Method for outputting stable structure of catalyst under specified reaction conditions based on machine learning

Abstract

The present invention provides a method for outputting a stable structure of a catalyst under specified reaction conditions based on machine learning. This method first constructs a machine learning potential MLP based on known structure-energy data, then searches for catalyst structures that conform to the thermodynamic relationship by using the genetic algorithm of the open-source platform ASE and the MLP. Finally, according to the search results with ΔG≤0, the stable structure of the catalyst that conforms to the thermodynamic relationship is obtained. By using thermodynamics for correction while searching for the most stable structure by the genetic algorithm, the present invention can determine a configurational space of the stable chemical ratio and structure of a catalyst according to the specified reaction conditions. Under two or more reaction conditions, the stable structure of the catalyst under complex working conditions can be obtained through spatial combination.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent chemical catalyst modeling, and relates to a method for outputting a stable structure of a catalyst under specified reaction conditions based on machine learning, which is applied to simulate / model a catalyst under specified reaction conditions. Background Art

[0002] First-principles calculations have become an important tool in the fields of material development, mechanism research, etc. The construction of the model of the research object directly determines the accuracy and credibility of the calculation results. However, the current mainstream construction of material simulation models still relies on artificial experience and semi-empirical modeling, which is seriously affected by human subjectivity. With the development of computational science, machine learning-accelerated genetic algorithms are used for structure prediction and screening, making it possible to predict material structures that are free from human subjectivity, efficient, and accurate.

[0003] In the field of catalytic materials, reaction conditions such as high temperature, high pressure, and multi-component gases often occur. However, in previous computational studies, due to the influence of computational costs, it is necessary to assume an "ideal model" for calculation. For example, metals or their alloy clusters / surfaces are used as independent computational research objects, without overall consideration of the influence of reaction conditions on the catalyst structure. Especially in the process of computational simulation modeling, the influence of reaction conditions on the catalyst model is often ignored. Generally, the models are under absolute zero temperature and vacuum conditions, far from the real experimental conditions, resulting in unreasonable models, incorrect calculation results, and waste of computational resources. Under reaction conditions, the "adsorption compound cluster" formed by the intermediate species (*O, *H, *CH x ) staying on the catalyst surface and the catalyst cluster is closer to the real state of the catalyst cluster during the reaction. Obviously, using such an "adsorption compound cluster" for calculation can better reveal the reaction mechanism. However, how to obtain a reasonable "adsorption compound cluster" model has become the primary problem before starting the calculation, and there are the following difficulties: ① There are many atoms and types, and the computational cost is high, or the existing computing power cannot perform the calculation. When considering the supported clusters on the carrier, the system may contain hundreds or thousands of atoms, or even tens of thousands of atoms, exceeding the system size that can be accurately calculated by current traditional calculations. ② Structure prediction is difficult. It is very difficult to infer a reasonable theoretical model for the equilibrium structure under the reaction conditions of the adsorbed species + cluster. Even if some in-situ techniques can provide details under reaction conditions, the models established therefrom often enter the "meta-stable structure with the lowest local energy" rather than the "stable structure with the global minimum".

[0004] Based on the above situation, if there is a method that can combine machine learning potential accelerated genetic algorithms with the simulation and modeling of catalysts, it is expected to achieve more realistic and accurate catalyst simulation and modeling, providing assistance for the research and development of catalysts. Summary of the Invention

[0005] In view of the problems in the above-mentioned prior art, the present invention provides a method for outputting a stable structure of a catalyst under specified reaction conditions based on machine learning. By using thermodynamics to correct while searching for the most stable structure through a genetic algorithm, a configuration space of the stable chemical ratio and structure of a catalyst can be determined according to the specified reaction conditions. Under two or more reaction conditions, the stable structure of the catalyst under complex working conditions can be obtained through spatial combination.

[0006] To achieve the above object, the present invention is implemented by a technical solution composed of the following technical measures.

[0007] The present invention provides a method for outputting a stable structure of a catalyst under specified reaction conditions based on machine learning, mainly including the following steps:

[0008] SⅠ. Based on structure-energy data, construct a machine learning potential MLP (Machine Learning Potentials, MLPs):

[0009] Ⅰ-1. Based on the known compound structures, use the first-principles calculated energy data corresponding thereto as labels to form a training set;

[0010] Ⅰ-2. Input the training set into the open-source software package AML (Active Machine Learning), and train it based on the open-source n2p2 high-dimensional neural network potential and the Committee Neural Network Potential (CNNP) to obtain the machine learning potential MLP;

[0011] Among them, the types of elements in the neural network committee need to be set according to the known compound structures included in the training set, and the required element bond combinations need to be selected;

[0012] SⅡ. Based on the genetic algorithm of the open-source platform ASE and use MLP to search for the catalyst structure that conforms to the thermodynamic relationship:

[0013] II-1. Specify the constituent elements and reaction conditions of the original catalyst, and predict the new elements that may be introduced into the original catalyst;

[0014] Ⅱ-2. Based on the constituent elements of the original catalyst, use the open-source platform ASE (Atomic Simulation Environment) to randomly generate multiple initial structures as the initial population of the genetic algorithm;

[0015] II-3. In the open-source platform ASE, use the machine learning potential MLP obtained in Step I as a calculator to perform a genetic algorithm search on the initial population, and perform thermodynamic correction data calibration calculations based on each search result to obtain ΔG;

[0016] Among them, the thermodynamic correction data calibration calculation to obtain ΔG is based on the known thermodynamic parameters and thermodynamic relationship formulas preset according to the reaction conditions, and uses the calculation results of the machine learning potential MLP to perform thermodynamic correction data calibration calculation to obtain ΔG;

[0017] II-4. If ΔG ≤ 0 corresponding to the search result in Step II-3, add the new elements predicted in Step II-1 to the initial structure of the initial population, and repeat Step II-3 in ascending order of the number of atoms of the added new elements until ΔG > 0;

[0018] III. According to the search results with ΔG ≤ 0 obtained in Step II, the stable structure of the catalyst that conforms to the thermodynamic relationship is obtained.

[0019] In this article, the known compound structures and their corresponding first-principles calculation energy data can be based on the content publicly disclosed in this technical field, such as the public database in this technical field, or the content summarized by those skilled in the art themselves.

[0020] In this article, the open-source software package AML (Active Machine Learning) is applied to automatically and data-drivenly construct a reference set for the training of neural network potentials. This open-source software package AML is sourced from the network open-source platform github.

[0021] In this article, the open-source n2p2 high-dimensional neural network potential is an off-the-shelf software for high-dimensional neural network potentials applied in computational physics and chemistry. This open-source n2p2 high-dimensional neural network potential is sourced from the network open-source platform github.

[0022] In one of the technical solutions, to obtain a better machine learning potential MLP, the number of the training set is preferably 300 - 350.

[0023] In this article, the types of elements in the neural network committee need to be set according to the known compound structures included in the training set, and the required element bonding combinations are selected. Among them, the selection of the required element bonding combinations is to select and combine the element bonding between the constituent elements of the specified original catalyst and the new elements that may be introduced into the original catalyst according to user needs.

[0024] In one of the technical solutions, to obtain a better machine learning potential MLP, the number of members in the neural network committee needs to be set, preferably set to 4 to 10.

[0025] In this article, the new elements that may be introduced into the original catalyst include predicting the elements of the intermediate species staying on the catalyst surface and the catalyst cluster elements.

[0026] In one of the technical solutions, to improve the search efficiency, the number of randomly generated initial structures should be controlled within 20 to 50.

[0027] In one of the technical solutions, to improve the search efficiency and effect, the genetic operations, mutation operations, and mutation rate in the genetic algorithm need to be set. Preferably, the genetic operation is set as cut and splice pairing, the mutation operations include mirroring, swapping, and random movement, and the mutation rate is set to 0.1 to 0.3.

[0028] It should be noted that the known thermodynamic parameters can be based on the content publicly disclosed in this technical field, such as the public database in this technical field, or the content summarized by those skilled in the art on their own.

[0029] To automate the correction calculation of the thermodynamic correction data, in one of the preferred technical solutions, it can be achieved by adding a parameter retrieval module in the open-source platform ASE to automatically obtain the known thermodynamic parameters from the network public database.

[0030] To better illustrate the present invention and provide a reference technical solution, the parameter retrieval module uses Python code. When the reaction conditions specified are an oxygen atmosphere and the corresponding gas thermodynamic parameters need to be obtained, based on the network public database NIST Webbook, the specific code is as follows:

[0031]

[0032]

[0033] To automate the correction calculation of the thermodynamic correction data, in one of the preferred technical solutions, it can be achieved by adding a thermodynamic correction data correction calculation module in the open-source platform ASE. This thermodynamic correction data correction calculation module is a conventional coding process based on the thermodynamic relationship formula under the specified reaction conditions, transforming the thermodynamic relationship formula into a thermodynamic relationship expression. For example, when the reaction conditions specified are an oxygen atmosphere and the corresponding gas thermodynamic relationship expression needs to be obtained, it is to transform the corresponding thermodynamic relationship formula ΔG O2 = ΔH O2 - TΔSO2 It is transformed into 0.5O2 + cat_Ox = cat_Ox + 1.

[0034] To better illustrate the present invention and provide a reference technical solution, the thermodynamic correction data calibration calculation module uses Python code. When the reaction condition is specified as an oxygen atmosphere, based on its gas thermodynamic parameters and the gas thermodynamic relation expression 0.5O2 + cat_Ox = cat_Ox + 1, the specific code is as follows:

[0035]

[0036]

[0037]

[0038]

[0039] The present invention has the following beneficial effects:

[0040] 1. The present invention provides a method for outputting the stable structure of a catalyst under specified reaction conditions based on machine learning. Using the machine learning potential MLP obtained by training as a tool for structure - energy calculation, it replaces the DFT (Density Functional Theory) used in the prior art and is used for fast and accurate calculation of large systems with multiple elements and multiple atoms; using a genetic algorithm to quickly find the globally most stable structure and combining machine learning to accelerate the genetic algorithm, significantly reducing the calculation cost.

[0041] 2. The present invention provides a method for outputting the stable structure of a catalyst under specified reaction conditions based on machine learning, which can automatically search for the stable structure of the catalyst under reaction conditions, thus solving the problem that the previous modeling ignored reaction conditions, making the subsequent modeling more reasonable and improving the scientific nature of the calculation.

[0042] 3. In one of the technical solutions, on the basis of the machine learning - accelerated genetic algorithm for searching the most stable structure of the catalytic material, the present invention further develops a parameter retrieval module for known thermodynamic parameters and a thermodynamic correction data calibration calculation module, enabling the entire search process to run automatically and reducing the labor cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic flowchart of a method for outputting the stable structure of a catalyst under specified reaction conditions based on machine learning in Embodiment 1 of the present invention.

[0044] Figure 2This is a line graph showing the relationship between the Pt4Sn4 clusters with different oxygen element doping contents in the stable structure of the catalyst output under specified reaction conditions based on machine learning in Embodiment 1 of the present invention and the ΔG obtained by calibration calculation.

[0045] Figure 3 This is the stable structure diagram of the catalyst finally obtained in Embodiment 1 of the present invention. Detailed implementation manners

[0046] To further understand the present invention, the preferred implementation manners of the present invention will be described below in conjunction with embodiments. However, it should be understood that these descriptions are only for further illustrating the features and advantages of the present invention, rather than limiting the claims of the invention. Those skilled in the art can draw on the content of this article and appropriately modify the parameters to implement. It should be particularly noted that all similar substitutions and modifications are obvious to those skilled in the art, and they are all regarded as included in the present invention. The methods and applications of the present invention have been described through preferred embodiments, and those skilled in the art can obviously make changes or appropriate modifications and combinations to the methods and applications described herein without departing from the content, spirit and scope of the present invention to implement and apply the technology of the present invention.

[0047] Embodiment 1

[0048] This embodiment demonstrates a method for outputting the stable structure of a catalyst based on machine learning for the Pt4Sn4 cluster under an 850K, 0.2atm O2 atmosphere, mainly including the following steps:

[0049] SⅠ. Based on the structure-energy data, construct a machine learning potential MLP (Machine Learning Potentials, MLPs):

[0050] Ⅰ-1. Based on Pt 12 Sn 12 , Pt9Sn9O6 and their known structures conforming to the molecular dynamics trajectory, use the first-principles calculation energy data corresponding to them as labels to form a training set, and the number of the training set is 300;

[0051] Ⅰ-2. Input the training set into the open-source software package AML (Active Machine Learning), and train it based on the open-source n2p2 high-dimensional neural network potential and the Committee Neural Network Potential (CNNP) to obtain the machine learning potential MLP;

[0052] Among them, the types of elements in the neural network committee need to be set according to the known compound structures included in the training set (Pt, Sn, O), and the required element bonding combinations need to be selected; the number of members in the neural network committee is set to 8;

[0053] SⅡ. Search for catalyst structures that conform to thermodynamic relationships using the genetic algorithm based on the open-source platform ASE and MLP:

[0054] II-1. Specify the Pt4Sn4 cluster and reaction conditions (under an O2 atmosphere of 850K and 0.2 atm), and predict the new element (O element) that may be introduced into the original catalyst;

[0055] Ⅱ-2. Based on the Pt4Sn4 cluster, randomly generate 20 initial structures as the initial population of the genetic algorithm using the open-source platform ASE (Atomic Simulation Environment);

[0056] Ⅱ-3. In the open-source platform ASE, use the machine learning potential MLP obtained in step I as a calculator, perform a genetic algorithm search on the initial population, and perform thermodynamic correction data calibration calculations based on each search result to obtain ΔG;

[0057] Among them, the genetic operations, mutation operations, and mutation rate in the genetic algorithm need to be set. It is preferably set that the genetic operation is cut and splice pairing, the mutation operations include mirroring, swapping, and random movement, and the mutation rate is set to 0.3;

[0058] The thermodynamic correction data calibration calculation to obtain ΔG is based on the known thermodynamic parameters and thermodynamic relationship formulas preset according to the reaction conditions, and uses the calculation results of the machine learning potential MLP for thermodynamic correction data calibration calculation to obtain ΔG;

[0059] Ⅱ-4. If ΔG ≤ 0 corresponding to the search result in step Ⅱ-3, add the O element predicted in step II-1 to the initial structure of the initial population, and repeat step Ⅱ-3 in ascending order of the number of O atoms added. When the number of O elements is 3, ΔG > 0, and the search stops;

[0060] Ⅲ. According to the search results with ΔG ≤ 0 obtained in step Ⅱ, the stable catalyst structures Pt4Sn4O and Pt4Sn4O2 that conform to the thermodynamic relationship are obtained.

[0061] In step II-3, the acquisition of the known thermodynamic parameters is achieved by adding a parameter retrieval module for automatically retrieving the known thermodynamic parameters from the publicly available online database in the open-source platform ASE. The parameter retrieval module uses Python code and is based on the publicly available online database NIST Webbook. The specific code is as follows:

[0062]

[0063] In step II-3, the correction calculation of the thermodynamic correction data is based on the known thermodynamic parameters obtained above and is achieved by adding a thermodynamic correction data correction calculation module in the open-source platform ASE. The thermodynamic correction data correction calculation module uses Python code and is based on the above-known thermodynamic parameters and the gas thermodynamic relation expression 0.5O2 + cat_Ox = cat_Ox+1. The specific code is as follows:

[0064]

[0065]

[0066]

[0067] An embodiment of the present invention also provides an electronic device, which can perform a method for outputting a stable structure of a catalyst under specified reaction conditions based on machine learning.

[0068] An embodiment of the present invention also provides a computer-readable storage medium, which is used to store program codes for executing the method for outputting a stable structure of a catalyst under specified reaction conditions based on machine learning in the foregoing embodiment.

[0069] An embodiment of the present invention also discloses a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to enable the computer device to execute Figure 1 the method shown.

[0070] When the above-mentioned method is implemented in the form of software functional units and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0071] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a predefined sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0072] More specific examples (non-exhaustive list) of computer-readable media include the following: electrical connection parts (electronic devices) having one or more wirings, portable computer disk cartridges (magnetic devices), random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memories), optical fiber devices, and portable compact disc read-only memories (CDROM). Additionally, a computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other appropriate processing when necessary, and then stored in a computer memory.

[0073] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following techniques known in the art or a combination thereof can be used: discrete logic circuits having logic gate circuits for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0074] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. A method for outputting a catalyst stable structure under specified reaction conditions based on machine learning, characterized in that The main steps include: SⅠ. Constructing machine learning potential MLP based on structure-energy data: Ⅰ-1. Based on the known compound structure, the corresponding first-principles calculated energy data is used as a label to form a training set; Ⅰ-2. Input the training set into the open source software package AML, and train it based on the open source n2p2 high-dimensional neural network potential and neural network committee to obtain the machine learning potential MLP; Among them, the element types in the neural network committee need to be set according to the known compound structures contained in the training set, and the required element bonding combination is selected; SⅡ, based on the genetic algorithm of the open source platform ASE and using MLP to search for catalyst structures that meet thermodynamic relationships: II-1. Given the constituent elements and reaction conditions of the original catalyst, predict the new elements that may be introduced into the original catalyst; Ⅱ-2. Based on the constituent elements of the original catalyst, the open source platform ASE is used to randomly generate multiple initial structures as the initial population of the genetic algorithm; II-3. Using the machine learning potential MLP obtained in step I as a calculator in the open source platform ASE, a genetic algorithm search is performed on the initial population, and a thermodynamic correction data correction calculation is performed based on each search result to obtain ΔG; Wherein, the thermodynamic correction data correction calculation to obtain ΔG is based on the known thermodynamic parameters preset by the reaction conditions and the thermodynamic relationship formula, and the thermodynamic correction data correction calculation to obtain ΔG is performed using the machine learning potential MLP calculation results; The thermodynamic correction data correction calculation is based on the known thermodynamic parameters obtained, and is achieved by adding a thermodynamic correction data correction calculation module to the open source platform ASE. The thermodynamic correction data correction calculation module is a conventional coding processing based on the thermodynamic relationship formula under the specified reaction conditions, which converts the thermodynamic relationship formula into a thermodynamic relationship expression; Ⅱ-4. If ΔG corresponding to the search result in step Ⅱ-3 is ≤ 0, then add the new element predicted in step II-1 to the initial structure of the initial population, and repeat step Ⅱ-3 in order of the number of atoms of the added new element from small to large until ΔG>0; III. Based on the search result of ΔG≤0 obtained in step II, the stable structure of the catalyst that conforms to the thermodynamic relationship is obtained.

2. The method according to claim 1, characterized in that: The number of the training sets is preferably 300-350.

3. The method according to claim 1, characterized in that: The number of members in the neural network committee needs to be set to 4 to 10.

4. The method according to claim 1, characterized in that: The number of randomly generated initial structures is 20 to 50.

5. The method according to claim 1, characterized in that: The genetic operation, mutation operation and mutation rate in the genetic algorithm need to be set. The genetic operation is set to splicing genetics, the mutation operations include mirroring, exchange, and random movement, and the mutation rate is set to 0.1-0.

3.

6. The method according to claim 1, characterized in that: A parameter retrieval module is added to the open source platform ASE to automatically obtain known thermodynamic parameters from the public database on the Internet.

7. The method for outputting a catalyst stable structure under specified reaction conditions based on machine learning as claimed in claim 1 is applied to the field of chemical catalyst modeling technology.

Images