A two-dimensional heterojunction monatomic catalyst structure-activity relationship prediction method and device

Abstract

The application discloses a two-dimensional heterojunction single-atom catalyst structure-effect relationship prediction method and device, and relates to the technical field of catalyst performance prediction. The method comprises the following steps: constructing a two-dimensional heterojunction single-atom catalyst system and a molecular adsorption configuration corresponding to different adsorbed molecules, obtaining characteristic parameters representing the structural characteristics and electronic properties of the catalyst and adsorption performance parameters through first-principle calculation, and constructing a correlation data set according to the types of the adsorbed molecules; training and screening machine learning models based on the correlation data set, determining key catalyst descriptors, establishing a structure-effect relationship prediction model, and predicting the adsorption performance of a to-be-tested catalyst by using the structure-effect relationship prediction model, so as to realize performance prediction and candidate catalyst screening of the two-dimensional heterojunction single-atom catalyst.

Description

Technical Field

[0001] This invention relates to the field of catalyst performance prediction technology, and more specifically to a method and apparatus for predicting the structure-activity relationship of two-dimensional heterojunction single-atom catalysts. Background Technology

[0002] Single-atom catalysts, with their high atom utilization, well-defined active sites, and good catalytic selectivity, are widely used in new energy catalysis fields such as water electrolysis and fuel cells. Graphene, with its good electrical conductivity and ability to provide anchoring sites for metal single atoms through defect formation, has become a commonly used single-atom catalytic system. Existing graphene-based single-atom catalysts typically modify the coordination environment of metal single atoms through heteroatom doping, lattice defects, or surface modification, thereby adjusting the adsorption performance of intermediates in catalytic reactions such as hydrogen evolution reaction and oxygen reduction reaction.

[0003] However, the aforementioned regulatory methods mainly rely on changes in the local coordination environment of metal single atoms, resulting in limited electronic regulation pathways. This can easily lead to differences in the coordination structure of active sites, making it difficult to stably and quantitatively regulate the electronic state of metal single atoms. Furthermore, traditional graphene supports lack additional electron supply mechanisms, limiting the range of electron occupancy state regulation for metal single atoms and hindering targeted regulation of the adsorption performance of different catalytic reaction intermediates.

[0004] On the other hand, although the electronic structure of single-atom catalysts can be tuned by introducing electron donor substrates, existing methods lack systematic modeling of the correlation between interfacial charge transfer in two-dimensional heterojunctions and the catalytic performance of single atoms. Existing methods for predicting catalytic performance mainly rely on electronic characteristics such as d-band centers and work functions, which are insufficient to fully characterize the impact of interfacial charge transfer after heterojunction formation on metal single atoms and their catalytic adsorption performance. Therefore, it is necessary to establish a method that can obtain parameters such as interfacial charge transfer in heterojunctions, metal coordination environment, and other relevant structural and property parameters, and use these parameters to build a predictive model for catalytic adsorption performance.

[0005] Furthermore, current catalyst development processes typically require manual sequential completion of catalyst structure construction, first-principles calculations, and result analysis. When different metal single atoms and catalytic reaction intermediates are involved, this process must be repeated, making it difficult to quickly generate large-scale structure and catalytic performance data. Simultaneously, the catalytic adsorption behavior of different adsorbent molecules varies, making it difficult to establish corresponding structure-adsorption performance correlations for different adsorption systems using a uniform model. Therefore, it is necessary to develop a method that can automatically generate the adsorption configuration of two-dimensional heterojunction single-atom catalysts, obtain structural and property data in batches, and train structure-activity relationship prediction models for different adsorbent molecules. Summary of the Invention

[0006] The purpose of this invention is to provide a method and apparatus for predicting the structure-activity relationship (SAR) of two-dimensional heterojunction single-atom catalysts. By constructing a two-dimensional heterojunction single-atom catalyst system and its corresponding molecular adsorption configuration, first-principles calculations are used to obtain the structural features, electronic properties, and other characteristic parameters of the catalyst system, as well as the corresponding adsorption performance parameters. Correlation datasets are constructed and machine learning models are trained for different adsorption molecule types, establishing a correlation between the structural features, electronic properties, and adsorption performance of the two-dimensional heterojunction single-atom catalyst. Furthermore, the optimal machine learning model is used to determine key catalytic descriptors and establish a SAR prediction model. This eliminates the need to repeatedly perform complete first-principles calculations for each catalyst to be predicted; the adsorption performance can be predicted based on key catalytic descriptors. Therefore, a unified data processing and prediction method is provided for the performance analysis and candidate catalyst screening of two-dimensional heterojunction single-atom catalysts under different adsorption systems.

[0007] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for predicting the structure-activity relationship of a two-dimensional heterojunction single-atom catalyst, the method comprising: A two-dimensional heterojunction single-atom catalyst system was constructed, and molecular adsorption configurations corresponding to different adsorbed molecules were constructed based on the two-dimensional heterojunction single-atom catalyst system. First-principles calculations were performed on the two-dimensional heterojunction single-atom catalyst system and its molecular adsorption configuration to obtain characteristic parameters that characterize the system structure and electronic properties of the two-dimensional heterojunction single-atom catalyst system, as well as the adsorption performance parameters corresponding to the molecular adsorption configuration. A correlation dataset was constructed based on characteristic parameters and adsorption performance parameters, and the correlation dataset was divided according to the type of adsorbed molecules. Multiple machine learning models were trained based on the correlation datasets corresponding to different types of adsorbed molecules. The optimal machine learning model was determined based on the model evaluation results, and the corresponding key catalytic descriptor was determined based on the optimal machine learning model. A structure-activity relationship prediction model for two-dimensional heterojunction single-atom catalysts was established based on key catalytic descriptors, and the adsorption performance of the two-dimensional heterojunction single-atom catalysts to be predicted was then used.

[0008] Secondly, the present invention also provides a device for predicting the structure-activity relationship of a two-dimensional heterojunction single-atom catalyst, the device comprising: The system construction module is used to construct a two-dimensional heterojunction single-atom catalyst system and to construct molecular adsorption configurations corresponding to different adsorbed molecules based on the two-dimensional heterojunction single-atom catalyst system. The configuration calculation module is used to perform first-principles calculations on the two-dimensional heterojunction single-atom catalyst system and molecular adsorption configuration, and to obtain characteristic parameters that characterize the system structure and electronic properties of the two-dimensional heterojunction single-atom catalyst system, as well as the adsorption performance parameters corresponding to the molecular adsorption configuration. The data partitioning module is used to construct correlation datasets based on feature parameters and adsorption performance parameters, and to partition the correlation datasets according to the type of adsorbed molecules. The model determination module is used to train multiple machine learning models based on the correlation datasets corresponding to different adsorption molecule types, determine the optimal machine learning model based on the model evaluation results, and determine the corresponding key catalytic descriptor based on the optimal machine learning model. The performance prediction module is used to establish a structure-activity relationship prediction model for two-dimensional heterojunction single-atom catalysts based on key catalytic descriptors, and to use the structure-activity relationship prediction model to predict the adsorption performance of the two-dimensional heterojunction single-atom catalyst to be predicted.

[0009] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the structure-activity relationship prediction method for two-dimensional heterojunction single-atom catalysts provided in the first aspect.

[0010] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the structure-activity relationship prediction method for two-dimensional heterojunction single-atom catalysts provided in the first aspect.

[0011] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the structure-activity relationship prediction method for two-dimensional heterojunction single-atom catalysts provided in the first aspect.

[0012] The beneficial effects of this invention are as follows: The method of this invention constructs a two-dimensional heterojunction single-atom catalyst system and its corresponding molecular adsorption configuration, uses first-principles calculations to obtain the structural features and electronic properties of the catalyst system, as well as the corresponding adsorption performance parameters, and constructs correlation datasets and trains machine learning models for different adsorption molecule types. This enables the establishment of a correlation between the structural features, electronic properties, and adsorption performance of the two-dimensional heterojunction single-atom catalyst. Furthermore, by determining key catalytic descriptors and establishing a structure-activity relationship prediction model through the optimal machine learning model, the adsorption performance can be predicted based on key catalytic descriptors without having to repeatedly perform complete first-principles calculations for each catalyst to be predicted. This provides a unified data processing and prediction method for the performance analysis and candidate catalyst screening of two-dimensional heterojunction single-atom catalysts under different adsorption systems.

[0013] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0014] Figure 1 This invention provides a structure-activity relationship prediction for a two-dimensional heterojunction single-atom catalyst according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a two-dimensional electronic compound / graphene single-atom catalyst structure according to an embodiment of the present invention; Figure 3 The test set R of the structure-property relationship prediction model shown in an embodiment of the present invention 2 Comparison diagram; Figure 4 Cross-validation of the structure-property relationship prediction model shown in an embodiment of the present invention. 2 Comparison diagram; Figure 5 This is a schematic diagram comparing the MAE test set of the structure-property relationship prediction model according to an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the prediction performance of the optimal model for structure-property relationship prediction on the training and testing sets according to an embodiment of the present invention. Figure 7 This is a schematic diagram illustrating the performance evaluation of the cross-validation of the optimal model for predicting structure-property relationships according to an embodiment of the present invention. Figure 8 This is a schematic diagram of the feature importance analysis of the structure-activity relationship prediction model SHAP according to an embodiment of the present invention; Figure 9 This invention provides a structure-activity relationship prediction for a two-dimensional heterojunction single-atom catalyst according to an embodiment of the present invention. Figure 10 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0015] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] It should be noted that references to "an embodiment," "embodiment," "example embodiment," etc., in this specification refer to the described embodiment including specific features, structures, or characteristics; however, not every embodiment must include these specific features, structures, or characteristics. Furthermore, such expressions do not refer to the same embodiment. Moreover, when describing specific features, structures, or characteristics in conjunction with embodiments, whether or not explicitly described, it is indicated that incorporating such features, structures, or characteristics into other embodiments is within the knowledge of those skilled in the art.

[0017] Furthermore, the technical features involved 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.

[0018] Example 1

[0019] like Figure 1 As shown, this embodiment provides a method for predicting the structure-activity relationship of a two-dimensional heterojunction single-atom catalyst. The two-dimensional heterojunction single-atom catalyst in this embodiment refers to a catalyst system formed by a two-dimensional electron compound, a two-dimensional single-atom support, and transition metal single atoms supported on the two-dimensional single-atom support. The two-dimensional electron compound serves as the electron donor substrate in the heterojunction, the two-dimensional single-atom support provides loading sites for the transition metal single atoms, and the transition metal single atoms form catalytically active sites. This embodiment uses a two-dimensional heterojunction single-atom catalyst composed of two-dimensional electron compounds Ca₂N, Y₂C, and graphene as an example to illustrate the construction, data acquisition, model training, and adsorption performance prediction process of the two-dimensional heterojunction single-atom catalyst.

[0020] S101, a two-dimensional heterojunction single-atom catalyst system was constructed, and molecular adsorption configurations corresponding to different adsorbed molecules were constructed based on the two-dimensional heterojunction single-atom catalyst system.

[0021] Two-dimensional heterojunctions refer to composite structures formed by the interfacial contact of two two-dimensional materials. In this embodiment, the two-dimensional heterojunction uses a two-dimensional electronic compound as the substrate and graphene as a single-atom carrier to form a sandwich structure. The two-dimensional electronic compound is a two-dimensional material with interlayer free electrons. In this embodiment, Ca2N and Y2C are specifically used. Graphene serves as a two-dimensional single-atom carrier, providing active sites capable of loading transition metal single atoms.

[0022] Specifically, the lattice structures of Ca2N, Y2C, and graphene were first optimized. Lattice structure optimization refers to stabilizing the lattice parameters and atomic positions of the materials through calculations, thereby obtaining optimized structures for subsequent heterostructure construction. In a specific example, the optimized lattice constants were 3.570 Å for Ca2N, 3.571 Å for Y2C, and 2.479 Å for graphene. After obtaining the optimized structures, cell expansion was performed on Ca2N, Y2C, and graphene. Cell expansion refers to replicating the original unit cell by a predetermined factor to match the lattice sizes of the different materials, thus enabling the construction of heterojunctions.

[0023] In a specific example, 2×2 Ca2N supercells and 3×3 graphene supercells were used for lattice matching, with the 2×2 Y2C supercell and the 3×3 graphene supercell employing the same matching method. Calculations showed that the lattice matching degree between the 2×2 Ca2N supercell and the 3×3 graphene supercell was 1%, and the Y2C and graphene used the same cell expansion matching method. A sandwich-type two-dimensional heterojunction was constructed using the 2×2 Ca2N supercell as the substrate and the 3×3 graphene supercell as the upper single-atom carrier, and further lattice structure optimization yielded a heterojunction lattice constant of 7.445 Å. The resulting structure is shown below. Figure 2 As shown, the two-dimensional electron compound is located in the lower layer, and graphene is located in the upper layer, together forming a two-dimensional heterojunction structure.

[0024] To avoid significant periodic interactions between adjacent single atoms, the two-dimensional heterostructure undergoes further cell expansion before constructing the single-atom loading sites. In this embodiment, the two-dimensional heterostructure is expanded to a 2×2 cell to obtain the basic model for single-atom loading. Subsequently, two adjacent carbon atoms are removed near the center of the graphene, creating double vacancies. A double vacancy is a defect site formed after the removal of two adjacent carbon atoms in the graphene lattice; this defect site is used to anchor transition metal single atoms.

[0025] After forming double vacancies, a structure generation model is used to iterate through the preset transition metal elements, replacing the corresponding transition metal atoms at the double vacancy positions, thereby forming multiple two-dimensional heterojunction single-atom catalyst systems. The structure generation model is used to automatically generate catalyst structures corresponding to different metal elements according to preset structural rules, thus avoiding the need to manually build structural models for each metal element. In this embodiment, the preset transition metal elements include Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Y, Zr, Nb, Mo, Tc, Ru, Rh, Pd, Ag, Cd, Hf, Ta, W, Re, Os, Ir, Pt, and Au, a total of 28 3d, 4d, and 5d transition metal elements. For each transition metal element, the structure generation model places the corresponding metal atom at the graphene double vacancy position, forming the corresponding two-dimensional heterojunction single-atom catalyst system.

[0026] After obtaining the two-dimensional heterojunction single-atom catalyst system, the molecular adsorption configuration corresponding to the two-dimensional heterojunction single-atom catalyst system was further constructed. The molecular adsorption configuration refers to the structure after gaseous molecules or catalytic reaction intermediates form a specific spatial relationship with transition metal single atoms in the two-dimensional heterojunction single-atom catalyst, and is used to simulate the adsorption state of gaseous molecules or catalytic reaction intermediates at the catalytic active sites.

[0027] Specifically, a structure generation model is used to arrange pre-defined gaseous molecules or catalytic reaction intermediates near transition metal single atoms, and corresponding adsorption configurations are formed according to the structures of different molecules. The adsorbed molecules involved include gaseous molecules or catalytic reaction intermediates such as O2, CO, CO2, CH3, CH4, N2, H, OH, O, OOH, and COOH, which can be used to characterize the adsorption behavior in catalytic reaction systems such as hydrogen evolution reaction (HER), oxygen evolution reaction (OER), oxygen reduction reaction (ORR), CO2 reduction reaction, ammonia synthesis, CO oxidation, methane activation reaction, and formic acid decomposition reaction. The structure generation model generates corresponding molecular adsorption configurations for different transition metal single atoms and different adsorbed molecules, thus obtaining a batch structure dataset for subsequent first-principles calculations.

[0028] S102, first-principles calculations were performed on the two-dimensional heterojunction single-atom catalyst system and its molecular adsorption configuration to obtain characteristic parameters that characterize the system structure and electronic properties of the two-dimensional heterojunction single-atom catalyst system, as well as the adsorption performance parameters corresponding to the molecular adsorption configuration.

[0029] First-principles calculations refer to calculation methods based on quantum mechanics principles, directly calculating the energy, electronic structure, and related physical properties of a system based on its atomic composition and structure. This embodiment uses a batch processing program to call a first-principles calculation program to perform batch calculations on multiple two-dimensional heterojunction single-atom catalyst systems and their corresponding molecular adsorption configurations.

[0030] Specifically, for the two-dimensional heterojunction single-atom catalyst system and its corresponding molecular adsorption configuration, structural optimization calculations are performed to bring the calculated structure to a stable state, and the optimized structural data and calculation results are output. For the graphene single-atom catalyst system without a two-dimensional electron compound substrate, which serves as a control, the two-dimensional electron compound substrate portion is directly removed from the optimized two-dimensional heterojunction single-atom catalyst system and its corresponding molecular adsorption configuration to obtain the substrate-free graphene single-atom catalyst system and its corresponding molecular adsorption configuration. First-principles static calculations are then performed to obtain the corresponding calculation data. For gas molecules or catalytic reaction intermediates under vacuum, first-principles calculations are performed to obtain the corresponding gas phase molecular calculation results; for transition metal elements, first-principles calculations are performed for their most stable phase to obtain the corresponding metal atom-related calculation results.

[0031] After obtaining the first-principles calculation results, characteristic parameters characterizing the structure and electronic properties of two-dimensional heterojunction single-atom catalysts are extracted from the results. These characteristic parameters are numerical parameters used to describe the catalyst structure, electronic properties, and elemental properties; their function is to convert different catalyst structures into data that can be processed by machine learning models.

[0032] First, structural parameters are extracted from the optimized structural data. These parameters characterize the spatial structure relationship between active sites and adsorbed molecules in the two-dimensional heterojunction single-atom catalyst. In this embodiment, the structural parameters include atomic coordinates, bond lengths between transition metal atoms and neighboring carbon atoms, bond lengths between transition metal atoms and the nearest neighboring atom in the adsorbed molecule, and bond lengths and bond angles of the adsorbed molecule or catalytic reaction intermediate itself. By extracting these parameters, the coordination environment of the transition metal single atom and the structural state of the adsorbed molecule can be converted into numerical features that can be used for model training.

[0033] Further property parameters are obtained. These parameters characterize the stability and electronic properties of the two-dimensional heterojunction single-atom catalyst system. In this embodiment, the formation energy is calculated based on the total energy of the two-dimensional heterojunction single-atom catalyst system and the total energy of the most stable phase of the corresponding transition metal element. The formation energy characterizes the energy characteristics of transition metal single atoms forming and being loaded into the two-dimensional heterojunction system. The d-band centers are obtained based on the electronic structure of the two-dimensional heterojunction single-atom catalyst system to characterize the relevant electronic structure characteristics of the transition metal single atoms. Charge analysis is performed on the calculation results using the Bader charge processing program to obtain the interface transfer charges between the two-dimensional electronic compound and graphene, as well as within the heterojunction system. The work function of the two-dimensional heterojunction single-atom catalyst system is obtained using the VASPKit post-processing program.

[0034] Interface transfer charge is a property parameter specifically used in this embodiment to describe the redistribution of electrons at the heterojunction interface. It characterizes the charge transfer from the two-dimensional electronic compound to graphene and the transition metal single atoms loaded on graphene after the formation of the heterojunction. By incorporating interface transfer charge into subsequent machine learning features, the model can no longer predict solely based on the electronic characteristics of the metal single atoms themselves, but can instead utilize the electronic modulation information generated at the heterojunction interface.

[0035] Furthermore, intrinsic characteristic parameters are obtained based on the elemental properties of the transition metal elements themselves. These intrinsic characteristic parameters characterize the elemental properties of the transition metal atoms that are independent of specific adsorption configurations. In this embodiment, the intrinsic characteristic parameters include atomic number, Pauling electronegativity, atomic radius, common ionic radius, and the number of d electrons. These intrinsic characteristic parameters, together with the aforementioned structural and property parameters, constitute the characteristic parameters used to describe two-dimensional heterojunction single-atom catalysts.

[0036] In addition to obtaining structural parameters, property parameters, and intrinsic characteristic parameters, it is also necessary to obtain adsorption performance parameters corresponding to the adsorption configurations of each molecule. Adsorption performance parameters are used to characterize the adsorption behavior of gaseous molecules or catalytic reaction intermediates on two-dimensional heterojunction single-atom catalysts. In this embodiment, the corresponding adsorption performance data are calculated based on the two-dimensional heterojunction single-atom catalyst system, the corresponding molecular adsorption configuration, and the total energy of the corresponding gaseous molecules under vacuum conditions. Thus, each two-dimensional heterojunction single-atom catalyst system and its corresponding adsorbed molecule possess corresponding structural parameters, property parameters, intrinsic characteristic parameters, and adsorption performance parameters, providing a data foundation for subsequently establishing a correlation dataset.

[0037] S103, construct an association dataset based on characteristic parameters and adsorption performance parameters, and divide the association dataset according to the type of adsorbed molecules.

[0038] A correlation dataset refers to a data set that stores the characteristic parameters and adsorption performance parameters of a specific two-dimensional heterojunction single-atom catalyst system and its corresponding adsorption configuration. It is used to establish the correlation between catalyst structure and properties and adsorption performance.

[0039] Specifically, the aforementioned structural parameters, property parameters, and intrinsic characteristic parameters are integrated and correlated with adsorption performance parameters corresponding to the same two-dimensional heterojunction single-atom catalyst system and the same molecular adsorption configuration. This results in a data record containing structural parameters, property parameters, intrinsic characteristic parameters, and adsorption performance parameters. Furthermore, the correlated datasets are divided according to the type of adsorbed molecule, for example, datasets corresponding to adsorbed molecules such as H, O2, CO, CO2, CH3, CH4, N2, OH, O, OOH, and COOH are formed, so that the structural, property, and adsorption performance data corresponding to the same adsorbed molecule are organized in the corresponding datasets.

[0040] After the above processing, the structural and property data obtained from the first-principles calculations in the previous stage are not stored independently, but are associated with the corresponding adsorption performance parameters and further incorporated into a dataset divided according to the type of adsorbed molecules, thus providing complete input for subsequent machine learning model training.

[0041] S104. Multiple machine learning models are trained based on the correlation datasets corresponding to different adsorbed molecule types. The optimal machine learning model is determined based on the model evaluation results, and the corresponding key catalytic descriptor is determined based on the optimal machine learning model.

[0042] A machine learning model is a computational model that uses the data relationship between obtained feature parameters and target values ​​for training to establish a mapping relationship between input features and target values. In this embodiment, structural parameters, property parameters, and intrinsic feature parameters are used as model feature values, and the corresponding adsorption performance is used as the model target value.

[0043] First, the corresponding correlation datasets are processed according to the type of adsorbed molecules, and each correlation dataset is divided into a training set and a test set. In a specific example, the datasets are randomly divided into an 80% and 20% ratio, with 80% of the data used as the training set and 20% as the test set. The training set is used to establish the relationship between features in the machine learning model and adsorption performance, while the test set is used to evaluate the prediction performance of the trained machine learning model on data not used in the training.

[0044] Subsequently, multiple machine learning models were used to train the training set. The machine learning models used in this embodiment include K-Nearest Neighbors, XGBoost, Decision Tree, LightGBM, AdaBoost, Gradient Boosting, Random Forest, Multilayer Perceptron, Support Vector Machine, Lasso Regression, Elastic Network, and Ridge Regression. For each type of adsorbed molecule, the structural parameters, property parameters, and intrinsic feature parameters from its corresponding correlation dataset were used as inputs, and the corresponding adsorption performance was used as the target value to train the aforementioned machine learning models.

[0045] After training, each machine learning model is tested using a test set, and the coefficient of determination R of the model is calculated. 2 Root mean square error (RMSE) and mean absolute error (MAE). Coefficient of determination R0 2 The root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the model's ability to explain the variation of the target value. These are used to assess the deviation between the model's predicted and actual values. Simultaneously, comparing the model evaluation results on the training and test sets determines whether the model is significantly overfitting. Furthermore, 5-fold cross-validation is performed on the machine learning model. The data is divided into five subsets, and one subset is used for validation while the remaining subsets are used for training, resulting in multiple sets of cross-validation results used to further evaluate the generalization ability of the machine learning model.

[0046] Based on the model evaluation results above, multiple machine learning models are comprehensively evaluated, and the optimal machine learning model that matches the corresponding adsorbed molecule type is determined from among them. Thus, an optimal machine learning model can be obtained for each type of adsorbed molecule.

[0047] After determining the optimal machine learning model, the corresponding key catalytic descriptors are further determined based on the optimal machine learning model. Key catalytic descriptors refer to feature parameters that make a high contribution to the prediction of adsorption performance in the machine learning model. They are used to extract features that are more closely related to adsorption performance from the large number of structural parameters, property parameters, and intrinsic feature parameters that have been obtained.

[0048] This embodiment employs the SHAP method for feature importance analysis. SHAP is a method used to interpret the prediction results of machine learning models. It obtains feature importance information by calculating the contribution of each input feature to the model output. Specifically, the correlation dataset corresponding to the adsorbed molecule type is input into the determined optimal machine learning model, and the SHAP values ​​corresponding to the structural parameters, property parameters, and intrinsic feature parameters are calculated. The importance of each feature parameter is ranked according to the SHAP values, and the key catalytic descriptors for the corresponding adsorbed molecule type are determined based on the ranking results.

[0049] In a specific example of HER adsorption system analysis, formation energy, the number of d electrons in metal atoms, and interfacial transfer charge exhibit high feature importance. Formation energy characterizes the loading characteristics of metal atoms in a two-dimensional heterojunction single-atom catalyst; the number of d electrons in metal atoms characterizes the intrinsic electronic properties of transition metal single atoms; and interfacial transfer charge characterizes the degree of interfacial electronic modulation between the two-dimensional electronic compound and graphene and its supported metal single atoms. It should be noted that the above specific feature importance ranking is used to illustrate one analysis result in this embodiment. The key catalytic descriptors actually determined for different adsorbed molecule types can be determined based on the feature importance analysis results of the corresponding correlation dataset and the optimal machine learning model.

[0050] S105, a structure-activity relationship prediction model for two-dimensional heterojunction single-atom catalysts was established based on key catalytic descriptors, and the adsorption performance of the two-dimensional heterojunction single-atom catalysts to be predicted was then used.

[0051] Structure-activity relationship (SOR) prediction models are prediction models that take key catalytic descriptors, which characterize the catalyst structure and properties and are screened through eigenvalue analysis, as input, and output the adsorption performance of the corresponding adsorbed molecules. Since the aforementioned key catalytic descriptors have been determined by the correlation dataset and the optimal machine learning model, the SOR prediction model inherits the structure-property-performance correlations obtained from the aforementioned data processing and model training.

[0052] Furthermore, the structural parameters, property parameters, and intrinsic characteristic parameters of the two-dimensional heterojunction single-atom catalyst to be predicted are obtained. Based on the type of adsorbed molecule to be predicted, key catalytic descriptors corresponding to that adsorbed molecule type are extracted from these characteristic parameters. For example, when it is necessary to predict the adsorption performance of a certain two-dimensional heterojunction single-atom catalyst for a specific gas-phase molecule or catalytic reaction intermediate, the corresponding characteristic parameters are obtained based on the structure and electronic properties of the catalyst, and then key catalytic descriptors corresponding to that adsorbed molecule type are selected from them.

[0053] Subsequently, the key catalytic descriptors are input into the structure-activity relationship (SAR) prediction model corresponding to the adsorption molecule type. The SAR prediction model then outputs the predicted adsorption performance of the two-dimensional heterojunction single-atom catalyst to be predicted. For multiple two-dimensional heterojunction single-atom catalysts to be predicted, the predicted adsorption performance of each catalyst can be obtained separately. Based on the prediction results, the multiple two-dimensional heterojunction single-atom catalysts to be predicted can be compared, and candidate two-dimensional heterojunction single-atom catalysts can be selected.

[0054] Through the above process, the initially constructed two-dimensional heterojunction single-atom catalyst system is subjected to first-principles calculations after adsorption configuration generation. The structural parameters, property parameters, intrinsic characteristic parameters, and adsorption performance parameters obtained from the first-principles calculations are further used to form a correlation dataset. The correlation dataset is then used for machine learning model training and model screening. The optimal machine learning model obtained from the screening is further used for key catalytic descriptor analysis. The obtained key catalytic descriptors are then used to establish a structure-activity relationship prediction model. Finally, the structure-activity relationship prediction model is used to predict the adsorption performance of the new two-dimensional heterojunction single-atom catalyst and to screen candidate catalysts. This forms a complete processing flow from catalyst structure construction, data acquisition, data correlation, model training, descriptor extraction to catalytic performance prediction.

[0055] In the first-principles calculations of this embodiment, the calculation parameters can be set according to the calculation characteristics of the two-dimensional heterojunction single-atom catalytic system to obtain the corresponding calculation results. When calling the VASP calculation software for first-principles calculations, ALGO=All can be set in the INCAR input file to optimize the Kohn-Sham orbitals using the conjugate gradient algorithm to obtain the electronic ground state; ISPIN=2 can be set to consider the spin state of the adsorption system; IVDW=12 can be set to consider the van der Waals interactions between the two-dimensional electronic compound and the graphene layer, between the graphene layer and the single atom, and between the adsorbed molecules; when it is necessary to simulate an ionic electrolyte environment, LSOL=.TRUE., ISOL=2, C_MOLAR=1.0, and R_ION=4.0 can be set to simulate the implicit solvent environment using VASPsol. When constructing the two-dimensional heterojunction single-atom catalyst system, a vacuum layer of not less than 20 Å can also be set to reduce the influence of long-range interactions between periodic structures on the calculation results. The above calculation parameters are specific calculation conditions in this embodiment. The calculation results generated are used to extract structural parameters, property parameters and adsorption performance parameters in the aforementioned manner, and then enter the subsequent correlation dataset and model training process.

[0056] Example 2

[0057] This embodiment uses the adsorption system corresponding to the hydrogen evolution reaction (HER) as an example to verify the established structure-activity relationship prediction model of the two-dimensional heterojunction single-atom catalyst.

[0058] First, the HER-related dataset was divided into training and testing sets, with the training set comprising 80% and the testing set comprising 20%. Twelve machine learning models were used for training: K-Nearest Neighbors, XGBoost, Decision Tree, LightGBM, AdaBoost, Gradient Boosting, Random Forest, Multilayer Perceptron, Support Vector Machine, Lasso Regression, Elastic Network, and Ridge Regression. After training, the determination coefficients R0 for each model were calculated using the testing set. 2 The root mean square error (RMSE) and mean absolute error (MAE) were used to evaluate the predictive ability of different models for hydrogen adsorption performance. For example... Figure 3 As shown, the test set R of the K-nearest neighbor model 2 The value is 0.707 for the LightGBM model on the test set R. 2 The value is 0.337, and the test set R for the other models is... 2 All values ​​are greater than 0.85. For models with better predictive performance, the generalization ability and overfitting are further evaluated by combining the prediction results from the training and test sets.

[0059] Subsequently, 5-fold cross-validation was performed on the models with better predictive performance. 5-fold cross-validation involves dividing the data into five subsets, using one subset sequentially for validation, and using the remaining subsets for training, to evaluate the model's stability under different data partitioning conditions. For example... Figure 4 As shown, the validation results indicate that the cross-validation determination coefficient R0 of the Lasso regression, elastic network, and ridge regression models is [value missing]. 2 All reached over 90%. For example... Figure 5 As shown, the MAE on the test set of the three models is further compared. The MAE of the ridge regression model is 0.189 eV, therefore the R-value of the combined model is... 2 Based on RMSE, MAE, and overfitting, the Ridge Regression model was determined as the optimal machine learning model for HER.

[0060] like Figure 6 As shown, further cross-validation of the ridge regression model is performed on its training set R. 2 The test set R is 95%. 2 The performance rate was 90%, and no significant performance difference was observed between the training and test sets; for example... Figure 7 As shown, in 5-fold cross-validation, the R-squared value of the prediction results for each test set is... 2 The accuracy rates are all around 90%, indicating that the ridge regression model has good predictive stability and generalization ability.

[0061] After determining the optimal machine learning model, the SHAP method is further used to perform feature importance analysis on the ridge regression model. SHAP is used to calculate the contribution of different input features to the model's prediction results, and the feature importance is ranked according to the SHAP values ​​corresponding to each feature. Figure 8 As shown, in the HER system, the formation energy E_form, the number of d electrons in metal atoms, and the interface transfer charge have high characteristic importance, ranking among the top important characteristics. Therefore, the formation energy E_form, the number of d electrons in metal atoms, and the interface transfer charge are identified as the key catalytic descriptors for the HER system. Among them, the formation energy E_form is used to characterize the loading and formation characteristics of transition metal single atoms, the number of d electrons in metal atoms is used to characterize the intrinsic electronic properties of transition metal single atoms, and the interface transfer charge is used to characterize the electronic regulation characteristics of the two-dimensional heterojunction interface.

[0062] Further analysis of the key catalytic descriptors using characteristic SHAP dependencies, characteristic SHAP distributions, and characteristic correlation matrices was conducted to verify the correlation between the key catalytic descriptors and hydrogen adsorption performance. Based on the analyzed key catalytic descriptors, a structure-activity relationship (SHR) prediction model was established for the HER (Heterochromatic Effect) profile. The formation energy E_form, the number of d electrons in the metal atom, and the interfacial transfer charge of the two-dimensional heterojunction single-atom catalyst to be predicted were input into the SHR prediction model to obtain the corresponding hydrogen adsorption performance prediction results. Based on the prediction results of multiple two-dimensional heterojunction single-atom catalysts to be predicted, candidate catalysts were screened.

[0063] The verification results of this embodiment show that the optimal machine learning model suitable for the HER system can be determined by multi-model comparison, the stability of the optimal machine learning model can be verified by cross-validation, and key catalytic descriptors can be determined from multiple feature parameters by SHAP feature importance analysis, providing a verification basis for establishing a structure-activity relationship prediction model for two-dimensional heterojunction single-atom catalysts using key catalytic descriptors.

[0064] Based on the same inventive concept, this application also provides a two-dimensional heterojunction single-atom catalyst structure-activity relationship prediction device for implementing the above-described method for predicting the structure-activity relationship of two-dimensional heterojunction single-atom catalysts. The solution provided by this device is similar to the solution described in the above-described method. Therefore, the specific limitations in one or more embodiments of the two-dimensional heterojunction single-atom catalyst structure-activity relationship prediction device provided below can be found in the limitations of the two-dimensional heterojunction single-atom catalyst structure-activity relationship prediction method described above, and will not be repeated here.

[0065] In one embodiment, such as Figure 9 As shown, a device for predicting the structure-activity relationship of a two-dimensional heterojunction single-atom catalyst is provided. The device includes: System construction module 30 is used to construct a two-dimensional heterojunction single-atom catalyst system and to construct molecular adsorption configurations corresponding to different adsorbed molecules based on the two-dimensional heterojunction single-atom catalyst system. Configuration calculation module 31 is used to perform first-principles calculations on the two-dimensional heterojunction single-atom catalyst system and molecular adsorption configuration, and obtain characteristic parameters that characterize the system structure and electronic properties of the two-dimensional heterojunction single-atom catalyst system, as well as the adsorption performance parameters corresponding to the molecular adsorption configuration. The data partitioning module 32 is used to construct a correlation dataset based on characteristic parameters and adsorption performance parameters, and to partition the correlation dataset according to the type of adsorbed molecules. The model determination module 33 is used to train multiple machine learning models based on the correlation datasets corresponding to different adsorption molecule types, determine the optimal machine learning model based on the model evaluation results, and determine the corresponding key catalytic descriptor based on the optimal machine learning model. The performance prediction module 34 is used to establish a structure-activity relationship prediction model for two-dimensional heterojunction single-atom catalysts based on key catalytic descriptors, and to use the structure-activity relationship prediction model to predict the adsorption performance of the two-dimensional heterojunction single-atom catalyst to be predicted.

[0066] This application also provides an electronic device, in some embodiments, referring to... Figure 10 As shown, the electronic device 700 includes an input unit 710, a memory 720, a processor 730, and an output unit 740. The memory 720 stores program instructions that can be executed on the processor 730. The processor 730 can execute the structure-activity relationship prediction method and / or technical solution for two-dimensional heterojunction single-atom catalysts based on the foregoing embodiments by calling the program instructions. The electronic device 700 can be a mobile terminal device such as a mobile phone or computer.

[0067] Furthermore, embodiments of this application also provide a computer-readable storage medium for storing a computer program that performs a method for predicting the structure-activity relationship of a two-dimensional heterojunction single-atom catalyst. For example, computer program instructions, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. The program instructions that invoke the methods of this application may be stored in a fixed or removable storage medium, and / or transmitted via data streams in broadcast or other signal carrying media, and / or stored in a storage medium that operates according to the program instructions.

[0068] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0069] The technical features of the above embodiments can be arbitrarily integrated. For the sake of brevity, not all possible integrations of the technical features in the above embodiments are described. However, as long as the integration of these technical features does not contradict each other, they should be considered to be within the scope of this specification.

[0070] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for predicting the structure-activity relationship of a two-dimensional heterojunction single-atom catalyst, characterized in that, The method includes: A two-dimensional heterojunction single-atom catalyst system was constructed, and molecular adsorption configurations corresponding to different adsorbed molecules were constructed based on the two-dimensional heterojunction single-atom catalyst system. First-principles calculations were performed on the two-dimensional heterojunction single-atom catalyst system and the molecular adsorption configuration to obtain characteristic parameters characterizing the system structure and electronic properties of the two-dimensional heterojunction single-atom catalyst system, as well as the adsorption performance parameters corresponding to the molecular adsorption configuration. A correlation dataset is constructed based on the feature parameters and the adsorption performance parameters, and the correlation dataset is divided according to the type of adsorbed molecules. Multiple machine learning models are trained based on the correlation datasets corresponding to different types of adsorbed molecules. The optimal machine learning model is determined based on the model evaluation results, and the corresponding key catalytic descriptor is determined based on the optimal machine learning model. Based on the key catalytic descriptor, a structure-activity relationship prediction model for two-dimensional heterojunction single-atom catalysts is established, and the adsorption performance of the two-dimensional heterojunction single-atom catalyst to be predicted is then used.

2. The method for predicting the structure-activity relationship of two-dimensional heterojunction single-atom catalysts as described in claim 1, characterized in that, The construction of the two-dimensional heterojunction single-atom catalyst system includes: The structures of two-dimensional electronic compounds and two-dimensional single-atom supports were optimized and lattice-matched respectively, and a heterojunction was constructed based on the lattice-matching results of the two-dimensional electronic compounds and two-dimensional single-atom supports. Single-atom loading sites are constructed in the two-dimensional single-atom support, and transition metal atoms are loaded onto the single-atom loading sites to form the two-dimensional heterojunction single-atom catalyst system. Based on the two-dimensional heterojunction single-atom catalyst system, different gaseous molecules or catalytic reaction intermediates are arranged near the transition metal atoms to generate corresponding molecular adsorption configurations.

3. The method for predicting the structure-activity relationship of two-dimensional heterojunction single-atom catalysts as described in claim 2, characterized in that, The acquisition of characteristic parameters characterizing the structure and electronic properties of the two-dimensional heterojunction single-atom catalyst includes: Structural and property parameters were extracted from the calculation results of the two-dimensional heterojunction single-atom catalyst system and the molecular adsorption configuration, and the intrinsic characteristic parameters of the transition metal atoms were obtained. The structural parameters are used to characterize the structural relationship between the transition metal atoms, the two-dimensional single-atom support, and the adsorbed molecules; the property parameters are used to characterize the stability and electronic properties of the two-dimensional heterojunction single-atom catalyst system; and the intrinsic characteristic parameters are used to characterize the elemental properties of the transition metal atoms. Based on the two-dimensional heterojunction single-atom catalyst system, the molecular adsorption configuration, and the calculation results of the corresponding gas phase molecules, the adsorption performance parameters corresponding to the molecular adsorption configuration are obtained.

4. The method for predicting the structure-activity relationship of two-dimensional heterojunction single-atom catalysts as described in claim 3, characterized in that, The aforementioned datasets, based on the correlations corresponding to different adsorbed molecule types, are used to train multiple machine learning models, including: The corresponding correlation datasets are divided into training data and test data according to the type of adsorbed molecules. Multiple machine learning models are used to train the training data, and the trained machine learning models are evaluated using the test data. Based on the prediction evaluation index of the machine learning model and the fitting performance during the training process, multiple machine learning models are comprehensively evaluated, and the optimal machine learning model for the corresponding adsorption molecule type is determined according to the comprehensive evaluation results. Cross-validation is performed on the optimal machine learning model to obtain the evaluation results of its generalization ability.

5. The method for predicting the structure-activity relationship of two-dimensional heterojunction single-atom catalysts as described in claim 4, characterized in that, The determination of the corresponding key catalytic descriptor based on the optimal machine learning model includes: Based on the optimal machine learning model, feature importance analysis is performed on the feature parameters in the related dataset; Based on the results of the feature importance analysis, the feature parameters are ranked in order of importance, and the feature parameters that meet the preset conditions for correlation with the adsorption performance are determined as the key catalytic descriptors for the corresponding adsorption molecule types. Based on the correlation between the key catalytic descriptors and the adsorption performance, a structure-activity relationship prediction model for the corresponding adsorption molecule type is established.

6. The method for predicting the structure-activity relationship of two-dimensional heterojunction single-atom catalysts as described in claim 5, characterized in that, The prediction of the adsorption performance of the two-dimensional heterojunction single-atom catalyst using the structure-activity relationship prediction model includes: Obtain the characteristic parameters corresponding to the two-dimensional heterojunction single-atom catalyst to be predicted, and extract the key catalytic descriptors corresponding to the type of adsorbed molecules to be predicted from the characteristic parameters; The key catalytic descriptor is input into the structure-activity relationship prediction model corresponding to the type of adsorbed molecule to be predicted, and the adsorption performance prediction result corresponding to the two-dimensional heterojunction single-atom catalyst to be predicted is obtained. Based on the predicted adsorption performance of multiple two-dimensional heterojunction single-atom catalysts, the predicted two-dimensional heterojunction single-atom catalysts are screened to obtain candidate catalysts.

7. A device for predicting the structure-activity relationship of a two-dimensional heterojunction single-atom catalyst, characterized in that, The device includes: The system construction module is used to construct a two-dimensional heterojunction single-atom catalyst system, and to construct molecular adsorption configurations corresponding to different adsorbed molecules based on the two-dimensional heterojunction single-atom catalyst system. The configuration calculation module is used to perform first-principles calculations on the two-dimensional heterojunction single-atom catalyst system and the molecular adsorption configuration to obtain characteristic parameters that characterize the system structure and electronic properties of the two-dimensional heterojunction single-atom catalyst system, as well as the adsorption performance parameters corresponding to the molecular adsorption configuration. The data partitioning module is used to construct a correlation dataset based on the feature parameters and the adsorption performance parameters, and to partition the correlation dataset according to the type of adsorbed molecules. The model determination module is used to train multiple machine learning models based on the correlation dataset corresponding to different adsorption molecule types, determine the corresponding optimal machine learning model based on the model evaluation results, and determine the corresponding key catalytic descriptor based on the optimal machine learning model. The performance prediction module is used to establish a structure-activity relationship prediction model for two-dimensional heterojunction single-atom catalysts based on the key catalytic descriptors, and to use the structure-activity relationship prediction model to predict the adsorption performance of the two-dimensional heterojunction single-atom catalyst to be predicted.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for predicting the structure-activity relationship of two-dimensional heterojunction single-atom catalysts according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for predicting the structure-activity relationship of a two-dimensional heterojunction single-atom catalyst according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for predicting the structure-activity relationship of two-dimensional heterojunction single-atom catalysts as described in any one of claims 1 to 6.

Images