Data-driven two-dimensional double-metal oxygen cluster carbon dioxide reduction photocatalyst material design method, system and medium

Abstract

The application relates to a data-driven two-dimensional bimetallic oxygen compound carbon dioxide reduction photocatalyst material design method, system and medium, wherein the design method comprises the following steps: S1, establishing an initial database; S2, cleaning and preprocessing the initial data, and determining the machine learning problem attributes of different target variables; S3, feature engineering and model training; S4, designing a new two-dimensional photocatalyst material, and predicting target attributes based on the model of S3; S5, pre-screening photocatalyst candidate materials based on the target attributes predicted in S4; and S6, DFT verification of the candidate materials. Compared with the prior art, the application can provide the influence trend of each feature on the target variable while quickly giving the expected two-dimensional photocatalyst material by learning the target variable data of the known two-dimensional material, so that the number of experiments for finding the two-dimensional photocatalyst is reduced, and the production process of the carbon dioxide reduction photocatalyst is accelerated.

Description

Technical Field

[0001] This invention relates to the field of solar photocatalysis, and in particular to a data-driven design method, system, and medium for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials. Background Technology

[0002] The extensive combustion of fossil fuels and industrial production processes generate large amounts of greenhouse gases, accelerating global warming and causing significant changes in the Earth's climate. Carbon dioxide is a major greenhouse gas, and finding suitable catalysts to convert it into chemical substances (such as CO, CH3OH, CH4, etc.) is key to solving this problem. Among various conversion methods, photocatalysis is an important approach, utilizing clean, safe, and renewable solar energy to drive catalysts for carbon dioxide conversion.

[0003] Two-dimensional materials have attracted much attention in the field of photocatalysis due to their excellent stability, unique light absorption properties, and high surface catalytic activity. A2B2X6, as a bimetallic compound, possesses a distinct two-dimensional layered structure, with each layer composed of A, B, and X atoms stacked together by van der Waals interactions. This two-dimensional layered structure provides a larger specific surface area, allowing more reactants to contact the catalyst surface, thereby improving reaction efficiency. The van der Waals interactions weaken the interlayer binding forces, making it easier for electrons and holes to migrate between layers and reducing the chance of carrier recombination. These characteristics strongly suggest that the two-dimensional A2B2X6 bimetallic compound has the potential to serve as a photocatalyst for carbon dioxide reduction.

[0004] Although there are various methods for synthesizing novel photocatalytic materials in the laboratory, such as sol-gel method and hydrothermal synthesis, these methods are often time-consuming, expensive, and have a very low trial-and-error rate. Blindly carrying out experiments may not meet the needs of large-scale catalyst synthesis.

[0005] The introduction of density functional theory (DFT) has provided researchers with a more efficient method for simulating and screening novel photocatalytic materials. However, DFT research often suffers from excessively high computational costs, slow speeds, and low efficiency when dealing with large-scale systems and complex materials. In recent years, with the development of materials informatics, the combination of machine learning computation and first-principles calculations has rapidly advanced in various fields and has proven to be an effective method for accelerating the materials design process.

[0006] In materials design, how to use machine learning methods to predict the photocatalytic performance of candidate materials and use the prediction results to screen out suitable photocatalysts has become one of the problems that those skilled in the art want to actively solve. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the existing technology and, by combining machine learning algorithms and first-principles calculations, establish an efficient and accurate method to explore novel two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalysts, and to provide a data-driven design method, system, and medium for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] The first aspect of this invention provides a data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials. The design method is implemented by combining machine learning algorithms and first-principles calculations, and specifically includes the following steps:

[0010] S1, Obtain information on the two-dimensional photocatalyst material system of interest, obtain the two-dimensional material data corresponding to the system information, the two-dimensional material data includes the stability information and band structure information of the two-dimensional material, and obtain the input features of the two-dimensional material to establish an initial dataset;

[0011] S2, the initial dataset is cleaned and preprocessed to obtain a modeling dataset for machine learning. The modeling dataset includes the input features, stability and band structure information of the two-dimensional material. The input features are used as variables, and the stability and band structure information are used as target variables. The problem attributes of machine learning are determined according to different target variables.

[0012] S3: Filter the input features of the modeling dataset to obtain the best feature subset, train and evaluate the machine learning model on the best feature subset to obtain the machine learning model, and perform interpretability analysis on the machine learning prediction model.

[0013] S4. By limiting the elements at each site, a theoretical two-dimensional carbon dioxide reduction photocatalyst dataset is created. The theoretical two-dimensional carbon dioxide reduction photocatalyst dataset is used as the input to the machine learning model obtained in S3 to obtain the predicted value of the target variable.

[0014] S5. Based on the predicted values ​​of the target variables, candidate photocatalyst materials are selected.

[0015] S6, perform DFT verification on the photocatalyst candidate materials to screen out materials that meet the conditions for photocatalytic carbon dioxide reduction.

[0016] Furthermore, in S1, the process of establishing the initial database includes the following sub-steps:

[0017] S1.1, Obtain the two-dimensional material system of interest, collect known stability and band structure information of two-dimensional materials from literature and public material databases. The stability information includes formation energy information and convex hull energy information. The band structure information includes PBE band gap information, HSE band gap information, VBM at the HSE level and CBM at the HSE level.

[0018] S1.2, Obtaining the input features of two-dimensional materials, specifically including: (1) Obtaining the space group of the material; (2) Generating stoichiometric features, elemental property statistical features, electronic structure property features, and ionic component property features through the Magpie method;

[0019] S1.3, the initial dataset is established by taking the input features and the stability and band structure information of the two-dimensional material.

[0020] Furthermore, S2 specifically includes the following sub-steps:

[0021] S2.1, based on machine learning, missing values ​​are processed for the target variable of the initial dataset, and outlier values ​​are processed for the input features to obtain a dataset for machine learning modeling. The modeling dataset includes input features, forming energy, convex hull energy, PBE bandgap, HSE bandgap, VBM at the HSE level, and CBM at the HSE level.

[0022] S2.2, determine the problem attributes of each target variable. Convex hull energy is used as the target variable for classification, with a threshold of 0.2 eV. Materials with a convex hull energy below 0.2 eV are considered stable and marked as 1; otherwise, they are marked as 0. Formation energy, PBE band gap, HSE band gap, VBM at the HSE level, and CBM at the HSE level are used as target variables for regression.

[0023] Furthermore, in S2.1, the upper bound of outliers is Q3 + 1.5 × (Q3 - Q1), and the lower bound of outliers is Q1 + 1.5 × (Q3 - Q1), where Q1 is the first quartile and Q3 is the third quartile.

[0024] Furthermore, S3 specifically includes the following sub-steps:

[0025] S3.1, perform Pierce correlation calculation on the features of the dataset obtained in S2, filter out features with high correlation, obtain a feature set with low collinearity, and use the RFE method based on machine learning to filter the features of the low collinearity feature set to obtain the optimal feature subset.

[0026] S3.2 For the convex hull energy classification problem, the automatic ML tool TPOT is used to perform an extensive search for the optimal feature subset and convex hull energy;

[0027] S3.3, For the formation energy problem, CrabNet is used as a machine learning model to predict the formation energy;

[0028] S3.4 For the bandgap regression problem, two regression models were established: one for predicting the HSE bandgap and the other for predicting the VBM. The model stacking technique was adopted, using the PBE bandgap as one of the input features of the HSE bandgap regression model, and the HSE bandgap as one of the input features of the VBM regression model.

[0029] S3.5 For the classification and regression problems mentioned above, the optuna software package is used to optimize the Bayesian parameters, and the SHAP method is used to perform interpretability analysis on the features of the machine learning prediction model, thereby improving the interpretability of the model and optimizing the model.

[0030] Furthermore, S3.4 specifically includes the following sub-steps:

[0031] S3.4.1, a PBE bandgap regression model based on the GGA function was trained based on the best feature subset obtained in S3. The PBE bandgap of the two-dimensional material dataset was predicted based on the PBE model to obtain new PBE bandgap values.

[0032] S3.4.2, based on the best feature subset obtained in S3 and the new PBE bandgap value obtained in S3.4.1, a model stacking technique is used to train the HSE bandgap regression model, and the HSE bandgap of the two-dimensional material dataset is predicted based on the HSE model to obtain the new HSE bandgap value;

[0033] S3.4.3, based on the optimal feature subset obtained from S3 and the new HSE bandgap value obtained from S3.4.3, uses model stacking technology to train the VBM regression model at the HSE level.

[0034] Furthermore, S4 specifically includes the following sub-steps:

[0035] S4.1, determine the element selection of A, B and X sites in A2B2X6;

[0036] S4.2, through arrangement, combination, and electrically neutral compounding, A2B2X6 candidate is generated, in which the B site is replaced by a combination of two different metal elements;

[0037] S4.3 uses the theoretical two-dimensional carbon dioxide reduction photocatalyst dataset as input to the machine learning model obtained in S3 to obtain the predicted value of the target variable of the dataset.

[0038] Furthermore, S5 specifically includes the following sub-steps:

[0039] S5.1, stability screening, screening indicators include formation energy less than 0 and convex hull energy less than 0.2;

[0040] S5.2, light absorption and reduction potential matching screening, the screening indexes include HSE band gap value between 1.5 and 3.2, valence band value less than -3.89, and conduction band value greater than -3.89;

[0041] S5.3, Environmental screening of materials, screening out candidates that do not contain heavy metals, precious metals, or toxic substances.

[0042] Furthermore, S6 specifically includes the following sub-steps:

[0043] S6.1, Construct the crystal structure of candidate materials;

[0044] S6.2, Optimize the crystal structure of candidate materials;

[0045] S6.3, Perform PBE self-consistent calculations on the optimized crystal structure;

[0046] S6.4, perform HSE self-consistent calculations based on the crystal structure obtained in S6.3;

[0047] S6.5, based on the crystal structure obtained in S6.4, performs HSE non-self-consistent calculations, including HSE band gap, VBM at the HSE level, CBM at the HSE level, Fermi level and DOS density of states, to screen materials that meet the conditions for photocatalytic carbon dioxide reduction, thus completing the entire photocatalyst material design process.

[0048] A second aspect of this invention provides a data-driven design system for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials, comprising:

[0049] The initial database establishment module acquires information on the two-dimensional photocatalyst material system of interest, acquires the two-dimensional material data corresponding to the system information, the two-dimensional material data includes the stability information and band structure information of the two-dimensional material, and acquires the input features of the two-dimensional material to establish the initial dataset;

[0050] The data preprocessing module cleans and preprocesses the initial dataset to obtain a modeling dataset for machine learning. The modeling dataset includes the input features, stability, and band structure information of the two-dimensional material. The input features are used as variables, and the stability and band structure information are used as target variables. The problem attributes of machine learning are determined based on different target variables.

[0051] The feature engineering and model acquisition module filters the input features of the modeling dataset to obtain the best feature subset, trains and evaluates the machine learning model on the best feature subset to obtain the machine learning model, and performs interpretability analysis on the machine learning prediction model.

[0052] The two-dimensional photocatalyst design module creates a theoretical two-dimensional carbon dioxide reduction photocatalyst dataset by defining the elements at each site. The theoretical two-dimensional carbon dioxide reduction photocatalyst dataset is used as input to a machine learning model to obtain the predicted value of the target variable.

[0053] A two-dimensional photocatalyst pre-screening module screens out candidate photocatalyst materials based on the predicted values ​​of the target variables;

[0054] The two-dimensional photocatalyst DFT verification module performs DFT verification on the photocatalyst candidate materials and screens out materials that meet the conditions for photocatalytic carbon dioxide reduction.

[0055] A third aspect of the present invention provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the above-described data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials.

[0056] Compared with the prior art, the present invention has the following technical advantages:

[0057] 1) This invention establishes a data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials, providing a direction for exploring two-dimensional photocatalysts with carbon dioxide reduction potential.

[0058] 2) This invention comprehensively considers the stability, band structure, and potential requirements of carbon dioxide reduction photocatalyst materials, and combines machine learning methods with first-principles calculations to design a rapid screening tool.

[0059] 3) By learning the target variable data of known two-dimensional materials, this invention can provide the influence trend of each feature on the target variable, and can also quickly give the desired two-dimensional photocatalyst material, so as to reduce the number of experiments in finding two-dimensional photocatalysts and accelerate the production process of carbon dioxide reduction photocatalysts. Attached Figure Description

[0060] Figure 1 This is a flowchart of the photocatalyst material design method in Example 1 of the present invention;

[0061] Figure 2 This is a flowchart of step 2 in the photocatalyst material design process of Example 1 of the present invention;

[0062] Figure 3 This is a flowchart of step 3 in the photocatalyst material design process of Example 1 of the present invention;

[0063] Figure 4This is a flowchart of step 4 in the photocatalyst material design process of Example 1 of the present invention;

[0064] Figure 5 This is a flowchart of step 5 in the photocatalyst material design process of Example 1 of the present invention;

[0065] Figure 6 This is a flowchart of step 6 in the photocatalyst material design process of Example 1 of the present invention;

[0066] Figure 7 This is a performance diagram of the convex hull energy classification model in this invention;

[0067] Figure 8 This is a performance graph of the HSE bandgap regression model in this invention;

[0068] Figure 9 This is a performance graph of the valence band regression model at the HSE level in this invention.

[0069] Figure 10 This is a site element selection diagram for the two-dimensional bimetallic oxo compounds in this invention. Detailed Implementation

[0070] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0071] 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. The technical features designed in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0072] Example 1

[0073] Figure 1 This is a flowchart of the photocatalyst material design method in Example 1 of the present invention;

[0074] like Figure 1 As shown, this embodiment provides a data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials, including the following steps:

[0075] Step 1: Based on the two-dimensional photocatalyst material system under study, collect known two-dimensional material data, including the stability and band structure information of the two-dimensional material, obtain the input characteristics of the two-dimensional material and the corresponding input characteristic parameters, and establish an initial dataset.

[0076] Specifically, the process of establishing a complete initial database includes the following sub-steps:

[0077] Step 1.1: Based on the two-dimensional material system under study, more than 1,000 stability and band structure information of the material were obtained from the C2DB two-dimensional material database. The stability information includes formation energy and convex hull energy. The band structure information includes PBE band gap, HSE band gap, VBM at the HSE level and CBM at the HSE level.

[0078] Step 1.2, the input features of the material include: (1) the space group of the material; (2) the stoichiometric features, elemental property statistical features, electronic structure property features, and ionic composition property features generated by the Magpie method;

[0079] Step 1.3: Build an initial dataset by combining the input features and the stability and band structure information of the two-dimensional material.

[0080] Step 2: Based on machine learning, the initial dataset is cleaned and preprocessed to obtain the modeling dataset for machine learning. The modeling dataset includes the input features, stability and band structure information of the two-dimensional material. The input features are used as variables, and the stability and band structure information are used as target variables. The problem attributes of machine learning are determined according to different target variables.

[0081] Figure 2 This is a flowchart of step 2 in the photocatalyst material design process of Example 1 of the present invention;

[0082] like Figure 2 As shown, step 2 includes the following sub-steps:

[0083] Step 2.1: Based on machine learning, the target variable of the initial dataset is processed for missing values, and the input features are processed for outliers, resulting in a dataset for machine learning modeling. This modeling dataset includes input features, forming energy, convex hull energy, PBE bandgap, HSE bandgap, VBM at the HSE level, and CBM at the HSE level. The upper bound of the outliers is Q3 + 1.5 × (Q3 - Q1), and the lower bound is Q1 + 1.5 × (Q3 - Q1), where Q1 is the first quartile and Q3 is the third quartile.

[0084] Step 2.2: Determine the problem attributes of each target variable, where convex hull energy is used as the target variable for classification. A threshold of 0.2 eV is used for convex hull energy classification. Materials with convex hull energies below 0.2 eV are considered stable and marked as 1; otherwise, they are marked as 0. Formation energy, PBE band gap, HSE band gap, VBM at the HSE level, and CBM at the HSE level are used as target variables for regression.

[0085] Step 3: Filter the input features of the dataset to obtain the best feature subset, train and evaluate the model using the best feature subset, and use machine learning algorithms to improve the model's predictive ability to obtain the machine learning model. Then, perform interpretability analysis on the machine learning prediction model.

[0086] Figure 3 This is a flowchart of step 3 in the photocatalyst material design process of Example 1 of the present invention;

[0087] like Figure 3 As shown, step 3 includes the following sub-steps:

[0088] Step 3.1: Perform Pierce correlation calculation on the features of the dataset obtained in Step 2, filter out features with high correlation, obtain a feature set with low collinearity, and use the RFE method based on machine learning to filter the features of the low collinearity feature set to obtain the optimal feature subset.

[0089] Step 3.2: For the convex hull energy classification problem, use the automated ML tool TPOT to perform an extensive search for the optimal feature subset and convex hull energy;

[0090] Step 3.3: For the formation energy problem, CrabNet is used as a machine learning model to predict the formation energy;

[0091] Step 3.4: For the bandgap regression problem, two regression models were established: one for predicting the HSE bandgap and the other for predicting the VBM. A model stacking technique was employed, using the PBE bandgap as one of the input features of the HSE bandgap regression model, and the HSE bandgap as one of the input features of the VBM regression model.

[0092] Step 3.5: For the classification and regression problems mentioned above, use the optuna software package to optimize the Bayesian parameters, and use the SHAP method to perform interpretability analysis on the features of the machine learning prediction model, thereby improving the interpretability of the model and optimizing the model.

[0093] like Figure 7 As shown, the following analysis is performed on the convex hull classification model: (1) the importance of model features is analyzed; (2) the confusion matrix is ​​plotted, and the F1 score is calculated to be 94.39%. The F1 calculation formula is: Where Precision is the accuracy, calculated using the following formula: Recall is the rate of return, and it is calculated using the following formula: (3) Plot the AUC curve and calculate the area under the curve.

[0094] like Figure 8 As shown, the following analysis is performed on the HSE bandgap regression model: (1) Output the model's shap value and output the importance of model features; (2) Calculate the model's accuracy, root mean square error, and mean absolute error.

[0095] like Figure 9 As shown, the comprehensive evaluation of the valence band regression model under the HSE level includes, but is not limited to, the following methods: (1) analyzing the importance of model features; (2) calculating the accuracy and mean absolute error of the model.

[0096] Step 4: By defining the elements at each site, a theoretical two-dimensional carbon dioxide reduction photocatalyst database is created. The dataset is then used as the input to the machine learning model obtained in Step 3 to obtain the predicted value of the target variable in the dataset.

[0097] Figure 4 This is a flowchart of step 4 in the photocatalyst material design process of Example 1 of the present invention;

[0098] like Figure 4 As shown, step 4 includes the following sub-steps:

[0099] Step 4.1: Determine the element selection for the A, B, and X sites of A2B2X6.

[0100] like Figure 10 As shown, alkali metals, alkaline earth metals, transition metals, main group metals and metalloids are considered as potential cations at the A and B sites, and chalcogens are considered as potential anions at the x site.

[0101] Step 4.2: A2B2X6 candidate compounds are generated by arranging, combining and making the compound electrically neutral, wherein the B site is replaced by a combination of two different metal elements.

[0102] Step 4.3: Use the dataset as input to the machine learning model obtained in Step 3 to obtain the predicted value of the target variable in the dataset;

[0103] Step 5: Screening. Based on the predicted values ​​of the target variable, candidate photocatalyst materials are screened out.

[0104] Figure 5 This is a flowchart of step 5 in the photocatalyst material design process of Example 1 of the present invention;

[0105] like Figure 5 As shown, step 5 includes the following sub-steps:

[0106] Step 5.1, stability screening, including formation energy less than 0 and convex hull energy less than 0.2;

[0107] Step 5.2, screening for matching light absorption and reduction potential: HSE band gap value between 1.5 and 3.2, valence band value less than -3.89, conduction band value greater than -3.89;

[0108] Step 5.3: Environmental screening of materials to identify candidates that do not contain heavy metals, precious metals, or toxic substances.

[0109] Step 6: Perform DFT verification on the candidate photocatalyst materials, screen out materials that meet the conditions for photocatalytic carbon dioxide reduction, and complete the entire photocatalyst material design process.

[0110] Figure 6 This is a flowchart of step 6 in the photocatalyst material design process of Example 1 of the present invention;

[0111] like Figure 6 As shown, step 6 includes the following sub-steps:

[0112] Step 6.1: Construct the crystal structure of the candidate material;

[0113] Step 6.2: Optimize the crystal structure of candidate materials;

[0114] Step 6.3: Perform PBE self-consistent calculations on the optimized crystal structure;

[0115] Step 6.4: Perform HSE self-consistent calculations based on the crystal structure obtained in Step 6.3;

[0116] Step 6.5: Based on the crystal structure obtained in Step 6.4, perform HSE non-self-consistent calculations, including HSE band gap, VBM at the HSE level, CBM at the HSE level, Fermi level and DOS density of states, screen out materials that meet the conditions for photocatalytic carbon dioxide reduction, and complete the entire photocatalyst material design process.

[0117] This embodiment also provides a data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials. The data-driven design system for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials described in Embodiment 1 includes an initial database establishment module, a data preprocessing module, a feature engineering and model acquisition module, a two-dimensional photocatalyst design module, a two-dimensional photocatalyst pre-screening module, and a two-dimensional photocatalyst DFT verification module.

[0118] The initial database establishment module, based on the studied two-dimensional photocatalyst material system, collects known two-dimensional material data, including the stability and band structure information of the two-dimensional materials, obtains the input characteristics of the two-dimensional materials and the corresponding input characteristic parameters, and establishes the initial dataset.

[0119] The data preprocessing module, based on machine learning, cleans and preprocesses the initial dataset to obtain the modeling dataset for machine learning. The modeling dataset includes the input features, stability, and band structure information of the two-dimensional material. The input features are used as variables, and the stability and band structure information are used as target variables. The problem attributes of machine learning are determined according to different target variables.

[0120] The feature engineering and model acquisition module filters the input features of the dataset to obtain the best feature subset, trains and evaluates the model on the best feature subset, uses machine learning algorithms to improve the model's predictive ability, obtains the machine learning model, and performs interpretability analysis on the machine learning prediction model.

[0121] The two-dimensional photocatalyst design module creates a theoretical two-dimensional carbon dioxide reduction photocatalyst database by defining the elements at each site. The dataset is then used as the input to the machine learning model obtained in step 3 to obtain the predicted value of the target variable in the dataset.

[0122] A two-dimensional photocatalyst pre-screening module screens photocatalyst candidate materials based on the predicted values ​​of the target variables.

[0123] The two-dimensional photocatalyst DFT verification module performs DFT verification on the photocatalyst candidate materials, screens out materials that meet the conditions for photocatalytic carbon dioxide reduction, and completes the entire photocatalyst material design process.

[0124] It should be understood that the modules of the data-driven two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst material design system described in Example 2 correspond to the steps of the photocatalyst material design method described in Example 1. Therefore, the operations, features, and advantages described above for each step of the method are also applicable to the modules of the system. For the sake of brevity, some operations, features, and advantages will not be repeated here.

[0125] This embodiment also proposes a computer-readable storage medium storing computer instructions for instructing a computer to execute the aforementioned data-driven design method for a two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst material. The storage medium can be an electronic medium, magnetic medium, optical medium, electromagnetic medium, infrared medium, or a semiconductor system or propagation medium. The storage medium can also include semiconductor or solid-state memory, magnetic tape, removable computer disk, random access memory (RAM), read-only memory (ROM), hard disk, and optical disc. Optical discs can include optical disc-read-only memory (CD-ROM), optical disc-read-write (CD-RW), and DVD.

[0126] As can be seen from the examples, the present invention establishes a data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials, providing a direction for exploring the identification of two-dimensional photocatalysts with carbon dioxide reduction potential.

[0127] As can be seen from the embodiments, the present invention comprehensively considers the stability, band structure, and potential requirements of carbon dioxide reduction photocatalyst materials, and combines machine learning methods with first-principles calculations to design a rapid screening tool.

[0128] As can be seen from the embodiments, by learning the target variable data of known two-dimensional materials, the present invention can provide the influence trend of each feature on the target variable, and can also quickly give the desired two-dimensional photocatalyst material, thereby reducing the number of experiments in finding two-dimensional photocatalysts and accelerating the production process of carbon dioxide reduction photocatalysts.

[0129] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials, characterized in that, The design method is implemented by combining machine learning algorithms and first-principles calculations based on DFT, and specifically includes the following steps: S1, Obtain information on the two-dimensional photocatalyst material system of interest, obtain the two-dimensional material data corresponding to the system information, the two-dimensional material data includes the stability information and band structure information of the two-dimensional material, and obtain the input features of the two-dimensional material to establish an initial dataset; S2, the initial dataset is cleaned and preprocessed to obtain a modeling dataset for machine learning. The modeling dataset includes the input features, stability and band structure information of the two-dimensional material. The input features are used as variables, and the stability and band structure information are used as target variables. The problem attributes of machine learning are determined according to different target variables. S3: Filter the input features of the modeling dataset to obtain the best feature subset, train and evaluate the machine learning model on the best feature subset to obtain the machine learning model, and perform interpretability analysis on the machine learning prediction model. S4. By limiting the elements at each site, a theoretical two-dimensional carbon dioxide reduction photocatalyst dataset is created. The theoretical two-dimensional carbon dioxide reduction photocatalyst dataset is used as the input to the machine learning model obtained in S3 to obtain the predicted value of the target variable. S5. Based on the predicted values ​​of the target variables, candidate photocatalyst materials are selected. S6, Perform DFT verification on the photocatalyst candidate materials to screen out materials that meet the conditions for photocatalytic carbon dioxide reduction; S2 specifically includes the following sub-steps: S2.1, based on machine learning, missing values ​​are processed for the target variable of the initial dataset, and outlier values ​​are processed for the input features to obtain a dataset for machine learning modeling. The modeling dataset includes input features, forming energy, convex hull energy, PBE bandgap, HSE bandgap, VBM at the HSE level, and CBM at the HSE level. S2.2, determine the problem attributes of each target variable. Among them, convex hull energy is used as the target variable for classification problems, and a threshold of 0.2 eV is used for convex hull energy classification. Formation energy, PBE band gap, HSE band gap, VBM at HSE level, and CBM at HSE level are used as target variables for regression problems. In S2.1, the upper bound for outliers is The lower bound for outliers is Q1 is the first quartile, and Q3 is the third quartile.

2. The data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials according to claim 1, characterized in that: In S1, the process of establishing the initial database includes the following sub-steps: S1.1, Obtain the two-dimensional material system of interest, collect known stability and band structure information of two-dimensional materials from literature and public material databases. The stability information includes formation energy information and convex hull energy information. The band structure information includes PBE band gap information, HSE band gap information, VBM at the HSE level and CBM at the HSE level. S1.2, Obtain the input features of the two-dimensional material, specifically including: (1) Obtain the spatial group of the material; (2) Generate stoichiometric characteristics, elemental property statistical characteristics, electronic structure property characteristics, and ionic component property characteristics using the Magpie method; S1.3, the initial dataset is established by taking the input features and the stability and band structure information of the two-dimensional material.

3. The data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials according to claim 1, characterized in that: S3 specifically includes the following sub-steps: S3.1, perform Pierce correlation calculation on the features of the dataset obtained in S2, filter out features with high correlation, obtain a feature set with low collinearity, and use the RFE method based on machine learning to filter the features of the low collinearity feature set to obtain the optimal feature subset. S3.2 For the convex hull energy classification problem, the automatic ML tool TPOT is used to perform an extensive search for the optimal feature subset and convex hull energy; S3.3, For the formation energy problem, CrabNet is used as a machine learning model to predict the formation energy; S3.4 For the bandgap regression problem, two regression models were established: one for predicting the HSE bandgap and the other for predicting the VBM. The model stacking technique was adopted, using the PBE bandgap as one of the input features of the HSE bandgap regression model, and the HSE bandgap as one of the input features of the VBM regression model. S3.5 For the classification and regression problems mentioned above, the optuna software package is used to optimize the Bayesian parameters, and the SHAP method is used to perform interpretability analysis on the features of the machine learning prediction model, thereby improving the interpretability of the model and optimizing the model.

4. The data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials according to claim 3, characterized in that: S3.4 specifically includes the following sub-steps: S3.4.1, a PBE bandgap regression model based on the GGA function was trained based on the best feature subset obtained in S3. The PBE bandgap of the two-dimensional material dataset was predicted based on the PBE model to obtain new PBE bandgap values. S3.4.2, based on the best feature subset obtained in S3 and the new PBE bandgap value obtained in S3.4.1, a model stacking technique is used to train the HSE bandgap regression model, and the HSE bandgap of the two-dimensional material dataset is predicted based on the HSE model to obtain the new HSE bandgap value; S3.4.3, based on the optimal feature subset obtained from S3 and the new HSE bandgap value obtained from S3.4.3, uses model stacking technology to train the VBM regression model at the HSE level.

5. The data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials according to claim 1, characterized in that: S4 specifically includes the following sub-steps: S4.1, determine the element selection of A, B and X sites in A2B2X6; S4.2, through arrangement, combination, and electrically neutral compounding, A2B2X6 candidate is generated, in which the B site is replaced by a combination of two different metal elements; S4.3 uses the theoretical two-dimensional carbon dioxide reduction photocatalyst dataset as input to the machine learning model obtained in S3 to obtain the predicted value of the target variable of the dataset.

6. The data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials according to claim 1, characterized in that: S5 specifically includes the following sub-steps: S5.1, stability screening, screening indicators include formation energy less than 0 and convex hull energy less than 0.2; S5.2, light absorption and reduction potential matching screening, the screening indexes include HSE band gap value between 1.5 and 3.2, valence band value less than -3.89, and conduction band value greater than -3.89; S5.3, Environmental screening of materials, screening out candidates that do not contain heavy metals, precious metals, or toxic substances.

7. The data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials according to claim 6, characterized in that: S6 specifically includes the following sub-steps: S6.1, Construct the crystal structure of candidate materials; S6.2, Optimize the crystal structure of candidate materials; S6.3, Perform PBE self-consistent calculations on the optimized crystal structure; S6.4, perform HSE self-consistent calculations based on the crystal structure obtained in S6.3; S6.5, based on the crystal structure obtained in S6.4, performs HSE non-self-consistent calculations, including HSE band gap, VBM at the HSE level, CBM at the HSE level, Fermi level and DOS density of states, to screen materials that meet the conditions for photocatalytic carbon dioxide reduction, thus completing the entire photocatalyst material design process.

8. A data-driven design system for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials, used to implement the design method as described in any one of claims 1 to 7, characterized in that, include: The initial database establishment module acquires information on the two-dimensional photocatalyst material system of interest, acquires the two-dimensional material data corresponding to the system information, the two-dimensional material data includes the stability information and band structure information of the two-dimensional material, and acquires the input features of the two-dimensional material to establish the initial dataset; The data preprocessing module cleans and preprocesses the initial dataset to obtain a modeling dataset for machine learning. The modeling dataset includes the input features, stability, and band structure information of the two-dimensional material. The input features are used as variables, and the stability and band structure information are used as target variables. The problem attributes of machine learning are determined based on different target variables. The feature engineering and model acquisition module filters the input features of the modeling dataset to obtain the best feature subset, trains and evaluates the machine learning model on the best feature subset to obtain the machine learning model, and performs interpretability analysis on the machine learning prediction model. The two-dimensional photocatalyst design module creates a theoretical two-dimensional carbon dioxide reduction photocatalyst dataset by defining the elements at each site. The theoretical two-dimensional carbon dioxide reduction photocatalyst dataset is used as input to a machine learning model to obtain the predicted value of the target variable. A two-dimensional photocatalyst pre-screening module screens out candidate photocatalyst materials based on the predicted values ​​of the target variables; The two-dimensional photocatalyst DFT verification module performs DFT verification on the photocatalyst candidate materials and screens out materials that meet the conditions for photocatalytic carbon dioxide reduction.

9. A storage medium containing computer-executable instructions, characterized in that, When executed by a computer processor, the storage medium containing the computer-executable instructions is used to perform the data-driven design method for two-dimensional bimetallic oxometalate carbon dioxide reduction photocatalyst materials as described in any one of claims 1 to 7.

Images