A method of adsorption energy prediction and related apparatus

Abstract

The embodiment of the application discloses a kind of adsorption energy prediction method and related device, at least involve machine learning in artificial intelligence etc., can be applied to catalysis field, environmental catalysis, energy conversion, chemical production, biomedical scene etc..The method provided in the application embodiment can save computing time, improve prediction accuracy, speed up the development process of new materials.Moreover, the adsorption energy prediction method comprises: obtaining the target structure and element characteristics of the material to be predicted;According to the target structure of the material to be predicted, target distance information and target angle information are obtained;The target structure characteristics of the material to be predicted are determined based on the target distance information and the target angle information;The target structure characteristics and element characteristics are processed based on the preset machine learning model, and a prediction label is obtained, which is used to indicate the adsorption energy of the adsorption molecule in the material to be predicted.

Description

Technical Field

[0001] This application relates to the field of computational chemistry, specifically to a method and related apparatus for predicting adsorption energy. Background Technology

[0002] A suitable catalyst is crucial for catalytic reactions to proceed, accelerating the reaction process and enabling reactions that would otherwise take a very long time to complete to proceed rapidly. Finding a suitable catalyst not only accelerates the reaction rate but also speeds up specific types of reactions, thus selectively producing only a particular product. For example, in the reduction of carbon dioxide using substances such as carbon dioxide and water as reactants, a series of chemical reactions occur, producing gaseous or liquid fuels such as methane, methanol, and ethanol. This series of chemical reactions is difficult to carry out under normal conditions, and the reaction pathways are complex, producing a wide variety of products, making it difficult to obtain pure products. Carbon monoxide is one of the important intermediate products in this series of reactions. Therefore, studying the adsorption energy of carbon monoxide and other adsorbed molecules on the catalyst surface can provide a deeper understanding of the underlying mechanism of the carbon dioxide reduction reaction, and thus help in finding catalysts with high selectivity and high conversion rates.

[0003] In related approaches, the generalized coordination number at the dopant atom sites is typically used to describe the catalyst's structural information, and a simple machine learning model is employed to predict the adsorption energy of the adsorbed molecules in the catalyst. However, the value of the generalized coordination number of the dopant atom is affected by its position, which often changes significantly before and after structural relaxation. Therefore, only by relaxing the initial structure of the crystal material to a stable structure through first-principles calculations can the position of the dopant atom be stabilized, thus ensuring that the value of the generalized coordination number remains unchanged.

[0004] In other words, the relevant schemes require first-principles calculations to relax the structure to a stable state before the generalized coordination number at the doped atom sites in the relaxed stable structure can be used as the input to the machine learning model to predict the adsorption energy. This inevitably makes the relevant schemes rely on time-consuming first-principles calculations to achieve structural relaxation. Moreover, when the initial structure before relaxation is directly used as the input to the machine learning model, the prediction accuracy will decrease due to the change in the generalized coordination number. Summary of the Invention

[0005] This application provides a method and related apparatus for predicting adsorption energy, which can not only save calculation time, but also improve the prediction accuracy of the adsorption energy of adsorbed molecules in the material to be predicted, thereby accelerating the research and development process of subsequent new materials.

[0006] In a first aspect, embodiments of this application provide a method for predicting adsorption energy. The method includes: acquiring the target structure and elemental characteristics of a material to be predicted, wherein the target structure represents the atomic structure obtained from a first structure of the material to be predicted with a target atom as the center and a preset cutoff radius, the first structure being obtained by symmetrically expanding the smallest repeating unit cell of the material to be predicted; acquiring target distance information and target angle information based on the target structure of the material to be predicted, wherein the target distance information indicates the distance between the target atom and each atom, and the target angle information indicates the angle formed between the target atom and the first atom, and between the target atom and the second atom, the first atom being any atom within the target structure, or, the first atom being any atom within a preset radius centered on the target atom, the preset radius being obtained from a preset cutoff radius; determining the target structural characteristics of the material to be predicted based on the target distance information and the target angle information; and processing the target structural characteristics and elemental characteristics based on a preset machine learning model to obtain a prediction label, the prediction label being used to indicate the adsorption energy of adsorbed molecules in the material to be predicted.

[0007] Secondly, embodiments of this application provide an adsorption energy prediction device. This adsorption energy prediction device includes, but is not limited to, terminal devices, servers, etc. The adsorption energy prediction device includes an acquisition unit and a processing unit. The acquisition unit is used to acquire the target structure and elemental characteristics of the material to be predicted. The target structure represents the range of atomic structures obtained from a first structure of the material to be predicted, centered on a target atom and with a preset cutoff radius. The first structure is obtained by symmetrically expanding the smallest repeating unit cell of the material to be predicted. The acquisition unit is used to acquire target distance information and target angle information based on the target structure of the material to be predicted. The target distance information indicates the distance between the target atom and each atom, and the target angle information indicates the angle formed between the target atom and the first atom, and between the target atom and the second atom. The first atom can be any atom within the target structure, or any atom within a preset radius centered on the target atom, where the preset radius is obtained from a preset cutoff radius. The processing unit is used to determine the target structural characteristics of the material to be predicted based on the target distance information and the target angle information. The processing unit is used to process the target structural features and elemental features based on a preset machine learning model to obtain a prediction label, which is used to indicate the adsorption energy of the material to be predicted.

[0008] In some optional examples, the processing unit is used to: determine a first value based on first distance information and a preset cutoff radius, wherein the first value indicates the contribution of the first atom to the adsorption energy, and the first distance information indicates the distance between the target atom and the first atom; determine a first structural feature of the material to be predicted based on the first value, target angle information, second distance information, and third distance information, wherein the first structural feature indicates the atomic structural relationship between the target atom, the first atom, and the second atom, the second distance information indicates the distance between the target atom and the second atom, and the third distance information indicates the distance between the first atom and the second atom; and determine a target structural feature of the material to be predicted based on the first structural feature.

[0009] In some alternative examples, the processing unit is used to: determine a second value based on the second distance information and a preset cutoff radius, the second value being used to indicate the contribution of the second atom to the adsorption energy; determine a third value based on the third distance information and a preset cutoff radius, the third value being used to indicate the contribution of the first atom and the second atom to the adsorption energy; and process the first value, target angle information, first distance information, second distance information, third distance information, second value, and third value based on a preset Gaussian model to obtain the first structural feature of the material to be predicted.

[0010] In some alternative examples, the processing unit is also configured to: determine a second structural feature of the material to be predicted based on a first value, the second structural feature indicating the atomic structure surrounding the target atom. The processing unit is configured to determine the target structural feature of the material to be predicted based on the first and second structural features.

[0011] In some other optional examples, the processing unit is also used to: process the first value and the first distance information based on a preset Gaussian model to obtain the third structural feature of the material to be predicted, the third structural feature being used to indicate the atomic structural relationship between the target atom and the first atom; and determine the target structural feature of the material to be predicted based on the first structural feature and the third structural feature.

[0012] In some other optional examples, the processing unit is also configured to: determine a second structural feature of the material to be predicted based on a first value, the second structural feature being used to indicate the atomic structure around the target atom; and determine a target structural feature of the material to be predicted based on the first structural feature, the second structural feature, and the third structural feature.

[0013] In some alternative examples, the processing unit is used to sum each first value to obtain a second structural feature of the material to be predicted.

[0014] In some alternative examples, the processing unit is used to: determine the thickness information of each layer of the target structure based on a preset cutoff radius and the number of layers into which the target structure is divided; determine the target distance based on the preset cutoff radius and the thickness information, the target distance being used to indicate the distance between the target atom and the boundary of each layer of the target structure; and determine a first value based on the first distance information, the target distance, and the thickness information.

[0015] In some alternative examples, the processing unit is used to: determine a second value based on the second distance information, the target distance, and the thickness information, the second value indicating the contribution of the second atom to the adsorption energy; determine a third value based on the third distance information, the target distance, and the thickness information, the third value indicating the contribution of the first atom and the second atom to the adsorption energy; and process the first value, the second value, the third value, and the target angle information based on a preset cosine function model to obtain the first structural feature of the material to be predicted.

[0016] In some alternative examples, the processing unit is further configured to: process the first value based on a preset cosine function model to obtain a third structural feature of the material to be predicted, wherein the third structural feature is used to indicate the atomic structural relationship between the target atom and the first atom. The processing unit is configured to determine the target structural feature of the material to be predicted based on the first structural feature and the third structural feature.

[0017] In some alternative examples, the atoms include dopant atoms and at least two matrix atoms. The acquisition unit is used to: acquire the elemental characteristics of the dopant atoms when the atom type of each of the at least two matrix atoms is the same as the atom type of the matrix atoms in the remaining structure, so as to obtain the elemental characteristics of the material to be predicted.

[0018] In some alternative examples, the acquisition unit is also used to: acquire the elemental characteristics of the doped atoms and the elemental characteristics of each matrix atom when the atom type of any matrix atom in at least two matrix atoms is different from the atom type of the matrix atoms in the remaining structure, so as to acquire the elemental characteristics of the material to be predicted.

[0019] In some alternative examples, the acquisition unit is further configured to: acquire the minimum repeating unit cell of the material to be predicted, wherein the minimum repeating unit cell includes dopant atoms, matrix atoms, and adsorbed molecules. The processing unit is configured to: determine a first crystal structure based on the minimum repeating unit cell, wherein the dopant atoms in the first crystal structure are replaced by matrix atoms, and the first crystal structure does not include adsorbed molecules; and extend the first crystal structure symmetrically in a preset first direction and a preset second direction, respectively, with the target atom as the center, to obtain the first structure.

[0020] In some alternative examples, the target atom may include a dopant atom, or the target atom may include an adsorbed molecule.

[0021] A third aspect of this application provides an adsorption energy prediction device, including: a memory, an input / output (I / O) interface, and a processor. The memory stores program instructions. The processor executes the program instructions in the memory to perform the adsorption energy prediction method corresponding to the embodiment of the first aspect described above.

[0022] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method corresponding to the embodiments of the first aspect described above.

[0023] The fifth aspect of this application provides a computer program product containing instructions that, when run on a computer or processor, causes the computer or processor to execute the method described above for performing the implementation method of the first aspect.

[0024] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0025] In this embodiment, the target structure and elemental characteristics of the material to be predicted are obtained. The target structure represents the atomic structure obtained from the first structure of the material to be predicted with the target atom as the center and a preset cutoff radius. The first structure is obtained by symmetrically expanding the smallest repeating unit cell of the material to be predicted. Then, target distance information and target angle information are obtained based on the target structure of the material to be predicted, and the target structural characteristics of the material to be predicted are determined based on the target distance information and the target angle information. The target distance information is used to indicate the distance between the target atom and each atom, and the target angle information is used to indicate the angle formed between the target atom and the first atom, and between the target atom and the second atom. Finally, the target structural characteristics and elemental characteristics are processed based on a preset machine learning model to obtain a prediction label, which is used to indicate the adsorption energy of the adsorbed molecules in the material to be predicted. In the above manner, the first structure is obtained after symmetrically expanding the smallest repeating unit cell of the material to be predicted, and the atomic structure is obtained from the first structure with the target atom as the center and a preset cutoff radius. This atomic structure is used as the target structure of the material to be predicted. Then, the target distance and angle information between the target atom and other atoms are directly determined from the target structure. This target distance and angle information are used to determine the target structural features, which, combined with elemental features, can be used as input to a pre-defined machine learning model. This means that the corresponding target structural features can be directly determined from the target structure of the material before relaxation, eliminating the need for time-consuming first-principles calculations to relax the target structure and obtain input for the pre-defined machine learning model, thus saving computation time. Furthermore, symmetric expansion of the smallest repeating unit cell ensures that the atomic positions in the expanded first structure remain unchanged, and consequently, the atomic positions in the final target structure also remain unchanged. Therefore, considering the distances and angles between atoms in the target structure to obtain the target structural features improves the accuracy of structural feature extraction. This enhances the prediction accuracy of the adsorption energy of adsorbed molecules in the material during subsequent model prediction, thereby accelerating the development of new materials. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A schematic diagram illustrating the translation and rotation of material according to an embodiment of this application is shown;

[0028] Figure 2 This paper shows a schematic diagram of the minimum repeating unit cell structure of the single-atom catalyst provided in the embodiments of this application;

[0029] Figure 3 A schematic diagram of the adsorption energy prediction process framework provided in an embodiment of this application is shown;

[0030] Figure 4 A flowchart of a method for predicting adsorption energy provided in an embodiment of this application is shown;

[0031] Figure 5 A schematic diagram illustrating the mechanism of action of the symmetry function provided in the embodiments of this application is shown;

[0032] Figure 6 A schematic diagram of the adsorption energy prediction device provided in an embodiment of this application is shown;

[0033] Figure 7 A schematic diagram of the hardware structure of the adsorption energy prediction device provided in the embodiments of this application is shown. Detailed Implementation

[0034] This application provides a method and related apparatus for predicting adsorption energy, which can not only save calculation time, but also improve the prediction accuracy of the adsorption energy of adsorbed molecules in the material to be predicted, thereby accelerating the research and development process of subsequent new materials.

[0035] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0036] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that implementations of the application described herein can be implemented, for example, in sequences other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0038] Catalytic reactions are ubiquitous in daily life. For example, every car is equipped with an exhaust purification system that converts toxic carbon monoxide and nitrogen oxides in exhaust gases into non-toxic carbon dioxide and nitrogen before emission. A suitable catalyst is crucial for catalytic reactions to occur. In the aforementioned exhaust purification reaction, platinum is the catalyst. The purpose of a catalyst is to accelerate the reaction process, allowing reactions that would normally take a very long time to complete to proceed rapidly.

[0039] For example, in the carbon dioxide reduction reaction, a series of chemical reactions using carbon dioxide, water, and other substances as reactants to produce gaseous or liquid fuels such as methane, methanol, and ethanol, are difficult to carry out under normal conditions. Furthermore, the reaction pathways are complex, and the products are numerous, making it impossible to obtain pure products and hindering their practical application. Therefore, a suitable catalyst can not only accelerate the reaction rate but also selectively accelerate a specific type of reaction, thus producing only a particular product. In this series of chemical reactions, carbon monoxide is one of the important intermediate products. Studying the adsorption energy of carbon monoxide on the catalyst surface can help us better understand the mechanism behind the carbon dioxide reduction reaction and thus find catalysts with high selectivity and high conversion rates.

[0040] With the research and advancement of artificial intelligence (AI) technology, AI technology is being researched and applied in many fields, such as smart homes, smart wearable devices, virtual assistants, smart speakers, self-driving cars, autonomous driving, drones, robots, smart healthcare, and smart customer service. It is believed that with the development of technology, AI technology will be applied in more fields and play an increasingly important role.

[0041] Therefore, in studying the adsorption energies of adsorbed molecules such as carbon monoxide on the surface of catalysts and other materials, structural and elemental features can be extracted from the target structure of the material to be predicted. Then, machine learning (ML) from the field of artificial intelligence can be used to predict the adsorption energies of the adsorbed molecules in the material. The predicted adsorption energies can then be used to select whether the material can be used as a catalyst with high selectivity and high conversion rate. It should be noted that the adsorption capacity of adsorbed molecules on the catalyst surface is generally used as the basis for judging the catalytic performance of the catalyst. If the adsorption energy is too high, it is not conducive to the detachment of product molecules from the catalyst surface; if the adsorption energy is too low, it is not conducive to the adsorption of reactants on the catalyst surface and the occurrence of the catalytic reaction.

[0042] It should be understood that the material to be predicted described may include, but is not limited to, single-atom catalysts, other types of catalysts, crystalline materials, etc., and this application does not specifically limit its description. The following description will only use single-atom catalysts as an example to illustrate the method for predicting adsorption energy provided in the embodiments of this application.

[0043] This application provides a method for predicting adsorption energy. The method for predicting adsorption energy provided in this application is based on artificial intelligence (AI). AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to obtain optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can react in a way similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making functions.

[0044] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech recognition, natural language processing, and machine learning / deep learning. In this application's embodiments, the main AI technologies involved include the aforementioned machine learning areas. For example, it may involve deep learning within machine learning (ML).

[0045] The adsorption energy prediction method provided in this application can be applied to adsorption energy prediction devices with data processing capabilities, such as terminal devices and servers. Terminal devices may include, but are not limited to, smartphones, desktop computers, laptops, tablets, smart speakers, in-vehicle devices, smartwatches, wearable smart devices, smart voice interaction devices, smart home appliances, and aircraft. Servers may be independent physical servers, server clusters or distributed systems composed of multiple physical servers, or cloud servers providing cloud computing services; this application does not impose specific limitations. Furthermore, the mentioned terminal devices and servers can be directly or indirectly connected via wired or wireless communication; this application does not impose specific limitations.

[0046] The aforementioned adsorption energy prediction device can perform the aforementioned machine learning. Machine learning is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and many other disciplines. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as neural networks.

[0047] The adsorption energy prediction method provided in this application uses an artificial intelligence model, which mainly involves the application of neural networks. The adsorption energy of adsorbed molecules in the material to be predicted is predicted through neural networks.

[0048] In related schemes, the positions of dopant atoms often change significantly before and after structural relaxation, and the generalized coordination number of the dopant atoms is affected by their positions. Therefore, first-principles calculations are needed to relax the structure to a stable state before the generalized coordination number at the dopant atom sites in the relaxed stable structure can be used as input to a machine learning model to predict the adsorption energy. However, this inevitably relies on time-consuming first-principles calculations for structural relaxation, and when the initial structure before relaxation is directly used as input to the machine learning model, the change in the generalized coordination number leads to a decrease in prediction accuracy.

[0049] Furthermore, by representing the position of each atom in the material to be predicted using a set of three-dimensional coordinates (such as the XYZ coordinate system), and by obtaining the elemental type of each atom, the material to be predicted can be uniquely identified. Although the material to be predicted contains atoms of different types, it is usually obtained by endlessly replicating a minimal repeating unit cell in a three-dimensional coordinate system. Therefore, describing the position of the atoms and the elemental type of the atoms in a minimal repeating unit cell is sufficient to completely describe the material to be predicted. Thus, the material to be predicted can be described from both its structure and elemental composition. In this way, the structural and elemental characteristics of the material to be predicted can be extracted and used as input information for a machine learning model to predict the corresponding properties of the material, namely, the adsorption energy of the adsorbed molecules in the material to be predicted.

[0050] As described above, the structural features of the material to be predicted can be described by the position coordinates of atoms, and a mapping relationship is formed between the structural features and the corresponding material to be predicted; that is, the structure of the material to be predicted can be uniquely determined by a set of position coordinates. However, since the position coordinates of atoms do not satisfy the invariance of translation or rotation, the position coordinates themselves cannot be directly used as input to a machine learning model. For example, Figure 1 This illustration shows a schematic diagram of the translation and rotation of material according to an embodiment of this application. For example... Figure 1 As shown, taking a single-atom catalyst as the material to be predicted as an example, after determining the XYZ coordinate system, the position coordinates of each atom in the material to be predicted can be determined. Furthermore, from... Figure 1 It can be seen that rotating the material to be predicted changes the position coordinates of each atom. Similarly, translating the material also changes the position coordinates of each atom. This change in atom position coordinates alters the structural features input into the machine learning model, causing the model to fit these changed structural features as features of a new material. However, it is clear that the physical and chemical properties of the material to be predicted should not change after translation or rotation, and the corresponding structural features should also remain unchanged. Therefore, inputting these unchanged structural features into the machine learning model will ensure that the predicted values ​​remain unchanged even after the overall translation or selection of the material. It should be noted that... Figure 1 The specific structure of the single-atom catalyst shown in the figure can be found in the following sections. Figure 2 The content shown in the image will be understood, and will not be elaborated upon here.

[0051] Based on this, in order to solve the aforementioned technical problems, embodiments of this application provide a method for predicting adsorption energy. This method for predicting adsorption energy can be applied in the field of catalysis, such as in oxidation reactions, electrocatalysis, photoelectrocatalysis, hydrogenation catalysis, lithium batteries, enzyme catalysis, and other chemical reaction fields. Indicatively, this method for predicting adsorption energy can also be applied in environmental catalysis, chemical production, energy conversion, organic synthesis, biomedicine, pharmaceuticals, and waste treatment, etc., but this application does not specifically limit its application.

[0052] Before describing in detail the method for predicting adsorption energy in this application, the dopant atoms, adsorbed molecules, and matrix atoms will be specifically described. For example, introducing atoms of other elements into a matrix material or substrate material can be understood as the dopant atoms described above. Furthermore, the matrix material or substrate material is primarily composed of matrix atoms. The described adsorbed molecules can be understood as intermediate products generated in a chemical reaction. Additionally, introducing single atoms of other elements onto the surface of a metallic material constitutes a single-atom catalyst, where the metallic material can be understood as the matrix material or substrate material. It should be noted that the described surface of the metallic material can be understood as the surface exposed by cutting an ideal metal unit cell at a certain cutting angle.

[0053] For example, Figure 2 A schematic diagram of the minimum repeating unit cell of the single-atom catalyst provided in this application is shown. Figure 2 As shown, taking a copper-boron single-atom catalyst as an example, (a) shows a top view of the copper-boron single-atom catalyst, and (b) shows a front view of the copper-boron single-atom catalyst. Figure 2 As shown in sections (a) and (b), the substrate material of this copper-boron single-atom catalyst is the surface of metallic copper, with boron atoms replacing copper atoms at a certain site on the surface as dopant atoms, thus forming the copper-boron single-atom catalyst. Furthermore, a carbon monoxide molecule is adsorbed on the surface of this copper-boron single-atom catalyst. This carbon monoxide molecule, as an adsorbed molecule, forms a chemical bond with the matrix atoms (i.e., copper atoms) on the surface of the single-atom catalyst.

[0054] in addition, Figure 2 The structure shown is the smallest repeating unit cell of this copper-boron single-atom catalyst. The actual structure of this copper-boron single-atom catalyst consists of a few dopant atoms scattered on an infinitely large metallic copper plane. These dopant atoms are far apart and do not interact with each other. Moreover, under normal circumstances, the dopant atoms have higher catalytic activity than the substrate material. Therefore, the properties of the matrix atoms near the dopant atoms are also affected, often exhibiting stronger catalytic performance than the undoped form.

[0055] Based on this, embodiments of this application provide a method for predicting adsorption energy. This method for predicting adsorption energy can be applied to... Figure 3 The process framework shown is as follows. Figure 3 As shown, the smallest repeating unit cell of the material to be predicted can first be obtained, for example, as mentioned above. Figure 2 The smallest repeating unit cell of the copper-boron single-atom catalyst mentioned in the text is used. Centered on the target atom, this smallest repeating unit cell is symmetrically extended along two preset directions to obtain a first structure. Then, with the target atom as the center and a preset cutoff radius (hereinafter denoted by Rc), the corresponding atomic structure range is extracted from this first structure, thus obtaining the target structure of the material to be predicted. The target structure of the material to be predicted can characterize the atomic situation around the target atom within the preset cutoff radius. Thus, by layering the target structure, structural features at different positions around the target atom are extracted using preset Gaussian and cosine models and other symmetry functions. Specifically, based on the target structure, corresponding target distance and target angle information can be obtained, and then the structural features of the material to be predicted can be extracted using preset Gaussian and cosine models and other symmetry functions. After obtaining the elemental features of the material to be predicted, these elemental features and structural features can be used as input to a preset machine learning model to predict the adsorption energy of the adsorbed molecules in the material to be predicted.

[0056] It should be noted that the target atom described can be a doped atom or an adsorbed molecule, as mentioned above. Figure 3 This illustration uses only the target atom as a doped atom as an example; no specific limitations are made in this application. The material to be predicted described can be any of the aforementioned... Figure 2 The copper-boron single-atom catalyst shown can also be other types of single-atom catalysts, crystalline materials, or other types of catalysts, etc., and this application does not specifically limit or describe them. In the following embodiments, only the target atom is used as the dopant atom and the material to be predicted is used as the catalyst. Figure 2 The method for predicting adsorption energy provided in this application is introduced using a copper-boron single-atom catalyst as an example.

[0057] Furthermore, the aforementioned cutoff radius can be understood as the range used in molecular dynamics to represent the force. If two atoms are within the cutoff radius, the potential function is effective; conversely, if two atoms are outside the cutoff radius, the potential function is ineffective. The preset cutoff radius provided in the embodiments of this application can be set according to requirements, for example, 12 angstroms. No specific restrictions are imposed.

[0058] The following describes a method for predicting adsorption energy provided by an embodiment of this application, with reference to the accompanying drawings. Figure 4A flowchart illustrating a method for predicting adsorption energy provided in an embodiment of this application is shown. Figure 4 As shown, the method for predicting adsorption energy may include the following steps:

[0059] 401. Obtain the target structure of the material to be predicted. The target structure is used to represent the atomic structure obtained from the first structure of the material to be predicted with the target atom as the center and a preset cutoff radius. The first structure is obtained by symmetry expansion of the smallest repeating unit cell of the material to be predicted.

[0060] In this example, after obtaining the material to be predicted, the minimum repeating unit cell can be determined from the material. This minimum repeating unit cell is symmetrically extended into a first structure on a plane formed by a preset first direction and a preset second direction; that is, the first structure is the structure obtained by symmetrically extending the minimum repeating unit cell of the material to be predicted. The described minimum repeating unit cell is understood as the smallest repeating unit that constitutes the structure of the material to be predicted, and this minimum repeating unit cell includes dopant atoms, matrix atoms, and adsorbed molecules.

[0061] For example, the minimal repeating unit cell can be symmetrically extended into a first structure in any of the following ways: determining the first crystal structure based on the minimal repeating unit cell, replacing the doped atoms in the first crystal structure with matrix atoms, and excluding adsorbed molecules. Then, the first crystal structure is symmetrically extended in a predetermined first direction and a predetermined second direction, centered on the target atom, to obtain the first structure. For example, as described above... Figure 2 Taking the smallest repeating unit cell of the copper-boron single-atom catalyst shown as an example, assuming that the plane formed by the first and second preset directions is the XY plane, the smallest repeating unit cell of the copper-boron single-atom catalyst can be replicated. Then, the adsorbed molecules in the replicated smallest repeating unit cell are deleted, and the dopant atoms are replaced with matrix atoms to obtain the first crystal structure. Thus, taking the target atom in the original smallest repeating unit cell as the center, the first crystal structure is extended 3×3 symmetrically on the XY plane, that is, the first crystal structure is repeatedly extended 3 times in the X direction and 3 times in the Y direction to obtain the first structure.

[0062] Alternatively, the target atom in the minimum repeating unit cell can be used as the center, and the minimum repeating unit cell can be symmetrically extended in a predetermined first direction and a predetermined second direction to obtain a second structure. Then, in this second structure, the adsorbed molecules in the minimum repeating unit cell during the symmetrical extension are deleted, and the corresponding dopant atoms are replaced with matrix atoms, so that the second structure retains only the dopant atoms and adsorbed molecules from the obtained minimum repeating unit cell, as well as the matrix atoms. For example, as described above... Figure 2Taking the minimum repeating unit cell of the copper-boron single-atom catalyst shown as an example, assuming the plane formed by the first and second preset directions is the XY plane, the minimum repeating unit cell of the copper-boron single-atom catalyst can be replicated. Then, the replicated minimum repeating unit cell is first extended 3×3 symmetryally on the XY plane, that is, the minimum repeating unit cell is extended three times in the X direction and three times in the Y direction to obtain the second structure. In this second structure, the adsorbed molecules in the replicated minimum repeating unit cell are deleted, and the dopant atoms are replaced with matrix atoms to obtain the first structure.

[0063] It should be understood that the first structure mentioned can describe the pure metallic surface formed around the smallest repeating unit cell to simulate the real structure of the material to be predicted. Furthermore, the aforementioned preset first direction can be understood as the X direction in the XY plane, and the preset second direction can be understood as the Y direction in the XY plane, etc., and this application does not impose any specific limitations. The mentioned 3×3 symmetry extension in the XY plane can, in practical applications, also be 5×5, 6×6, etc., and this application does not impose any specific limitations.

[0064] In this way, after expanding the smallest repeating unit cell to obtain the first structure, the atomic structure within the first structure, centered on the target atom and within a predetermined cutoff radius, can be selected as the target structure of the material to be predicted. For example, using the dopant atom as the target atom, the atomic structure within the first structure with a cutoff radius of 12 angstroms can be selected as the target structure.

[0065] It should be noted that since the interaction between atoms decreases with increasing distance between them, it can be approximated that atoms outside the preset cutoff radius have no effect on the target atom, while atoms within the preset cutoff radius interact with the target atom. Therefore, the structural environment around the target atom can be fully described by the positions of atoms within the preset cutoff radius.

[0066] 402. Obtain the elemental characteristics of the material to be predicted.

[0067] In this example, elemental features can be understood as features that are independent of the specific material of the material to be predicted, but only related to the element type. For example, the atoms in the target structure may include dopant atoms and at least two matrix atoms, as well as adsorbed molecules. Then, if the atom type of each matrix atom in the at least two matrix atoms is the same as the atom type of the matrix atoms in the remaining structures (i.e., the element types of the matrix atoms are consistent), the elemental features of the dopant atoms are obtained, and these elemental features are used as the elemental features of the material to be predicted. Conversely, if the atom type of any matrix atom in the at least two matrix atoms is different from the atom type of the matrix atoms in the remaining structures (i.e., the element types of any matrix atom in the target structure are different from the element types of the matrix atoms in the remaining structures), then the elemental features of the dopant atoms and the elemental features of each matrix atom are obtained, and these elemental features of the dopant atoms and the elemental features of each matrix atom are used as the elemental features of the material to be predicted. The elemental features of each matrix atom described here include the elemental features of the matrix atoms in the target structure and the elemental features of the matrix atoms in the remaining structures.

[0068] It should be noted that the described elemental characteristics may include, but are not limited to, one or more of the following characteristics: period number, group number, atomic number, relative atomic mass, atomic radius, electron affinity, first ionization energy, electronegativity, number of p or d orbitals, number of valence electrons, and d-band center energy. In practical applications, elemental characteristics may also be other types of characteristics, which are not specifically limited in the embodiments of this application.

[0069] 403. Obtain target distance information and target angle information based on the target structure of the material to be predicted. The target distance information is used to indicate the distance between the target atom and each atom, and the target angle information is used to indicate the angle between the target atom and the first atom and the target atom and the second atom.

[0070] In this example, the first atom can be any atom within the target structure, and thus the target atom and the preset cutoff radius R can be considered. c The structural relationship between all other atoms within the target atom. Alternatively, the first atom can be any atom within a predetermined radius centered on the target atom; that is, it can also consider only the structural relationship between the target atom and other atoms within a predetermined radius, rather than the predetermined cutoff radius R. c All atoms within the range. The preset radius is obtained from the preset cutoff radius. It should be noted that the second atom described is located in the same structural range as the first atom, for example, if the first atom is within the preset cutoff radius R. c When any atom is within the range, the second atom is also the R. cThe second element is any atom within the range that does not overlap with the first atom. Similarly, when the first atom is any atom within the preset radius, the second element is also any atom within the preset radius that does not overlap with the first atom; no specific limitation is made here.

[0071] Based on this, the target distance information can be obtained by determining the distance between the target atom and each atom according to the target structure. Similarly, by taking the target atom as the origin and connecting the target atom with the first atom and the second atom, and by measuring the angles formed between the connection between the target atom and the first atom and the connection between the target atom and the second atom, the target angle information can be obtained.

[0072] 404. Determine the target structural features of the material to be predicted based on target distance information and target angle information.

[0073] In this example, after obtaining the target distance and target angle information, the target structural features of the material to be predicted can be determined based on this information. For example, different symmetry functions can be used to determine the target structural features based on the target distance and target angle information. It should be noted that the described symmetry functions are essentially a set of Gaussian functions, which act as filters to perform hierarchical filtering of the target structure. Different Gaussian functions extract structural features at different locations around the target atom in different ways. This symmetry function can be a symmetry function constructed using a localized cosine function as a filter, or a symmetry function constructed using a Gaussian function and a truncation function as filters. The symmetry function described using a Gaussian function and a truncation function as filters can further include Gaussian function-cosine truncation function, Gaussian function-arctangent truncation function, etc. This application does not specifically limit the details.

[0074] For example, in determining the target structural features of the material to be predicted based on target distance information and target angle information, this can be achieved as follows: A first value is determined based on first distance information and a preset cutoff radius. This first value indicates the contribution of the first atom to the adsorption energy of the adsorbed molecule, and the first distance information indicates the distance between the target atom and the first atom. Then, the distance between the target atom and the second atom, i.e., the second distance information, is also calculated; and the distance between the first atom and the second atom, i.e., the third distance information, is calculated. Thus, the first structural feature of the material to be predicted is determined based on the first value, the target angle information, and the second and third distance information. The first structural feature indicates the atomic structural relationships between the target atom, the first atom, and the second atom. Therefore, the first structural feature can be used as the target structural feature of the material to be predicted.

[0075] In some examples, where a Gaussian function and a cutoff function are used as filters for the symmetry function, the first structural feature of the material to be predicted is obtained by processing based on a first value, target angle information, second distance information, and third distance information. Specifically, this can be achieved by determining a second value based on the second distance information and a preset cutoff radius, and determining a third value based on the third distance information and a preset cutoff radius. The second value indicates the contribution of the second atom to the adsorption energy, and the third value indicates the contributions of the first and second atoms to the adsorption energy. Then, the first value, target angle information, first distance information, second distance information, third distance information, second value, and third value are processed based on a preset Gaussian model to obtain the first structural feature of the material to be predicted.

[0076] It should be noted that the first, second, and third values ​​described above can be obtained using the truncation function f. c This is calculated using the truncation function f. c The main goal is to make the contribution of atoms outside the cutoff radius to the adsorption energy approach zero, and this cutoff function f c It remains smooth near the cutoff radius. The cutoff function f c The form of expression can be understood by referring to the following formula (1) or (2), that is:

[0077]

[0078] or,

[0079]

[0080] Where i represents the target atom, j represents the first atom, and R ij This refers to the distance between the target atom i and the first atom j, i.e., the first distance information. Additionally, R... c f is the preset cutoff radius. c (R ij ) represents the first value. Additionally, formula (1) shows f... c For the cosine-cut-off function, formula (2) shows f c This is the arctangent truncation function.

[0081] Therefore, after obtaining the preset cutoff radius R c and the first distance information R ij After that, the R can be... c and R ij By inputting the cosine cutoff function shown in formula (1) or the arctangent cutoff function shown in formula (2), the first value f can be obtained. c (R ijSimilarly, the obtained preset cutoff radius R can be... c and the second distance information R ik Input the cutoff function f shown in formula (1) or (2) above. c In this way, the second value f can be obtained. c (R ik ), where the second distance information R ik This represents the distance between the target atom i and the second atom k. For the third value f... c (R jk Alternatively, the obtained preset cutoff radius R can be used in the same way. c And the third distance information R jk Input the cutoff function f shown in formula (1) or (2) above. c It was obtained from the middle.

[0082] Thus, after calculating the first value f c (R ij ), second value f c (R ik ) and the third value f c (R jk Then, the target angle information and the first distance information R described above can be combined. ij Second distance information R ik And the third distance information R jk The input information is used as the input information for the preset Gaussian model, and the first structural feature is calculated through the preset Gaussian model. For example, the preset Gaussian model is represented by the following formula (3), that is:

[0083]

[0084] Where, θ ijk Represents the target's included angle information, R s η, λ, and ξ are all adjustable parameters. This is represented as the first structural feature.

[0085] In other examples, a second structural feature of the material to be predicted can be determined based on a first value, and then a target structural feature of the material to be predicted can be determined based on the first and second structural features, for example, by combining the first and second structural features to obtain the target structural feature. For instance, after calculating each first value, each first value can be summed to obtain the second structural feature of the material to be predicted. Right now Where, N atomThis represents the number of atoms other than the target atom i within the preset cutoff radius, such as the aforementioned first atom j and second atom k. It should be noted that the second structural feature... It indicates the atomic structure surrounding the target atom. Furthermore, the first structural feature mentioned here can be referenced to the first structural feature described above. The content will be understood in detail here, and will not be elaborated upon further.

[0086] In other examples, the first value and first distance information can be processed based on a preset Gaussian model to obtain the third structural feature of the material to be predicted. Then, the target structural feature of the material to be predicted can be determined based on this first and third structural feature; for example, the first structural feature can be combined with the second structural feature to obtain the target structural feature. It should be noted that the preset Gaussian model mentioned here is represented as follows: Where, N atom This represents the number of atoms other than the target atom i within the preset cutoff radius, such as the first atom j and the second atom k mentioned above. This is the third structural feature.

[0087] For example, in the process of determining the target structural features, the aforementioned first structural feature can also be included. Second structural features and third structural features By combining the features, the structural characteristics of the target can be determined.

[0088] It should be noted that the mechanism of action of the symmetry functions using Gaussian and truncation functions as filters mentioned above can be found in [reference needed]. Figure 5 Please refer to the diagram shown for clarification. Figure 5 The image shows a one-dimensional atomic chain, where atoms are scattered in a single direction, such as atom 0 (the aforementioned target atom), atom 1, atom 2, atom 3, and atom 4, etc. The adjustable parameter R... s By selecting different values ​​for η, λ, and ξ, we can obtain Figure 5 The Gaussian functions with different means and variances are shown. Figure 5 It can be seen that the distance between atom 1 and the extreme point of Gaussian function 1 is closer than that between atom 2 and the extreme point of Gaussian function 1. The value of Gaussian function 1 is larger at atom 1. Therefore, compared to atom 2, atom 1 contributes more to the symmetry function filtered by Gaussian function 1 and the cutoff function. Thus, the symmetry function filtered by Gaussian function 1 and the cutoff function extracts the structural information around the target atom, i.e., near atom 1. It represents the structural information around the target atom, within a certain distance R from the target atom. 01The number of atoms at a given location is determined by the Gaussian function. Similarly, Gaussian function 2 and Gaussian function 3 will also extract structural information at corresponding distances from the target atom. The larger the variance of the Gaussian function, the wider the range of distances from which it can extract structural information, but the less refined the representation.

[0089] Besides using the Gaussian function and truncation function mentioned above as the symmetry function constructed by the filter, the symmetry function constructed by the local cosine function can also be used to determine the target structural features. For example, in the process of determining the first value based on the first distance information and the preset truncation radius, the thickness information r of each layer of the target structure can be determined based on the preset truncation radius and the number of layers into which the target structure is divided. Among them, R c Let R be the preset cutoff radius, and N be the number of layers into which the target structure is divided. Then, based on this preset cutoff radius R... c The target distance d is determined by the thickness information r. m The target is at a distance of d m It can indicate the distance between the target atom and the boundary of each layer of the target structure after partitioning. Specifically, d m =d m-1 +r, where, when m=1, d m =0; when m=N, d m =R c Finally, the first value is determined based on the first distance information, the target distance, and the thickness information.

[0090] Similarly, as with determining the first value, the second value can be determined based on the second distance information, the target distance, and the thickness information. This second value indicates the contribution of the second atom to the adsorption energy. A third value can also be determined based on the third distance information, the target distance, and the thickness information. This third value indicates the contributions of both the second and first atoms to the adsorption energy.

[0091] Thus, determining the first structural feature of the material to be predicted based on the first value, the target angle information, the second distance information, and the third distance information can be achieved in the following way: the first value, the second value, the third value, and the target angle information are processed based on a preset cosine function model to obtain the first structural feature of the material to be predicted.

[0092] It should be noted that the first, second, and third values ​​mentioned in this example can be obtained using the local cosine function f. m The local cosine function f is calculated. m The form of expression can be understood by referring to the following formula (4), that is:

[0093]

[0094] Among them, Rij The distance between target atom i and the first atom j, i.e., the first distance information; d m Where r is the target distance, and f is the thickness information. m (R ij ) represents the first value.

[0095] Therefore, after obtaining the target distance d m Thickness information r and first distance information R ij After that, the d can be m r and R ij The local cosine function f shown in formula (4) above is input. m In this way, we can obtain the first value f. m (R ij Similarly, the obtained d can be... m r and R ik The local cosine function f shown in formula (4) above is input. m In this way, the second value f can be obtained. m (R ik ). and the obtained d m r and R jk The local cosine function f shown in formula (4) above is input. m In this way, the third value f can be obtained. m (R jk )f m (R jk ).

[0096] Thus, after calculating the first value f m (R ij ), second value f m (R ik ) and the third value f m (R jk Afterwards, the target angle information described above can be used as input information for a preset cosine function model, thereby calculating the first structural feature through the preset cosine function model. For example, the preset cosine function model is expressed as shown in formula (5) below, namely:

[0097]

[0098] in, This is represented as the first structural feature.

[0099] For example, the first value can also be processed based on a preset pre-function model to obtain the third structural features of the material to be predicted. Right now Where, N atomThis is expressed as the number of atoms other than the target atom i within the preset cutoff radius, such as the aforementioned first atom j and second atom k, etc. Then, based on the first structural feature... and third structural features The target structural features of the material to be predicted are determined. The described third structural feature... It indicates the structural configuration of the atoms surrounding the target atom.

[0100] It should be noted that the local cosine function mentioned above, used as a filter to construct a symmetric function, can be understood as a simplified form of the Gaussian function and the truncation function mentioned earlier used as filters to construct symmetric functions. The difference lies in the use of the local cosine function f. m When calculating the corresponding value, it is only necessary to consider the distance from the target atom (d). m-1 d m The atoms within the range are sufficient; it is not necessary to consider the preset cutoff radius R in symmetry functions constructed using Gaussian functions and cutoff functions as filters. c All atoms within. Furthermore, the local cosine function, used as a filter, is discontinuous at a predetermined cutoff radius, directly taking a value of 0, making its computation speed faster than that of the Gaussian function and the cutoff function used as filters.

[0101] It should be noted that the above mainly describes the construction of symmetry functions centered on doped atoms. In practical applications, symmetry functions can also be constructed centered on adsorbed molecules. The operation process is the same as that of constructing symmetry functions centered on doped atoms, and this application does not make specific limitations.

[0102] 405. Based on the preset machine learning model, the target structural features and elemental features are processed to obtain prediction labels. The prediction labels are used to indicate the adsorption energy of adsorbed molecules in the material to be predicted.

[0103] In this example, after obtaining the elemental features in step 402 and the target structural features in step 404, these target structural features and elemental features can be used as input information for a preset machine learning model to predict the corresponding label. For example, the material to be predicted can be characterized by the target structural features and elements. The actual material database can be represented in tabular form, with each row representing a feature of the material to be predicted. The number of samples in the entire dataset is represented by the number of rows in the table, and the number of columns in the table can be used to represent the number of features (i.e., target structural features and elemental features) used for each material to be predicted. In this way, the target structural features and elemental features of all the materials to be predicted can be used as input information for the preset machine learning model.

[0104] It should be noted that the specific number of structural features included in the target structural feature can be calculated by adjusting the aforementioned parameters Rs, η, λ, and ξ, resulting in different values ​​for the first, second, and third structural features. Similarly, by adjusting the value of the aforementioned parameter N, different values ​​for the first and third structural features can also be calculated. In this way, the first, second, and third structural features with different values ​​are combined to form the target structural feature. For example, R... s The values ​​of parameters such as η, λ, ζ, and N can be found in Table 1 below, i.e.:

[0105] Table 1. Parameter values ​​for several symmetry functions

[0106]

[0107] Where N2 and N3 are the N values ​​taken when extracting the third structural feature and the first structural feature, respectively, in the symmetric function constructed using the local cosine function as a filter.

[0108] Thus, as can be seen from the arrangement of parameters in Table 1, the symmetry function constructed using the Gaussian function and the truncation function as filters includes 1 second structural feature, 25 third structural features, and 50 first structural features, for a total of 76 structural features; the symmetry function constructed using the local cosine function as a filter includes 12 third structural features and 64 first structural features, for a total of 76 structural features.

[0109] For elemental information, the aforementioned elemental characteristics can be used for characterization, such as the 11 elemental characteristics. For example, for boron, its atomic number is 5 and its period number is 2, and so on, the elemental characteristics of the material to be predicted can be determined.

[0110] Therefore, regarding the aforementioned Figure 2 The single-atom catalyst shown can be characterized using 11 + 76 = 87 features, without considering the elemental characteristics of the matrix atoms. If, for this single-atom catalyst, the dopant atoms and adsorbed molecules are selected as the centers, and the corresponding surrounding target structural features are extracted, the number of target structural features becomes 152. Therefore, the single-atom catalyst can be described using 152 + 11 = 163 features. It should be noted that the above explanation also applies to the case considering the elemental characteristics of the matrix atoms; further details are omitted here.

[0111] The pre-defined machine learning model described may include, but is not limited to, gradient boosting regression (GBR) models or fully connected layers, etc., and this application does not specify any particular limitation.

[0112] This application also experimentally verifies the adsorption energy prediction method of this application, which can improve prediction accuracy and save computation time. Specifically:

[0113] For example, a dataset of single-atom catalysts using copper atoms as the matrix atoms and carbon monoxide as the adsorbed molecules was used. This dataset contains 75 different structures, each consisting of a copper crystal surface with five different crystal plane indices, different carbon monoxide adsorption sites, and different dopant atom sites. The minimum repeating unit cell contains at least 38 atoms (35 copper atoms + 1 dopant atom + 1 carbon monoxide molecule). After structural expansion of this minimum repeating unit cell, the minimum and maximum number of atoms are 326 and 326 respectively. For each structure, the dopant atom sites were replaced with 41 different dopant elements, resulting in a total of 3075 samples. Each sample was described using either the 87 or 163 features mentioned above, forming a sample matrix of size 3075×87 or 3075×163.

[0114] First, we tested the time consumed in constructing structural features using the three symmetry functions mentioned above in this application (i.e., Localized cos, Gaussian-cos, and Gaussian-tanh). The test results are shown in Table 2 below:

[0115] Table 2. Time consumed in constructing structural features

[0116]

[0117]

[0118] Since the initial structure (i.e., the smallest repeating unit cell) is used for prediction, these 3075 samples actually contain only 75 different structures. Therefore, it is only necessary to construct the structural features of these 75 samples with different structures, and then copy them 40 times to obtain the structural features of all samples. It can be seen that for the shortest method, it takes an average of only 0.3 seconds to construct the structural features of one sample, and the longest time is only 12.6 seconds.

[0119] Based on this, the adsorption energy of carbon monoxide was further predicted using these structural features, and compared with the prediction results of a series of graph neural networks in existing schemes on this dataset. For each algorithm, the training set:validation set:test set = 6:2:2, and the dataset was divided 10 times using 10 random number seeds from 0 to 9, with 10 tests performed for each, and the MAE of the 10 prediction results was averaged. During training, the training set labels were normalized using the mean and standard deviation of the training set labels, so the predicted values ​​output by the model need to be denormalized to obtain their true prediction values. For the three models that predict based on the preset machine learning model GBR, their input features were standardized, and this table shows the results of material characterization using 163 features (i.e., constructing structural features centered on adsorption atoms and dopant atoms, respectively). The performance of different algorithms and training time are shown in Table 3 below:

[0120] Table 3 shows the test results of the symmetry function-based model and the classical graph neural network on the single-atom catalyst dataset.

[0121]

[0122]

[0123] As shown in Table 3 above, the machine learning model built based on the structural features extracted by symmetry functions outperforms the SchNet and CGCNN models, exhibiting better stability and less susceptibility to poor performance under certain partitioning methods. Furthermore, although the performance of the machine learning model based on the structural features extracted by symmetry functions is inferior to DimeNet and DimeNet++, its training time is significantly shorter than that of DimeNet and DimeNet++. Moreover, its testing accuracy has reached the requirement of an absolute error within the range of 0.1-0.15 eV, making it suitable for quickly determining the adsorption energy of a single-atom catalyst for a specific adsorbed molecule, thereby assessing its suitability for catalyzing related chemical reactions.

[0124] Furthermore, the structural features extracted based on symmetry functions are discussed. Table 4 below shows the results of predicting the adsorption energy using GBR as the preset machine learning model when extracting features using only dopant atoms (i.e., extracting 76 target structural features and 11 elemental features, for a total of 87 features), extracting features using both dopant atoms and adsorbed molecules (i.e., 163 features), and extracting features using only adsorbed molecules (i.e., 87 features).

[0125] Table 4. Results of adsorption energy prediction using different features.

[0126]

[0127]

[0128] As shown in Table 4 above, the position of the dopant atom plays a crucial role in predicting the adsorption energy of single-atom catalysts. Therefore, by expanding the structure around the dopant atom and extracting structural features using the aforementioned symmetry function, and then using the extracted target structural and elemental features as input to a pre-defined machine learning model, the prediction accuracy of the adsorption energy can be improved. Furthermore, by saving computation time and improving the prediction accuracy of the adsorption energy of adsorbed molecules, this approach can further accelerate the development of new materials, shorten the development time, and ultimately create new materials that meet specific needs.

[0129] In this embodiment, the target structure and elemental characteristics of the material to be predicted are obtained. The target structure represents the atomic structure obtained from the first structure of the material to be predicted with the target atom as the center and a preset cutoff radius. The first structure is obtained by symmetrically expanding the smallest repeating unit cell of the material to be predicted. Then, target distance information and target angle information are obtained based on the target structure of the material to be predicted, and the target structural characteristics of the material to be predicted are determined based on the target distance information and the target angle information. The target distance information is used to indicate the distance between the target atom and each atom, and the target angle information is used to indicate the angle formed between the target atom and the first atom, and between the target atom and the second atom. Finally, the target structural characteristics and elemental characteristics are processed based on a preset machine learning model to obtain a prediction label, which is used to indicate the adsorption energy of the adsorbed molecules in the material to be predicted. In the above manner, the first structure is obtained after symmetrically expanding the smallest repeating unit cell of the material to be predicted, and the atomic structure is obtained from the first structure with the target atom as the center and a preset cutoff radius. This atomic structure is used as the target structure of the material to be predicted. Then, the target distance and angle information between the target atom and other atoms are directly determined from the target structure. This target distance and angle information are used to determine the target structural features, which, combined with elemental features, can be used as input to a pre-defined machine learning model. This means that the corresponding target structural features can be directly determined from the target structure of the material before relaxation, eliminating the need for time-consuming first-principles calculations to relax the target structure and obtain input for the pre-defined machine learning model, thus saving computation time. Furthermore, symmetric expansion of the smallest repeating unit cell ensures that the atomic positions in the expanded first structure remain unchanged, and consequently, the atomic positions in the final target structure also remain unchanged. Therefore, considering the distances and angles between atoms in the target structure to obtain the target structural features improves the accuracy of structural feature extraction. This enhances the prediction accuracy of the adsorption energy of adsorbed molecules in the material during subsequent model prediction, thereby accelerating the development of new materials.

[0130] The foregoing primarily describes the solutions provided by the embodiments of this application from a methodological perspective. It is understood that to achieve the above functions, corresponding hardware structures and / or software modules are included to execute each function. Those skilled in the art should readily recognize that, based on the modules and algorithm steps described in conjunction with the embodiments disclosed in this application, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0131] This application embodiment can divide the device into functional modules according to the above method example. For example, each function can be divided into its own functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0132] The adsorption energy prediction device in the embodiments of this application will be described in detail below. Figure 6 This is a schematic diagram of the adsorption energy prediction device provided in the embodiments of this application. Figure 6 As shown, the adsorption energy prediction device may include an acquisition unit 601 and a processing unit 602.

[0133] The acquisition unit 601 is used to acquire the target structure and elemental characteristics of the material to be predicted. The target structure represents the range of atomic structures obtained from the first structure of the material to be predicted, centered on the target atom and with a preset cutoff radius. The first structure is obtained by symmetric expansion of the smallest repeating unit cell of the material to be predicted. The acquisition unit 601 is used to acquire target distance information and target angle information based on the target structure of the material to be predicted. The target distance information indicates the distance between the target atom and each atom, and the target angle information indicates the angle formed between the target atom and the first atom, and between the target atom and the second atom. The first atom can be any atom within the target structure, or any atom within a preset radius centered on the target atom. The preset radius is obtained from the preset cutoff radius. The processing unit 602 is used to determine the target structural characteristics of the material to be predicted based on the target distance information and the target angle information. The processing unit 602 is used to process the target structural characteristics and elemental characteristics based on a preset machine learning model to obtain prediction labels, which indicate the adsorption energy of the material to be predicted.

[0134] In some optional examples, the processing unit 602 is configured to: determine a first value based on first distance information and a preset cutoff radius, wherein the first value indicates the contribution of the first atom to the adsorption energy and the first distance information indicates the distance between the target atom and the first atom; determine a first structural feature of the material to be predicted based on the first value, target angle information, second distance information, and third distance information, wherein the first structural feature indicates the atomic structural relationship between the target atom, the first atom, and the second atom, the second distance information indicates the distance between the target atom and the second atom, and the third distance information indicates the distance between the first atom and the second atom; and determine a target structural feature of the material to be predicted based on the first structural feature.

[0135] In some alternative examples, the processing unit 602 is used to: determine a second value based on the second distance information and a preset cutoff radius, the second value being used to indicate the contribution of the second atom to the adsorption energy; determine a third value based on the third distance information and a preset cutoff radius, the third value being used to indicate the contribution of the first atom and the second atom to the adsorption energy; and process the first value, target angle information, first distance information, second distance information, third distance information, second value, and third value based on a preset Gaussian model to obtain the first structural feature of the material to be predicted.

[0136] In some alternative examples, processing unit 602 is further configured to: determine a second structural feature of the material to be predicted based on a first value, the second structural feature indicating the atomic structure surrounding the target atom. Processing unit 602 is configured to determine the target structural feature of the material to be predicted based on the first structural feature and the second structural feature.

[0137] In some alternative examples, the processing unit 602 is also configured to: process the first value and the first distance information based on a preset Gaussian model to obtain the third structural feature of the material to be predicted, the third structural feature being used to indicate the atomic structural relationship between the target atom and the first atom; and determine the target structural feature of the material to be predicted based on the first structural feature and the third structural feature.

[0138] In some alternative examples, the processing unit 602 is also configured to: determine a second structural feature of the material to be predicted based on a first value, the second structural feature being used to indicate the atomic structure around the target atom; and determine a target structural feature of the material to be predicted based on the first structural feature, the second structural feature, and the third structural feature.

[0139] In some alternative examples, the processing unit 602 is used to sum each first value to obtain a second structural feature of the material to be predicted.

[0140] In some alternative examples, the processing unit 602 is used to: determine the thickness information of each layer of the target structure based on a preset cutoff radius and the number of layers into which the target structure is divided; determine the target distance based on the preset cutoff radius and the thickness information, the target distance being used to indicate the distance between the target atom and the boundary of each layer of the target structure; and determine a first value based on the first distance information, the target distance, and the thickness information.

[0141] In some alternative examples, the processing unit 602 is used to: determine a second value based on the second distance information, the target distance, and the thickness information, the second value indicating the contribution of the second atom to the adsorption energy; determine a third value based on the third distance information, the target distance, and the thickness information, the third value indicating the contribution of the first atom and the second atom to the adsorption energy; and process the first value, the second value, the third value, and the target angle information based on a preset cosine function model to obtain the first structural feature of the material to be predicted.

[0142] In some alternative examples, the processing unit 602 is further configured to: process the first value based on a preset cosine function model to obtain a third structural feature of the material to be predicted, wherein the third structural feature is used to indicate the atomic structural relationship between the target atom and the first atom. The processing unit 602 is configured to determine the target structural feature of the material to be predicted based on the first structural feature and the third structural feature.

[0143] In some alternative examples, the atoms include dopant atoms and at least two matrix atoms. The acquisition unit 601 is used to: acquire the elemental characteristics of the dopant atoms when the atom type of each of the at least two matrix atoms is the same as the atom type of the matrix atoms in the remaining structure, thereby acquiring the elemental characteristics of the material to be predicted.

[0144] In some alternative examples, the acquisition unit 601 is also used to: acquire the elemental characteristics of the doped atoms and the elemental characteristics of each matrix atom when the atom type of any matrix atom in at least two matrix atoms is different from the atom type of the matrix atoms in the remaining structure, so as to acquire the elemental characteristics of the material to be predicted.

[0145] In some alternative examples, the acquisition unit 601 is further configured to: acquire the minimum repeating unit cell of the material to be predicted, wherein the minimum repeating unit cell includes dopant atoms, matrix atoms, and adsorbed molecules. The processing unit 602 is configured to: determine a first crystal structure based on the minimum repeating unit cell, wherein the dopant atoms in the first crystal structure are replaced by matrix atoms, and the first crystal structure does not include adsorbed molecules; and extend the first crystal structure symmetrically in a preset first direction and a preset second direction, respectively, with the target atom as the center, to obtain the first structure.

[0146] In some alternative examples, the target atom may include a dopant atom, or the target atom may include an adsorbed molecule.

[0147] The above describes the adsorption energy prediction device in the embodiments of this application from the perspective of modular functional entities. The following describes the adsorption energy prediction device in the embodiments of this application from the perspective of hardware processing. Figure 7This is a schematic diagram of the adsorption energy prediction device provided in the embodiments of this application. The adsorption energy prediction device can vary considerably due to differences in configuration or performance. The adsorption energy prediction device may include at least one processor 701, a communication line 707, a memory 703, and at least one communication interface 704.

[0148] The processor 701 may be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (server IC), or one or more integrated circuits used to control the execution of the program of the present application.

[0149] Communication line 707 may include a path for transmitting information between the aforementioned components.

[0150] Communication interface 704 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.

[0151] The memory 703 can be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions. The memory can exist independently and be connected to the processor via communication line 707. The memory can also be integrated with the processor.

[0152] The memory 703 stores computer execution instructions for implementing the scheme of this application, and the processor 701 controls the execution. The processor 701 executes the computer execution instructions stored in the memory 703, thereby realizing the adsorption energy prediction method provided in the above embodiments of this application.

[0153] Optionally, the computer execution instructions in the embodiments of this application may also be referred to as application code, and the embodiments of this application do not specifically limit this.

[0154] In a specific implementation, as one example, the adsorption energy prediction device may include multiple processors, such as... Figure 7Processors 701 and 702 are described herein. Each of these processors may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. A processor here may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0155] In a specific implementation, as one embodiment, the adsorption energy prediction device may further include an output device 705 and an input device 706. The output device 705 communicates with the processor 701 and can display information in various ways. The input device 706 communicates with the processor 701 and can receive input from the target object in various ways. For example, the input device 706 may be a mouse, a touch screen device, or a sensing device, etc.

[0156] The aforementioned adsorption energy prediction device can be a general-purpose device or a dedicated device. In specific implementations, the adsorption energy prediction device can be a server, terminal, or other similar device. Figure 7 Devices with similar structures. The embodiments of this application do not limit the type of adsorption energy prediction device.

[0157] It should be noted that Figure 7 The processor 701 can invoke computer execution instructions stored in the memory 703 to cause the adsorption energy prediction device to perform actions such as... Figure 4 The method in the corresponding method embodiment.

[0158] Specifically, Figure 6 The function / implementation process of the processing unit 602 can be achieved through... Figure 7 The processor 701 in the memory calls computer execution instructions stored in the memory 703 to implement the function. Figure 6 The function / implementation process of the acquisition unit 601 can be achieved through... Figure 7 It is implemented using the 704 communication interface.

[0159] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0160] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0161] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0162] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0163] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0164] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0165] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, they can be implemented in whole or in part in the form of a computer program product.

[0166] A computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, they generate, in whole or in part, the processes or functions according to embodiments of this application. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions may be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., SSDs), etc.

[0167] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for predicting adsorption energy, characterized in that, include: The target structure and elemental characteristics of the material to be predicted are obtained. The target structure is used to represent the atomic structure obtained from the first structure of the material to be predicted with the target atom as the center and a preset cutoff radius. The first structure is obtained by obtaining the minimum repeating unit cell of the material to be predicted. The minimum repeating unit cell includes dopant atoms, matrix atoms, and adsorbed molecules. The dopant atoms and matrix atoms are different types of elements. A first crystal structure is determined based on the minimum repeating unit cell, wherein the doped atoms in the first crystal structure are replaced by the matrix atoms, and the first crystal structure does not include the adsorbed molecules; Centered on the target atom, the first crystal structure is symmetrically extended in a preset first direction and a preset second direction to obtain the first structure; the target atom includes the dopant atom; the elemental characteristics include the elemental characteristics of the dopant atom; Target distance information and target angle information are obtained based on the target structure of the material to be predicted. The target distance information is used to indicate the distance between the target atom and each atom. The target angle information is used to indicate the angle between the target atom and the first atom and the target atom and the second atom. The first atom is any atom within the target structure, or the first atom is any atom within a preset radius centered on the target atom. The preset radius is obtained from the preset cutoff radius. Determining the target structural features of the material to be predicted based on the target distance information and the target angle information includes: determining a first value based on first distance information and a preset cutoff radius, wherein the first value indicates the contribution of the first atom to the adsorption energy, and the first distance information indicates the distance between the target atom and the first atom; determining a first structural feature of the material to be predicted based on the first value, the target angle information, the second distance information, and the third distance information, wherein the first structural feature indicates the atomic structural relationship between the target atom, the first atom, and the second atom, the second distance information indicates the distance between the target atom and the second atom, and the third distance information indicates the distance between the first atom and the second atom; and determining the target structural features of the material to be predicted based on the first structural feature. The target structural features and elemental features are processed based on a preset machine learning model to obtain a prediction label, which is used to indicate the adsorption energy of adsorbed molecules in the material to be predicted.

2. The method according to claim 1, characterized in that, The step of determining the first structural feature of the material to be predicted based on the first value, the target angle information, the second distance information, and the third distance information includes: A second value is determined based on the second distance information and the preset cutoff radius. The second value is used to indicate the contribution of the second atom to the adsorption energy. A third value is determined based on the third distance information and the preset cutoff radius. The third value is used to indicate the contribution of the first atom and the second atom to the adsorption energy. Based on a preset Gaussian model, the first value, the target angle information, the first distance information, the second distance information, the third distance information, the second value, and the third value are processed to obtain the first structural feature of the material to be predicted.

3. The method according to claim 2, characterized in that, The method further includes: Based on the first value, a second structural feature of the material to be predicted is determined, wherein the second structural feature is used to indicate the atomic structure around the target atom; Determining the target structural features of the material to be predicted based on the first structural feature includes: The target structural features of the material to be predicted are determined based on the first structural feature and the second structural feature.

4. The method according to claim 2, characterized in that, The method further includes: Based on a preset Gaussian model, the first value and the first distance information are processed to obtain the third structural feature of the material to be predicted. The third structural feature is used to indicate the atomic structure relationship between the target atom and the first atom. Determining the target structural features of the material to be predicted based on the first structural feature includes: The target structural features of the material to be predicted are determined based on the first structural feature and the third structural feature.

5. The method according to claim 4, characterized in that, The method further includes: Based on the first value, a second structural feature of the material to be predicted is determined, wherein the second structural feature is used to indicate the atomic structure around the target atom; Determining the target structural features of the material to be predicted based on the first structural feature and the third structural feature includes: The target structural features of the material to be predicted are determined based on the first structural feature, the second structural feature, and the third structural feature.

6. The method according to claim 3 or 5, characterized in that, Determining the second structural feature of the material to be predicted based on the first value includes: Summing each of the first values ​​yields the second structural feature of the material to be predicted.

7. The method according to claim 1, characterized in that, Determining the first value based on the first distance information and the preset cutoff radius includes: The thickness information of each layer of the target structure is determined based on the preset cutoff radius and the number of layers into which the target structure is divided. The target distance is determined based on the preset cutoff radius and the thickness information, and the target distance is used to indicate the distance between the target atom and the boundary of each layer of the target structure; The first value is determined based on the first distance information, the target distance, and the thickness information.

8. The method according to claim 7, characterized in that, Based on the first value, the target angle information, the second distance information, and the third distance information, the first structural feature of the material to be predicted is determined, including: A second value is determined based on the second distance information, the target distance, and the thickness information. The second value is used to indicate the contribution of the second atom to the adsorption energy. A third value is determined based on the third distance information, the target distance, and the thickness information. The third value is used to indicate the contribution of the first atom and the second atom to the adsorption energy. The first value, the second value, the third value, and the target angle information are processed based on a preset cosine function model to obtain the first structural feature of the material to be predicted.

9. The method according to claim 7 or 8, characterized in that, The method further includes: The first value is processed based on a preset cosine function model to obtain the third structural feature of the material to be predicted. The third structural feature is used to indicate the atomic structure relationship between the target atom and the first atom. Determining the target structural features of the material to be predicted based on the first structural feature includes: The target structural features of the material to be predicted are determined based on the first structural feature and the third structural feature.

10. The method according to any one of claims 1 to 5, 7 to 8, characterized in that, The atoms in the target structure include dopant atoms and at least two matrix atoms. The elemental characteristics of the material to be predicted are obtained, including: When the atom type of each of the at least two matrix atoms is the same as the atom type of the matrix atoms in the remaining structure, the elemental characteristics of the doped atoms are obtained to obtain the elemental characteristics of the material to be predicted.

11. The method according to claim 10, characterized in that, The method further includes: When the atom type of any of the at least two matrix atoms is different from the atom type of the matrix atoms in the remaining structure, the elemental characteristics of the doped atom and the elemental characteristics of each matrix atom are obtained to obtain the elemental characteristics of the material to be predicted.

12. The method according to claim 1, characterized in that, The target atom also includes the adsorbed molecule.

13. An adsorption energy prediction device, characterized in that, include: An acquisition unit is used to acquire the target structure and elemental characteristics of the material to be predicted. The target structure is used to represent the range of atomic structures obtained from the first structure of the material to be predicted with the target atom as the center and a preset cutoff radius. The first structure is obtained by acquiring the minimum repeating unit cell of the material to be predicted. The minimum repeating unit cell includes dopant atoms, matrix atoms, and adsorbed molecules. The dopant atoms and matrix atoms are different types of elements. A first crystal structure is determined based on the minimum repeating unit cell, wherein the doped atoms in the first crystal structure are replaced by the matrix atoms, and the first crystal structure does not include the adsorbed molecules; Centered on the target atom, the first crystal structure is symmetrically extended in a preset first direction and a preset second direction to obtain the first structure; the target atom includes the dopant atom; the elemental characteristics include the elemental characteristics of the dopant atom; The acquisition unit is used to acquire target distance information and target angle information based on the target structure of the material to be predicted. The target distance information is used to indicate the distance between the target atom and each atom. The target angle information is used to indicate the angle between the target atom and the first atom and between the target atom and the second atom. The first atom is any atom within the target structure, or the first atom is any atom within a preset radius centered on the target atom. The preset radius is obtained from the preset cutoff radius. A processing unit is configured to determine the target structural features of the material to be predicted based on the target distance information and the target angle information, including: determining a first value based on first distance information and a preset cutoff radius, wherein the first value indicates the contribution of the first atom to the adsorption energy, and the first distance information indicates the distance between the target atom and the first atom; determining a first structural feature of the material to be predicted based on the first value, the target angle information, the second distance information, and the third distance information, wherein the first structural feature indicates the atomic structural relationship between the target atom, the first atom, and the second atom, the second distance information indicates the distance between the target atom and the second atom, and the third distance information indicates the distance between the first atom and the second atom; and determining the target structural features of the material to be predicted based on the first structural feature. The processing unit is used to process the target structural features and the elemental features based on a preset machine learning model to obtain a prediction label, which is used to indicate the adsorption energy of the material to be predicted.

14. The apparatus according to claim 13, characterized in that, The atoms in the target structure include dopant atoms and at least two matrix atoms, and the acquisition unit is specifically used for: When the atom type of each of the at least two matrix atoms is the same as the atom type of the matrix atoms in the remaining structure, the elemental characteristics of the doped atoms are obtained to obtain the elemental characteristics of the material to be predicted.

15. An adsorption energy prediction device, characterized in that, include: Input / output (I / O) interface, processor, and memory, wherein program instructions are stored in the memory; The processor is configured to execute program instructions stored in the memory to perform the method as described in any one of claims 1 to 12.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes instructions that, when executed on a computer device, cause the computer device to perform the method as described in any one of claims 1 to 12.

17. A computer program product, characterized in that, The computer program product includes instructions that, when executed on a computer device, cause the computer device to perform the method as described in any one of claims 1 to 12.

Images