Model training method and device for simulating water decomposition reaction

Abstract

The invention discloses a model training method and device for simulating a water decomposition reaction. The method comprises the following steps: acquiring structural parameters and experimental characterization data of a sample material; constructing a surface model corresponding to the sample material based on the structure parameters and the experimental characterization data; based on a dynamic evolution process of an AIMD simulation surface model in a water decomposition reaction environment, obtaining first target continuous change data, and determining the first target continuous change data as an initial training set; determining second target continuous change data from the initial training set by adopting a time step sampling strategy; inputting the second target continuous change data into a neural network for model training to obtain an inter-atomic interaction potential function model; the interatomic interaction potential function model is used for simulating a target material, the calculation speed can be greatly increased on the premise of guaranteeing certain precision, and the method is suitable for simulation of a long time scale and a large system.

Description

Technical Field

[0001] The present application relates to the field of molecular simulation computational technology in materials science, and in particular to a model training method and device for simulating water decomposition reactions. Background Art

[0002] In actual application environments where metal materials come into contact with water (such as electrochemical reactions, corrosion protection, energy conversion systems, etc.), the structure and properties of the metal / water interface have a decisive influence on the performance of the system. Currently, there are a variety of theoretical models used to explain the behavior of the metal / water interface, especially the experimental results obtained under ultra-high vacuum conditions, which can be reasonably explained by the existing theoretical framework. However, actual application scenarios often involve normal pressure, impurity-containing, and liquid water environments, which are significantly different from idealized ultra-high vacuum conditions. As a result, the interface structure description obtained under idealized conditions is limited in applicability under real working conditions.

[0003] Among related technologies, AIMD (Ab Initio Molecular Dynamics) is used to simulate metal / water interfaces. This method explicitly accounts for the presence of liquid water molecules and their complex interactions with the metal surface at the atomic scale, thereby more realistically reflecting actual interfacial behavior. Although AIMD offers high accuracy in revealing microscopic mechanisms, it is computationally expensive and limited in both time and space, making it difficult to study interfacial behavior over long periods of time and across large systems.

[0004] Furthermore, to further expand the spatial and temporal scales of simulations, the team introduced ML (machine learning) methods to construct efficient interatomic potential function models for metal / water and metal oxide / water interfaces. The ML models trained on AIMD data not only retain high physicochemical fidelity but also significantly improve simulation efficiency, enabling the study of interfacial dynamics in larger systems and over longer timescales.

[0005] However, taking the interaction between platinum and water as an example, the exchange process between the bulk water and the interface region takes about 500 picoseconds, a time scale far beyond the capabilities of the AIMD method, which can usually only perform simulations at the level of tens of picoseconds. Since the AIMD simulation time is not long enough for the system to reach thermodynamic equilibrium, the results of liquid water microstructure and its related electrochemical properties obtained based on this short time scale simulation may not be accurate or representative enough. In other words, the existing AIMD simulation cannot fully capture the complete dynamic exchange process of water molecules between the interface and the bulk phase, resulting in limitations in the description of the interface structure and the corresponding electrochemical properties, making it difficult to fully reflect the actual situation under actual working conditions. Summary of the Invention

[0006] In view of this, the present application provides a model training method and device for simulating water decomposition reactions, which are used to solve the problem of low simulation efficiency in the prior art.

[0007] The purpose of this application can be achieved through the following technical solutions: The first aspect of the present application is to provide a model training method for simulating a water splitting reaction, comprising: Obtain structural parameters and experimental characterization data of sample materials; Construct a surface model corresponding to the sample material based on structural parameters and experimental characterization data; Based on AIMD simulation of the dynamic evolution process of the surface model in the water decomposition reaction environment, the first target continuously changing data is obtained and the first target continuously changing data is determined as the initial training set; A time step sampling strategy is used to determine the second target continuously changing data from the initial training set; Based on the continuous change data of the second target, the data is input into the neural network for model training to obtain the potential function model of the interaction between atoms; ML-MD simulations of target materials are performed based on the interatomic interaction potential function model.

[0008] In an optional embodiment, the method further includes: Determine the minimum energy path of the reaction system of the target material; Determine transition state structures and reaction energy barriers based on minimum energy paths; Reveal atomic-scale reaction mechanisms based on transition state structures and reaction energy barriers.

[0009] In an optional embodiment, the method further includes: Couple the interatomic interaction potential function model with the pre-set coarse-grained model to build a cross-scale computational framework; Based on the cross-scale computing framework, the reactant mass transfer, reactant surface adsorption and reactant-product desorption processes corresponding to the target material are simulated to obtain the kinetic parameters of the target material.

[0010] In an optional embodiment, the method further includes: The computational accuracy and generalization ability of the interatomic interaction potential function model are verified based on multi-dimensional indicators of the interatomic interaction potential function model.

[0011] In an optional embodiment, the method further includes: Determine the element doping ratio, surface active site distribution, and stability optimization parameters during ML-MD simulations; The catalyst was prepared based on the element doping ratio, surface active site distribution and stability optimization parameters.

[0012] In an optional embodiment, the method further includes: The catalyst was subjected to microstructural characterization.

[0013] In an optional embodiment, the method further includes: The electrochemical performance of the catalyst was tested.

[0014] The second aspect of the present application is to provide a model training device for simulating a water decomposition reaction, comprising: An acquisition module, used to obtain structural parameters and experimental characterization data of sample materials; A construction module for constructing a surface model corresponding to the sample material based on structural parameters and experimental characterization data; A first simulation module is used to simulate the dynamic evolution process of the surface model in a water decomposition reaction environment based on AIMD to obtain first target continuously changing data, and determine the first target continuously changing data as an initial training set; A determination module, configured to determine second target continuously changing data from an initial training set using a time step sampling strategy; A training module is used to input the continuously changing data of the second target into the neural network for model training to obtain an interatomic interaction potential function model; The second simulation module is used to perform ML-MD simulation on the target material based on the interatomic interaction potential function model.

[0015] The third aspect of the present application is to provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the method of the first aspect when executing the computer program.

[0016] The fourth aspect of the present application is to provide a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method of the first aspect.

[0017] Compared with the prior art, the model training method for simulating water decomposition reaction provided in the present application obtains the structural parameters and experimental characterization data of the sample material; constructs a surface model corresponding to the sample material based on the structural parameters and experimental characterization data; simulates the dynamic evolution process of the surface model in the water decomposition reaction environment based on AIMD to obtain first target continuously changing data, and determines the first target continuously changing data as the initial training set; adopts a time step sampling strategy to determine the second target continuously changing data from the initial training set; inputs the second target continuously changing data into the neural network for model training to obtain an interatomic interaction potential function model; uses the interatomic interaction potential function model to simulate the target material, which can greatly improve the calculation speed while ensuring a certain accuracy, and is suitable for simulation of long time scales and large systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 A schematic flow chart of a model training method for simulating a water splitting reaction provided in an embodiment of the present application; Figure 2 A structural block diagram of a model training device for simulating water decomposition reaction provided in an embodiment of the present application; Figure 3 A structural block diagram of an electronic device for implementing a model training method for simulating a water decomposition reaction provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0021] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0022] It should be understood that in the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "Including A, B and / or C" means including any one, any two, or any three of A, B, and C.

[0023] It should be understood that in the embodiments of the present application, "B corresponding to A," "B corresponding to A," "A corresponds to B," or "B corresponds to A" means that B is associated with A and B can be determined based on A. Determining B based on A does not mean determining B based solely on A; B can also be determined based on A and / or other information.

[0024] To solve the technical problems existing in the related art, the embodiments of the present application provide a model training method and device for simulating water decomposition reaction.

[0025] The model training method for simulating the water decomposition reaction provided in the embodiments of the present application can be executed by an electronic device, which can be a terminal or a server. The terminal can be a terminal device such as a smart phone, a tablet computer, or a laptop computer. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. It is understandable that the present application does not specifically limit the execution entity of the model training method for simulating the water decomposition reaction.

[0026] Before describing in detail the model training method for simulating the water splitting reaction provided in the embodiments of the present application, the terms involved in the present application are first explained: High-entropy oxide (HEO) refers to an oxide with a configurational entropy greater than or equal to 1.5R. The specific calculation formula is as follows:

[0027] Where S is the mixing configuration entropy of the high entropy oxide, N is the number of element types, x i is the mole fraction of the i-th component, and R is the universal gas constant.

[0028] LAMMPS (Large-scale Atomic / Molecular Massively Parallel Simulator): A classic molecular dynamics program focused on materials modeling, LAMMPS offers potential functions for simulating solid-state materials (metals and semiconductors), soft matter (biomolecules and polymers), and coarse-grained or mesoscopic systems. It can be used for atomic-scale modeling but also serves more generally as a platform for parallel particle simulations at the atomic, mesoscopic, or continuum scales.

[0029] AIMD: It is a molecular dynamics simulation method based on the basic principles of quantum mechanics. Its core feature is to directly solve the interaction forces between atoms through first-principles calculations (that is, without relying on empirical parameters), thereby simulating the motion trajectory of atoms or molecules.

[0030] NEB (Nudged Elastic Band) method: In the fields of computational chemistry and materials science, it refers to a numerical simulation algorithm used to find transition states in chemical / physical processes (such as reaction paths and phase change paths). Its core idea is to "nudging" path points on the potential energy surface through an elastic band model to converge to the minimum energy path.

[0031] The technical solution of the present application is described in detail below through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments described below are used to explain the technical solution of the present application and are not intended to be used as limitations for actual use.

[0032] In order to solve the technical problems existing in the related art, the embodiment of the present application provides a model training method for simulating water decomposition reaction, such as Figure 1 As shown, Figure 1This is a flow chart of an example of a model training method for simulating a water splitting reaction provided in an embodiment of the present application. It should be noted that the steps shown can be performed in a different logical order than that shown in the flow chart of the method. The method may include the following steps S101 to S106.

[0033] Step S101: Acquire structural parameters and experimental characterization data of sample materials.

[0034] In an optional embodiment, the material capable of promoting the water splitting reaction includes at least one of the following materials: a metal, a metal oxide, a carbon-based material, a sulfide, a phosphide, a nitride, and a perovskite-type oxide. For example, the sample material is a high-entropy oxide. It should be noted that in addition to the exemplary materials given above, other materials may also be used, and this application is not limited to this.

[0035] In an optional embodiment, the structural parameters of the material mainly refer to those parameters that can define the arrangement of atoms or molecules inside the material and its macroscopic morphology. The structural parameters include at least one of the following parameters: lattice parameters, atomic positions, space groups, defect information, thermal vibration parameters and microstructural characteristics. In addition, other structural parameters may also be included, which is not limited in this application.

[0036] Experimental characterization data is data obtained through various experimental means, which is used to verify the structural parameters of the material and further explore the performance of the material. Experimental characterization data includes at least one of the following parameters: X-ray diffraction: used to determine the crystal structure and lattice constant. Transmission electron microscopy: can directly observe the atomic arrangement at the nanoscale, which is especially useful for surface and interface research. Scanning tunneling microscopy: can observe the surface morphology at the atomic level, and can obtain information on the local electron state density. X-ray photoelectron spectroscopy: used to analyze the surface element composition and chemical state. Raman spectroscopy: can be used to study molecular vibration modes and indirectly reflect the bonding characteristics in the crystal structure. Of course, the experimental characterization data can also be other experimental characterization data, which is not limited in this application.

[0037] Step S102: constructing a surface model corresponding to the sample material based on the structural parameters and experimental characterization data.

[0038] It should be noted that if the sample material is a metal oxide, the surface model may be a metal oxide surface model.

[0039] In an optional embodiment, constructing a surface model corresponding to the sample material based on the structural parameters and experimental characterization data specifically includes the following steps: The main exposed crystal plane of the sample material is determined based on experimental characterization data and structural parameters; the selected unit cell is cut along a preset direction to generate a surface model corresponding to the sample material.

[0040] In this step, a surface model corresponding to the sample material is constructed based on the structural parameters and experimental characterization data to ensure that the surface model is consistent with the actual crystal structure and surface exposed crystal features.

[0041] Step S103: Based on the AIMD simulation, the dynamic evolution process of the surface model in the water decomposition reaction environment is obtained to obtain first target continuously changing data, and the first target continuously changing data is determined as the initial training set.

[0042] In an optional embodiment, the water decomposition reaction includes an electrocatalytic oxygen evolution reaction and an electrocatalytic hydrogen evolution reaction.

[0043] In an optional embodiment, the dynamic evolution process of the surface model under the water decomposition reaction environment is simulated based on AIMD to obtain first target continuously changing data, which specifically includes the following steps: Place the constructed surface model into the AIMD model, ensuring that the surface model is under the correct temperature and pressure conditions; set simulation conditions such as simulation time step and total simulation time; execute the simulation, the program will update the atomic position at each time step, and obtain the energy and force field of the system by solving the Schrödinger equation.

[0044] In an optional embodiment, the first target continuous change data includes at least one of the following data: atomic type change data, atomic coordinate change data, system energy change data, and atomic force field change data.

[0045] Step S104: Determine the second target continuously changing data from the initial training set using a time step sampling strategy.

[0046] In an optional embodiment, the second target continuous change data includes: the first target continuous change data including at least one of an energy mutation point, a bonding reconstruction point, and a reactant and product transformation point.

[0047] In an optional embodiment, a time step sampling strategy is used to determine the second target continuously changing data from the initial training set, specifically including the following steps: extracting frame configurations at fixed intervals; dynamically adjusting the sampling frequency according to the energy or structure change rate; performing energy mutation point detection, bonding reconstruction point detection, and / or reactant / product conversion point identification detection to obtain the second target continuously changing data.

[0048] In this step, not only can the information content of the dataset be improved, but also the burden of redundant data on model training can be reduced.

[0049] Step S105: inputting the second target continuously changing data into the neural network for model training to obtain an interatomic interaction potential function model.

[0050] In an optional embodiment, the neural network can be a graph neural network or a neural network potential. In addition, according to actual needs, the neural network can also be other neural networks, which is not limited in this application.

[0051] In a specific embodiment, the multi-dimensional indicator includes at least one of the following indicators: energy prediction error, force field convergence, and reaction energy barrier reproducibility.

[0052] In another optional embodiment, the model training method for simulating water decomposition reaction provided in the embodiment of the present application also includes: verifying the computational accuracy and generalization ability of the interatomic interaction potential function model based on multi-dimensional indicators of the interatomic interaction potential function model.

[0053] Step S106: performing ML-MD (Machine Learning Molecular Dynamics) simulation on the target material based on the interatomic interaction potential function model.

[0054] In an embodiment of the present application, the structural parameters and experimental characterization data of the sample material are obtained; a surface model corresponding to the sample material is constructed based on the structural parameters and experimental characterization data; the dynamic evolution process of the surface model in the water decomposition reaction environment is simulated based on AIMD to obtain first target continuously changing data, and the first target continuously changing data is determined as the initial training set; the time step sampling strategy is used to determine the second target continuously changing data from the initial training set; the second target continuously changing data is input into the neural network for model training to obtain the interatomic interaction potential function model. Using the interatomic interaction potential function model to simulate the target material can greatly improve the calculation speed while ensuring a certain accuracy, and is suitable for simulations of long time scales and large systems. It can also track the dynamic evolution behavior of the surface atoms of the catalytic material in the electrochemical reaction environment at the nanoscale time scale.

[0055] In another optional embodiment, the model training method for simulating the water splitting reaction provided in the embodiment of the present application further includes: Determine the minimum energy path of the reaction system of the target material; determine the transition state structure and reaction energy barrier based on the minimum energy path; reveal the atomic-scale reaction mechanism based on the transition state structure and reaction energy barrier.

[0056] It should be noted that the minimum energy path refers to the path with the lowest energy from reactants to products, which usually includes one or more transition states. The transition state structure refers to the point with the highest energy in the reaction process, which determines the rate of the reaction. The reaction energy barrier refers to the energy difference from the reactants to the transition state, which reflects the difficulty of the reaction. Revealing the atomic-scale reaction mechanism means that by analyzing these paths and structures, we can deeply understand the specific reaction steps occurring on the surface of the catalytic material, including microscopic processes such as adsorption, dissociation, and recombination. In a specific embodiment, the minimum energy path of the reaction system of the target material is determined by the NEB method. The NEB method is used to find the optimal path from reactants to products to ensure that the energy changes in each step are accurately captured.

[0057] In a specific embodiment, determining the transition state structure and reaction energy barrier based on the minimum energy path specifically includes the following steps: The transition state structure is identified based on the minimum energy path, and the reaction energy barrier is calculated based on the transition state structure.

[0058] Identifying the highest energy point in a reaction pathway, known as the transition state, is crucial for understanding the reaction mechanism. Calculating the reaction energy barrier quantifies the energy barrier for a reaction to occur, helping to assess the likelihood and rate of a reaction.

[0059] In another optional embodiment, the model training method for simulating the water splitting reaction provided in the embodiment of the present application further includes: Couple the interatomic interaction potential function model with the pre-set coarse-grained model to build a cross-scale computational framework; Based on the cross-scale computing framework, the reactant mass transfer, reactant surface adsorption and reactant-product desorption processes corresponding to the target material are simulated to obtain the kinetic parameters of the target material.

[0060] In a specific embodiment, the kinetic parameter comprises at least one of the following parameters: reaction rate constant and conversion frequency.

[0061] In this embodiment, the kinetic parameters and electrochemical data such as the polarization curve and Tafel slope obtained by experimental measurement are combined to cross-validate the catalytic activity and stability of the target material.

[0062] In another optional embodiment, the model training method for simulating the water splitting reaction provided in the embodiment of the present application further includes: Determine the element doping ratio, surface active site distribution and stability optimization parameters during the ML-MD simulation process; prepare the catalyst based on the element doping ratio, surface active site distribution and stability optimization parameters.

[0063] In a specific embodiment, the catalyst is prepared by high-temperature solid-phase synthesis and sol-gel methods based on element doping ratio, surface active site distribution and stability optimization parameters.

[0064] In another optional embodiment, the model training method for simulating the water splitting reaction provided in the embodiment of the present application further includes: The catalyst was subjected to microstructural characterization.

[0065] In a specific embodiment, the catalyst is subjected to microstructural characterization, which specifically includes the following steps: analyzing the surface morphology and particle distribution of the catalyst by scanning electron microscopy, determining the crystal phase composition and the degree of lattice distortion by X-ray diffraction, and analyzing the atomic-scale structural characteristics by combining transmission electron microscopy and selected area electron diffraction.

[0066] In another optional embodiment, the model training method for simulating the water splitting reaction provided in the embodiment of the present application further includes: The electrochemical performance of the catalyst was tested.

[0067] In a specific embodiment, the electrochemical performance test of the catalyst includes the following steps: using a three-electrode system to obtain the oxygen evolution reaction polarization curve by linear sweep voltammetry on an electrochemical workstation, and analyzing the interfacial reaction kinetics by electrochemical impedance spectroscopy; evaluating the material stability by constant current cycling method, and testing the device-level performance parameters in an anion exchange membrane or proton exchange membrane electrolyzer.

[0068] Among them, performance parameters include voltage efficiency, gas production purity and life attenuation rate.

[0069] Traditional DFT (first principles, Density Functional Theory) methods require thousands of core hours for a single surface catalytic reaction simulation, and the system size is typically limited to 200 atoms. The method provided in this application can reduce single-step calculation time from seconds to milliseconds while maintaining DFT-level accuracy. Furthermore, the method provided in the examples of this application supports nanosecond molecular dynamics (ML-MD) simulations of 10,000-atom supercells, revealing the dynamics of interfacial water molecule migration.

[0070] In the model training method for simulating the water decomposition reaction provided in the embodiment of the present application, in order to solve the problem of configuration space explosion caused by the heterogeneity of the local chemical environment of multi-atoms, a quantitative characterization system based on smooth overlapping atomic position descriptors or atomic center symmetry functions is constructed; by analyzing the bond angle distribution, coordination number variation coefficient and valence charge transfer amount of each element within a local radius of 1-5 Å, the coordination environment differences of the surface active sites are accurately encoded, so that the discrimination between different chemical environments is ≥95%.

[0071] Machine learning molecular dynamics (ML-MD) is used to replace traditional empirical force fields to achieve nanosecond-level simulation of surface evolution processes. By real-time tracking the kinetic correlation between oxygen vacancy formation energy and surface partial order parameter, the influence mechanism of non-equilibrium behaviors such as phase segregation on catalytic activity is quantified, and the simulation accuracy reaches the quantum mechanical level (force field error ≤ 0.1 eV / Å) (electron volts).

[0072] In another specific embodiment, in order to address the defects of the traditional micro-motion elastic band method that requires a preset reaction path and a single calculation takes ≥72 hours (taking a 100-atom system as an example), an autonomous path-finding algorithm based on machine learning potential function coupled with reinforcement learning is developed; a Markov decision process is established, with the atomic force gradient as the reward function, and a Q-learning strategy is used to dynamically explore multiple candidate paths on the potential energy surface, automatically screening the optimal reaction channel with an energy barrier ≤0.8 eV.

[0073] In another optional embodiment, the problem that classical molecular dynamics cannot accurately describe the double-layer structure of the electrode-electrolyte interface is solved, and a pre-trained neural network potential is used to analyze the interfacial water molecule orientation distribution function and hydrogen bond lifetime in real time; the dynamic hydrogen bond configuration (bond length 1.5-2.0 Å, bond angle 155°-175°) between the reaction intermediates and water molecules is captured through a graph convolutional network, and the effect of the hydrogen bond network stability on the adsorption free energy is quantitatively output.

[0074] In another optional embodiment, to address the defect that the traditional trial-and-error method requires the synthesis of ≥200 material samples (R&D cycle ≥6 months), a D-optimal experimental design is used to construct a sample screening model; based on the confidence interval of material properties predicted by Gaussian process regression, key component points on the Pareto front are preferentially selected, thereby reducing the demand for experimental samples and shortening the R&D cycle.

[0075] In order to solve the systematic error of >0.5 eV between DFT prediction and measured performance, an in-situ characterization data fusion channel was established, which provided real-time access to the local electronic structure data of synchrotron radiation X-ray absorption spectrum and the lattice strain field distribution of four-dimensional scanning transmission electron microscopy. The machine learning potential function was dynamically corrected through a graph neural network weighted by an attention mechanism to reduce the root mean square error between the theoretical adsorption energy and the experimental value.

[0076] Based on D-optimal experimental design or uncertainty sampling strategy, an iterative data generation loop is constructed. The importance of configurations is evaluated through the prediction error of the machine learning model, and samples that contribute most to the improvement of potential function accuracy are prioritized for DFT calculations, achieving targeted allocation of computing resources and reducing redundant calculations by ≥70%. By integrating electronic-scale DFT energy / force field data, atomic-scale molecular dynamics trajectory data, and experimental-scale X-ray photoelectron spectroscopy, extended X-ray absorption fine structure and other experimental characterization data, a unified encoding feature space is constructed. Through transfer learning and multi-task training mechanisms, the generalization ability of machine learning models for complex chemical environments is improved, so that the average absolute error between the predicted results and the experimental values ​​is ≤ 0.03 eV / atom.

[0077] Through D-optimal design or Bayesian optimization algorithms, we intelligently screen key samples from high-dimensional combinatorial spaces, reducing DFT computational complexity by over 70%. We also employ negative data mining strategies to build a sample database and enhance model generalization capabilities.

[0078] By combining electronic structure data calculated by density functional theory, atomic trajectory data from molecular dynamics simulations, and microstructural information obtained from experimental characterization (including but not limited to element valence state distribution from X-ray photoelectron spectroscopy and local coordination environment from extended X-ray absorption fine structure), a radial basis function-based interpolation algorithm or Gaussian process regression model is used to discretely sample ≥1000 feature points in the reaction coordinate space (such as the two-dimensional bond length-bond angle plane). Through energy gradient optimization and transition state verification, a continuously differentiable three-dimensional potential energy surface is generated, where the potential energy value error is ≤0.05 eV and the force field root mean square error is ≤0.1 eV / Å, which is used to accurately describe the energy evolution process in the catalytic reaction path.

[0079] Metadynamics and reinforcement learning are integrated to accelerate the simulation of oxygen vacancy migration and interfacial water molecular dynamics, extending the timescale to the nanosecond level. A hybrid quantum-classical simulation is employed, with an interatomic interaction potential function model for the interface reaction zone and a coarse-grained model for the bulk phase, enabling simulation of catalytic processes at the 10,000-atom level.

[0080] Pre-trained neural network potentials or graph neural network potentials are used to replace traditional quantum mechanics calculations, improving the speed of single-point energy / force field calculations while maintaining the accuracy of electronic structure. After initializing the linear interpolation path between reactants and products, the intermediate image position is iteratively optimized using the conjugate gradient method, where the atomic force calculation is performed by the interatomic interaction potential function model, and the path convergence condition is set to an average atomic force threshold of ≤0.05 eV / Å. The converged path is transition state verified, and ≥5 candidate paths are searched in parallel. Finally, the path with the lowest reaction energy barrier is selected as the optimal catalytic reaction channel, which has a lower energy barrier than the path discovered by the traditional density functional theory (DFT)-NEB method.

[0081] Corresponding to the model training method for simulating water decomposition reaction provided in the embodiment of the present application, the embodiment of the present application also provides a model training device for simulating water decomposition reaction, such as Figure 2 Shown, including: An acquisition module 201 is used to obtain structural parameters and experimental characterization data of a sample material; A construction module 202 is used to construct a surface model corresponding to the sample material based on the structural parameters and experimental characterization data; A first simulation module 203 is configured to simulate the dynamic evolution of the surface model in a water decomposition reaction environment based on AIMD to obtain first target continuously changing data, and determine the first target continuously changing data as an initial training set; A determination module 204 is configured to determine second target continuously changing data from the initial training set using a time step sampling strategy; The training module 205 is used to input the continuously changing data of the second target into the neural network for model training to obtain the interatomic interaction potential function model.

[0082] The second simulation module 206 is used to perform ML-MD simulation on the target material based on the interatomic interaction potential function model.

[0083] In an optional embodiment, the method further includes: Determine the minimum energy path of the reaction system of the target material; Determine transition state structures and reaction energy barriers based on minimum energy paths; Reveal atomic-scale reaction mechanisms based on transition state structures and reaction energy barriers.

[0084] In an optional embodiment, the method further includes: Couple the interatomic interaction potential function model with the pre-set coarse-grained model to build a cross-scale computational framework; Based on the cross-scale computing framework, the reactant mass transfer, reactant surface adsorption and reactant-product desorption processes corresponding to the target material are simulated to obtain the kinetic parameters of the target material.

[0085] In an optional embodiment, the method further includes: The computational accuracy and generalization ability of the interatomic interaction potential function model are verified based on multi-dimensional indicators of the interatomic interaction potential function model.

[0086] In an optional embodiment, the method further includes: Determine the element doping ratio, surface active site distribution, and stability optimization parameters during ML-MD simulations; The catalyst was prepared based on the element doping ratio, surface active site distribution and stability optimization parameters.

[0087] In an optional embodiment, the method further includes: The catalyst was subjected to microstructural characterization.

[0088] In an optional embodiment, the method further includes: The electrochemical performance of the catalyst was tested.

[0089] Corresponding to the model training method for simulating water decomposition reaction provided in the embodiment of the present application, the embodiment of the present application also provides an electronic device for executing the model training method for simulating water decomposition reaction, such as Figure 3 As shown, the electronic device includes: a processor 301; and a memory 302, which is used to store a program for a model training method for simulating a water decomposition reaction. After the device is powered on and the program for the model training method for simulating a water decomposition reaction is run by the processor, the following steps are performed: Obtain structural parameters and experimental characterization data of sample materials; Construct a surface model corresponding to the sample material based on structural parameters and experimental characterization data; Based on AIMD simulation of the dynamic evolution process of the surface model in the water decomposition reaction environment, the first target continuously changing data is obtained and the first target continuously changing data is determined as the initial training set; A time step sampling strategy is used to determine the second target continuously changing data from the initial training set; Based on the continuous change data of the second target, the data is input into the neural network for model training to obtain the potential function model of the interaction between atoms; ML-MD simulations of target materials are performed based on the interatomic interaction potential function model.

[0090] Corresponding to the model training method for simulating the water splitting reaction provided in the embodiment of the present application, the embodiment of the present application also provides a computer-readable storage medium storing a program for the model training method for simulating the water splitting reaction, which is executed by a processor to perform the following steps: Obtain structural parameters and experimental characterization data of sample materials; Construct a surface model corresponding to the sample material based on structural parameters and experimental characterization data; Based on AIMD simulation of the dynamic evolution process of the surface model in the water decomposition reaction environment, the first target continuously changing data is obtained and the first target continuously changing data is determined as the initial training set; A time step sampling strategy is used to determine the second target continuously changing data from the initial training set; Based on the continuous change data of the second target, the data is input into the neural network for model training to obtain the potential function model of the interaction between atoms; ML-MD simulations of target materials are performed based on the interatomic interaction potential function model.

[0091] Corresponding to the model training method for simulating the water splitting reaction provided in the embodiment of the present application, the embodiment of the present application also provides a computer program comprising instructions. When the program is executed by a computer, the instructions cause the computer to perform the following steps: Obtain structural parameters and experimental characterization data of sample materials; Construct a surface model corresponding to the sample material based on structural parameters and experimental characterization data; Based on AIMD simulation of the dynamic evolution process of the surface model in the water decomposition reaction environment, the first target continuously changing data is obtained and the first target continuously changing data is determined as the initial training set; A time step sampling strategy is used to determine the second target continuously changing data from the initial training set; Based on the continuous change data of the second target, the data is input into the neural network for model training to obtain the potential function model of the interaction between atoms; ML-MD simulations of target materials are performed based on the interatomic interaction potential function model.

[0092] It should be noted that for a detailed description of the model training device, electronic device, computer-readable storage medium, and computer program for simulating water decomposition reactions provided in the embodiments of the present application, reference can be made to the relevant description of the model training method embodiment for simulating water decomposition reactions provided in the embodiments of the present application, and no further details will be given here.

[0093] Although the present application is disclosed as above with the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

[0094] In a typical configuration, an electronic device includes one or more processors (Central Processing Unit), input / output interfaces, network interfaces, and memory.

[0095] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory. Memory is an example of a computer-readable medium.

[0096] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can be implemented using any method or technology to store information. The information can be computer-readable operations, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, programmable analog modules (PAMs), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RANM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPRM), flash memory or other memory technologies, compact disc read-only memory (CDROM), digital versatile disc (DVCD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include non-transitory media such as modulated data signals and carrier waves.

[0097] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, compact disc read-only memory, optical storage, etc.) containing computer-usable program code.

[0098] Although the present application is disclosed as above with the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A model training method for simulating water splitting reaction, characterized in that, include: Obtain structural parameters and experimental characterization data of sample materials; constructing a surface model corresponding to the sample material based on the structural parameters and experimental characterization data; Simulating the dynamic evolution process of the surface model in a water decomposition reaction environment based on AIMD to obtain first target continuously changing data, and determining the first target continuously changing data as an initial training set; Determining second target continuously changing data from the initial training set using a time step sampling strategy; Based on the continuous change data of the second target, the data is input into the neural network for model training to obtain the potential function model of the interaction between atoms; The target material is simulated by ML-MD based on the interatomic interaction potential function model.

2. The model training method for simulating water decomposition reaction according to claim 1, characterized in that Also includes: Determining the minimum energy path of the reaction system of the target material; determining a transition state structure and a reaction energy barrier based on the minimum energy path; The atomic-scale reaction mechanism is revealed based on the transition state structure and the reaction energy barrier.

3. The model training method for simulating water decomposition reaction according to claim 1, characterized in that Also includes: Coupling the interatomic interaction potential function model with a preset coarse-grained model to construct a cross-scale computing framework; Based on the cross-scale computing framework, the reactant mass transfer, reactant surface adsorption and reactant-product desorption processes corresponding to the target material are simulated to obtain the kinetic parameters of the target material.

4. The model training method for simulating water splitting reaction according to claim 1, characterized in that Also includes: The calculation accuracy and generalization ability of the interatomic interaction potential function model are verified based on the multi-dimensional indicators of the interatomic interaction potential function model.

5. The model training method for simulating water splitting reaction according to claim 1, characterized in that Also includes: Determine the element doping ratio, surface active site distribution, and stability optimization parameters during ML-MD simulations; A catalyst is prepared based on the element doping ratio, surface active site distribution and stability optimization parameters.

6. The model training method for simulating water splitting reaction according to claim 5, characterized in that, Also includes: The catalyst was subjected to microstructural characterization.

7. The model training method for simulating water splitting reaction according to claim 5, characterized in that Also includes: The electrochemical performance of the catalyst was tested.

8. A model training device for simulating water decomposition reaction, characterized in that: include: An acquisition module, used to obtain structural parameters and experimental characterization data of sample materials; A construction module, configured to construct a surface model corresponding to the sample material based on the structural parameters and experimental characterization data; a first simulation module, configured to simulate the dynamic evolution of the surface model in a water decomposition reaction environment based on AIMD to obtain first target continuously changing data, and determine the first target continuously changing data as an initial training set; a determination module, configured to determine second target continuously changing data from the initial training set using a time step sampling strategy; A training module, configured to input the continuously changing data of the second target into a neural network for model training to obtain an interatomic interaction potential function model; The second simulation module is used to perform ML-MD simulation on the target material based on the interatomic interaction potential function model.

9. An electronic device, characterized in that: The invention comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the model training method for simulating the water decomposition reaction according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the model training method for simulating a water decomposition reaction according to any one of claims 1 to 7 is implemented.

Images