High-flux ORR and OER electrocatalyst screening method based on constant potential

Abstract

The invention belongs to the technical field of materials, and particularly relates to a high-flux ORR and OER electrocatalyst screening method based on constant potential. Comprising the following steps: establishing and optimizing an ORR and OER electrocatalyst initial crystal structure model, and matching actual crystal characteristics; performing energy minimization calculation by using a density functional theory to obtain a thermodynamic stable structure; constructing an electrolyte environment, regulating the pH value, and applying a wide-range potential to form an adaptive model; drawing a Bayer diagram, labeling adsorption species, and establishing association between adsorption and working conditions; calculating charge density and other verification models, outputting performance, simulating dynamic adsorption to evaluate stability, and determining long-term stable species. According to the scheme, the problem that ORR and OER high-throughput screening working condition authenticity and structural diversity are difficult to consider in electrocatalyst screening is solved, virtual high-throughput screening is achieved through analog calculation, introduction of potential, solvent and other working conditions and by means of machine learning, the cycle is shortened, the machine-hour cost is reduced, the performance prediction accuracy is improved, and the structure-activity relationship is constructed; and electrocatalyst research and development are accelerated.

Description

TECHNICAL FIELD

[0001] The present scheme belongs to the technical field of materials, and particularly relates to a high-throughput ORR and OER electrocatalyst screening method based on constant potential. BACKGROUND

[0002] High-throughput screening of oxygen reduction reaction (ORR) and oxygen evolution reaction (OER) electrocatalysts is a core requirement for material development in the field of energy conversion, which lies in quickly screening high-performance materials from a large number of candidate systems through batch modeling, calculation and analysis, and providing support for device optimization such as fuel cells, metal-air batteries, etc. At present, there are high-throughput screening schemes for energy materials, such as the data-driven low-dimensional energy material high-throughput screening method disclosed in Chinese patent CN116168782A, which obtains substrate structure information, constructs a two-dimensional electrocatalyst model, combines a database with machine learning, simulates an electrochemical reaction mechanism and calculates an overpotential, and provides a material model and an electrochemical evaluation index for a user, thereby reducing the dependence on first-principle calculations or supercomputer experience and improving the screening efficiency of low-dimensional energy materials.

[0003] However, the above method has significant limitations in ORR and OER electrocatalyst screening. On the one hand, traditional experimental trial and error takes 3-12 months for single catalyst optimization in a closed loop of empirical formula, synthesis, characterization and testing, and the search space expands exponentially after combining multiple components, and is affected by synthesis batch differences, atmosphere fluctuations, etc., resulting in poor data repeatability and difficulty in systematically exploring the influence of working conditions on ORR / OER performance. On the other hand, traditional DFT static calculation only simulates the intrinsic environment in vacuum, obtains parameters such as adsorption energy by solving the cell model and electronic structure, ignores factors such as potential and solvent, and cannot reflect dynamic effects such as interface water molecule reconstruction and ion-specific adsorption, resulting in a prediction error of ORR / OER overpotential often exceeding 200 mV. Although the method introduces data-driven and machine learning, it does not adequately consider key working condition factors such as constant potential regulation and solvation effects specific to ORR / OER, and the database focuses on intrinsic materials, providing limited support for modification strategies such as doping and defects, making it difficult to accurately capture the dynamic charge distribution on the catalyst surface and the change in adsorption free energy of intermediates in actual reactions.

[0004] These defects make it difficult to meet the high-efficiency screening demand of ORR and OER electrocatalysts in electrocatalyst screening, resulting in the inability to balance the authenticity of working conditions and the diversity of structures in high-throughput screening methods. SUMMARY

[0005] The purpose of the scheme is to provide a high-throughput ORR and OER electrocatalyst screening method based on constant potential, so as to solve the problem that the efficient screening of ORR and OER electrocatalysts in electrocatalyst screening is difficult to meet, and the real working condition and structural diversity cannot be considered in the high-throughput screening method.

[0006] In order to achieve the above purpose, the scheme provides a high-throughput ORR and OER electrocatalyst screening method based on constant potential, comprising the following steps:

[0007] S10: Based on the chemical composition of the target electrocatalytic material, a reference material is derived from a database; an initial crystal structure model is constructed based on the reference material, and basic parameters including lattice parameters and atomic coordinates are extracted to obtain a structure data set to be screened;

[0008] S20: The initial crystal structure model corresponding to the structure data set to be screened is preprocessed to obtain atomic coordinates and pseudo-potential files of the structure to be screened; the atomic coordinates are optimized to obtain the structure to be screened in the lowest energy state; then the electronic self-consistent calculation is performed on the structure to be screened; the solvent effect is introduced into the screening framework of the structure data set to be screened after processing according to the structure after the electronic self-consistent calculation by using the solvation method to simulate the electrolyte environment, and the surface atomic interaction adsorption energy corrected by the solvent effect is obtained;

[0009] S30: Based on the surface atomic interaction adsorption energy, the stability of the catalyst to be screened is measured by using indexes including formation energy and dissolution potential, and the implicit solvent model is enabled in the measurement process to obtain a stable structure;

[0010] S40: For the stable structure, a constant potential calculation method is used in the implicit solvent model to adjust the electrode potential and monitor the surface net charge change in real time, and the simulation result of the surface charge distribution of the catalyst is obtained;

[0011] S50: Collect the data generated in the constant potential calculation process, including the surface net charge distribution under different electrode potentials and the change of intermediate adsorption free energy; use the data as input features, use overpotential as a target variable of catalytic performance index, use machine learning tools to screen features of the target variable, and mine key physical descriptors; based on the key physical descriptors, an efficacy relationship model of catalytic performance and descriptors is constructed, and the performance of new electrocatalytic materials is predicted by using the efficacy relationship model.

[0012] The principle and technical effects of this scheme are as follows: An initial crystal structure model is constructed by obtaining reference materials from a database. Through structure optimization and electronic self-consistent calculations, an energy-stable structure to be screened is obtained. A solvation model is introduced to simulate the electrolyte environment to correct the adsorption energy. Stable structures are then retained through a stability coarse screening process. Subsequently, the surface charge distribution and intermediate adsorption free energy changes under actual operating conditions are simulated using the constant potential method. Finally, machine learning tools are used to mine key physical descriptors and construct an efficacy relationship model. This scheme achieves high-throughput, low-cost screening of electrocatalysts, improves the accuracy of performance prediction, and shortens the screening process by replacing some experimental operations with virtual screening, incorporating operating conditions such as potential and solvent, and efficiently processing data.

[0013] Specifically, firstly, traditional experiments require 3–12 months for each dopant-calcination-test cycle. With combinations of various elements, proportions, calcination temperatures, and atmospheres, the search space expands exponentially. Simultaneously, uncontrollable factors such as batch-to-batch variations, impurities, and atmospheric fluctuations contribute to poor data repeatability and high costs. This proposed solution simulates the ORR / OER electrocatalyst reaction, enabling high-throughput screening of electrocatalysts under simulated conditions. It can complete thousands of virtual screenings within 3 months, significantly shortening the experimental cycle and eliminating batch-to-batch uncertainties. The time cost is reduced by several times compared to experimental costs, achieving low-cost, repeatable catalyst optimization.

[0014] Secondly, traditional DFT methods, when simulating electrocatalytic reactions, often only consider the intrinsic state of the vacuum environment, neglecting factors such as solvent and potential. This leads to an incomplete consideration of solvation effects, especially the interaction between solvent molecules and the catalyst surface under complex operating conditions, resulting in ORR / OER overpotential prediction errors exceeding 200mV. This proposed solution utilizes a solvation model (e.g., setting the solvent dielectric constant to 80 and the Debye length...) Combining this method with the constant potential method, which incorporates factors such as potential, solvent, and solute, allows for a more realistic simulation of the reaction environment and accurately reflects dynamic effects such as interfacial water molecule reconstruction and ion-specific adsorption. For example, the calculation of the adsorption energy of -OH groups on the catalyst surface in the OER reaction, by incorporating solvation energy correction, significantly reduces the deviation from the actual reaction, thus improving the accuracy of ORR / OER catalytic performance prediction.

[0015] Further, the conventional method is difficult to establish a clear structure-activity relationship from a large amount of structure and performance data. In the research of electrocatalytic materials, a large amount of data needs to be calculated and processed, and a large amount of data is difficult to directly convert into an effective performance prediction model, and it is difficult to directly build the exact relationship between structure and performance. The scheme introduces machine learning tools to quickly screen out key descriptors such as surface charge density from data such as surface net charge distribution and bond length change under different potentials, which can save half of the research time and build the performance relationship between catalytic performance and descriptors, providing a more scientific basis for the design of new catalysts, while quickly screening out key physical descriptors and reducing redundant calculations. For example, without calculating all performance parameters one by one, the overpotential can be predicted by descriptor correlation, which can efficiently predict the performance change of electrocatalysts under actual working conditions, significantly shorten the research time, provide accurate theoretical support for the design and optimization of new efficient electrocatalysts, and accelerate the research and development process of ORR / OER electrocatalytic materials.

[0016] In summary, the scheme solves the problem of difficulty in considering the authenticity of ORR and OER high-throughput screening conditions and structural diversity in electrocatalyst screening. Through simulation calculation, the inclusion of potential and solvent conditions, and the use of machine learning, virtual high-throughput screening is realized, the cycle is compressed, the cost is reduced, the performance prediction accuracy is improved, the structure-activity relationship is built, and the electrocatalyst research and development is accelerated.

[0017] Further, the specific calculation parameters and correction methods of structure optimization in step S20 are: the self-consistent convergence criterion of the electronic step is set to EDIFF = 1 x 10 -5 eV, the force convergence criterion of the ion step is set to The plane wave cutoff energy is set to ENCUT = 520 eV; Gaussian broadening is adopted with a broadening coefficient of 0.05 eV; spin polarization is turned on; the structure optimization mode is full relaxation, and the optimization algorithm adopts the conjugate gradient method with an ion step of 0.5 fs; the DFT-D3 damping scheme is used for van der Waals correction; ADDGRID = T is enabled to improve the charge density grid; GGA+U correction is applied to the atoms, and the Hubbard U value of the transition metal element is automatically set according to the element type.

[0018] The relative strict convergence criterion of the electronic step and the ion step is which balances the accuracy and efficiency while ensuring the calculation accuracy of the electronic state density and the atomic position, and avoiding the waste of calculation resources caused by excessive iteration; the plane wave cutoff energy 520 eV can fully cover the electronic broadening needs of transition metal valence electrons (such as 3d, 4s orbitals) and non-metal atoms, while avoiding the calculation burden brought by redundant basis sets.

[0019] The conjugate gradient method combined with the full relaxation mode can not only efficiently converge to the lowest energy structure, but also adapt to catalyst structures containing complex defects (such as double vacancies and heteroatom doping) such as the surface modification structure of Ti3C2 MXene, avoiding being trapped in local energy minima, and is more universal than single optimization algorithm.

[0020] The combination of DFT-D3 damping scheme and ADDGRID=T not only accurately describes the weak intermolecular interactions in layered or porous catalysts (such as MOFs materials) through van der Waals correction, but also eliminates numerical integration errors by improving the accuracy of the charge density grid, for example, in the calculation of the adsorption energy of OOH* intermediate on the surface of Fe-N4 catalyst, the combination of the two makes the error much smaller than the single correction, and the combined effect is much better than the additive effect of the single correction.

[0021] Further, when simulating the electrolyte environment by using the solvation method, the enabling mode and solvent characteristic parameters of the implicit solvent model are configured by script, including the solvent dielectric constant and Debye length, and the solvent characteristic parameters are coordinated with the structure optimization parameters in step S20 to adapt to the simulation requirements of the electrolyte environment and improve the accuracy of the solvation effect simulation and the adaptability to the subsequent screening steps.

[0022] The script configuration of the present scheme avoids the parameter deviation of manual setting, improves the consistency of solvation simulation parameters in the large number of model data sets to be screened, and compared with traditional manual configuration, improves the batch processing efficiency and adapts to the large-scale demand of high-throughput screening. On the other hand, the coordination of solvent characteristic parameters and structure optimization parameters (such as the adaptation of solvent dielectric constant and plane wave cutoff energy, and the coordination of Debye length and van der Waals correction) breaks through the limitations of traditional simulation in which structure optimization and solvation effect are processed separately, and solves the problem of electronic state calculation distortion in high dielectric environment. For example, in the ORR screening of phosphorus-doped carbon-based (P-C) catalyst, ADDGRID=T in step S20 improves the accuracy of the charge density grid, and the script-configured Debye length The simulation of the electrostatic shielding of ions in the electrolyte on the P active site, the coordination of the two makes the adsorption energy calculation error of *O intermediate on the P site from 0.15 eV to 0.04 eV; at the same time, the coordination of solvent dielectric constant and GGA+U correction accurately captures the electron density deviation of carbon skeleton induced by P doping, which is difficult to achieve in traditional separate parameter setting simulation, significantly improving the reliability of subsequent S30 step stability rough screening and S40 step constant potential calculation.

[0023] Further, the calculation framework of DFT+U is configured by a script, which cooperates with the GGA+U correction of step S20 to apply GGA+U correction to atoms, wherein the Hubbard U value of the transition metal element is automatically set according to the element type, the electrode potential adjustment range is set to -1.5 to 1.5 V, and the potential change caused by the change in the number of electrons is a step; the simulation accuracy of the surface charge distribution and the adaptability to subsequent steps are comprehensively improved.

[0024] The calculation framework of DFT+U is configured by a script, which realizes multiple synergies. On the one hand, DFT+U cooperates with the GGA+U correction of step S20 to automatically set the Hubbard U value of the transition metal according to the element type, avoiding errors caused by U value faults in electronic structure calculation, making the description of the d-orbital electron state of the transition metal (such as Mn and Co) coherent, improving the accuracy of surface net charge monitoring, and providing a coherent electronic state basis for subsequent performance analysis. On the other hand, the electrode potential adjustment range is set to -1.5 to 1.5 V, and the potential change caused by the change in the number of electrons is a step; the simulation accuracy of the surface charge distribution and the adaptability to subsequent steps are comprehensively improved. The present scheme links the two together, solving the pain point that the electrolyte concentration needs to be recalibrated in the traditional constant potential simulation.

[0025] Further, when collecting data in the constant potential method solvation simulation process, the data includes basic parameters and surface atom dynamic characteristic parameters; wherein the basic parameters include Relaxation, E0, Fermi, SHIFT, deltaQ and U_SHE, E, the surface atom characteristic parameters include atomic displacement, bond length change, electronegativity and first ionization energy; and the data are correspondingly associated with electrode potential iteration parameters and surface net charge monitoring results to enrich the data dimension and provide multi-dimensional input for subsequent machine learning to build the efficacy relationship model.

[0026] The present scheme collects basic parameters (such as Relaxation, E0, etc.) and surface atom dynamic characteristic parameters (atomic displacement, bond length change, etc.), and correspondingly associates them with electrode potential iteration parameters and surface net charge monitoring results, on the one hand, the rich data dimension breaks through the limitation of traditional dependence on electronic state parameters, enabling the subsequent machine learning model to capture the coupling effect of electronic structure and atomic dynamics, improving the comprehensiveness of the structure-activity relationship model; on the other hand, the data association mechanism unexpectedly unlocks the mining of hidden rules of dynamic structure, charge distribution and catalytic performance decay, which is difficult to achieve when only electronic or structural data is processed in the field.

[0027] Further, the collected data is processed using the SISSO machine learning tool, based on the basic parameters, surface atom dynamic characteristic parameters and associated data, the feature combination order is set to 1-2 order, the key physical descriptors are automatically screened, and the efficacy relationship model is constructed; the machine learning tool cooperates with the associated data, directly calls the associated storage potential, charge and structure data, without additional format conversion.

[0028] The associated data (such as basic parameters, surface atom dynamic characteristic parameters and potential, charge and structure associated data) is processed by the SISSO machine learning tool, the 1-2 order feature combination order is set, the cooperation of the tool and the associated data avoids additional format conversion, improves the calling efficiency of the data, and the 1-2 order feature combination reduces the redundant calculation of high-order features, reduces the memory occupation, and adapts to the resource constraints of high-throughput screening. The automatic screening of key physical descriptors breaks through the limitations of artificial preset features and captures the implicit association of low-order features.

[0029] Further, after constructing the efficacy relationship model, k-fold cross-validation and preset independent test set verification are used for model verification; according to the verification result, the efficacy relationship model is optimized, including adjusting the feature combination order of the SISSO tool, removing redundant features with a weight less than a preset weight threshold, and re-calling the associated storage potential, charge and structure data for iterative training.

[0030] The double verification in the scheme effectively avoids the limitations of single verification, the k-fold cross-validation reduces the data partition bias, and the independent test set verification ensures that the model has generalization ability for new catalysts not involved in training. Compared with traditional single verification, the risk of overfitting is reduced. The linkage of feature order adjustment and redundant feature removal not only reduces invalid calculation, but also has the advantage of low-order feature cross-system stability. The 1-order low-order feature reduces the prediction weight fluctuation amplitude in different catalyst systems, and significantly improves the universality of the efficacy relationship model.

[0031] Further, before constructing the efficacy relationship model, the preliminary screened structure data set is preprocessed, based on the stability index including formation energy and solubility potential, combined with the basic parameters and surface atom dynamic characteristic parameters to set the screening conditions, and remove structures that do not meet the screening conditions and basic performance requirements; the preprocessed data set is used for model verification and optimization, reducing the interference of redundant data on iterative training.

[0032] The stability index (formation energy, dissolution potential), basic parameters and surface atomic dynamic characteristic parameters are used to preprocess the structure data set, remove structures that do not meet the requirements, directly reduce redundant data, shorten the iteration training time of the efficiency relationship model verification and optimization, and solve the problem of waste of computing resources in traditional full-quantity data training. The screened data set is characterized by high stability, and the correlation of its characteristic parameters (such as bond length change) is significantly enhanced, making the efficiency relationship model more accurate in capturing the implicit coupling rules of stability and activity binding, which is difficult to achieve when relying solely on activity parameters.

[0033] Further, when performing constant potential calculation on the preprocessed stable structure data set, the key nodes covering the electrode potential range of-1.5 to 1.5 V are simulated to simulate the electrocatalytic reaction under acidic, neutral and alkaline conditions, and the surface net charge distribution, intermediate adsorption free energy and bond length change data under different potential and working condition combinations are recorded and stored in association with the stability index during preprocessing.

[0034] The combination calculation of multiple potentials and multiple working conditions breaks through the limitations of traditional single condition testing, can fully capture the performance differences of catalysts in different application scenarios (such as fuel cells, water electrolysis, and seawater hydrogen production), and avoids screening bias caused by single working condition. The association storage of data and stability index builds a multi-dimensional data chain of potential, working condition, structure dynamics and stability, enabling the subsequent efficiency relationship model to mine cross-dimensional implicit rules, which is difficult to achieve when only storing performance data or stability data.

[0035] Further, the stable structures after constant potential calculation are screened, and static performance indicators including potential, current density, and dynamic performance indicators including activity decay rate and selectivity retention rate after cycling are used to form comprehensive screening conditions. The surface net charge distribution, intermediate adsorption free energy and bond length change data under different potential and working condition combinations are used to set screening thresholds, the associated stored preprocessing stability index is called, and structures that do not meet the comprehensive screening conditions are removed to form a comprehensive data set that integrates stability, static performance and dynamic performance.

[0036] Through comprehensive screening of static performance indicators (including potential, Gibbs free energy and starting point) and dynamic performance indicators, combined with surface net charge distribution, intermediate adsorption free energy and bond length change data under different potential and working condition combinations, and associated with the preprocessed stability index, the limitations of traditional single indicator screening are broken through. Static indicators ensure that the instantaneous performance of the catalyst under a specific working condition meets the standard (overpotential less than 0.3 V), and the stability is high (formation energy is negative), and the two work together to improve the actual application adaptation rate of the screened catalyst. The association of multiple parameters and stability index reveals the strong correlation between surface zero charge potential and Gibbs free energy, which is a rule that is easily overlooked when only focusing on horizontal charge methods. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 A flowchart of a high-throughput ORR and OER electrocatalyst screening method based on constant potential in an embodiment of the present application.

[0038] Figure 2 A technical flowchart of a high-throughput ORR and OER electrocatalyst screening method based on constant potential in an embodiment of the present application. DETAILED DESCRIPTION

[0039] The concept and technical effects of the present application will be described in detail below with reference to the embodiments, so as to fully understand the purpose, features and effects of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all embodiments. Based on the embodiments of the present application, other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application:

[0040] As shown in Figure 1 The high-throughput ORR and OER electrocatalyst screening method based on constant potential includes the following steps:

[0041] S10: Based on the chemical composition of the target electrocatalytic material, the reference material is derived from the database; based on the reference material, the initial crystal structure model is constructed, and the basic parameters including lattice parameters and atomic coordinates are extracted to obtain the structure data set to be screened;

[0042] S20: The initial crystal structure model corresponding to the structure data set to be screened is preprocessed to obtain the atomic coordinates and pseudo-potential file of the structure to be screened; the atomic coordinates are optimized to obtain the structure to be screened in the lowest energy state; then the electronic self-consistent calculation is performed on the structure to be screened; the solvent method is used to simulate the electrolyte environment according to the structure after the electronic self-consistent calculation, and the solvent effect is included in the screening framework of the structure data set after processing to include the solvent effect, to obtain the surface atomic interaction adsorption energy corrected by the solvent effect;

[0043] S30: Based on the surface atomic interaction adsorption energy, the stability of the catalyst to be screened is measured by using the indicators including formation energy and dissolution potential, and the implicit solvent model is enabled in the measurement process to obtain the stable structure;

[0044] S40: For the stable structure, the constant potential calculation method is used in the implicit solvent model to adjust the electrode potential and monitor the surface net charge change in real time to obtain the simulation results of the dynamic distribution of the surface charge of the catalyst;

[0045] S50: Collecting data generated in the constant potential calculation process, including surface net charge distribution under different electrode potentials, intermediate adsorption free energy change and surface atom dynamic parameters; taking the data as input features, taking overpotential as the catalytic performance index as the target variable, using machine learning tools to screen features of the target variable, and mining key physical descriptors; constructing a performance relationship model of catalytic performance and descriptors based on the key physical descriptors, and predicting the performance of new electrocatalytic materials using the performance relationship model.

[0046] Specifically, as shown in Figure 2 The constant potential-based high-throughput ORR and OER electrocatalyst screening method can further include the following steps:

[0047] S10: Material model construction, including script mkalldir.sh.

[0048] S11: Import reference materials. Based on the chemical composition of the target electrocatalytic material, the reference materials are exported from the Materials project database.

[0049] S12: Modeling (VESTA). Use the material simulation software VESTA to construct an initial crystal structure model, determine the basic parameters such as lattice parameters and atomic coordinates, and obtain the structure data set to be screened.

[0050] S20: Theoretical calculation, including scripts GGA+U.sh, Gr-magmon.sh.

[0051] S21: Structure optimization. The conditions for the structure optimization calculation include: electronic step self-consistent convergence criterion ≤ 1 × 10 -5 eV, Plane wave cutoff energy 520 eV; Gaussian broadening is used with a broadening coefficient of 0.05 eV; spin polarization is turned on; the structure optimization mode is full relaxation; the optimization algorithm uses the conjugate gradient method; the ion step is 0.5 fs, and the van der Waals correction uses the DFT-D3(BJ) damping scheme; ADDGRID=T (T is TRUE, which is a parameter setting in the VESTA software) is enabled to improve the charge density grid; GGA+U correction is applied to the atoms.

[0052] S22: Electronic self-consistency. The conditions for the electronic self-consistency calculation include: electronic step self-consistent convergence criterion ≤ 1 × 10 -6 eV; plane wave cutoff energy 520 eV; Gaussian broadening is used with a broadening coefficient of 0.05 eV; spin polarization is turned on; the structure remains fixed (IBRION = -1); the van der Waals correction uses the DFT-D3(BJ) damping scheme; ADDGRID=T is enabled to improve the charge density grid.

[0053] S23: Solvation simulation. The conditions of solvation calculation include: electronic step self-consistent convergence criterion < 1 x 10 -6 eV; plane wave cutoff energy 520 eV; using tetrahedron smearing (ISMEAR = -5) with smearing coefficient 0.05 eV; spin polarization is on; structure is fixed (IBRION = -1, NSW = 0); van der Waals correction uses DFT-D3 (BJ) damping scheme; implicit solvent model is enabled (LSOL = T) with solvent dielectric constant 80 and Debye length ADDGRID = T is enabled to improve the charge density grid.

[0054] S30: Material screening.

[0055] S31: Stability preliminary screening. The stability of the catalyst to be screened is measured using indicators such as formation energy, solubility potential, etc., to complete the rough screening.

[0056] S40: Theoretical calculation, including script mkconstant.sh.

[0057] S41: Constant potential calculation. For the relatively stable structure obtained by rough screening, in the solvation model, the constant potential calculation method is used, such as DFT+U combined with Nernst equation, to adjust the electrode potential and monitor the surface net charge change in real time, to simulate the dynamic distribution of the surface charge of the catalyst in the process of electrocatalytic reaction. Adjust the surface groups of the screened material, such as -OH, -O, and calculate the total energy of the optimized structure in the solvation model, to screen the stable structure with the lowest energy. For the key intermediates of the target electrocatalytic reaction, such as ORR / OER, under the conditions of constant potential and solvation, the adsorption free energy (ΔG) change of each intermediate on the catalyst surface is calculated, considering the modification of solvation energy on ΔG.

[0058] S50: Result analysis, including scripts getoptout.sh, dist-all.py, getcpresults.sh and getscfOSZICAR.sh.

[0059] S51: Data collection. Collect various data obtained in the constant potential method solvation simulation process, including the net charge distribution on the catalyst surface under different electrode potentials, the adsorption free energy change of the reaction intermediates, and the dynamic change parameters of the surface atoms, such as atomic displacement, bond length change, electronegativity, first ionization energy, etc.

[0060] S52: Feature selection. Using machine learning tools such as SISSO, the collected data is used as input features, and the catalytic performance indicators such as overpotential are used as target variables for feature selection.

[0061] S53: Find descriptors. By machine learning algorithm to mine the potential law in the data, filter out the physical descriptors that have important influence on catalytic performance, such as surface charge density, charge change of specific atoms, adsorption energy after solvation energy correction, etc.

[0062] S54: Build efficacy relationship. Build the efficacy relationship model between catalytic performance and physical descriptors, which is used to predict the performance of new electrocatalytic materials and provide theoretical guidance for catalyst design and optimization.

[0063] More specifically, the script mkalldir.sh includes: traversing all files under the current directory with the extension.xsd, extracting the file name of each file, removing the.xsd extension from the file name, taking the file name as the folder name, creating a corresponding folder for each file, and moving the.xsd file to its folder. Traverse all folders under the current directory, enter each folder in turn, and call the vaspkit tool by piping in the "106" "*.xsd" instruction for processing. After completion, return to the parent directory.

[0064] The script GGA+U.sh includes: reading the 6th line of the POSCAR file to obtain the element arrangement information. Traverse all files starting with "INCAR", delete lines containing "LDAU", "LDAUL", "LDAUU", "LDAUJ", "LDAUTYPE", "LMAXMIX", "#ELEMENTS", and clear the old related configuration. Check if the elements in POSCAR contain transition metals (specified elements such as Sc, Ti, etc.); if yes, add "##GGA+U##" annotation and "LDAU=T" in all "INCAR*" files; if not, add the "#LDAU=T" annotation. According to the element order in POSCAR, combined with the preset associative array (store the LDAUL, LDAUU values of transition metals, and the default values -1 and 0.0 for non-transition metals), generate the parameter lines of LDAUL, LDAUU, LDAUJ (set to 0.0); At the same time, add "LDAUTYPE=2" "LMAXMIX=4" and element order annotation "#ELEMENTS=...", append these contents to all "INCAR*" files, complete the DFT+U parameter configuration.

[0065] The script Gr-magmon.sh includes: setting the MAGNETIC_ELEMENTS array containing transition metal elements as the basis for judging whether the element has a magnetic moment. Extract the element species from the 6th line of the POSCAR file and the corresponding number of each element from the 7th line, and store them in the ELEMENTS and NUMBERS arrays, respectively. Traverse the element array, if the element belongs to the list of magnetic elements, set the magnetic moment value to 4; otherwise, set it to 0, and generate a string in the format of "MAGMOM=number of elements magnetic moment value number of elements magnetic moment value..." in combination with the number of elements. Traverse all files starting with INCAR, first delete the existing MAGMOM line in it; if the file contains the NSW line, insert the generated MAGMOM string in the next line; if it does not contain the NSW line, append the MAGMOM string to the end of the file. After completion, prompt "MAGMOM successfully added to INCAR files.".

[0066] The script mkconstant.sh includes: setting a value list values containing -2, -1.5, -1, …, 2 as input parameters for subsequent calculations. For each value val in values, create a folder named "e${val}" (such as e-2, e0.5, etc.), copy the current directory's POSCAR file into the folder, and enter the folder. Run vaspkit in the folder and input 103 to extract the total valence electron number; by adding the total valence electron number to the current val value, calculate the new NELECT (valence electron number) parameter value. Create 3 INCAR files (INCAR-1, INCAR-2, INCAR-3) to configure different VASP calculation parameters (such as SYSTEM, ENCUT, ISMEAR, etc.), and the NELECT parameter uses the new_nelect value calculated above; at the same time, create 2 KPOINTS files (KPOINTS-1, KPOINTS-2) to define different k-point grids (3x3x1 and 5x5x1). Execute the two external scripts GGA+U.sh and Gr-magmon.sh in the current folder to complete the relevant preprocessing or configuration. Return to the parent directory, and after all numerical processing is complete, traverse the INCAR files in all folders using the grep command to extract and output the NELECT parameter value.

[0067] The script getoptout.sh includes: obtaining the path of the current working directory by the pwd command, and storing it in the current_dir variable. Loop through all the subfolders in the current directory (path "$current_dir" / * / ). For each subfolder, determine whether there is a subfolder named opt (-d "${dir}opt"). If the opt subfolder exists, further define opt_dir as the path of the opt folder, and opt_file as the path of the opt.out file under the opt folder; check if opt.out exists, if it exists, output "=====Source: ${dir}=====", and display the last three lines (tail-n3) of the file, followed by an empty line separation; if opt.out does not exist, output "Warning: ${opt_file} does not exist".

[0068] The script dist-all.py includes: importing numpy (numerical calculation), os (system operation), glob (file search), sys (command line parameter processing) libraries to provide support for subsequent operations. Read the file starting with CONTCAR (compatible with POSCAR format), extract key information, get the scaling factor from the first line of the file, extract the lattice vectors from the second to fourth lines and calculate the lattice matrix combined with the scaling factor; extract the element type from the fifth line, and extract the number of atoms of each element from the sixth line and calculate the total number of atoms; read the fractional coordinates of the atoms from the eighth line, convert to array format and return the lattice matrix and atomic coordinates. Based on the lattice matrix and atomic coordinates, the distance between two specified atoms is calculated. First, convert the fractional coordinates to Cartesian coordinates (by multiplying with the lattice matrix); calculate the Cartesian coordinate vector difference of the two atoms, combined with the periodic boundary condition (adjust the vector to the minimum periodic range); finally, calculate the vector length to get the atomic distance.

[0069] Main program logic: check if the command line argument is 3 (script name + two atom numbers), if not, prompt usage ("python batch_bond_length.py atom number 1 atom number 2") and exit; extract the two input atom numbers (integers); use glob to find all files in the current directory starting with "CONTCAR", sort and store; if no such file is found, prompt information and exit; traverse each CONTCAR file: try to call read_poscar to get the lattice and coordinates, then use calculate_distance to calculate the distance between the two atoms, output the file name and the distance with four decimal places; if an error occurs, output the file name and the error message.

[0070] The script getcpresults.sh includes: creating or emptying the output file cpresult.txt, writing the table header "Folder t Relaxation t E0 t Fermi t SHIFT t deltaQ t U_SHE t E", defining the parameter items to be recorded. Loop processing the specified e-2 / to e2 / total of 9 folders, first remove the slash at the end of the folder name, and use the folder name as the starting identifier of each row record.

[0071] Extract and calculate parameters: check the relax / relax.out file under each folder, if it contains "reached required accuracy", record it as "reached", otherwise record it as "no"; extract the last line E0 value from the solvation / OSZICAR file, keep 4 decimal places, if the file does not exist, record it as N / A; extract the last line fermi value from the solvation / OUTCAR file, if the file does not exist, record it as N / A; extract the last line SHIFT value from the solvation / solvation.out file, keep 4 decimal places, if the file does not exist, record it as N / A; directly extract from the folder name (remove the prefix "e", such as e-2 corresponds to -2); calculate according to the formula "-4.44-Fermi-SHIFT", if Fermi or SHIFT is N / A, the result is nan; calculate according to the formula "E0-Fermi*deltaQ", if E0 or Fermi is N / A, the result is nan.

[0072] Splice each parameter into a row and append to cpresult.txt; use the column command to align the text, generate an aligned table file and replace the original file, and finally output the contents of the file.

[0073] The script getscfOSZICAR.sh includes: get the path of the current working directory through the pwd command, and store it in the current_dir variable. Loop through all the subfolders in the current directory (path "$current_dir" / * / ). For each subfolder, determine whether a subfolder named vacuum exists (-d"${dir}vacuum"). If the vacuum subfolder exists, further define opt_dir as the path of the vacuum folder, and opt_file as the path of the OSZICAR file under the vacuum folder; check if the OSZICAR exists, if it exists, output "====== Source: ${dir} ========" and display the last three lines (tail-n3) of the file, followed by an empty line; if the OSZICAR does not exist, output "Warning: ${opt_file} does not exist".

[0074] In conventional techniques, potentiostatic calculations and solvation simulations are often treated as independent steps. Potentialostatic methods are commonly used for testing the electrochemical performance of single catalysts, while solvation models are primarily used for static correction of adsorption energies. These two methods are rarely dynamically coupled in high-throughput screening. However, this approach deeply integrates potentiostatic control (real-time adjustment of electrode potential and monitoring of net surface charge) with solvation models (implicit solvent dielectric constant, Debye length parameters). This allows for a closed-loop process during batch screening, simultaneously realizing potential changes, dynamic surface charge distribution, real-time correction of solvation energies, and updates to intermediate adsorption free energy. This enables the adsorption energy calculations for key ORR / OER intermediates (such as OOH* and OH*) to dynamically respond to potential changes. This dynamic coupling is not a simple superposition but rather incorporates the interaction between electrochemical potential and the solvent environment into the high-throughput process through a unified computational framework, addressing the limitation of static solvation correction in conventional screening, which cannot match dynamic potential changes.

[0075] Currently, existing high-throughput screening schemes for energy materials, while incorporating machine learning, primarily rely on static structural features (such as lattice parameters and defect types) to construct descriptors. In this scheme, however, the input features for machine learning not only include static structural parameters (atomic coordinates, bond lengths) but also dynamic parameters coupled with constant potential and solvation (such as the net surface charge distribution at different potentials and the rate of change of solvation energy with potential). Machine learning tools (SISSO tools) are used to mine dynamic descriptors such as the correlation between surface charge density and potential response coefficients, and the relationship between solvation-corrected adsorption energy and intermediate bond energy. This machine learning model, which uses dynamic changes in operating conditions as core features, breaks through the conventional mindset of using static structure as the sole input in high-throughput screening. While machine learning is used for material prediction in this field, it is generally difficult to consider using dynamic operating condition parameters such as potential and solvent as the core of the descriptor. This scheme, through the aforementioned method, achieves accurate capture of the nonlinear relationship between structure, operating conditions, and performance. For example, it can directly predict the overpotential difference of a doped catalyst at 1.2V and 1.5V, which is impossible with static descriptors.

[0076] The existing high-throughput screening scheme for energy materials involves the construction of defects and heterojunction structures, but does not verify them in actual working conditions (such as electric potential and solvent). In this scheme, under the constant potential and solvation model, the stable defect / doping structure screened in the preliminary screening is simultaneously tested for its activity (intermediate adsorption energy) and stability (dissolution potential) under different electric potentials and different solvent dielectric constants, and a three-dimensional correlation model of defect concentration, electric potential and activity is established through machine learning. This way of verifying the structure diversity (defects / doping) and the authenticity of the working conditions (electric potential / solvent) through high-throughput calculation and machine learning breaks the fragmented mode of conventional screening, which either studies the effect of structure on performance alone or studies the effect of working conditions on performance alone. Although there are techniques for studying defects or working conditions in this field, it is difficult to think of including both as two-dimensional variables in the same high-throughput framework under conventional conditions. However, the above mechanism can quickly locate the optimal activity of a certain type of defect under a specific electric potential interval, for example, it is found that the ORR activity of nitrogen-doped graphene at 0.8V is much higher than that at 1.0V, which is a conclusion that cannot be efficiently obtained through single-dimensional screening.

[0077] In step S20, the specific calculation parameters and correction methods for structure optimization are as follows: the self-consistent convergence criterion for electronic steps is set to EDIFF = 1 x 10 -5 eV, and the force convergence criterion for ion steps is set to The plane wave cutoff energy is set to ENCUT = 520 eV; Gaussian broadening is used with a broadening coefficient of 0.05 eV; spin polarization is enabled; the structure optimization mode is full relaxation, and the optimization algorithm uses the conjugate gradient method with an ion step size of 0.5 fs; the van der Waals correction uses the DFT-D3 damping scheme; ADDGRID = T is enabled to improve the charge density grid; and GGA+U correction is applied to the atoms, with the Hubbard U value for transition metal elements automatically set according to the element type.

[0078] Specifically, in mkconstant.sh, the variables EDIFF = 1e-5 (corresponding to the electronic step convergence criterion), EDIFFG = -0.02 (ion step force convergence criterion), and ENCUT = 520 (plane wave cutoff energy) are defined, and the sed command is used to batch write the calculation input files (such as the INCAR file accompanying the POSCAR file) of each initial model in the structure dataset to be screened (step S10) to ensure consistent parameter calling. The plane wave cutoff energy of 520 eV is determined based on the energy of the highest valence electron of the target material (such as the 4d orbital energy of Ru, which is about 510 eV) plus a 10 eV safety threshold to avoid deviations in the description of electronic wave functions due to insufficient energy; the Gaussian broadening coefficient of 0.05 eV is automatically calculated by the script associated with the temperature parameter (300K) to match the thermal disturbance conditions of the actual reaction.

[0079] Scripted parameter configuration reduces human input error (error rate from 5% to 0.1%), while supporting batch optimization of structure data sets to be screened (S10 step), reducing the parameter deployment time of 1000 groups of structures from 2 hours to 10 minutes, which is highly adaptable to high-throughput screening requirements (S50 step).

[0080] Element-specific setting of Hubbard U value in GGA+U correction (e.g. Fe takes 4eV, Ni takes 3eV), cooperates with the monitoring of surface net charge in S40 step, accurate electronic state description makes the error of surface charge density from to Conjugate gradient method cooperates with 0.5fs ion step length, when dealing with interlayer sliding of layered catalysts (such as LiNiO2), the convergence speed is greatly improved compared with traditional 1fs step length, and the interlayer distance error is smaller than traditional way, avoiding the calculation deviation of subsequent adsorption energy caused by interlayer misplacement.

[0081] Among them, when simulating the electrolyte environment by using solvation method, the enabling mode and solvent characteristic parameters of implicit solvent model are configured by script, including solvent dielectric constant and Debye length, and the solvent characteristic parameters are cooperated with the structure optimization parameters in step S20 to adapt to the simulation requirements of electrolyte environment, improve the accuracy of solvation effect simulation and the adaptability to subsequent screening steps.

[0082] Specifically, LSOL=T is defined in the script to enable solvation simulation, EB_K=80 (i.e. the dielectric constant of water solution electrolyte), The ion shielding length of 0.1mol / L electrolyte is simulated, and the above parameters are bound with the structure optimization parameters in step S20 through the correlation function built in the script.

[0083] Scripted parameter binding makes the structure optimized in step S20 (such as TiO2 containing oxygen vacancies) in solvation simulation, improves parameter matching, shortens the batch simulation time of structure, and is compatible with the data input format of machine learning in step S50, reduces the data format conversion error. Parameter cooperation can suppress the electronic state drift in high dielectric environment, i.e. the adsorption energy calculation of OOH intermediate in ORR on Cu-doped MnO2 surface, because EB_K cooperates with GGA+U (such as U=4eV for Cu), the temperature-dependent error of adsorption energy is reduced from 0.1eV to 0.03eV, which provides support for the applicability of constant potential calculation in step S40 under variable temperature conditions.

[0084] The calculation framework of DFT+U is configured by a script, which cooperates with the GGA+U correction of step S20, the Hubbard U value of the transition metal element is automatically set according to the element type, the electrode potential adjustment range is set to-1.5 to 1.5V, and each 0.1V is a step iteration; when the surface net charge change is monitored in real time, the Debye length of the implicit solvent model is associated with the ion activity term in the Nernst equation, and when the Debye length changes due to the change of the electrolyte ion concentration, the activity coefficient calculation of the Nernst equation is automatically corrected by the script to improve the simulation accuracy of the dynamic distribution of the surface charge and the adaptability to the subsequent steps.

[0085] Specifically, the script reads the transition metal Hubbard U value (such as 4eV for Fe and 3eV for Ni) from the GGA+U parameter file of step S20 and directly assigns it to the DFT+U calculation module to ensure the consistency of the U value across steps; the potential iteration parameters POTENTIAL_START=0, POTENTIAL_END=1.6, and STEP=0.1 are set to realize the potential adjustment of each 0.1V step through a for loop.

[0086] The scripted U value transfer reduces the calculation deviation of the d-band center of the Fe-Ni composite catalyst from 0.1eV to 0.03eV, which cooperates with the structure optimization parameters (such as ENCUT=520eV) of step S20 to avoid the calculation fault of the electronic state.

[0087] The data collected during the constant potential method solvation simulation includes basic parameters and surface atom dynamic characteristic parameters; the basic parameters include Relaxation, E0, Fermi, SHIFT, deltaQ, and U_SHE, E, and the surface atom dynamic characteristic parameters include atomic displacement, bond length change, electronegativity, and first ionization energy; and the data is associated with the electrode potential iteration parameters and the surface net charge monitoring results to enrich the data dimension and provide multi-dimensional input for the subsequent machine learning to build the efficacy relationship model.

[0088] The script extracts the basic parameters from the constant potential simulation output file (such as OUTCAR, OSZICAR), where Relaxation corresponds to the structure relaxation step number, E0 is the zero-point energy, Fermi is the Fermi level, SHIFT is the energy level displacement, deltaQ is the surface net charge change, U_SHE is the potential relative to the standard hydrogen electrode, and E is the total energy; the surface atom dynamic characteristic parameters are calculated through the atomic coordinate trajectory file (XDATCAR), the atomic displacement takes the root mean square of the coordinate difference before and after optimization, the bond length change tracks the bond length fluctuation value of the active site (such as the transition metal-oxygen bond), the electronegativity adopts the Pauling scale, and the first ionization energy references the standard value of the element.

[0089] The script built-in correlation index mechanism binds each set of data with a unique identifier (including electrode potential value, Debye length, and surface net charge value) to the -1.5 to 1.5 potential iteration step, and the surface net charge monitoring result one by one, to generate a structured data table (CSV format).

[0090] In one specific embodiment of the present scheme, the automation extraction shortens the collection time of 1000 sets of data from 8 hours to 40 minutes, reduces the data omission rate from 12% to 0.5%, and is directly compatible with the input format of the machine learning model, reducing data conversion loss. The time sequence correlation of dynamic parameters (bond length change) and deltaQ can reflect the structural fatigue of the catalyst. When the bond length change / cycle and deltaQ fluctuation > 0.03e, it is predicted that the catalyst activity will decay by 20% after 500 cycles, which is a rule that cannot be captured by traditional electronic state data, providing a new dimension for catalyst durability screening.

[0091] The collected data are processed using a SISSO machine learning tool, based on basic parameters, surface atomic dynamic characteristic parameters, and associated data, with a feature combination order of 1-2 orders, automatic selection of key physical descriptors, and construction of an efficacy relationship model; the machine learning tool cooperates with the associated data to directly call the associated potential, charge, and structure data without additional format conversion.

[0092] The script specifies the input data as an associated structured CSV file, including basic parameters, surface atomic dynamic characteristic parameters, and potential, charge, and structure associated data, sets maxorder = 2 to limit the feature combination order to 1-2 orders, nfeature = 50 to control the maximum number of features, and cv_folds = 5 to enable cross-validation to avoid overfitting. The script is directly compatible with the getcpresults.sh output format, automatically maps data fields (such as deltaQ to surface net charge and bond length change to structure dynamics) through the data_link function, and does not require manual format conversion.

[0093] In one specific embodiment of the present scheme, the script call shortens the feature selection time of 1000 sets of data from 8 hours to 2.5 hours, and is linked with the potential iteration step data to ensure that the features match the 0.1V step potential change, reducing the prediction error of the efficacy relationship model for overpotential. The linear correlation between the 1st order feature electronegativity x U_SHE and the transition metal Hubbard U value improves the prediction accuracy of the efficacy relationship model for Fe, Co, and other transition metal catalysts, without the need for high-order feature calculation, reducing memory usage.

[0094] Wherein, after constructing the efficacy relationship model, k-fold cross-validation and pre-set independent test set verification are used for model verification; according to the verification result, the efficacy relationship model is optimized, including adjusting the feature combination order of the SISSO tool, removing redundant features with a weight less than a pre-set weight threshold, and re-calling the associated stored potential, charge and structure data for iterative training.

[0095] Specifically, the script calls the initial model output by the SISSO tool, sets k = 5-fold cross-validation, randomly divides each fold of data to ensure that each fold contains different potential interval samples, randomly extracts from the associated data, contains catalyst structures not involved in training, and pre-set independent test set; Set the weight threshold to 2%, remove features with a weight < 2% (such as some high-order cross terms) through the feature_prune function; the feature combination order adjustment range is 1-2 order, when the root mean square error (RMSE) fluctuation of cross-validation is > 0.04V, automatically reduce the order from 2 order to 1 order, and retrain the associated stored potential, charge and structure data (CSV format), iterate until RMSE ≤ 0.03V.

[0096] In one specific embodiment of the present scheme, the script is compatible with the output format of sisso_run.sh, and the verification and optimization time of 1000 groups of data is reduced from 10 hours to 3 hours, and is linked with the potential iteration step (0.1V), ensuring that the test set covers the full potential range, and improving the generalization of the efficacy relationship model. The stability of 1st order features is significantly improved in cross-electrolyte systems (such as from 0.1 mol / L HCIO4 to 1 mol / L KOH), reducing the prediction weight fluctuation and further reducing the prediction error of the efficacy relationship model for catalyst performance in different acid-base environments.

[0097] Wherein, before constructing the efficacy relationship model, the preliminary screened structure data set is pre-processed, based on the stability index containing formation energy and dissolution potential, combined with basic parameters and surface atomic dynamic characteristic parameters to set the screening conditions, and remove structures that do not meet the screening conditions and basic performance requirements; the pre-processed data set is used for model verification and optimization, reducing the interference of redundant data on iterative training.

[0098] The script extracts formation energy, dissolution potential and basic parameters (such as Relaxation, E, etc.) and surface atomic dynamic characteristic parameters (such as bond length change, atomic displacement, etc.) from the associated data, and sets the screening threshold, wherein the screening threshold includes formation energy ≤ 1.2 eV / atom, dissolution potential ≥ -0.6 V vs SHE, Relaxation step number ≤ 500 steps (i.e. excluding unstable structures that are difficult to relax), (i.e. exclude samples with dramatic structural fluctuations). The script automatically labels and removes structures beyond the threshold by the data_filter function, and stores the remaining data in JSON format according to stability indicators, base parameters and dynamic parameters, which is directly compatible with the efficacy relationship model verification and optimization process without format conversion.

[0099] In one specific embodiment of the present scheme, automated preprocessing reduces the invalid structure removal time of 1000 groups of data from 3 hours to 40 minutes, and cooperates with the feature removal step to reduce the total time consumption of the efficacy relationship model iterative training. In the screened data set, a strong negative correlation (correlation coefficient of -0.75) is formed with the electronegativity difference, which provides a quantitative basis for regulating the stability of the catalyst through the electronegativity of the element, and improves the key feature weight of the SISSO tool screening, and reduces the prediction error of the efficacy relationship model on stability.

[0100] Among them, when the pre-processed stable structure data set is calculated by constant potential, the key nodes covering the electrode potential range of-1.5 to 1.5V 0-1.6V simulate the electrocatalytic reaction under acidic, neutral and alkaline conditions, and record the surface net charge distribution, intermediate adsorption free energy and bond length change data under different potential and working condition combinations, and store them in association with the stability indicators during preprocessing.

[0101] Specifically, the script calls the pre-processed stable structure data set, sets the key potential nodes of-1.5 to 1.5V to-1.5V, -1V, -0.5V, 0V, 0.5V, 1V, 1.5V, simulates acidic (i.e. pH=0, 0.1mol / L HCIO4), neutral (i.e. pH=7, 0.1mol / L NaCl), and alkaline (i.e. pH=14, 0.1mol / L KOH) conditions, and completes the calculation through the DFT+U framework.

[0102] The surface net charge distribution is obtained from the charge density analysis module of the OUTCAR file, the adsorption free energy of the intermediate (such as *OH, *O) is calculated by combining the energy file with the solvation correction, and the bond length change is calculated by the XDATCAR trajectory file to calculate the average fluctuation value of the metal-oxygen bond at the active site. All data are associated with stability indicators (including formation energy and dissolution potential) through a unique structure identifier (including structure ID, potential value, and pH value), and stored as a SQLite database, with fields including structure ID, potential, pH, net charge, adsorption energy, bond length change, formation energy, and dissolution potential.

[0103] In one specific embodiment of the present scheme, the script is compatible with the output format of preprocess_data.sh, and the enhanced data association storage improves the efficiency of subsequent model calls. The sensitivity of bond length change to potential under alkaline conditions (pH = 14) is several times that of acidity, which can be used as an implicit indicator of catalyst operating adaptability, providing a new dimension for cross-condition screening.

[0104] Among them, the stable structure after constant potential calculation is fine screened, static performance indicators including potential, current density and target product faraday efficiency, and dynamic performance indicators including activity decay rate and selectivity retention rate after cycling are used to form comprehensive screening conditions, combined with surface net charge distribution, intermediate adsorption free energy and bond length change data under different potentials and conditions to set screening thresholds, call associated storage preprocessing stability indicators, eliminate structures that do not meet the comprehensive screening conditions, and form a comprehensive data set integrating stability, static performance and dynamic performance.

[0105] In one specific embodiment of the present scheme, the script calls the associated storage of constant potential calculation data (including surface net charge distribution, intermediate adsorption free energy, bond length change under different potentials and conditions) and preprocessing stability indicators (including formation energy, dissolution potential), and sets a comprehensive screening threshold, i.e., the static performance indicator is overpotential ≤ 0.3 V.

[0106] The script automatically matches data fields (including structure ID associated potential, condition, stability indicator) through the multi_criteria function, batch eliminates structures that do not meet any indicator, and integrates the stability, static performance and dynamic performance data of the remaining structures into a CSV format comprehensive data set, including structure ID, formation energy, dissolution potential, overpotential, decay rate, net charge fluctuation and bond length increment, which is compatible with the efficacy relationship model verification and optimization process.

[0107] The above only describes the embodiments of the present application, and the well-known specific structures and properties in the scheme are not described in detail. It should be noted that for those skilled in the art, without departing from the structure of the present application, some modifications and improvements can be made, which should also be considered as the protection scope of the present application, and these will not affect the effect and practicality of the patent. The protection scope claimed in this application should be subject to the content of its claims, and the specific embodiments and the like in the specification can be used to explain the content of the claims.

Claims

1. A high-throughput ORR and OER electrocatalyst screening method based on constant potential, characterized in that, Includes the following steps: S10: Based on the chemical composition of the target electrocatalytic material, the reference material is derived from the database; An initial crystal structure model is constructed based on the reference material, and basic parameters including lattice parameters and atomic coordinates are extracted to obtain a dataset of structures to be screened. S20: Preprocess the initial crystal structure model corresponding to the dataset of structures to be screened to obtain atomic coordinates containing the structures to be screened and generate pseudopotential files in batches; The atomic coordinates are optimized to obtain the structure to be screened in the lowest energy state; then electronic self-consistent calculation is performed on the structure to be screened; the solvation method is used to simulate the electrolyte environment based on the structure after electronic self-consistent calculation, and the solvent effect is incorporated into the data set of the structure to be screened. After processing, it is incorporated into the solvation effect screening framework to obtain the adsorption energy of surface interatomic interaction corrected by the solvent effect. S30: Based on the adsorption energy of surface interatomic interactions, the stability of the catalyst to be screened is measured by indicators including formation energy and solubility potential. An implicit solvent model is used in the measurement process to obtain a stable structure. S40: For stable structures, a constant potential calculation method is used in the implicit solvent model to adjust the electrode potential and monitor the change of net surface charge in real time to obtain the simulation results of the dynamic distribution of catalyst surface charge. S50: Collect data generated during the constant potential calculation process, including the surface net charge distribution under different electrode potentials and the change in adsorption free energy of intermediates. Using the data as input features and overpotential as the catalytic performance index as the target variable, machine learning tools are used to filter features of the target variable and mine key physical descriptors. Based on the key physical descriptors, a model of the relationship between catalytic performance and the descriptor's efficacy is constructed, and the performance of novel electrocatalytic materials is predicted using the efficacy relationship model.

2. The high-throughput ORR and OER electrocatalyst screening method based on constant potential according to claim 1, characterized in that: The specific calculation parameters and correction methods for structural optimization in step S20 are as follows: the electronic step self-consistent convergence criterion is set as EDIFF = 1 × 10 -5 eV, the ion step convergence criterion is set as The plane wave cutoff energy was set to ENCUT = 520 eV; Gaussian broadening was used with a broadening factor of 0.05 eV; spin polarization was enabled; the structure optimization mode was full relaxation, the optimization algorithm used the conjugate gradient method, and the ion step size was 0.5 fs; van der Waals correction was performed using the DFT-D3 damping scheme; ADDGRID = T was enabled to improve the charge density grid; GGA+U correction was applied to atoms, where the Hubbard U value for transition metal elements was automatically set according to the element type.

3. The high-throughput ORR and OER electrocatalyst screening method based on constant potential according to claim 2, characterized in that: When simulating the electrolyte environment using the solvation method, the activation mode of the implicit solvent model and solvent characteristic parameters, including solvent dielectric constant and Debye length, are configured through a script. The solvent characteristic parameters are coordinated with the structure optimization parameters in step S20 to adapt to the requirements of electrolyte environment simulation, improve the accuracy of solvation effect simulation and adaptability to subsequent screening steps.

4. The high-throughput ORR and OER electrocatalyst screening method based on constant potential according to claim 3, characterized in that: The DFT+U calculation framework is configured by script, where DFT+U works in conjunction with the GGA+U correction in step S20 to automatically set the Hubbard U value of the transition metal, set the electrode potential adjustment range to -1.5 to 1.5V, and use the potential change caused by the change in the number of electrons as the step size; thus comprehensively improving the accuracy of surface charge distribution simulation and its adaptability to subsequent steps.

5. The high-throughput ORR and OER electrocatalyst screening method based on constant potential according to claim 4, characterized in that: When collecting data during the solvation simulation process using the constant potential method, the data includes basic parameters and surface atomic dynamic characteristic parameters. The basic parameters include Relaxation, E0, Fermi, Fermi_SHIFT, deltaQ, U_SHE, and E, while the surface atomic dynamic characteristic parameters include atomic displacement and bond length changes. Furthermore, the data is correlated with the electrode potential iteration parameters and the surface net charge monitoring results to enrich the data dimensions and provide multi-dimensional input for subsequent machine learning to construct an efficacy relationship model.

6. The high-throughput ORR and OER electrocatalyst screening method based on constant potential according to claim 5, characterized in that: The collected data is processed using SISSO machine learning tools. Based on basic parameters, surface atomic dynamic characteristic parameters, and associated data, the feature combination order is set to 1-2, key physical descriptors are automatically selected, and a power relationship model is constructed. The machine learning tools work in conjunction with the associated data to directly call the associated stored potential, charge, and structural data without additional format conversion.

7. The high-throughput ORR and OER electrocatalyst screening method based on constant potential according to claim 6, characterized in that: After constructing the efficacy relationship model, k-fold cross-validation and a preset independent test set were used to validate the model. Based on the validation results, the efficacy relationship model was optimized, including adjusting the feature combination order of the SISSO tool, removing redundant features with weights less than the preset weight threshold, and re-calling the associated stored potential, charge, and structure data for iterative training.

8. The high-throughput ORR and OER electrocatalyst screening method based on constant potential according to claim 7, characterized in that: Before constructing the efficacy relationship model, the preliminary screening of the structural dataset is preprocessed. Based on the stability index including formation energy and dissolution potential, screening conditions are set in combination with basic parameters and surface atomic dynamic characteristic parameters to remove structures that do not meet the screening conditions and basic performance requirements. The preprocessed dataset is used for model validation and optimization, reducing the interference of redundant data on iterative training.

9. The high-throughput ORR and OER electrocatalyst screening method based on constant potential according to claim 8, characterized in that: When performing constant potential calculations on the preprocessed stable structure dataset, key nodes covering the electrode potential range of -1.5 to 1.5V are covered. Electrocatalytic reactions under acidic, neutral, and alkaline conditions are simulated. Data on surface net charge distribution, intermediate adsorption free energy, and bond length changes under different combinations of potentials and conditions are recorded and stored in association with the stability indicators during preprocessing.

10. The high-throughput ORR and OER electrocatalyst screening method based on constant potential according to claim 9, characterized in that: The stable structures calculated under constant potential are finely screened. Static performance indicators including potential, current density, and Faraday efficiency of the target product, as well as dynamic performance indicators including activity decay rate and selectivity retention rate after cycling, are used to form comprehensive screening conditions. Screening thresholds are set by combining surface net charge distribution, intermediate adsorption free energy, and bond length change data under different potentials and operating conditions. The preprocessed stability indicators stored in association are called to eliminate structures that do not meet the comprehensive screening conditions, forming a comprehensive dataset that integrates stability, static performance, and dynamic performance.

Images