Method, apparatus, electronic device and medium for accelerating material structure search

Abstract

The present invention discloses a method for accelerating the search of material structures, including: searching for potential energy surface structures through a global potential energy surface search method; calculating the potential energy and atomic forces corresponding to the extracted potential energy surface structures by using first-principles calculations and collecting them as an initial training data set; training a machine learning potential energy model representing the potential energy surface of the system through a machine learning model; combining the global potential energy surface search method to perform screening sampling on the potential energy surface of the system, updating the training data set and the machine learning potential energy model; performing high-throughput search on the target system structure through the global potential energy surface search method, predicting the structure potential energy and performing structure screening; performing first-principles calculations on the screened structures to obtain the corresponding potential energy and atomic forces, using the potential energy and atomic forces as a test data set to test the prediction performance of the machine learning potential energy model, and calculating the formation energy of the structure through the potential energy. The present invention also discloses a device, an electronic device and a medium for accelerating the search of material structures.

Description

Technical Field

[0001] The present invention relates to the technical fields of computational chemistry and physics. More specifically, the present invention particularly relates to a method, device, electronic device, and medium for accelerating the search for material structures. Background Art

[0002] Semiconductor alloy materials are a very important type of functional materials. Such materials are composite semiconductor materials obtained by alloying two or more single semiconductor materials. The alloying method can make various properties of semiconductor alloy materials, such as band structures, band gaps, and the movement of electrons and holes, continuously change among the corresponding constituent materials, thereby having greater flexibility and tunability. By adjusting the composition ratios of different types of semiconductors, semiconductor alloy materials can achieve purposeful property regulation, and thus have broad application prospects in the fields of optoelectronic devices, solar cells, sensors, etc. In the development process of semiconductor alloy materials, structure search is a very important part, which is one of the key factors affecting the regulation of their performance. However, semiconductor alloy materials have a huge search space in research, and their complex structures and compositions make structure search very difficult. In existing semiconductor alloy material structure search methods, most use heuristic algorithms or high-throughput calculation methods based on first principles, and there are certain limitations in terms of calculation efficiency and search space. Therefore, it is very important to develop an accurate and fast semiconductor alloy material structure search method.

[0003] The method of machine learning potential (mlp), as a productive combination of machine learning and computational simulation, is becoming a new tool in the field of large-scale material simulation to accelerate material discovery. Based on a known high-precision electronic structure calculation data set, the computational complexity is reduced by characterizing the local structure of atoms, and a single mapping of the local atomic structure to atomic energy and atomic force is constructed to achieve the fitting of the first-principles potential energy surface. The fitting of the first-principles potential energy surface can accelerate the calculation of material potential energy, relate the potential energy of materials to chemical compositions, reveal the potential physical properties of materials, and provide important guidance for the design and discovery of materials. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, device, electronic device, and medium for accelerating the search for material structures to overcome the defects existing in the prior art.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] A method for accelerating the search for material structures, comprising:

[0007] Searching for the potential energy surface structure by a global potential energy surface search method;

[0008] Extract some structures from the searched potential energy surface structures, calculate the potential energy and atomic forces corresponding to the extracted potential energy surface structures using first-principles calculations, and collect them as the initial training dataset;

[0009] Based on the initial training dataset, a machine learning potential energy model characterizing the potential energy surface of the system is obtained through training with a machine learning model;

[0010] Based on the machine learning potential energy model and combined with the global potential energy surface search method, perform screening sampling on the potential energy surface of the system to update the training dataset and the machine learning potential energy model;

[0011] Based on the updated training dataset and machine learning potential energy model, perform high-throughput search on the target system structure through the global potential energy surface search method to predict the structure potential energy and perform structure screening;

[0012] Perform first-principles calculations on the screened structures to obtain the corresponding potential energy and atomic forces, use the potential energy and atomic forces as the test dataset to test the prediction performance of the machine learning potential energy model, and calculate the formation energy of the structure through the potential energy.

[0013] Furthermore, after calculating the formation energy of the structure according to the potential energy, draw the formation energy - configuration image.

[0014] Furthermore, the steps of performing screening sampling on the potential energy surface of the system based on the machine learning potential energy model and combined with the global potential energy surface search method to update the training dataset and the machine learning potential energy model specifically include:

[0015] Extract structure data based on the initial training dataset;

[0016] Perform global potential energy surface search simulation based on the extracted structure data using the machine learning potential energy model to obtain a simulation trajectory;

[0017] Screen some structures on the simulation trajectory for first-principles calculations to obtain the potential energy and atomic forces corresponding to the part of the structures, and add the potential energy and atomic forces to the training dataset to update the training dataset;

[0018] Based on the updated training dataset, after training with the machine learning potential energy model, obtain an updated machine learning potential energy model.

[0019] Furthermore, screening some structures on the simulation trajectory specifically means: performing clustering screening on the simulation trajectory to obtain some structures.

[0020] Further, the step of screening and sampling the potential energy surface of the system based on the machine learning potential energy model and in combination with the global potential energy surface search method to update the training data set and the machine learning potential energy model is iterated multiple times.

[0021] Further, the potential energy type in the machine learning potential energy model includes deep potential energy, and the global potential energy surface search method includes the random surface walking method.

[0022] Further, based on the updated training data set and the machine learning potential energy model, high-throughput search of the target system structure is carried out through the global potential energy surface search method to predict the structural potential energy and perform structural screening, which specifically includes:

[0023] Based on the global potential energy surface structure search of the machine learning potential energy model, the global potential energy surface structure is obtained;

[0024] Based on the global potential energy surface structure and the corresponding machine learning potential energy prediction potential, screening is carried out with a given energy threshold;

[0025] Based on the structures obtained by screening with the energy threshold, clustering screening is carried out.

[0026] The present invention also provides a device for accelerating the search of material structures, including:

[0027] A potential energy surface structure search module, configured to search for the potential energy surface structure through the global potential energy surface search method;

[0028] An initial training data set acquisition module, configured to extract some structures from the searched potential energy surface structures, calculate the potential energy and atomic forces corresponding to the extracted potential energy surface structures by using the first principle and collect them as the initial training data set;

[0029] A machine learning potential energy model training module, configured to train a machine learning potential energy model representing the potential energy surface of the system through a machine learning model based on the initial training data set;

[0030] A training data set update and structure clustering screening module, configured to perform screening and sampling on the potential energy surface of the system based on the machine learning potential energy model and in combination with the global potential energy surface search method to update the training data set and the machine learning potential energy model;

[0031] A structure energy screening module, configured to perform high-throughput search of the target system structure through the global potential energy surface search method based on the updated training data set and the machine learning potential energy model to predict the structural potential energy and perform structural screening;

[0032] A structure formation energy acquisition module is used to perform first-principles calculations on the screened structures to obtain the corresponding potential energy and atomic forces, use the potential energy and atomic forces as a test data set to test the prediction performance of the machine learning potential energy model, and calculate the formation energy of the structure through the potential energy.

[0033] The present invention also provides an electronic device, including:

[0034] A memory that stores execution instructions;

[0035] A processor that executes the execution instructions stored in the memory, so that the processor executes the method for accelerating the search for material structures described above.

[0036] The present invention also provides a readable storage medium, in which execution instructions are stored, and when the execution instructions are executed by a processor, they are used to implement the method for accelerating the search for material structures described above.

[0037] Compared with the prior art, the advantages of the present invention are as follows: Compared with the traditional method of constructing a machine learning potential energy model, the present invention combines iterative update and screening sampling to assist in the construction of the potential energy model, reduces the structural redundancy of the training data set of the machine learning potential energy model, and reduces the consumption of computing resources; Compared with the high-throughput calculation based on first principles, the time scale can be increased by more than 1000 times. At the same time, the training data set is labeled by first-principles calculation, ensuring the accuracy of numerical calculation, and ensuring the accuracy of the potential energy surface while increasing the time scale; Compared with the high-throughput calculation based on first principles, the number of structures can also be increased by more than 1000 times. At the same time, the use of the global potential energy surface search method expands the scope of structure search and accelerates the high-throughput structure screening. Description of the Drawings

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0039] Figure 1 is a flowchart of the method for accelerating the search for material structures of the present invention.

[0040] Figure 2 is a schematic diagram of the method flow for screening and sampling the system potential energy surface in an embodiment of the present invention to update the training data set and the machine learning potential energy model.

[0041] Figure 3Schematic diagram of the method for combining machine learning potential energy and global potential energy surface search to screen and sample, update the dataset and machine learning potential energy, and accelerate structure search according to an embodiment of the present invention.

[0042] Figure 4 is an embodiment of the present invention using In x Ga 1-x N training dataset update and screening sampling result schematic diagram.

[0043] Figure 5 is an embodiment of the present invention using In x Ga 1-x N structure search combined with machine learning potential energy model error schematic diagram.

[0044] Figure 6 Device for accelerating material structure search based on machine learning potential energy and global potential energy surface according to an embodiment of the present invention. Specific embodiments

[0045] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0046] Figure 1 This embodiment discloses a method S100 for accelerating material structure search, including the following steps:

[0047] Step S102: Search for the potential energy surface structure by the global potential energy surface search method.

[0048] Step S104: Extract part of the structures from the potential energy surface structures searched in step S102, calculate the potential energy and atomic forces corresponding to the extracted potential energy surface structures using first-principles calculations, and collect them as the initial training dataset; through steps S102 and S104, the initial structures are obtained by the global potential energy surface search method, and the method for calculating the interatomic interaction forces includes the first-principles calculation method.

[0049] Step S106: Based on the initial training dataset obtained in step S104, train a machine learning potential energy model representing the potential energy surface of the system through a machine learning model to represent the potential energy surface E (x).

[0050] Step S108: Based on the machine learning potential energy model and combined with the global potential energy surface search method, perform screening sampling on the potential energy surface of the system to update the training dataset and the machine learning potential energy model; Figure 2 Shows a specific implementation of step S108, such as Figure 2As shown, step S108 includes the following steps:

[0051] Step S1082: Extract structural data based on the initial training dataset.

[0052] Step S1084: Perform a global potential energy surface search simulation based on the extracted structural data using a machine learning potential energy model to obtain a simulation trajectory. Applicable machine learning potential energy types include, but are not limited to, DeepPotential (DP), and applicable global potential energy surface search methods include, but are not limited to, the Stochastic Surface Walking (SSW) method.

[0053] Step S1086: Screen some structures on the simulation trajectory for first-principles calculations to obtain the potential energy and atomic forces corresponding to the some structures, and add the potential energy and atomic forces to the training dataset to update the training dataset; The method of iteratively updating the dataset includes randomly selecting simulated structures from the training dataset or obtaining them through an active learning method. The screening sampling method includes obtaining based on the deviation between machine learning potential energy models and using a structure clustering method.

[0054] Step S1088: Based on the updated training dataset, after training through a machine learning potential energy model, obtain an updated machine learning potential energy model.

[0055] It should be noted that step S108 can be repeatedly executed through this specific implementation manner to obtain an updated training dataset and an updated machine learning potential energy surface.

[0056] Step S110: Based on the updated training dataset and the machine learning potential energy model, perform a high-throughput search for the target system structure through a global potential energy surface search method to predict the structural potential energy and perform structure screening, specifically including:

[0057] Perform a global potential energy surface structure search based on the machine learning potential energy model to obtain a global potential energy surface structure;

[0058] Perform screening with a given energy threshold based on the global potential energy surface structure and the corresponding machine learning potential energy prediction potential;

[0059] Perform clustering screening on the structures obtained by energy threshold screening.

[0060] Step S112: Perform first-principles calculations on the screened structures to obtain the corresponding potential energy and atomic forces, use the potential energy and atomic forces as a test dataset to test the prediction performance of the machine learning potential energy model, and calculate the formation energy of the structure through the potential energy to draw a formation energy - configuration image.

[0061] Figure 3Schematic diagram of a method for combining machine learning potential energy and global potential energy surface search to perform screening sampling, update the dataset and machine learning potential energy, and accelerate structure search. First, the potential energy surface structure is obtained through the global potential energy surface search method, and some structures in the potential energy surface structure are calculated using first-principles calculations and marked as the initial dataset. On this basis, a machine learning potential energy model for the target system is obtained through model training. Screening sampling of the structure is performed by combining the machine learning potential energy model and the global potential energy surface search method to update the dataset and the machine learning potential energy model. After obtaining a relatively good dataset, a machine learning potential energy model finally used to predict the material energy is constructed. The final potential energy model is used for structure search and screening, and the formation energy is obtained through the structure energy, which can be used to draw the formation energy - configuration image.

[0062] Figure 4 is a schematic diagram of the results of updating and screening sampling of the In x Ga 1-x N training dataset according to an embodiment of the present invention. In this embodiment, screening sampling of the system potential energy surface is performed by combining the machine learning potential energy model and the global potential energy surface search method to update the training dataset and the machine learning potential energy model. The peak shape of the distribution of the atomic force errors between the machine learning potential energy models gradually narrows with the increase in the number of iterations, and the mean error converges rapidly. As Figure 4 shown. (a) Distribution diagram of atomic force errors between machine learning potential energy models in different iteration processes. (b) Evolution of the mean atomic force error between machine learning potential energy models and the distribution ratio of each set of structures with the number of iterations. (c) Two-dimensional potential energy surface contour diagram of the In x Ga 1-x N training dataset.

[0063] Figure 5 is a schematic diagram of the structure search and machine learning potential energy model error of In x Ga 1-x N according to an embodiment of the present invention. In this embodiment, the first-principles calculation based on the PBE functional is used to describe the potential energy surface. The root mean square error of the atomic energy of the machine learning potential energy model training dataset is on the order of 2.9E-3 eV, and the root mean square error of the atomic force is on the order of 7.3E-2 eV / Å. The error of the test dataset is even smaller. Combining the machine learning potential energy model and the global potential energy surface search enables high-throughput structure screening on the order of 1.0E7. As Figure 5 shown, (a) Schematic diagram of high-throughput structure screening; (b) Comparison of the average atomic energy of the structures obtained by the machine learning potential energy model and the first-principles calculation; (c) Comparison of the atomic forces of the structures obtained by the machine learning potential energy model and the first-principles calculation; (d) Formation energy - configuration image of the structures obtained by screening.

[0064] Figure 6 Device 1000 for accelerating material structure search based on machine learning potential energy and global potential energy surface according to an embodiment of the present invention, including: global potential energy surface structure search module 1002, initial training data set acquisition module 1004, machine learning potential energy training module 1006, training data set update and screening sampling module 1008, structure screening module 1010, machine learning potential energy testing module 1012, structure formation energy acquisition module 1014.

[0065] Potential energy surface structure search module 1002 is used to search for potential energy surface structures by the global potential energy surface search method;

[0066] Initial training data set acquisition module 1004 is used to extract some structures from the searched potential energy surface structures, calculate the potential energy and atomic forces corresponding to the extracted potential energy surface structures by first-principles calculation and collect them as the initial training data set; through the global potential energy surface structure search module 1002 and the initial training data set acquisition module 1004, the initial structure is obtained by the global potential energy surface search method, and the way of calculating the interatomic interaction force includes the first-principles calculation method.

[0067] Machine learning potential energy model training module 1006 is used to train a machine learning potential energy model E(x) representing the potential energy surface of the system based on the initial training data set obtained by the initial training data set acquisition module 1004 through machine learning model training.

[0068] Training data set update and structure clustering screening module 1008 is used to perform screening sampling on the potential energy surface of the system based on the machine learning potential energy model and in combination with the global potential energy surface search method to update the training data set and the machine learning potential energy model; the training data set update and structure clustering screening module 1008 specifically realizes the following functions:

[0069] Extract structure data based on the training data set.

[0070] Perform global potential energy surface search simulation based on the machine learning potential energy model on the structure data to obtain a simulation trajectory. Applicable types of machine learning potential energy include but are not limited to Deep Potential (DP), and applicable global potential energy surface search methods include but are not limited to the Stochastic Surface Walking (SSW) method.

[0071] Screen some structures on the simulated trajectory for first-principles calculations to obtain the potential energy and atomic forces corresponding to the part of the structures, and add the potential energy and atomic forces to the training dataset to update the training dataset. The method of iteratively updating the dataset includes randomly selecting simulated structures from the training dataset or obtaining them through an active learning method. The screening sampling method includes obtaining based on the deviation between machine learning potential energy models and using a structure clustering method.

[0072] Based on the updated training dataset, after training through a machine learning potential energy model, an updated machine learning potential energy model is obtained. This specific implementation can be repeatedly executed multiple times to obtain an updated training dataset and an updated machine learning potential energy model.

[0073] The structure energy screening module 1010 is used to perform high-throughput search for the target system structure through a global potential energy surface search method based on the training dataset updated by the structure energy screening module 1010 and the training dataset and machine learning potential energy model updated by the structure clustering screening module 1008, so as to predict the structure potential energy and perform structure screening.

[0074] The structure formation energy acquisition module 1014 is used to perform first-principles calculations on the screened structures to obtain the corresponding potential energy and atomic forces, use the potential energy and atomic forces as a test dataset to test the prediction performance of the machine learning potential energy model, and calculate the formation energy of the structure through the potential energy.

[0075] As Figure 6 shown, the present invention also provides an electronic device, including:

[0076] A memory 1300 for storing execution instructions;

[0077] A processor 1200 for executing the execution instructions stored in the memory 1300, so that the processor 1300 executes the method for accelerating the search of the material structure in any of the above items.

[0078] According to another aspect of the present invention, a readable storage medium is provided. The readable storage medium stores execution instructions, and when the execution instructions are executed by the processor 1300, they are used to implement the method for accelerating the search of the material structure in any of the above items.

[0079] The present invention constructs a machine learning potential energy model by combining the machine learning potential energy method and the global potential energy surface search method to accelerate the structure search and obtain the formation energy of the structure. Starting from a reasonable initial structure set, a machine learning potential energy model is constructed by sampling based on the global potential energy surface search method, so as to use the machine learning potential energy model to accelerate the structure search and obtain the structure formation energy. The potential energy surface information of the target system is derived from quantum mechanics calculations, and the machine learning potential energy model can quickly and accurately repeat the structural energy and atomic forces on the potential energy surface through fitting. Based on the machine learning potential energy model, the structural energy calculation is accelerated, thereby accelerating the structure search and obtaining the structure formation energy. The present invention can achieve rapid search of structures and is applicable to a variety of complex systems. It can be used for high-throughput screening of structures in complex systems to accelerate material development.

[0080] Based on the idea of screening potential energy surface sampling, the present invention clusters and screens the structures obtained by the global potential energy surface search method, reduces the redundancy of the sampled structures, and reduces the consumption of computing resources. At the same time, methods such as iterative learning and active learning are used to iteratively update the machine learning data set to achieve the purpose of accurately predicting the potential energy surface of the target system. Finally, based on the accurate machine learning potential energy surface obtained, high-throughput calculations are accelerated through global potential energy surface search, structural energy screening, and structural clustering to achieve structure search and obtain the structure formation energy. The present invention has the following technical advantages:

[0081] 1. Compared with the traditional method of constructing a machine learning potential energy model, the present invention combines iterative update and screening sampling to assist in the construction of the potential energy model, reduces the structural redundancy of the training data set of the machine learning potential energy model, and reduces the consumption of computing resources.

[0082] 2. Compared with the high-throughput calculation based on first principles, the time scale can be increased by more than 1000 times. At the same time, the training data set is labeled by first principles calculation, ensuring the accuracy of numerical calculation. While the time scale is increased, the accuracy of the potential energy surface is ensured.

[0083] 3. Compared with the high-throughput calculation based on first principles, the number of structures can also be increased by more than 1000 times. At the same time, the use of the global potential energy surface search method expands the structure search range and accelerates high-throughput structure screening.

[0084] Figure 6FIG. shows an example diagram of an apparatus implemented in hardware of a processing system. The apparatus may include corresponding modules for performing each or several of the steps in the above flowchart. Therefore, each step or several steps in the above flowchart may be performed by the corresponding modules, and the apparatus may include one or more of these modules. The modules may be one or more hardware modules specifically configured to perform the corresponding steps, or implemented by a processor configured to perform the corresponding steps, or stored in a computer-readable medium for implementation by a processor, or implemented by a certain combination.

[0085] The hardware structure may be implemented using a bus architecture. The bus architecture may include any number of interconnecting buses and bridges, depending on the specific application of the hardware and overall design constraints. Bus 1100 connects various circuits including one or more processors 1200, a memory 1300, and / or hardware modules together. Bus 1100 may also connect various other circuits 1400 such as peripheral devices, voltage regulators, power management circuits, external antennas, etc.

[0086] Bus 1100 may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one connecting line is shown in this figure, but it does not mean that there is only one bus or one type of bus.

[0087] Any process or method description represented in a flowchart or otherwise described herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a specific logical function or process, and the scope of the preferred embodiments of the present invention includes additional implementations, where functions may be performed in an order not shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain. A processor executes the various methods and processes described above. For example, the method embodiments disclosed in the present invention can be implemented as a software program tangibly embodied in a machine-readable medium, such as a memory. In some embodiments, part or all of the software program can be loaded and / or installed via the memory and / or a communication interface. When the software program is loaded into the memory and executed by the processor, one or more steps of the methods described above can be performed. Alternatively, in other embodiments, the processor can be configured to execute one of the above methods by any other suitable means (e.g., by means of firmware).

[0088] The logic and / or steps represented in a flowchart or otherwise described herein can be embodied in any readable storage medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other systems that can fetch instructions from and execute instructions.

[0089] As used in this specification, a "readable storage medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the readable storage medium include the following: an electrical connection having one or more wires (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the readable storage medium can even be paper or other suitable medium on which a program can be printed, as the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a memory.

[0090] It should be understood that each part of the present invention can be implemented by hardware, software, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0091] Those of ordinary skill in the art of this technology can understand that all or part of the steps for implementing the above embodiments can be completed by a program instructing relevant hardware. The program for accelerating the material structure search method can be stored in a readable storage medium. When this program is executed, it includes one or a combination of the steps of the method embodiments.

[0092] In addition, in each embodiment of the present invention, each functional unit can be integrated into a processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a readable storage medium. The storage medium can be a read-only memory, a disk, an optical disc, etc.

[0093] In the description of this specification, the description with reference to terms such as "one embodiment / way", "some embodiments / ways", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment / way or example are included in at least one embodiment / way or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment / way or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments / ways or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments / ways or examples described in this specification and the features of different embodiments / ways or examples.

[0094] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" can explicitly or implicitly include at least one of these features. In the description of the present application, the meaning of "a plurality" is at least two, such as two, three, etc., unless otherwise specifically and clearly defined.

[0095] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, the patent owner may make various deformations or modifications within the scope of the appended claims, and as long as they do not exceed the protection scope described in the claims of the present invention, they should be within the protection scope of the present invention.

Claims

1. A method for accelerating the search of material structures, characterized in that, Comprising: Searching for the potential energy surface structure by the global potential energy surface search method; Extracting partial structures from the searched potential energy surface structures, calculating the potential energy and atomic forces corresponding to the extracted potential energy surface structures using first principles and collecting them as the initial training dataset; Based on the initial training dataset, training a machine learning potential energy model representing the potential energy surface of the system through a machine learning model; Based on the machine learning potential energy model and combined with the global potential energy surface search method, performing screening sampling on the potential energy surface of the system to update the training dataset and the machine learning potential energy model; Based on the updated training dataset and machine learning potential energy model, performing high-throughput search for the target system structure through the global potential energy surface search method to predict the structure potential energy and perform structure screening; Performing first principles calculation on the screened structures to obtain the corresponding potential energy and atomic forces, using the potential energy and atomic forces as the test dataset to test the prediction performance of the machine learning potential energy model, and calculating the formation energy of the structure through the potential energy; The step of performing screening sampling on the potential energy surface of the system based on the machine learning potential energy model and combined with the global potential energy surface search method to update the training dataset and the machine learning potential energy model specifically includes: Extracting structure data based on the initial training dataset; Performing global potential energy surface search simulation based on the machine learning potential energy model on the extracted structure data to obtain a simulation trajectory; Screening partial structures on the simulation trajectory for first principles calculation to obtain the potential energy and atomic forces corresponding to the partial structures, and adding the potential energy and atomic forces to the training dataset to update the training dataset; Based on the updated training dataset, after training through the machine learning potential energy model, obtaining an updated machine learning potential energy model; The potential energy type in the machine learning potential energy model includes deep potential energy, and the global potential energy surface search method includes the random surface walking method; Performing high-throughput search for the target system structure through the global potential energy surface search method based on the updated training dataset and machine learning potential energy model to predict the structure potential energy and perform structure screening specifically includes: Searching for the global potential energy surface structure based on the global potential energy surface structure search of the machine learning potential energy model; Performing screening with a given energy threshold based on the global potential energy surface structure and the corresponding machine learning potential energy prediction potential energy; Performing clustering screening on the structures screened based on the energy threshold.

2. The method for accelerating the search of the material structure according to claim 1, wherein Drawing a formation energy - configuration image after calculating the formation energy of the structure according to the potential energy; 3. The method for accelerating the search of a material structure according to claim 2, wherein Specifically, screening partial structures on the simulation trajectory means: performing clustering screening on the simulation trajectory to obtain partial structures.

4. The method for accelerating the search of a material structure according to claim 1, wherein, The step of performing screening sampling on the potential energy surface of the system based on the machine learning potential energy model and combined with the global potential energy surface search method to update the training dataset and the machine learning potential energy model is iterated multiple times.

5. An apparatus for accelerating the search of material structures, characterized in that, Comprising: A potential energy surface structure search module for searching for the potential energy surface structure by the global potential energy surface search method; An initial training dataset acquisition module for extracting partial structures from the searched potential energy surface structures, calculating the potential energy and atomic forces corresponding to the extracted potential energy surface structures using first principles and collecting them as the initial training dataset; A machine learning potential energy model training module, which is used to train a machine learning potential energy model representing the potential energy surface of the system based on the initial training data set through a machine learning model; A training data set update and structure clustering and screening module, which is used to perform screening sampling on the potential energy surface of the system based on the machine learning potential energy model and in combination with the global potential energy surface search method to update the training data set and the machine learning potential energy model; A structure energy screening module, which is used to perform high-throughput search on the target system structure through the global potential energy surface search method based on the updated training data set and machine learning potential energy model to predict the structure potential energy and perform structure screening; A structure formation energy acquisition module, which is used to perform first-principles calculation on the screened structure to obtain the corresponding potential energy and atomic force, use the potential energy and atomic force as a test data set to test the prediction performance of the machine learning potential energy model, and calculate the formation energy of the structure through the potential energy; The steps of performing screening sampling on the potential energy surface of the system based on the machine learning potential energy model and in combination with the global potential energy surface search method to update the training data set and the machine learning potential energy model specifically include: Extracting structure data based on the initial training data set; Performing global potential energy surface search simulation based on the extracted structure data using the machine learning potential energy model to obtain a simulation trajectory; Screening some structures on the simulation trajectory for first-principles calculation to obtain the potential energy and atomic force corresponding to the part of the structure, and adding the potential energy and atomic force to the training data set to update the training data set; Based on the updated training data set, after training through a machine learning potential energy model, an updated machine learning potential energy model is obtained; The potential energy type in the machine learning potential energy model includes deep potential energy, and the global potential energy surface search method includes the random surface walking method; Performing high-throughput search on the target system structure through the global potential energy surface search method based on the updated training data set and machine learning potential energy model to predict the structure potential energy and perform structure screening specifically includes: Performing global potential energy surface structure search based on the machine learning potential energy model to obtain the global potential energy surface structure; Performing screening with a given energy threshold based on the global potential energy surface structure and the corresponding machine learning potential energy prediction potential energy; Performing clustering screening on the structures screened based on the energy threshold.

6. An electronic device, characterized in that, Including: A memory that stores execution instructions; A processor that executes the execution instructions stored in the memory, so that the processor executes the method for accelerating material structure search according to any one of claims 1 to 4.

7. A readable storage medium, characterized in that, Execution instructions are stored in the readable storage medium, and when the execution instructions are executed by a processor, they are used to implement the method for accelerating material structure search according to any one of claims 1 to 4.

Images