volume (PONAV) inaccessible to the probe molecule, accessible porosity (POAV_AV) of the probe molecule and accessible porosity (PONAV_AV) inaccessible to the probe molecule.
[0055] A multivariate linear model machine learning algorithm is used to establish a structure-activity relationship model between the above 17 structural features of zeolite molecular sieves and the adsorption amount of electron gas by zeolite molecular sieves in step 1. The existing zeolite-electron gas dataset contains 240 sample data, i.e., 240 pure silicon zeolite molecular sieve data, a structural feature dataset (data of zeolite molecular sieves containing 17 features), and label data (the label data is the adsorption amount data of zeolite molecular sieves for electron gas in step 1).
[0056] The implementation of the linear model is provided by the scikit-learn Python library. During the training process, the Z-score is used to standardize both the structural feature data and the label data. Then all the data (i.e., the feature data and the label data are standardized) are divided into training sets and test sets in a ratio of 4:1, and each experiment is repeated 10 times to take the average value. The regression model has an independent set of evaluation indicators, and the performance evaluation indicator is R 2 Value, R 2 The closer the value is to 1, the better the model performance.
[0057] Step 3: Use a linear model to measure the relative importance of each feature, obtain the most important structural feature that affects the adsorption amount of different molecules, and screen the zeolite molecular sieve based on the most important structural feature.
[0058] Figure 2 This is the performance result of the linear model obtained in Example 1 of the present invention on the data set.
[0059] Example 2
[0060] The difference between Example 2 and Example 1 is that a random forest model is used to replace the linear model in Example 1. The implementation of the random forest model is provided by the scikit-learn Python library.
[0061] Figure 3 This is the performance result of the random forest model obtained in Example 2 of the present invention on the data set.
[0062] contrast Figure 2 and Figure 3 It can be seen that the linear model does not have good prediction results for most molecules, and the random forest model is better than the linear model in predicting the adsorption amount and achieves good results.
[0063] Take C2F6 and B2H6 as an example:
[0064] like Figure 4 As shown, the relative importance of C2F6 descriptors to adsorption selectivity is AV_VF>AV>GCD>Density>LCD_c.
[0065] like Figure 5 As shown, the relative importance of B2H6 descriptors to adsorption selectivity is Density>AV>AV_VF>POAV>GCD.
[0066] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.
[0067] The above is an exemplary description of the embodiments of the present invention. However, the protection scope of the present invention is not limited to the above embodiments. Any modification, equivalent substitution, improvement, etc. made by those skilled in the art within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for screening zeolite molecular sieves that adsorb electron gas, characterized in that: The method comprises the following steps: (1) Using high-throughput computational simulation to obtain the adsorption amount of electron gas by zeolite molecular sieves; (2) Use machine learning algorithms to model the structural characteristics of zeolite molecular sieves and the adsorption of electron gases; (3) Based on the random forest model and / or linear model, the most important structural features that affect the adsorption of different electron gases are obtained to screen zeolite molecular sieves.
2. The method according to claim 1, characterized in that In step (1), the electronic gas is selected from at least one of AsH3, B2H6, BCl3, C2F6, ClF3, HBr, NF3, PF3, SF6, PH3, SiH4 or GeH4.
3. The method according to claim 1 or 2, characterized in that: In step (1), the geometric parameters of the electron gas are calculated by density functional theory, and the RESP charge of the electron gas and the DDEC6 charge of the atoms on the zeolite molecular sieve are calculated. Then, using the above-calculated parameters and UFF force field parameters, the adsorption process of the electron gas on the molecular sieve on the zeolite is simulated by grand canonical Monte Carlo, thereby obtaining the adsorption amount of the electron gas on the molecular sieve on the zeolite.
4. The method according to any one of claims 1 to 3, characterized in that: In step (2), the structural characteristics of the zeolite molecular sieve are calculated using the Zeo++ program to obtain a structural characteristic data set. Preferably, in step (2), the structural characteristics include pore size, density (Density), accessible surface area (ASA), accessible pore volume (AV), accessible porosity (AV_VF) and molecular pore volume that can be occupied by probe molecules.
5. The method according to claim 4, characterized in that The apertures include a global maximum cavity diameter (GCD), a cavity limit diameter (PLD) and a maximum cavity diameter (LCD), as well as PLDs and LCDs in three crystal axis directions (PLD_a, PLD_b, PLD_c, LCD_a, LCD_b and LCD_c). Preferably, the probe molecule-occupiable molecular pore volume includes probe molecule accessible occupiable pore volume (POAV), probe molecule inaccessible occupiable pore volume (PONAV), probe molecule accessible porosity (POAV_AV) and probe molecule inaccessible accessible porosity (PONAV_AV).
6. The method according to any one of claims 1 to 5, characterized in that: In step (2), a structure-activity relationship model between the structural characteristics of the zeolite molecular sieve and the adsorption amount of the electron gas by the zeolite molecular sieve in step (1) is established. Preferably, the structural features include pore size GCD, PLD, LCD, PLD_a, PLD_b, PLD_c, LCD_a, LCD_b, LCD_c, density, accessible surface area (ASA) of zeolite molecular sieve materials, accessible pore volume (AV), accessible porosity (AV VF), accessible pore volume (POAV) of probe molecules, occupied pore volume (PONAV) inaccessible to probe molecules, porosity (POAV_AV) and accessible porosity (PONAV_AV) inaccessible to probe molecules.
7. The method according to any one of claims 1 to 6, characterized in that: In step (2), the machine learning algorithm includes a multivariate linear model or a random forest model for modeling, and the index for model performance evaluation is R 2 value.
8. The method according to any one of claims 1 to 7, characterized in that: Step (2) also includes establishing a data set, which includes sample data, structural feature data and label data; wherein the sample data is zeolite molecular sieve data, the structural feature data set contains zeolite molecular sieve material data having the structural features, and the label data is the adsorption amount data of the zeolite molecular sieve for electron gas in step (1).
9. The method according to any one of claims 1 to 8, characterized in that: Preferably, in step (2), the structural feature data and the label data are standardized by Z-score during the training process, and then all the data are divided into a training set and a test set, and the experiment is repeated multiple times to take the average value. Preferably, all the data are divided into a training set and a test set in a ratio of 4:
1.
10. The method according to any one of claims 1 to 9, characterized in that: In step (3), the random forest Gini importance index is used to obtain the most important structural features that affect the adsorption amount of different molecules.