Method for screening g-C3N4 immobilized monatomic catalyst

Abstract

The invention relates to the technical field of catalyst screening, in particular to a method for screening a g-C3N4 immobilized monatomic catalyst. The specific method comprises the following steps: placing a doping element at a doping site of a g-C3N4 substrate, and carrying out structure optimization to screen out a pre-screened catalyst meeting the electronic locality; through a density functional theory and an XGBoost model, the significant key features are used as input, formation energy is used as output, and a primarily screened catalyst is screened out; performing CO2 reduction reaction catalytic activity evaluation on the primarily-screened catalyst through adsorption free energy change so as to screen out a secondarily-screened catalyst; and performing hydrogen evolution reaction catalytic activity evaluation on the re-screened catalyst to screen out a catalyst for synthesis gas production, a CO2 reduction reaction catalyst and a hydrogen evolution reaction catalyst. The method provided by the invention solves the problem that the reliability of a screening result is affected due to low screening efficiency, long time consumption and low prediction accuracy existing in an existing monatomic catalyst screening method.

Description

Technical Field

[0001] The present invention relates to the technical field of catalyst screening, and in particular to a method for screening gC 3 N 4 Methods for supported single-atom catalysts. Background Art

[0002] At present, the existing single-atom catalyst screening methods mainly rely on the following technologies: experimental screening, DFT-based computational screening, and pure machine learning screening.

[0003] The experimental screening method requires a lot of material synthesis and characterization, which is not only time-consuming and costly. It usually requires repeated testing under multiple experimental conditions, and the screening results have a certain degree of uncertainty. It can be seen that the experimental screening method has the defects of low efficiency, inability to handle a large number of potential catalyst combinations, and inability to quickly find efficient catalyst candidates.

[0004] Computational screening based on DFT is a widely used method in theoretical calculations, which can accurately predict the electronic structure and reactivity of materials. However, the DFT calculation process is complex and consumes a lot of computing resources. It can be seen that computational screening based on DFT, because each material combination requires high-precision ab initio calculations, the cost of screening a large number of single-atom catalysts is too high and the speed is slow, making it difficult to quickly screen out efficient catalysts in large-scale material libraries.

[0005] Pure machine learning screening mainly uses machine learning models, which can mine potential catalyst combinations from large-scale data. However, pure machine learning methods rely on high-quality training data sets. Data sets in the field of materials science are generally scarce, especially high-quality and comprehensive experimental data. It can be seen that pure machine learning screening lacks sufficient high-quality data, which can easily lead to overfitting of the model and inaccurate predictions.

[0006] Currently, gC 3 N 4 With its unique two-dimensional structure, especially the rich six-fold cavities and non-equivalent distribution of nitrogen atoms, it has become an ideal support for single-atom catalysts, especially CO 2 Reduction reaction, hydrogen evolution reaction, etc. This structure provides a stable anchoring point for the single atom, ensuring its efficient fixation and good electronic localization. 3 N 4 The large specific surface area also makes the catalytic reaction more active, which can effectively improve the selectivity of the reaction. However, the existing single-atom catalyst screening methods have low screening efficiency, long time consumption, and low prediction accuracy, which easily affects the reliability of the screening results. Summary of the invention

[0007] In order to solve the problems of low screening efficiency, long time consumption and low prediction accuracy in the existing single-atom catalyst screening methods, which affect the reliability of the screening results, the present invention aims to provide a screening method for gC 3 N 4 The method of immobilized single-atom catalysts has been used to improve gC 3 N 4 While improving the screening efficiency of system catalysts, it also improves the reliability of prediction results.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows.

[0009] The present invention provides a method for screening gC 3 N 4 The method of immobilizing a single atom catalyst comprises the following steps:

[0010] According to gC 3 N 4 The two-dimensional structure of the substrate is obtained by gC 3 N 4 Doping site of substrate; place doping element on gC 3 N 4 The doping sites of the substrate are structurally optimized to screen out catalysts that meet the electronic localization requirements and obtain pre-screening results; the formation energy of some catalysts in the pre-screening results is calculated, key feature data related to the formation energy is collected to establish a formation energy-key feature data set, and a first prediction model is constructed so that the first prediction model can learn the nonlinear relationship between the key features and the formation energy; feature importance analysis is performed on the first prediction model to generate a feature importance pie chart; significance analysis is performed on the feature importance pie chart to determine significant key features; catalysts with negative formation energy are screened out using the significant key features to obtain preliminary screening results; calculation The adsorption free energy changes of all catalysts in the preliminary screening results were collected, and the core feature data related to the adsorption free energy changes were collected to establish the adsorption free energy change-core feature data set, and a second prediction model was constructed so that the second prediction model could learn the nonlinear relationship between the key features and the adsorption free energy changes; the SHAP value was calculated based on the results of the second prediction model training, and the feature marginal contribution analysis of the core features was performed based on the SHAP value to determine the significant core features; based on the second prediction model, the adsorption free energy change and limiting potential of the preliminary screening catalysts were predicted using the significant core features, which were used to perform CO on the preliminary screening catalysts. 2 The catalytic activity of the reduction reaction was evaluated, and the catalysts with the absolute value of the adsorption free energy change ≤0.8eV and the limiting potential ≤0.8V were screened out to obtain the re-screening results; the catalytic activity of the hydrogen evolution reaction was evaluated for the catalysts in the re-screening results to screen out the catalysts suitable for synthesis gas production, CO 2 Catalyst for reduction reaction or hydrogen evolution reaction.

[0011] Preferably, gC 3 N 4 The doping sites of the substrate are 2 C substitution sites, 3 N substitution sites and 2 H interstitial sites. Preferably, the doping element is at least one of Zr, Ti, Mn, Fe, Co, Ni, Cu, Zn, Ru, Rh, Pd, Pt, Al, Li, B, Si, S, O, P and Cl.

[0012] Preferably, the key features related to formation energy are doping site, atomic radius, number of valence electrons, electron affinity, chemical potential, atomic number, electronegativity, first ionization energy, atomic weight and number of internal electrons; significant key features are doping site and atomic radius.

[0013] Preferably, the method for constructing the first prediction model is: based on the formation energy-key feature data set, using the root mean square error, determination coefficient and mean absolute error as evaluation indicators, optimizing the hyperparameters of the model, thereby learning the nonlinear relationship between the key features and the formation energy, and constructing and training the first prediction model with the best prediction accuracy.

[0014] Preferably, the method for constructing the second prediction model is: based on the adsorption free energy change-core feature data set, using the root mean square error, determination coefficient and mean absolute error as evaluation indicators, the hyperparameters of the model are optimized, so as to learn the nonlinear relationship between the core features and the adsorption free energy change, and construct and train a second prediction model with high interpretability.

[0015] Preferably, the core features related to the change in adsorption free energy include single-atom features and catalyst features; single-atom features are atomic radius, atomic number, single-atom chemical potential, valence electrons, first ionization energy and electron affinity; catalyst features are charge transfer amount, binding energy, formation energy, Fermi level and band gap; significant core features are band gap and binding energy.

[0016] Preferably, suitable for synthesis gas production, CO 2 The method of the catalyst for the reduction reaction or hydrogen evolution reaction is: according to CO 2 The competitive relationship between the adsorption free energy change of the reduction reaction and the hydrogen evolution reaction was studied. The catalytic activity of the catalysts in the re-screening results was evaluated for the hydrogen evolution reaction, CO 2 Evaluation of reduction reaction activity and analysis of suitability for synthesis gas production; CO 2 The change in adsorption free energy of the reduction reaction is recorded as ΔG_CO 2 RR; the change in adsorption free energy of hydrogen evolution reaction is recorded as ΔG_H; when the catalyst ΔG_CO 2 When RR is lower than ΔG_H, the catalyst tends to promote CO 2Reduction reaction of CO 2 Reduction reaction catalyst; when the ΔG_H of the catalyst is lower than ΔG_CO 2 RR, the catalyst is a hydrogen evolution catalyst that tends to promote the hydrogen evolution reaction; when the ΔG_CO 2 When RR is close to ΔG_H, the catalyst has the function of promoting CO 2 Syngas catalysts for reduction and hydrogen evolution reactions.

[0017] The present invention provides a method for screening gC 3 N 4 A system for a single-atom catalyst supported on a substrate, the system comprising: a processor and a memory, the memory storing a computer program that can be run on the processor; wherein the computer program, when executed by the processor, implements the screening gC 3 N 4 Steps of the method for immobilizing single atom catalysts.

[0018] The present invention provides a computer-readable storage medium, wherein a data processing program is stored on the computer-readable storage medium, and when the data processing program is executed by a processor, the screening gC is implemented. 3 N 4 Steps of the method for immobilizing single atom catalysts.

[0019] Beneficial effects of the present invention:

[0020] 1. The method of the present invention is mainly used to screen gC 3 N 4 The supported single-atom catalyst is first based on gC 3 N 4 The two-dimensional structure of the catalyst was used to determine the doping site, and the candidate catalysts that met the electron localization requirements were screened out through structural optimization. Then, a prediction model was established using machine learning technology to predict the formation energy and adsorption free energy changes of the catalysts through key feature data, thereby evaluating the CO 2 Finally, the catalytic activity of the selected catalysts for hydrogen evolution reaction is evaluated to screen out catalysts suitable for different catalytic reactions. The method of the present invention, combined with the efficient prediction of the machine learning model, can quickly screen out potential efficient catalysts, and combined with the feature importance analysis, can further narrow the screening range, significantly reduce the sample size of experimental and computational verification, and improve gC 3 N 4 While improving the screening efficiency of system catalysts, it also improves the reliability of prediction results.

[0021] 2. In the screening process, the present invention combines a variety of key features that affect catalyst performance through a prediction model, and through SHAP analysis, identifies the contribution of important features such as formation energy and adsorption energy to catalytic performance, thereby effectively improving the accuracy of screening. Compared with traditional screening methods, the method of the present invention can effectively avoid the problem of insufficient data in pure machine learning methods and ensure the reliability of prediction results.

[0022] 3. The screening method of the present invention can be used according to different reaction requirements, such as CO 2 Different screening conditions are set for reduction reaction, hydrogen evolution reaction and synthesis gas production. For example, by using the criteria of limiting potential and free energy change, catalysts with both high reaction rate and good selectivity can be specifically screened. For catalyst design of multiple reactions, such as synthesis gas generation, the present invention can screen out catalysts that can simultaneously catalyze CO 2 Catalysts for reduction and hydrogen evolution reactions, achieving efficient utilization of reaction conditions.

[0023] 4. The method of the present invention is not limited to CO 2 The reduction reaction can also be applied to other catalytic reactions at gas-solid interfaces, such as efficient screening of catalysts for hydrogen evolution reaction and synthesis gas generation. The method of the invention has high scalability and provides broad application prospects for the design of new catalysts. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is gC 3 N 4 The electronic localization function diagram and the schematic diagram of the substitution position and interstitial position. Among them, (a) is gC 3 N 4 The electronic localization function diagram of gC 3 N 4 Schematic diagram of substitution and interstitial positions; orange balls represent N atoms; blue balls represent C atoms.

[0025] Figure 2 E is the predicted value of the XGBoost model. f The training set of predicted values ​​and E calculated by DFT f The test set of calculated values ​​and E f The training set of predicted values ​​and E f Comparison results of the test set of calculated values ​​and E based on the XGBoost model f Feature importance analysis results. (a) is the coefficient of determination, mean absolute error, and root mean square error of various machine learning models; (b) is the E predicted by the XGBoost model. f The training set of predicted values ​​and the E calculated by DFT fComparison results of the test set of calculated values; (c) is the E based on the XGB model f Feature importance analysis results. Train R 2 represents the coefficient of determination of the training set; Train MAR represents the mean absolute error of the training set; Train MSR represents the root mean square error of the training set; Test R 2 It represents the coefficient of determination of the test set; Test MAR represents the mean absolute error of the test set; Test MSR represents the root mean square error of the test set. The full name of machine learning in English is Machine Learning, abbreviated as ML. XGB, ABR, KNR, LR, MLP, PLS, RR, SVR, RFR, GBR represent different machine learning algorithms or models. Doping site, Radii, Out_e, Aff, Chemi_pot, Atom_N, N m ,I 1 They represent key features related to formation energy. Among them, Doping site represents the doping site; Radii represents the atomic radius; Out_e represents the number of outer electrons; Aff represents the electron affinity; Chemi_pot represents the chemical potential; Atom_N represents the atomic number; N m Indicates electronegativity; I 1 represents the first ionization energy; Atom_W represents the atomic weight; In_e represents the number of internal electrons.

[0026] Figure 3 For SAC@gC 3 N 4 CO on the catalyst surface 2 Schematic diagram of the mechanism and reaction pathway of the reduction reaction and CO 2 Limiting potential diagram of reduction reaction. (a) is the limiting potential diagram of SAC@gC 3 N 4 CO on the catalyst surface 2 Schematic diagram of the mechanism and reaction pathway of the reduction reaction; (b) is CO 2 Diagram of the limiting potential of a reduction reaction.

[0027] Figure 4The training set of ΔG predicted values ​​predicted by the XGBoost prediction model, the test set of ΔG calculated values ​​calculated by DFT, the comparison results of the training set of ΔG predicted values ​​and the test set of ΔG calculated values, and the ΔG feature importance analysis results based on the XGBoost prediction model. Among them, (a) is the determination coefficient, mean absolute error, and root mean square error value of various machine learning models; (b) is the comparison result of the training set of ΔG predicted values ​​predicted by the XGBoost prediction model and the test set of ΔG calculated values ​​calculated by DFT; (c) is the ΔG feature importance analysis result based on the XGB prediction model.

[0028] Figure 5 SHAP summary diagrams of the XGB prediction model for ΔG1, ΔG2, ΔG3, and ΔG4. (a) is the SHAP summary diagram of the XGB prediction model for ΔG1; (b) is the SHAP summary diagram of the XGB prediction model for ΔG2; (c) is the SHAP summary diagram of the XGB prediction model for ΔG3; (d) is the SHAP summary diagram of the XGB prediction model for ΔG4. g , Radii, Out_e, Aff, μ, Atom_N, E b 、E f 、E fermi , Bader, I 1 They represent key features related to formation energy. g represents band gap; Radii represents atomic radius; Out_e represents the number of outer electrons; Aff represents electron affinity; μ represents monatomic chemical potential; Atom_N represents atomic number; E b Represents binding energy; E f represents the formation energy; E fermi represents the Fermi level; Bader represents the charge transfer amount; I 1 represents the first ionization energy.

[0029] Figure 6 For SAC@gC 3 N 4 CO on the catalyst surface 2 Competition diagram between reduction reaction and hydrogen evolution reaction. 2 RR stands for CO 2 Reduction reaction, HER stands for hydrogen evolution reaction.

[0030] Figure 7 Screening gC provided by one embodiment of the present invention 3 N 4 Schematic diagram of the supported single-atom catalyst. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0032] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0033] Density Functional Theory, the full name in English is Density Functional Theory, referred to as DFT. Vienna Ab initio Simulation Package, the full name in English is Vienna Ab initio Simulation Package, referred to as VASP. Generalized Gradient Approximation, the full name in English is Generalized Gradient Approximation, referred to as GGA. PBE functional, the full name in English is Perdew-Burke-Ernzerhof functional, the Chinese expression is Perdew-Burke-Ernzerhof functional. VESTA, the full name in English is Visualization for Electronic and STructural Analysis, the Chinese expression is Visualization Tool for Electronic and Structural Analysis. Projected Augmented Wave, the full name in English is Projected Augmented Wave, referred to as PAW. gC 3 N 4 Represents graphite phase carbon nitride or graphite-like phase carbon nitride. Single-atom catalyst, English full name Single-Atom Catalysts, referred to as SAC. CO 2 RR stands for carbon dioxide reduction reaction; HER stands for hydrogen evolution reaction.

[0034] The present invention mainly provides a method for screening gC by density functional theory and machine learning. 3 N 4 For CO 2 Single atom catalyst method for reduction reaction, hydrogen evolution reaction and synthesis gas generation.

[0035] In this paper, the Vienna ab initio simulation package was used to calculate the electronic properties of all structures, the PBE functional under the generalized gradient approximation was used to describe the electronic interaction, and VESTA was used to build the model. The pseudopotential used the projected augmented wave method, and the energy cutoff was set to 500 eV to ensure the balance between the accuracy of the calculation results and the calculation time. The k-point grid was 2×2×1 for the integration of the Brillouin zone, and the z direction was increased. The geometric relaxation process of the structure is based on the residual force per atom being less than The convergence standard of electron energy is 10 -5 In addition, van der Waals interactions are treated using DFT-D3 semiempirical corrections to ensure the accuracy of weak interactions.

[0036] Density functional theory was used to calculate the formation energy of the catalyst and to evaluate the influence of dopant atoms on gC 3 N 4 Thermodynamic stability on the surface. The calculation formula of formation energy is:

[0037] For alternative models: E f =E(XgC 3 N 4 )-E(gC 3 N 4 )-μ(X)+μ(N)+μ(C); for the gap model: E f =E(XgC 3 N 4 )-E(gC 3 N 4 )-μ(X); where E f represents the formation energy;

[0038] E(X 3 N 4 ) represents the gC of doping element X 3 N 4 The total energy of the system; E(gC 3 N 4 ) is pure gC 3 N 4 is the total energy of the doping element, μ(X) is the chemical potential of a single doping element X, μ(N) is the chemical potential of a single N atom, and μ(C) is the chemical potential of a single C atom.

[0039] In gC 3 N 4 The binding energy E of the doping element introduced into the system b , the calculation formula of binding energy is:

[0040] E b =E(XgC 3 N 4 )-E(X)-E(gC 3 N 4 ), where E b represents binding energy; E(XgC 3 N 4 ) represents the gC of doping element X 3 N 4 The total energy of the system; E(X) represents the energy from E(XgC 3 N4 ) to remove gC 3 N 4 Energy without structural optimization; E(gC 3 N 4 ) indicates that from E(XgC 3 N 4 ) without structural optimization after removing X. According to the definition of binding energy, E b The more negative the value, the stronger the bonding between the doping element X and the substrate.

[0041] Adsorption energy calculation for evaluating CO 2 and H 2 The adsorption condition on the catalyst surface ensures that the catalyst has good adsorption capacity for the target molecules. The adsorption energy calculation formula is:

[0042] E ads =E(YXgC 3 N 4 )-E(XgC 3 N 4 )-E Y ; Among them, E ads represents the adsorption energy;

[0043] E(X 3 N 4 ) represents the gC of adsorbed substance Y in doping element X 3 N 4 The total energy of the system; E(XgC 3 N 4 ) represents the gC of doping element X 3 N 4 The total energy of the system, E Y represents the total energy of adsorbed species Y in the free state. For example, adsorbed species Y is CO 2 or H 2 .

[0044] According to CO 2 The reaction paths of RR and HER use the change in Gibbs free energy to describe the thermodynamic feasibility of the reaction. The zero-point energy and entropy change are calculated by vibration frequency calculation. The calculation formula for the change in Gibbs free energy is: ΔG = E ads -TΔS+ΔZPE; where ΔG represents the change in Gibbs free energy; E ads represents adsorption energy; T represents temperature; ΔS represents entropy change; ΔZPE represents zero-point energy. 2For RR and HER, the entropy change and zero-point energy of the reaction intermediates are calculated using vibrational frequency data, and the entropy change of gaseous molecules is from the NIST database. All calculations are performed at a constant pressure of 0 GPa and a temperature of 298.15 K.

[0045] In addition, based on a large number of density functional theory calculation results, the present invention constructs a variety of machine learning models, including extreme gradient boosting regression, whose full name is XGBoost Regression, abbreviated as XGB; K nearest neighbor regression, whose full name is K-Nearest Neighbors Regression, abbreviated as KNR; support vector regression, whose full name is Support Vector Regression, abbreviated as SVR; random forest regression, whose full name is Random Forest Regression, abbreviated as RFR; multilayer perceptron, whose full name is Multilayer Perceptron, abbreviated as MLP; gradient boosting regression, whose full name is GradientBoosting Regression, abbreviated as GBR; linear regression, whose full name is Linear Regression, abbreviated as LR; partial least squares regression, whose full name is Partial Least Squares Regression, abbreviated as PLSR; ridge regression, whose full name is Ridge Regression, abbreviated as RR; AdaBoost regression, AdaBoost Regression, abbreviated as ABR. The present invention combines density functional theory (DFT) with machine learning to accelerate the screening process of single-atom catalysts, overcome many defects of existing screening methods, and has the significant advantages of high-efficiency screening capability, precision and reliability, strong targeting, and wide application range. The method of the present invention uses high-quality DFT data as a training set, and the machine learning model can effectively predict the performance of new catalyst combinations, especially in CO 2 The XGB model successfully predicts the activity of catalysts, and combined with the SHAP analysis method, it further analyzes the specific contribution of key features to the activity of catalysts, helping to optimize the screening results. SHAP, the full name of which is SHapley Additive Explanations, is translated into Shapley Additive Explanations in Chinese.

[0046] Feature Engineering: There are 10 key features related to formation energy, namely doping site, atomic radius, denoted as Radii; number of valence electrons, denoted as Out_e; electron affinity, denoted as Aff; chemical potential, denoted as Chemi_pot; number of atoms, denoted as Atom_N; electronegativity, denoted as N m ; The first ionization energy, denoted as I 1; atomic weight, denoted as Atom_W; number of internal electrons, denoted as In_e. There are 11 core features related to the change in adsorption free energy, which are divided into two categories: single-atom features and catalyst features. Single-atom features include atomic radius, denoted as Radii; atomic number, denoted as Atom_N; single-atom chemical potential, denoted as μ; number of valence electrons, denoted as Out_e; first ionization energy, denoted as I 1 ; Electron affinity, denoted as Aff. Catalyst characteristics include charge transfer amount, denoted as bader; binding energy, denoted as E b ; Formation energy, denoted as E f ; Fermi level, denoted as E fermi ; Band gap, denoted as E g .

[0047] Model training: The data set is divided into a training set and a test set in a ratio of 8:2. Through 500 iterations of training, the model is ensured to have high accuracy and generalization ability. The model performance is evaluated by the root mean square error, determination coefficient and mean absolute error. The specific calculation formula is as follows:

[0048]

[0049] Among them, R 2 represents the coefficient of determination; MSR represents the root mean square error; MAR represents the mean absolute error; y i Indicates the calculated DFT value; Represents the machine learning prediction value obtained by prediction; The closer the values ​​of MSR and MAR are to 0, the higher the R 2 The closer the value is to 1, the better the prediction performance of the machine learning model.

[0050] The technical solution of the present invention is further described below through specific embodiments.

[0051] In the following examples, the methods described are conventional methods unless otherwise specified; the reagents and materials described are commercially available unless otherwise specified.

[0052] like Figure 7 , a method for screening gC 3 N 4 The method of immobilizing a single atom catalyst comprises the following steps:

[0053] Step 1, according to gC 3 N 4 The two-dimensional structure of the substrate is obtained by gC 3 N 4 Doping site of substrate; place doping element on gC 3 N 4The doping sites of the substrate are structurally optimized to screen out pre-screening catalysts that meet the electron localization requirements and obtain pre-screening results.

[0054] C 3 N 4 The doping site of the substrate is located at the substitution site or interstitial site, gC 3 N 4 The substrate has 7 doping sites, such as Figure 1 , gC 3 N 4 The doping site of the substrate is at least one of 2 C substitution sites, 3 N substitution sites and 2 H interstitial sites. The doping element is a metal element or a non-metal element, wherein the metal element is at least one of Zr, Ti, Mn, Fe, Co, Ni, Cu, Zn, Ru, Rh, Pd, Pt, Al and Li; the non-metal element is at least one of B, Si, S, O, P and Cl. The pre-screened catalyst has 140 types. gC 3 N 4 The supported single atom catalyst is denoted as SAC@gC 3 N 4 . gC 3 N 4 As an ideal support for single-atom catalysts, it can be used to increase CO 2 Performance of reduction reactions.

[0055] like Figure 1 Figure (a) shows the gC 3 N 4 The electron localization function diagram of the substrate is used to reveal the localized electron distribution characteristics of atoms or molecules. The full name of the electron localization function in English is Electron Localization Function, abbreviated as ELF. The value range of the electron localization function diagram is usually between 0 and 1, and different value ranges correspond to different electron localizations. Among them, ELF = 1 corresponds to complete localization, that is, there are a lot of electrons distributed, usually corresponding to the red area. ELF = 0.5 corresponds to the pairing probability of the electron gas type, that is, the electrons here are bonding electrons, usually corresponding to the yellow area. ELF = 0 corresponds to complete electron delocalization, that is, there are no electrons here, usually corresponding to the blue area. By analyzing the electron localization function diagram, the local distribution of valence electrons can be intuitively obtained, and then the bonding characteristics between atoms or molecules can be judged. For example, in metallic bonds, electrons show a relatively uniform distribution characteristic.

[0056] C 3 N 4With its unique two-dimensional structure, especially the rich six-fold cavity and non-equivalent nitrogen atom distribution, it becomes an ideal carrier for single-atom catalysts. This structure provides a stable anchoring point for single atoms, ensuring their efficient fixation and good electronic localization. 3 N 4 The large specific surface area also makes the catalytic reaction more active and can effectively improve the selectivity of the reaction. Figure 1 Figure (b), gC 3 N 4 There are 5 substitution positions and 2 interstitial positions, the 5 substitution positions are represented by C1, C2, N1, N2 and N3 respectively; the 2 interstitial positions are represented by H1 and H2 respectively. The doping elements are metal elements and non-metal elements.

[0057] According to the doping elements and gC 3 N 4 The substitution and interstitial positions of different doping types are screened out. 3 N 4 Single atom catalyst. The present invention embodiment is based on gC 3 N 4 By combining 5 substitution sites and 2 interstitial sites with 14 metal elements and 6 non-metal elements, 140 SAC@gC with different doping types were systematically screened. 3 N 4 Catalyst with good electron localization.

[0058] By 3 N 4 Doping metal and non-metal single atoms on the substrate can effectively control its electronic structure and catalytic activity. Doping with metal elements can usually enhance the gC 3 N 4 The electron transfer ability of gC is improved, which accelerates the adsorption and conversion process of reactants on the catalyst surface. The doping of non-metallic elements is mainly used to optimize the distribution of active sites and improve the selectivity and stability of the reaction. 3 N 4 The six-fold cavity can provide a stable single-atom structure, prevent its aggregation, and reduce CO 2 The energy barrier in the reduction reaction ensures the efficient catalytic reaction. 3 N 4 Theoretically, single-atom catalysts with this structure are CO 2 The reduction reaction provides an efficient and low-energy solution for CO 2 Great potential for conversion into high value-added chemicals.

[0059] Step 2, calculate the formation energy of some pre-screened catalysts in the pre-screening results, collect key feature data related to the formation energy to establish a formation energy-key feature data set, and construct a first prediction model so that the first prediction model can learn the nonlinear relationship between the key features and the formation energy; perform feature importance analysis on the first prediction model to generate a feature importance pie chart; perform significance analysis on the feature importance pie chart to determine significant key features; use the significant key features to screen out catalysts with negative formation energy to obtain the initial screening results. The specific method is as follows:

[0060] Set hyper parameters: According to experience or experiments, set the hyper parameters of the first XGBoost prediction model, such as n_estimators, which indicates the number of trees; learning_rate, which indicates the learning rate; max_depth, which indicates the maximum depth of the tree; reg_alpha and reg_lambda, which indicate regularization parameters, etc. In the embodiment of the present invention, these parameters are set to n_estimators = 10, learning_rate = 0.781, max_depth = 5, reg_alpha = 2, and reg_lambda = 1.

[0061] The method of constructing the first XGBoost prediction model is: based on the formation energy-key feature data set, using root mean square error, determination coefficient and mean absolute error as evaluation indicators, the model's hyperparameters, such as tree depth, learning rate, number of weak learners, etc., are optimized to learn the nonlinear relationship between key features and formation energy, and to construct and train the first XGBoost prediction model with the best prediction accuracy.

[0062] There are 10 key features related to formation energy, namely, doping site, atomic radius, denoted as Radii; number of valence electrons, denoted as Out_e; electron affinity, denoted as Aff; chemical potential, denoted as Chemi_pot; number of atoms, denoted as Atom_N; electronegativity, denoted as N m ; The first ionization energy, denoted as I 1 ; atomic weight, denoted as Atom_W; number of internal electrons, denoted as In_e, to construct the feature pool.

[0063] The formation energy of some catalysts calculated by density functional theory was used to collect 10 key features related to the formation energy of some catalysts to form a data set. The data set was divided into a training set and a test set according to a ratio of 8:2. The training set was used to train the first XGBoost prediction model, while the test set was used to evaluate the prediction performance of the first XGBoost prediction model.

[0064] Model training: The first XGBoost prediction model is trained using the training set data through the calculation formulas of root mean square error, determination coefficient and mean absolute error, so that it can learn the nonlinear relationship between key features and formation energy. The prediction performance of the first XGBoost prediction model is evaluated on the test set, and indicators such as determination coefficient, mean absolute error and root mean square error are calculated. These indicators are used to measure the accuracy and reliability of model predictions.

[0065] Model evaluation: Based on the constructed first XGBoost prediction model, the key features of the test set are used as input independent variables, and the formation energy is used as the output dependent variable. The formation energy prediction value predicted by the first XGBoost prediction model is obtained through the constructed first XGBoost prediction model. A significance analysis is performed based on the feature importance pie chart obtained based on the formation energy prediction value to obtain significant key features. Figure 2 As shown in Figure (c), doping sites and atomic radius were identified as significant key features, accounting for 52.2% and 33.7% respectively.

[0066] In addition, the formation energy calculated by DFT is plotted as the horizontal axis, and the formation energy predicted by the first XGBoost prediction model is plotted as the vertical axis. Figure 2 The comparison results are shown in Figure (b). Based on the comparison results, the prediction ability of the first XGBoost prediction model is intuitively evaluated to verify the reliability and generalizability of the model.

[0067] To ensure SAC@gC 3 N 4 The stability and efficiency of the catalyst in long-term use, the embodiment of the present invention uses the XGBoost model to accurately predict the formation energy. The XGBoost model has strong nonlinear processing capabilities and can accurately predict the formation energy of the catalyst in E f The prediction shows extremely high accuracy and reliability. f The relevant key features were used to successfully establish SAC@gC using the XGBoost model. 3 N 4 Key features in the catalyst structure and E f Then, based on the prediction and feature importance analysis results of the XGBoost model, SAC@gC 3 N 4 The structure of the catalyst is optimized to improve its stability and performance. Based on feedback and new data from actual applications, the hyperparameters and feature sets of the XGBoost model are continuously adjusted and optimized to improve its prediction accuracy and generalization ability.

[0068] like Figure 2As shown in (a) of FIG. 1 , XGB, ABR, KNR, LR, MLP, PLS, RR, SVR, RFR, GBR, etc. represent different machine learning algorithms or models. In the embodiment of the present invention, the XGBoost model is mainly used, referred to as XGB. Figure 2 Figure (b) shows the E predicted by the XGBoost model. f The value is similar to the E calculated by DFT f Comparison of the calculated values, where the data points are closely distributed along the diagonal, shows the highly accurate prediction ability of the XGBoost model. This shows that the XGBoost model predicts E f The value has a higher precision.

[0069] R on the training and test sets 2 , MAR and MSR scores are 0.9664 / 0.9120, 0.5218eV / 0.7354eV and 0.5403eV / 0.8348eV respectively, indicating that the XGBoost model shows excellent fitting ability and prediction accuracy when processing complex data. This excellent performance is attributed to the reasonable settings of hyperparameters such as n_estimators=10, learning_rate=0.781 and max_depth=5, which help the model achieve a balance between complexity and overfitting risk. In addition, the regularization parameters reg_alpha=2 and reg_lambda=1 effectively suppress the complexity of the model, avoid overfitting, and thus improve the generalization ability of the model. Through these fine parameter tuning, the XGBoost model is able to capture E f Maintain stable and accurate predictions when dealing with complex nonlinear relationships, ensuring SAC@gC 3 N 4 Accurate modeling of catalysts.

[0070] In addition, the feature importance analysis of the XGBoost model further revealed that E f The main driving factors of Figure 2 As shown in Figure (c), the doping site and atomic radius are the factors affecting E f The two most critical features of the single-atom catalyst account for 52.2% and 33.7% respectively. This shows that the combination of different doping sites and atoms directly determines the stability and performance of the single-atom catalyst. f Prediction and powerful feature importance analysis capabilities, XGBoost model for SAC@gC 3 N 4 Catalyst structure optimization and CO 2 RR application provides strong support.

[0071] In step 2, the screening condition for catalyst screening using formation energy is: screening out catalysts with negative formation energy, indicating that the system is thermodynamically stable. The screening result is: from the 140 pre-screened catalysts in step 1, 35 catalysts with thermodynamic stability were obtained through the formation energy screening in step 2. The embodiment of the present invention performs a significance analysis based on the feature importance pie chart predicted by the first XGBoost prediction model to obtain significant key features; the prediction results of the predicted catalyst are obtained by the first XGBoost prediction model, and compared with the formation energy calculation results obtained by density functional theory calculations to verify the reliability and generalizability of the model.

[0072] Step 3, calculate the adsorption free energy change of the catalyst in all the preliminary screening results, collect the core feature data related to the adsorption free energy change to establish the adsorption free energy change-core feature data set, and construct a second prediction model so that the second prediction model can learn the nonlinear relationship between the key features and the adsorption free energy change; calculate the SHAP value according to the training results of the second prediction model, and perform feature marginal contribution analysis on the core features according to the SHAP value to determine the significant core features; based on the second prediction model, use the significant core features to predict the adsorption free energy change and limiting potential of the preliminary screening catalyst, which is used to perform CO on the preliminary screening catalyst. 2 The catalytic activity of the reduction reaction was evaluated, and the catalysts with the absolute value of the adsorption free energy change ≤ 0.8 eV and the limiting potential ≤ 0.8 V were screened out to obtain the re-screening results.

[0073] Step 3.1, constructing a second prediction model so that the second prediction model can learn the nonlinear relationship between the key features and the change in adsorption free energy.

[0074] The method for constructing the second XGBoost prediction model is: based on the adsorption free energy change-core feature data set, using the root mean square error, determination coefficient and mean absolute error as evaluation indicators, the model's hyperparameters, such as tree depth, learning rate, number of weak learners, etc., are optimized to learn the nonlinear relationship between the core features and the adsorption free energy change, and to construct and train a highly interpretable second XGBoost prediction model. The specific method is as follows:

[0075] The 35 SAC@gC selected in step 2 were collected 3 N 4 The 11 core characteristic data of the catalyst related to the change in adsorption free energy are divided into single-atom characteristic data and catalyst characteristic data. Among them, the single-atom characteristics include atomic radius, denoted as Radii; atomic number, denoted as Atom_N; single-atom chemical potential, denoted as μ; valence electrons, denoted as Out_e; first ionization energy, denoted as I 1; and electron affinity, denoted as Aff. Catalyst characteristics include the amount of charge transfer, denoted as bader; binding energy, denoted as E b ; Formation energy, denoted as E f ; Fermi level, denoted as E fermi ; Band gap, denoted as E g .

[0076] 35 SAC@gC species were calculated by density functional theory. 3 N 4 The adsorption free energy change of the catalyst is used as the target variable; 11 core feature data related to the adsorption free energy change of all the pre-screened catalysts are collected to form an adsorption free energy change-core feature data set, which is divided into a training set and a test set according to a ratio of 8:2. The training set is used to train the second XGBoost prediction model and adjust the model parameters to optimize the prediction performance. The test set is used to evaluate the prediction performance of the second XGBoost prediction model.

[0077] Model training: The second XGBoost prediction model is trained using the training set data through the calculation formulas of root mean square error, determination coefficient and mean absolute error, so that it can learn the nonlinear relationship between the core features and the change in adsorption free energy. Figure 4 As shown in Figures (a) and (b), the R of the second XGBoost prediction model on the training set and the test set 2 The values ​​are 0.9664 and 0.9120 respectively, which are significantly better than other models and show higher prediction accuracy and stability.

[0078] Model evaluation: The prediction performance of the first XGBoost prediction model was evaluated on the test set, and indicators such as the coefficient of determination, mean absolute error, and root mean square error were calculated to ensure that the second XGBoost prediction model showed good prediction accuracy and stability on both the training set and the test set.

[0079] According to the model evaluation results of the second XGBoost prediction model, a feature importance pie chart is obtained; a significance analysis is performed on the feature importance pie chart to obtain significant core features. Figure 4 Figure (c) shows the feature importance pie chart of the second XGBoost prediction model, revealing the factors that affect CO 2 There are four significant core features of RR activity: band gap, binding energy, atomic number and atomic radius. Among them, the band gap of 32.4% and the binding energy of 24.1% have the greatest impact on thermodynamic stability and electron transport capacity. A smaller band gap makes electron transition and conduction smoother, effectively improving catalytic efficiency, while the binding energy optimizes the reaction process by balancing the adsorption of reactants and the desorption of products.

[0080] Step 4.2, calculate the SHAP value according to the training result of the second prediction model, and perform feature marginal contribution analysis on the core features according to the SHAP value to determine the significant core features.

[0081] In the embodiment of the present invention, the SHAP method is used to calculate the marginal contribution value of each core feature to the prediction result, that is, the SHAP value, such as Figure 5 , and draw a SHAP summary diagram to show the significance of each core feature on the prediction results. Analyze the distribution and position of different features in the SHAP summary diagram to determine the impact of each core feature on the prediction of adsorption free energy. According to the positive and negative values ​​and magnitude of the SHAP value, explain the positive or negative impact of each feature on the prediction results. Analyze the different features in CO 2 Mechanisms of action during adsorption, hydrogenation catalysis and desorption.

[0082] The embodiment of the present invention combines the second XGBoost prediction model with the SHAP method to obtain the specific marginal contribution of each core feature in the catalytic process for accurately predicting SAC@gC 3 N 4 Catalyst CO 2 The reduction reaction activity is used to further screen the type of dopant. At the same time, the catalytic performance of the screened catalysts is evaluated and verified to obtain the accuracy of the above screening method. The present invention successfully predicts 35 SAC@gC 3 N 4 The change in adsorption free energy of the catalyst is an effective way to evaluate the CO 2 RR activity.

[0083] In order to more accurately quantify the contribution of each feature to the prediction results, such as Figure 5 SHAP summary plots were used to evaluate the significance of 11 core features that are closely related to geometric and electronic properties that affect adsorption free energy. Figure 5 In the figure, all features are arranged in descending order by the sum of SHAP values, and the distribution of SHAP values ​​is shown on the horizontal axis. Each point represents a catalyst, the color of the point represents the numerical value of the corresponding catalyst feature, and the position of the point represents the SHAP value of the corresponding catalyst feature. It is worth noting that among all the features, binding energy, atomic radius, ionization energy, and band gap are usually at the top of the summary graph and are widely distributed, indicating that they have a great influence on the prediction of adsorption free energy.

[0084] In CO 2 During the adsorption process, Figure 5In Figure (a), larger binding energies have positive SHAP values, which means that the binding energy of the atoms is positively correlated with the predicted adsorption free energy value. The binding energy reflects the interaction strength between the catalyst surface and the reactants, so larger binding energies are consistent with higher predicted adsorption free energy values.

[0085] In CO 2 In the catalytic hydrogenation process, Figure 5 In Figures (b) and (c), smaller ionization energies have negative SHAP values, which means that the ionization energy of the atom is positively correlated with the predicted adsorption free energy value. The ionization energy reflects the ability of the catalyst to lose electrons. Higher ionization energy is not conducive to the transfer of electrons and thus hinders the adsorption of CO. 2 The hydrogenation catalysis of , resulting in a larger ΔG. Similarly, larger binding energy, chemical potential, and atomic radius correspond to negative SHAP values, indicating a negative correlation between these parameters and the predicted adsorption free energy values. This means that enhanced binding energy and increased atomic radius may lead to a significant improvement in the stability of the reactants on the catalyst surface, thereby reducing their adsorption free energy.

[0086] In addition, in CO 2 In the desorption process, Figure 5 Figure (d) shows the relationship with CO 2 The opposite relationship is observed for adsorption, where larger binding energies correspond to negative SHAP values, indicating that weaker interactions make CO more easily released on the catalyst surface, thereby improving the desorption efficiency. Figure 5 The band gap values ​​of purple and dark blue balls appear in the area where ΔG is close to 0, which shows that a moderate band gap is more conducive to the transmission of electrons, thereby reducing ΔG and promoting CO 2 The occurrence of RR.

[0087] Step 4.3, based on the second prediction model, the significant core features are used to predict the adsorption free energy change and limiting potential of the primary screening catalysts, which are used to perform CO 2 The catalytic activity of the reduction reaction was evaluated, and the catalysts with the absolute value of the adsorption free energy change ≤ 0.8 eV and the limiting potential ≤ 0.8 V were screened out to obtain the re-screening results.

[0088] The present invention embodiment is through 35 kinds of thermodynamically stable gC 3 N 4 Single-atom catalysts for CO 2 Evaluation of the catalytic activity of the reduction reaction to further screen the dopant type. 2 When measuring the catalytic activity of reduction reactions, the change in adsorption free energy is a commonly used important indicator because it can reliably describe the activity of different materials. The specific method is:

[0089] Draw a reaction mechanism diagram: Figure 3 Figure (a) shows the SAC@gC 3 N 4 CO on the catalyst surface 2 Schematic diagram of the mechanism and reaction pathway of the reduction reaction. CO is clearly marked in the figure. 2 There are four key steps: adsorption, COOH intermediate generation, CO formation and CO desorption. 2 The four key steps of the reduction reaction correspond to CO 2 The changes in adsorption free energy during the processes of adsorption, COOH intermediate generation, CO formation, and CO desorption are named ΔG1, ΔG2, ΔG3, and ΔG4, respectively.

[0090] CO 2 Adsorption: CO 2 Molecules and SAC@gC 3 N 4 The active sites on the catalyst surface combine to form adsorbed CO 2 This process can be expressed as CO 2 +*+2(H + +e - )→CO 2 *+2(H + +e - ); where * represents SAC@gC 3 N 4 The catalyst surface can react with CO 2 Active site for binding. COOH intermediate generation: adsorbed CO 2 At SAC@gC 3 N 4 The catalyst surface is reduced to COOH intermediate through hydrogenation reaction. The specific reaction is CO 2 *+2(H + +e - )→COOH*+(H + +e - ). CO formation: COOH intermediate is further reduced to CO through hydrogenation reaction. The reaction formula is COOH*+(H + +e - )→CO*+H 2 O(l). CO desorption: The generated CO is removed from SAC@gC 3 N 4 Desorption on the catalyst surface. The expression is CO*+H 2 O→CO+*+H 2 O(l).

[0091] Calculate the change in adsorption free energy: For each key step, calculate the corresponding change in adsorption free energy, which is recorded as ΔG 1 , ΔG 2 , ΔG 3 and ΔG 4 These adsorption free energy changes are obtained by density functional theory calculations or experimental measurements.

[0092] Analyze the change in adsorption free energy: Analyze the change in adsorption free energy of each key step to determine the CO 2 With SAC@gC 3 N 4 Whether the bonding strength of the catalyst surface is moderate. According to the Sabatier principle, a catalyst with an absolute value of ΔG close to 0 is selected to ensure effective adsorption of CO 2 It can also release the generated CO smoothly. 2 During the reduction reaction, CO 2 With SAC@gC 3 N 4 The binding strength of the catalyst surface is crucial for each reaction step. If the binding strength is too weak, initial adsorption and activation will become difficult, while if the binding strength is too strong, the active sites will be occupied, thereby hindering the release of CO. Therefore, the ideal CO 2 The reduction reaction catalyst should follow the Sabatier principle; the Sabatier principle is also called the methanogenic principle. This means that CO 2 The adsorption should be close to thermoneutrality, that is, the absolute value of ΔG should be as close to 0 as possible. Only in this case, SAC@gC 3 N 4 Catalysts can effectively adsorb CO 2 , and can smoothly release the generated CO, thus achieving efficient CO 2 Reduction reaction.

[0093] Plotting the limiting potential: Based on the change in free energy of the potential-determining step, the limiting potential is calculated to evaluate the CO 2 Reduction reaction activity. The limiting potential is calculated as: UL = -ΔG max / e, where UL represents the limit potential; ΔG max represents the maximum value of the adsorption free energy change; e represents the charge of the electron. The limiting potential diagram drawn is as follows: Figure 3 (b) shows different SAC@gC 3 N 4 The limiting potential values ​​of the catalysts. Comparison of limiting potentials: In the limiting potential diagram, the different SAC@gC 3 N 4UL value of the catalyst. Select catalysts with lower UL values ​​because these catalysts have a higher CO 2 The limiting potential is lower during the reduction reaction and less external energy is required.

[0094] Screening of candidate dopants: Based on the conditions that the absolute value of ΔG is close to 0 and the UL value is low, candidate dopants with excellent catalytic activity and higher reaction efficiency are screened. Figure 3 Figure (b) shows 35 SAC@gC 3 N 4 UL value of the catalyst. It can be clearly seen that when the |ΔG| value is less than 0.8 eV, 11 candidate dopants can be screened out, namely Zr(H1), Zr(H2), Ti(H1), Ti(H2), Co(H1), Ni(H2), Si(H1), P(H1), P(H2), Al(H1) and Al(H2). Figure 3 It can be clearly seen from Figure (b) that the overpotentials of the 11 candidate dopants are low, which means that these dopants have a high 2 The reduction reaction requires less external energy, so it has excellent catalytic activity and higher reaction efficiency. 2 Reduction reaction screening, the screening conditions are: according to the absolute value of the adsorption free energy change, screen out catalysts with ΔG absolute value ≤ 0.8eV and limiting potential ≤ 0.8V. This condition indicates that the catalyst is in CO 2 In the RR process, it complies with the Sabatier principle and has effective adsorption and desorption behavior. Screening results: From 35 catalysts, 11 catalysts were obtained by screening the absolute value of ΔG and the limiting potential.

[0095] Step 4: evaluate the catalytic activity of the catalysts in the rescreening results for hydrogen evolution reaction, so as to select catalysts suitable for synthesis gas production, CO 2 Catalyst for reduction reaction or hydrogen evolution reaction.

[0096] Screen out suitable for synthesis gas production, CO 2 The method of the catalyst for the reduction reaction or hydrogen evolution reaction is: according to CO 2 The competitive relationship between the adsorption free energy change of the reduction reaction and the hydrogen evolution reaction was studied. The catalytic activity of the catalysts in the re-screening results was evaluated for the hydrogen evolution reaction, CO 2 Evaluation of reduction reaction activity and analysis of suitability for synthesis gas production; CO 2 The change in adsorption free energy of the reduction reaction is recorded as ΔG_CO 2 RR; the change in adsorption free energy of hydrogen evolution reaction is recorded as ΔG_H; when the catalyst ΔG_CO 2 When RR is lower than ΔG_H, the catalyst tends to promote CO 2Reduction reaction of CO 2 Reduction reaction catalyst; when the ΔG_H of the catalyst is lower than ΔG_CO 2 RR, the catalyst is a hydrogen evolution catalyst that tends to promote the hydrogen evolution reaction; when the ΔG_CO 2 When RR is close to ΔG_H, the catalyst has the function of promoting CO 2 Syngas catalysts for reduction and hydrogen evolution reactions.

[0097] like Figure 6 Figure (a) and (b) show that some catalysts are 2 Some catalysts showed high conversion ability in RR, some catalysts showed excellent catalytic activity in HER, and some catalysts showed low conversion ability in CO 2 It exhibits excellent catalytic performance in both RR and HER reactions.

[0098] Screening results: According to the HER performance, the remaining 11 single-atom catalysts were divided into three categories: one is a catalyst suitable for syngas production, which has both CO 2 RR and HER catalytic activity; one is focused on CO 2 Highly efficient catalyst for reduction reaction, mainly used for CO 2 The results provide a clear direction for the development of catalysts with different catalytic activities, which are suitable for different industrial application scenarios such as carbon dioxide emission reduction, clean energy conversion and synthesis gas production.

[0099] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for screening single-atom catalysts supported on g-C3N4, characterized in that: The following steps are involved: According to the two-dimensional structure of the g-C3N4 substrate, the doping sites of the g-C3N4 substrate are obtained; the doping elements are placed at the doping sites of the g-C3N4 substrate, and the structure is optimized to screen out catalysts that meet the electronic localization requirements and obtain pre-screening results; Calculating the formation energy of some catalysts in the pre-screening results, collecting key feature data related to the formation energy to establish a formation energy-key feature data set, and constructing a first prediction model so that the first prediction model can learn the nonlinear relationship between the key feature and the formation energy; Perform feature importance analysis on the first prediction model and generate a feature importance pie chart; Perform significance analysis on the feature importance pie chart to identify significant key features; Using significant key features, catalysts with negative formation energy were screened out to obtain preliminary screening results; Calculating the adsorption free energy changes of the catalysts in all the initial screening results, collecting the core feature data related to the adsorption free energy changes, so as to establish an adsorption free energy change-core feature data set, and constructing a second prediction model so that the second prediction model can learn the nonlinear relationship between the key features and the adsorption free energy changes; Calculate the SHAP value according to the training result of the second prediction model, and perform feature marginal contribution analysis on the core features according to the SHAP value to determine the significant core features; Based on the second prediction model, the significant core features are used to predict the adsorption free energy change and limiting potential of the primary screened catalysts, which are used to evaluate the catalytic activity of the primary screened catalysts for CO2 reduction reaction, and to screen out catalysts with an absolute value of adsorption free energy change ≤0.8eV and a limiting potential ≤0.8V to obtain the re-screening results; The catalytic activity of the catalysts in the re-screening results is evaluated for hydrogen evolution reaction to screen out catalysts suitable for synthesis gas production, CO2 reduction reaction or hydrogen evolution reaction.

2. The method for screening g-C3N4-supported single-atom catalysts according to claim 1, characterized in that: The doping sites of g-C3N4 substrate are 2 C substitution sites, 3 N substitution sites and 2 H interstitial sites.

3. The method for screening g-C3N4-supported single-atom catalysts according to claim 1, characterized in that: The doping element is at least one of Zr, Ti, Mn, Fe, Co, Ni, Cu, Zn, Ru, Rh, Pd, Pt, Al, Li, B, Si, S, O, P and Cl.

4. The method for screening g-C3N4-supported single-atom catalysts according to claim 1, characterized in that: The key features related to formation energy are doping site, atomic radius, number of valence electrons, electron affinity, chemical potential, atomic number, electronegativity, first ionization energy, atomic weight and number of internal electrons; the significant key features are doping site and atomic radius.

5. The method for screening g-C3N4-supported single-atom catalysts according to claim 1, characterized in that: The method for constructing the first prediction model is: According to the formation energy-key feature data set, the root mean square error, determination coefficient and mean absolute error are used as evaluation indicators to optimize the hyperparameters of the model, so as to learn the nonlinear relationship between the key features and the formation energy, and to construct and train the first prediction model with the best prediction accuracy.

6. The method for screening g-C3N4-supported single-atom catalysts according to claim 1, characterized in that: The method for constructing the second prediction model is: According to the adsorption free energy change-core feature data set, the root mean square error, determination coefficient and mean absolute error are used as evaluation indicators to optimize the model's hyperparameters, thereby learning the nonlinear relationship between the core features and the adsorption free energy change, and constructing and training a second prediction model with high interpretability.

7. The method for screening g-C3N4-supported single-atom catalysts according to claim 1, characterized in that: The core features related to the change in adsorption free energy include single-atom features and catalyst features; The single-atom characteristics are atomic radius, atomic number, single-atom chemical potential, valence electrons, first ionization energy and electron affinity; the catalyst characteristics are charge transfer amount, binding energy, formation energy, Fermi level and band gap; the significant core characteristics are band gap and binding energy.

8. The method for screening g-C3N4-supported single-atom catalysts according to claim 1, characterized in that: The method for screening catalysts suitable for synthesis gas production, CO2 reduction reaction or hydrogen evolution reaction is: According to the competitive relationship between the adsorption free energy changes of CO2 reduction reaction and hydrogen evolution reaction, the catalysts in the re-screening results were evaluated for their catalytic activity in hydrogen evolution reaction, CO2 reduction reaction activity, and suitability for synthesis gas production; The adsorption free energy change of CO2 reduction reaction is recorded as ΔG_CO2RR; the adsorption free energy change of hydrogen evolution reaction is recorded as ΔG_H; When the ΔG_CO2RR of the catalyst is lower than ΔG_H, the catalyst is a CO2 reduction reaction catalyst that tends to promote the CO2 reduction reaction; When the ΔG_H of the catalyst is lower than the ΔG_CO2RR, the catalyst is a hydrogen evolution catalyst that tends to promote the hydrogen evolution reaction; When ΔG_CO2RR of the catalyst is close to ΔG_H, the catalyst is a synthesis gas catalyst that promotes both CO2 reduction reaction and hydrogen evolution reaction.

9. A system for screening single-atom catalysts supported on g-C3N4, characterized in that: The system comprises: processor; a memory having stored thereon a computer program executable on the processor; Wherein, when the computer program is executed by the processor, the steps of the method for screening g-C3N4-supported single-atom catalysts as described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a data processing program, and when the data processing program is executed by a processor, the steps of the method for screening g-C3N4-immobilized single-atom catalysts as described in any one of claims 1 to 8 are implemented.

Images