Crystal property prediction method, device, electronic device and storage medium

Abstract

Embodiments of the present invention disclose a crystal property prediction method, device, electronic device, and storage medium. The method includes: obtaining the atomic positions and atomic types of initial atoms in a target crystal, and obtaining a trained target crystal property prediction model; determining a target atomic path based on the atomic positions of the initial atoms and a preset cutoff radius, and determining the path atomic distance of the target atomic path; determining the target path features corresponding to the target atomic paths based on the atomic positions and atomic types of the initial atoms; inputting the target path features and path atomic distances corresponding to the target atomic paths into the target crystal property prediction model, and determining the property prediction results of the target crystal based on the output results of the target crystal property prediction model. The technical solution of the embodiments of the present invention can improve the accuracy of crystal property prediction.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of machine learning, and in particular to a crystal property prediction method, device, electronic device, and storage medium. Background Art

[0002] The prediction of crystal properties is very important for the screening and discovery of materials.

[0003] However, the existing crystal property prediction methods have the problem of low accuracy, which needs to be solved. Summary of the Invention

[0004] Embodiments of the present invention provide a crystal property prediction method, device, electronic device, and storage medium to improve the accuracy of crystal property prediction.

[0005] According to one aspect of the present invention, a method for predicting crystal properties is provided, which may include:

[0006] Obtaining the atomic positions and atomic types of initial atoms in the target crystal, and obtaining a trained target crystal property prediction model;

[0007] Determining a target atomic path constructed based on the initial atoms according to the atomic position of the initial atoms and a preset cutoff radius, and determining a path atomic distance between path atoms on each target atomic path in the target atomic path;

[0008] Determine target path features corresponding to target atomic paths based on the atomic positions and atomic types of the initial atoms;

[0009] The target path characteristics and path atomic distances corresponding to the target atomic paths are input into the target crystal property prediction model, and the property prediction results of the target crystal are determined according to the output results of the target crystal property prediction model.

[0010] According to another aspect of the present invention, a crystal property prediction device is provided, which may include:

[0011] A target crystal property prediction model acquisition module is used to obtain the atomic positions and atomic types of initial atoms in the target crystal and to obtain a trained target crystal property prediction model;

[0012] a path atom distance determination module, configured to determine a target atom path constructed based on the initial atoms according to the atomic position of the initial atoms and a preset cutoff radius, and to determine the path atom distance between the path atoms on each target atom path in the target atom path;

[0013] A target path feature determination module is used to determine target path features corresponding to target atomic paths based on the atomic positions and atomic types of the initial atoms;

[0014] The property prediction result determination module is used to input the target path characteristics and path atomic distances corresponding to the target atomic paths into the target crystal property prediction model, and determine the property prediction results of the target crystal based on the output results of the target crystal property prediction model.

[0015] According to another aspect of the present invention, an electronic device is provided, which may include:

[0016] at least one processor; and

[0017] a memory communicatively connected to at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by at least one processor. The computer program is executed by the at least one processor so that the at least one processor implements the crystal property prediction method provided by any embodiment of the present invention when executing the computer program.

[0019] According to another aspect of the present invention, a computer-readable storage medium is provided, on which computer instructions are stored. The computer instructions are used to enable a processor to implement the crystal property prediction method provided by any embodiment of the present invention when executed.

[0020] The technical solution of the embodiment of the present invention obtains the atomic position and atomic type of the initial atoms in the target crystal, and obtains a trained target crystal property prediction model; determines the target atomic path constructed based on the initial atoms according to the atomic position of the initial atoms and a preset cutoff radius, and determines the path atomic distance between the path atoms on each target atomic path in the target atomic path; determines the target path features corresponding to the target atomic paths based on the atomic position and atomic type of the initial atoms; inputs the target path features and path atomic distance corresponding to the target atomic paths into the target crystal property prediction model, and determines the property prediction result of the target crystal according to the output result of the target crystal property prediction model. The technical solution of the embodiment of the present invention takes into account that the distance between the atoms constituting the crystal will also affect the properties of the crystal, and thus determines the property prediction result of the target crystal based on the path atomic distance, so that the predicted property prediction result is more accurate, thereby improving the accuracy of the crystal property prediction.

[0021] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of a crystal property prediction method provided in Example 1 of the present invention;

[0023] Figure 2 This is a flow chart of a crystal property prediction method provided in the second embodiment of the present invention;

[0024] Figure 3 This is a flow chart of a crystal property prediction method provided in Example 3 of the present invention;

[0025] Figure 4 This is a flow chart of a crystal property prediction method provided in the fourth embodiment of the present invention;

[0026] Figure 5 is a flowchart of an optional example of a crystal property prediction method provided in the fourth embodiment of the present invention;

[0027] Figure 6 is a structural block diagram of a crystal property prediction device provided in a fifth embodiment of the present invention;

[0028] Figure 7 It is a schematic structural diagram of an electronic device for implementing the crystal property prediction method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. The situations of "target", "original", etc. are similar and will not be repeated here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0031] Example 1

[0032] Figure 1 This is a flow chart of a crystal property prediction method provided in the first embodiment of the present invention. This embodiment is applicable to crystal property prediction. The method can be performed by the crystal property prediction device provided in the embodiment of the present invention. The device can be implemented in software and / or hardware and can be integrated into an electronic device, such as various user terminals or servers.

[0033] See also Figure 1 The method of the embodiment of the present invention specifically includes the following steps:

[0034] S110 , obtaining the atomic positions and atomic types of initial atoms in the target crystal, and obtaining a trained target crystal property prediction model.

[0035] Among them, the target crystal is the crystal for which property prediction is required. The target crystal is composed of at least one atom. The atom mentioned in the embodiment of the present invention refers to a microscopic material unit such as an atom, an ion or a molecule that can constitute a crystal.

[0036] Specifically, the atomic position and atomic type of at least one initial atom in the target crystal may be obtained, and a trained target crystal property prediction model may be obtained.

[0037] In an embodiment of the present invention, a target crystal can be periodically expanded according to its structure, and at least one atom obtained after the periodic expansion of the target crystal can be selected as an initial atom as needed or randomly. The at least one initial atom includes atoms within at least one complete period of the periodic expansion of the target crystal. For example, after the periodic expansion of the target crystal, atoms within a complete period of the target crystal can be randomly selected, and then at least one atom or at least one period of atoms can be selected from adjacent periods of the complete period, and each of the selected atoms can be used as the initial atom.

[0038] In an embodiment of the present invention, the atomic position and atomic type of at least one initial atom in the target crystal are obtained. The atomic position can be the coordinates of the atom. For example, the target crystal can be placed in a coordinate system to obtain the coordinates of the atoms constituting the target crystal. For another example, a coordinate system can be constructed with a certain atom on the target crystal as the coordinate origin, and then the coordinates of the atoms constituting the target crystal in the coordinate system are obtained. The atomic position can also be the relative orientation and distance of the atom relative to a certain reference object. In an embodiment of the present invention, the type of the atomic position is not specifically limited. The atomic type can be the element type of the atom, for example, the atomic type can be an oxygen type, an aluminum type, or a silicon type, etc. The atomic type can also be the particle type of the atom, for example, the atomic type can be a sodium ion type or a silicon atom type, etc.

[0039] In an embodiment of the present invention, a target crystal property prediction model capable of predicting properties of a target crystal can be pre-trained. The target crystal property prediction model can, for example, predict properties such as potential energy properties, melting point properties, or density properties of the target crystal. In an embodiment of the present invention, the properties of the target crystal that can be predicted are not specifically limited. In an embodiment of the present invention, the type of predicted property of the target crystal can be determined by different parameters of the target crystal property prediction model and / or different types of training samples used to train the target crystal property prediction model.

[0040] In an embodiment of the present invention, a target crystal property prediction model is pre-trained through the following steps: obtaining the atomic position and atomic type of at least one sample initial atom in the sample crystal, and obtaining the attribute label of the sample crystal; determining at least one sample atomic path constructed based on the at least one sample initial atom according to the atomic position and sample cutoff radius of the at least one sample initial atom, and determining the sample path atomic distance between at least one sample path atom on each sample atomic path in the at least one sample atomic path; determining the sample path features corresponding to the at least one sample atomic path respectively based on the atomic position and atomic type of the at least one sample initial atom; taking the sample path features, sample path atomic distance and attribute label of the sample crystal corresponding to the at least one sample atomic path respectively as a group of training samples; based on multiple groups of training samples, training the original crystal property prediction model to be trained that can predict the target crystal properties to obtain the target crystal property prediction model.

[0041] For example, in an embodiment of the present invention, the loss function corresponding to the target crystal property prediction model may use the mean absolute error, that is:

[0042]

[0043] Among them, Out i It can be understood as the output of the model; target i It can be understood as the attribute label of the sample crystal; l can be understood as the average error margin of the output result; m can be understood as the number of samples; i can be understood as the i-th sample.

[0044] S120 , determining a target atomic path constructed based on the initial atoms according to the atomic position of the initial atoms and a preset cutoff radius, and determining a path atomic distance between path atoms on each target atomic path in the target atomic path.

[0045] The preset cutoff radius can be understood as a pre-set distance value. It can be understood that the energy and other properties of each atom in the atoms constituting the target crystal may be determined by other atoms on its corresponding path. For example, the energy E of atom i i =E(R i , P R (i)), where R i For atom i, P R (i) represents the other atoms on the path corresponding to atom i, and E(.) represents the model corresponding to energy. Since the properties of atoms may not affect each other after the distance between atoms exceeds a certain value, a preset truncation radius can be set according to requirements or properties such as crystal density to construct at least one target atomic path based on the preset truncation radius.

[0046] Specifically, based on the atomic position of at least one initial atom and a preset cutoff radius, at least one target atomic path constructed based on at least one initial atom can be determined, and the path atomic distance between at least one path atom on each target atomic path in at least one target atomic path can be determined.

[0047] In an embodiment of the present invention, for each target atom path in at least one target atom path, the target atom path corresponds to an initial atom in at least one initial atom, and the target atom path may include at least one path atom determined from at least one initial atom. For each path atom in at least one path atom, the distance between the path atom and the path atoms before and after it on the target atom path is not greater than a preset cutoff radius. The target atom path includes the initial atom corresponding to the target atom path, and the initial atom is the first path atom on the target atom path. The number of target atom paths is the same as the number of atoms in a complete cycle after the target crystal is periodically unfolded. That is, the target atom path can be understood as a path that can determine the energy and other properties of the initial atom corresponding to the target atom path. The path atom can be understood as the initial atom located on the target atom path, and can determine the atom of the initial atom corresponding to the target atom path where it is located. It should be noted that the same initial atom can serve as a path atom for different target atom paths.

[0048] In an embodiment of the present invention, at least one target atomic path constructed based on at least one initial atom can be determined according to the atomic position of at least one initial atom and a preset cutoff radius, and the path atom distance between at least one path atom on each target atom path in at least one target atom path can be determined. The path atom distance can be understood as the distance between each path atom in at least one path atom and the path atoms before and after it on the target atom path where it is located.

[0049] S130 : Determine target path features corresponding to target atomic paths based on the atomic positions and atomic types of the initial atoms.

[0050] Among them, the target path characteristics can be understood as the characteristics of the target atomic path.

[0051] Specifically, target path features corresponding to at least one target atomic path may be determined based on the atomic position and atomic type of at least one initial atom.

[0052] S140 , inputting the target path features and the path atom distances respectively corresponding to the target atomic paths into a target crystal property prediction model, and determining the property prediction result of the target crystal according to the output result of the target crystal property prediction model.

[0053] Specifically, the target path characteristics and path atomic distance corresponding to at least one target atomic path can be input into the target crystal property prediction model, and the property prediction result of the target crystal can be determined according to the output result of the target crystal property prediction model.

[0054] In an embodiment of the present invention, the target path characteristics and path atom distances corresponding to at least one target atomic path can be input into a target crystal property prediction model, and the property prediction results of the target crystal can be determined based on the output results of the target crystal property prediction model. It should be noted that the output results of the target crystal property prediction model may not be directly the property prediction results of the target crystal. For example, the output results can be the probability values ​​of different property results of the target crystal, and the probability values ​​of the property results can be compared to determine the property prediction results of the target crystal based on the comparison results. In an embodiment of the present invention, there is no specific limitation on the content type of the output results of the target crystal property prediction model.

[0055] The technical solution of the embodiment of the present invention obtains the atomic position and atomic type of the initial atoms in the target crystal, and obtains a trained target crystal property prediction model; determines the target atomic path constructed based on the initial atoms according to the atomic position of the initial atoms and a preset cutoff radius, and determines the path atomic distance between the path atoms on each target atomic path in the target atomic path; determines the target path features corresponding to the target atomic paths based on the atomic position and atomic type of the initial atoms; inputs the target path features and path atomic distance corresponding to the target atomic paths into the target crystal property prediction model, and determines the property prediction result of the target crystal according to the output result of the target crystal property prediction model. The technical solution of the embodiment of the present invention takes into account that the distance between the atoms constituting the crystal will also affect the properties of the crystal, and thus determines the property prediction result of the target crystal based on the path atomic distance, so that the predicted property prediction result is more accurate, thereby improving the accuracy of the crystal property prediction.

[0056] An optional technical solution determines the target path features corresponding to the target atomic paths respectively based on the atomic position and atomic type of the initial atom, including: for each target atomic path in the target atomic path and each path atom in the path atoms on the target atomic path, determining the periodic atomic distance between the path atom and other atoms according to the atomic position of the initial atom, and determining the initial atomic features of the path atom according to the periodic atomic distance and the feature extraction function, wherein the other atoms include the initial atoms other than the path atom in the initial atoms within the period in which the path atom is located; determining the atomic type of the path atom according to the atomic type of the initial atom, and determining the representation vector of the path atom from the target matrix according to the atomic type of the path atom, wherein the target matrix includes the representation vectors corresponding to the atomic types of the initial atoms respectively; determining the target atomic features of the path atom according to the initial atomic features and the representation vector; determining the target path features of the target atomic path according to the target atomic features corresponding to the path atoms respectively.

[0057] Specifically, based on the atomic position and atomic type of at least one initial atom, target path characteristics corresponding to at least one target atomic path are determined, including: for each target atomic path in at least one target atomic path and each path atom in at least one path atom on the target atomic path, the periodic atomic distance between the path atom and other atoms is determined according to the atomic position of at least one initial atom, and the initial atomic characteristics of the path atom are determined according to the periodic atomic distance and at least one feature extraction function, wherein the other atoms include initial atoms other than the path atom in at least one initial atom within the period in which the path atom is located; the atomic type of the path atom is determined according to the atomic type of at least one initial atom, and the representation vector of the path atom is determined from the target matrix according to the atomic type of the path atom, wherein the target matrix includes representation vectors corresponding to the atomic types of at least one initial atom; the target atomic characteristics of the path atom are determined according to the initial atomic characteristics and the representation vector; the target path characteristics of the target atomic path are determined according to the target atomic characteristics corresponding to at least one path atom.

[0058] In an embodiment of the present invention, for each target atom path in at least one target atom path and each path atom in at least one path atom on the target atom path, the atomic position of the path atom and the atomic positions of other atoms can be determined based on the atomic position of at least one initial atom, and the other atoms include initial atoms other than the path atom in at least one initial atom within the period in which the path atom is located. It should be noted that if there are atoms other than initial atoms in the other atoms within the same period as the path atom, the atomic positions of the other atoms other than the initial atoms can also be obtained. Then, based on the atomic position of the path atom and the atomic positions of other atoms within the same period as the path atom, the distance between the path atom and the other atoms is determined, and the distance is used as the periodic atomic distance. The initial atomic features of the path atom are determined based on the periodic atomic distance and at least one feature extraction function; specifically, the initial atomic features of the path atom can be determined based on the periodic atomic distance, the atomic position of at least one initial atom, the atomic positions of other atoms, and at least one feature extraction function. It should be noted that if the number of feature extraction functions is at least two, the features corresponding to each feature extraction function can be determined separately according to the periodic atomic distance, and then the features corresponding to each feature extraction function can be spliced ​​or fused to obtain the initial atomic features.

[0059] In an embodiment of the present invention, the atomic type of the path atom can be determined based on the atomic type of at least one initial atom, and the representation vector of the path atom can be determined from the target matrix based on the atomic type of the path atom. The target matrix can be a pre-set matrix or a randomly initialized matrix. In an embodiment of the present invention, there is no specific limitation on the method of obtaining the target matrix. The initial atomic features and the representation vector are then spliced ​​or fused to obtain the target atomic features of the path atom. The target path features of the target atomic path are determined based on the target atomic features corresponding to at least one path atom. The scheme of the embodiment of the present invention introduces distance-related parameters, and through the joint action of multiple features, a target path feature with higher accuracy can be obtained.

[0060] For example, for the target atomic path P in at least one target atomic path n , target atomic path P n There are l path atoms on it, and the target atom path P n The i-th path atom on can be represented as path atom R i , R i It can also be understood as the atomic position of the i-th path atom, using the atomic position R of the i-th path atom i As the representation of the i-th path atom, the path atom R is determined based on the atomic position of at least one initial atom. i The periodic atomic distance between atoms and other atoms. Determine the feature extraction function as G 2 and G 4 .G 2 The function expression is:

[0061]

[0062] G 4 The function expression is:

[0063]

[0064] Among them, R ij is the path atom R i and other atoms R j The periodic atomic distance between ik is the path atom R i and other atoms R k The periodic atomic distance between jk are other atoms h and other atoms R k The periodic atomic distances between other atoms R j and other atoms R k are not the same atoms; η is the width of the Gaussian function; R s is the distance the center of the Gaussian function moves; Rs represents the angular resolution; λ∈[-1,1]; θ ijk is the path atom R i , other atoms R i and other atoms R k The angle between ijk =acos((R ij ·R ik ) / (|R ij |·|R ik |)). It should be noted that η, R s , ζ and λ and other parameters can be defined according to needs.

[0065] It should be noted that f c (.) is the truncation function, with f c (R ij ) as an example, the truncation function is defined as follows:

[0066]

[0067] Among them, R c The preset cutoff radius.

[0068] In the embodiment of the present invention, according to the periodic atomic distance, the atomic position of at least one initial atom, the atomic positions of other atoms, and the feature extraction function G 2 and G 2 , determine the path atom R i And through the feature extraction function G 2 and G 4 The features G(R i ) 2 and G(R i ) 4 , and then G(R i ) 2 and G(R i ) 4 Splicing to get the initial atomic features f i =[G 2 (R i );G 4 (R i )], where f i ∈R d Then determine the path atom R according to the atomic type of at least one initial atom i The atom type and the path atom R i The atom type of the path atom R is determined from the randomly initialized matrix M i The representation vector M(R i )∈R m . After the initial atomic feature f iAnd the vector M(R i ) to obtain the path atom R i The target atomic feature x i =[f i ;M(R i )]. The length of the target atomic feature is the length of the initial atomic feature f i And the vector M(R i Finally, the target path feature X = {x1, ..., x2, ..., x3, ..., x4, ..., x5, ..., x6, ..., x7, ..., x8, ..., x9, ... ... i ..., x l}.

[0069] Example 2

[0070] Figure 2 It is a flow chart of another crystal property prediction method provided in the second embodiment of the present invention. This embodiment is optimized based on the above-mentioned technical solutions. In this embodiment, optionally, according to the atomic position of the initial atom and the preset cutoff radius, the target atomic path constructed based on the initial atom is determined, including: determining the target atoms from the initial atoms, wherein the number of target atoms is the same as the number of initial atoms in any complete cycle after the target crystal is periodically expanded, and the target atoms in the target atoms are not equivalent points to each other; for each target atom in the target atoms, according to the preset cutoff radius, the atomic position corresponding to the target atom and the atomic position of the initial atom, determining the path atoms from the initial atoms; composing the path atoms into the target atomic path corresponding to the target atom to obtain the target atomic paths corresponding to the target atoms. Among them, the explanations of the terms that are the same or corresponding to the above-mentioned embodiments are not repeated here.

[0071] See also Figure 2 The method of this embodiment may specifically include the following steps:

[0072] S210: Obtain the atomic positions and atomic types of initial atoms in the target crystal, and obtain a trained target crystal property prediction model.

[0073] S220. Determine target atoms from the initial atoms, wherein the number of target atoms is the same as the number of initial atoms in any complete period after periodically unfolding the target crystal, and each target atom in the target atoms is not an identical point to another.

[0074] Among them, the target atom can be understood as the initial atom corresponding to the target atom path, and the target atom is the first path atom on its corresponding target atom path; the target atom can also be understood as the atom that can determine the crystal properties.

[0075] Specifically, at least one target atom can be determined from at least one initial atom, wherein the number of the at least one target atom is the same as the number of initial atoms in any complete cycle after periodic expansion of the target crystal, and the target atoms in the at least one target atom are not equivalent points to each other.

[0076] It is understood that after periodically unfolding the target crystal, at least one complete cycle can be obtained. The atoms at the same position in each complete cycle are the same. The atoms at the same position in each complete cycle can be considered as equivalent points. The geometric environment and material environment of the equivalent points can be the same. In an embodiment of the present invention, at least one target atom can be determined from at least one initial atom. The number of the at least one target atom is the same as the number of initial atoms in any complete cycle after periodically unfolding the target crystal, and the target atoms in the at least one target atom are not equivalent points to each other. For example, if at least one initial atom is all atoms within three complete periods, and if there are 5 atoms within a complete period, then at least one initial atom is a1, b1, c1, d1, e1, a2, b2, c2, d2, e2, a3, b3, c3, d3, and e3, respectively. A1, b1, c1, d1, and e1 are atoms within a complete period, a2, b2, c2, d2, and e2 are atoms within a complete period, and a3, b3, c3, d3, and e3 are atoms within a complete period. It is possible to determine a1, b1, c1, d1, and e1 as target atoms, and it is also possible to determine a1, b2, c3, d1, and e2 as target atoms. In an embodiment of the present invention, the target atoms can be determined as needed, and the target atoms can also be determined based on the geometric environment and material environment corresponding to each initial atom; and so on. S230: For each target atom in the target atoms, determine the path atoms from the initial atoms based on the preset cutoff radius, the atomic position corresponding to the target atom, and the atomic position of the initial atom.

[0077] Specifically, for each target atom in the at least one target atom, at least one path atom may be determined from the at least one initial atom according to a preset cutoff radius, an atomic position corresponding to the target atom, and an atomic position of the at least one initial atom.

[0078] In an embodiment of the present invention, for each target atom in at least one target atom, based on a preset cutoff radius, the atomic position corresponding to the target atom, and the atomic position of at least one initial atom, at least one initial atom that can affect the energy and other properties of the target atom can be determined from the at least one initial atom as at least one path atom.

[0079] S240 , grouping the path atoms into target atom paths corresponding to the target atoms, so as to obtain target atom paths corresponding to the target atoms respectively.

[0080] Specifically, at least one path atom may be combined into a target atom path corresponding to a target atom, so as to obtain a target atom path corresponding to at least one target atom.

[0081] For example, if the atomic positions of N initial atoms are {R1, R2..., R N}, if there are n target atoms, then n target atom paths P = {P1, P2...P n}, then the mth target atomic path can be recorded as Among them, l can be understood as the number of path atoms on the corresponding target atom path.

[0082] S250 , determining the path atom distance between the path atoms on each target atom path in the target atom path.

[0083] S260: Determine target path features corresponding to target atomic paths based on the atomic positions and atomic types of the initial atoms.

[0084] S270: Input the target path features and path atom distances corresponding to the target atomic paths into the target crystal property prediction model, and determine the property prediction result of the target crystal according to the output result of the target crystal property prediction model.

[0085] The technical solution of the embodiment of the present invention is to determine the target atoms from the initial atoms, wherein the number of target atoms is the same as the number of initial atoms in any period after the target crystal is periodically expanded, and the target atoms in the target atoms are not equivalent points to each other; for each target atom in the target atoms, the path atoms are determined from the initial atoms according to the preset cutoff radius, the atomic position corresponding to the target atom, and the atomic position of the initial atom; the path atoms are combined into the target atom paths corresponding to the target atoms to obtain the target atom paths corresponding to the target atoms. In the embodiment of the present invention, the target atoms that can determine the properties of the target crystal can be accurately determined, and the target atom paths corresponding to the target atoms that can affect the properties of the target atoms can be accurately determined.

[0086] An optional technical solution determines path atoms from initial atoms based on a preset cutoff radius, the atomic position corresponding to the target atom, and the atomic position of the initial atom, including: determining, from the initial atoms, a first neighbor atom that is within the preset cutoff radius of the target atom and is closest to the target atom based on the atomic position corresponding to the target atom and the atomic position of the initial atom; determining, from the initial atoms, a second neighbor atom that is within the preset cutoff radius of the first neighbor atom and is closest to the first neighbor atom based on the atomic position corresponding to the first neighbor atom and the atomic position of the initial atom; and at least the target atom, the first neighbor atom, and the second neighbor atom are respectively set as path atoms corresponding to the target atom to obtain the path atoms corresponding to the target atom.

[0087] Specifically, at least one path atom is determined from at least one initial atom according to a preset cutoff radius, the atomic position corresponding to the target atom, and the atomic position of at least one initial atom, including: determining a first neighbor atom that is located within the preset cutoff radius of the target atom and is closest to the target atom from at least one initial atom according to the atomic position corresponding to the target atom and the atomic position of at least one initial atom; determining a second neighbor atom that is located within the preset cutoff radius of the first neighbor atom and is closest to the first neighbor atom from at least one initial atom according to the atomic position corresponding to the first neighbor atom and the atomic position of at least one initial atom; and at least the target atom, the first neighbor atom, and the second neighbor atom are respectively set as path atoms corresponding to the target atom to obtain at least one path atom corresponding to the target atom.

[0088] In an embodiment of the present invention, the initial distances between the target atom and at least one initial atom can be determined based on the atomic position corresponding to the target atom and the atomic position of at least one initial atom. Each initial distance can be compared with a preset cutoff radius to obtain initial distances that are not greater than the preset cutoff radius. The smallest initial distance is then selected from the initial distances that are not greater than the preset cutoff radius, and the initial atom corresponding to the smallest initial distance is used as the first neighbor atom; or the smallest initial distance is selected from the initial distances, and the smallest initial distance is compared with the preset cutoff radius. If the smallest initial distance is not greater than the preset cutoff radius, the initial atom corresponding to the smallest initial distance is used as the first neighbor atom; if there are at least two first neighbor atoms, one of the first neighbor atoms can be randomly selected and updated as the only first neighbor atom.

[0089] In an embodiment of the present invention, a method for determining a second neighbor atom that is located within a preset cutoff radius of the first neighbor atom and is closest to the first neighbor atom from at least one initial atom based on the atomic position corresponding to the first neighbor atom and the atomic position of at least one initial atom may be consistent with the above-mentioned method for determining a first neighbor atom that is located within a preset cutoff radius of the target atom and is closest to the target atom from at least one initial atom based on the atomic position corresponding to the target atom and the atomic position of at least one initial atom, and will not be described in detail here.

[0090] In an embodiment of the present invention, a third neighbor atom that is located within a preset cutoff radius of the second neighbor atom and is closest to the second neighbor atom can be determined from the at least one initial atom based on the atomic position corresponding to the second neighbor atom and the atomic position of the at least one initial atom, and so on. At least the target atom, the first neighbor atom, the second neighbor atom, and the third neighbor atom that may be obtained by analogy are respectively regarded as path atoms corresponding to the target atom, so as to obtain at least one path atom corresponding to the target atom, thereby achieving accurate determination of the path atoms.

[0091] For example, the first target atom R1 can be determined first, and the truncation radius R of the target atom R1 can be determined from at least one initial atom. c The first neighbor atom R2 that is closest to the target atom R1 is determined from at least one initial atom and the first neighbor atom R2 is determined within the cutoff radius R c The second neighbor atom R3 that is closest to the first neighbor atom R2 is obtained; and so on, a total of n atoms are determined, thereby obtaining at least one path atom {R1, ..., R n}.

[0092] Based on the above scheme, another optional technical scheme is to determine the path atom distance between the path atoms on each target atom path in the target atom path, including: determining the path atom distance between the path atoms on the target atom path corresponding to the target atom based at least on the path atom distance between the target atom and the first neighbor atom, and the path atom distance between the first neighbor atom and the second neighbor atom.

[0093] Specifically, determining the path atom distance between at least one path atom on each target atom path in at least one target atom path includes: determining the path atom distance between at least one path atom on the target atom path corresponding to the target atom based at least on the path atom distance between the target atom and the first neighbor atom, and the path atom distance between the first neighbor atom and the second neighbor atom.

[0094] In an embodiment of the present invention, the distance between the target atom and the first neighbor atom can be determined as the path atom distance, and the distance between the first neighbor atom and the second neighbor atom can be used as the path atom distance. Similarly, for at least one path atom on each target atom path in at least one target atom path, the distance between the path atom and the path atom following the path atom on the target atom path can be used as the path atom distance, thereby determining the path atom distance between at least one path atom on the target atom path corresponding to the target atom, thereby realizing the determination of the path atom distance.

[0095] Example 3

[0096] Figure 3 : This is a flow chart of another crystal property prediction method provided in Example 3 of the present invention. This embodiment is optimized based on the above-mentioned technical solutions. In this embodiment, optionally, before inputting the target path features and path atom distances corresponding to the target atomic paths into the target crystal property prediction model, it also includes: inverting each target path feature in each target path feature, and updating the inverted result as the target path feature. Among them, the explanation of the terms that are the same as or corresponding to the above-mentioned embodiments will not be repeated here.

[0097] See also Figure 3 The method of this embodiment may specifically include the following steps:

[0098] S310: Obtain the atomic positions and atomic types of initial atoms in the target crystal, and obtain a trained target crystal property prediction model.

[0099] S320 , determining a target atomic path constructed based on the initial atoms according to the atomic position of the initial atoms and a preset cutoff radius, and determining a path atomic distance between path atoms on each target atomic path in the target atomic path.

[0100] S330 : Determine target path features corresponding to target atomic paths based on the atomic positions and atomic types of the initial atoms.

[0101] S340: Reverse each target path feature among the target path features, and update the inversion result as the target path feature.

[0102] In an embodiment of the present invention, considering that for each target path feature in each target path feature, it is necessary to determine the target atom feature corresponding to the target atom path in the target crystal property prediction model according to each target atom feature in the target path feature. The target atom feature is also the feature of the target atom corresponding to the target atom path, and the target atom is also the first path atom on the target atom path. Since the target crystal property prediction model is processed in the order of path atoms, in order to enable the target crystal property prediction model to finally obtain the feature of the first path atom, that is, the feature of the target atom, and to enable the features of other path atoms on the target atom path except the first path atom to affect the first path atom, that is, the target atom, each target path feature in each target path feature can be reversed, and the inversion result after inversion can be updated to the target path feature.

[0103] For example, referring to the above examples, the mth target atom path can be The target path feature X={x1,...,x i ..., x l}, and the inverted result X={x l ,...,x i ..., x1}.

[0104] S350: Input the target path features and path atom distances corresponding to the target atomic paths into a target crystal property prediction model, and determine the property prediction result of the target crystal according to the output result of the target crystal property prediction model.

[0105] The technical solution of an embodiment of the present invention is to invert each of the target path features and the path atom distances corresponding to the target atomic paths before inputting them into the target crystal property prediction model, and then update the inverted results as the target path features. In this embodiment of the present invention, by inverting each of the target path features, the subsequent target crystal property prediction model can accurately obtain the characteristics of each target atom, thereby obtaining an output result based on the characteristics of each target atom, thereby ensuring the accuracy of the output result.

[0106] Example 4

[0107] Figure 4It is a flow chart of another crystal property prediction method provided in the fourth embodiment of the present invention. This embodiment is optimized based on the above-mentioned technical solutions. In this embodiment, optionally, the target crystal property prediction model determines the output result through the following steps: for each target atomic path in the target atomic path, according to the path atom distance corresponding to the target atomic path, determine the path atom weights corresponding to the path atoms on the target atomic path respectively; according to the target path characteristics corresponding to the target atomic path and the path atom weights corresponding to the path atoms respectively, determine the implicit state of the target atomic path; according to the implicit state corresponding to the target atomic path respectively, determine the output result. Among them, the explanation of the terms that are the same as or corresponding to the above-mentioned embodiments will not be repeated here.

[0108] See also Figure 4 The method of this embodiment may specifically include the following steps:

[0109] S410: Obtain the atomic positions and atomic types of initial atoms in the target crystal, and obtain a trained target crystal property prediction model.

[0110] S420 , determining a target atomic path constructed based on the initial atoms according to the atomic position of the initial atoms and a preset cutoff radius, and determining a path atomic distance between path atoms on each target atomic path in the target atomic path.

[0111] S430 : Determine target path features corresponding to target atomic paths based on the atomic positions and atomic types of the initial atoms.

[0112] S440: Reverse each target path feature among the target path features, and update the inversion result as the target path feature.

[0113] S450 , inputting the target path features and path atom distances corresponding to the target atomic paths into the target crystal property prediction model, so as to execute the following S460 - S480 through the target crystal property prediction model.

[0114] S460 . For each target atom path in the target atom path, determine the path atom weights corresponding to the path atoms on the target atom path according to the path atom distances corresponding to the target atom path.

[0115] Among them, the path atom weight can be understood as the weight corresponding to the path atom.

[0116] Specifically, for each target atom path in at least one target atom path, the path atom weight corresponding to at least one path atom on the target atom path may be determined according to the path atom distance corresponding to the target atom path.

[0117] For example, in combination with the above example of determining the target path characteristics of the target atomic path, the target crystal property prediction model can be an improved Gate Recurrent Unit (GRU) model, which determines the path atom weight according to the following formula and the path atom distance:

[0118]

[0119] Among them, w t，t-1 is the path atom R t R t The corresponding path atom weight; R t , R t-1 is the path atom R t and path atom R t-1 The distance between them; t can be understood as the order of the target atom features corresponding to the path atoms in the target path features after reversing each target path feature and updating the reversal result as the target path feature, for example, the target atom path P a There are five path atoms a, b, c, d, and e, and the target atom path P a Indicated as P a ={R a , R b , R c , R d , R e}, and its corresponding target path feature XX={x a , x b , x c , x d , x e}, then reverse it, and update the reversed result to obtain the target path feature X={x e , x d , x c , x b , x a}, if R t R b , then R t-1 R c .

[0120] S470 , determining the implicit state of the target atomic path according to the target path features corresponding to the target atomic path and the path atom weights corresponding to the path atoms.

[0121] The hidden state can be understood as a hidden state that can obtain relevant information of the target atom corresponding to the target atom path, or can be understood as a hidden state of the target atom corresponding to the target atom path.

[0122] Specifically, the implicit state of the target atomic path may be determined according to the target path feature corresponding to the target atomic path and the path atom weight corresponding to at least one path atom.

[0123] For example, in combination with the above examples, the hidden state of the path atom can be determined according to the target path features corresponding to the target atom path and the path atom weights corresponding to the path atoms, as well as the following formula:

[0124] h t =f(X, w t，t-1 *h t-1 );

[0125] Among them, h t It can be understood as the hidden state of the path atom. It should be noted that the above formula can be understood as one of the input parameters h of the function that solves the hidden state in the GRU model. t-1 Modify to w t，t-1 *h t-1 .

[0126] According to the above formula, each target path feature in each target path feature is inverted, and after the inversion result is updated as the target path feature, the order of the target atom features corresponding to the path atoms in the target path feature is calculated in sequence. For each path atom in each path atom, the hidden state corresponding to the path atom can be used to determine the hidden states of other path atoms on the target atom path except the path atom, and the hidden state of the last calculated path atom h is used as the hidden state of the path atom. l As the implicit state of the target atom path.

[0127] S480: Determine the output result according to the implicit states corresponding to the target atomic paths.

[0128] Specifically, the output result may be determined according to the implicit state corresponding to at least one target atomic path.

[0129] For example, in combination with the above examples, the implicit state h l The input is into the fully connected layer in the target crystal property prediction model to obtain the target state feature logits of the target atom corresponding to the target atomic path.

[0130] According to the above process, the target state characteristics of the target atoms corresponding to each target atomic path can be obtained, and then the output results can be determined according to the following formula:

[0131]

[0132] Among them, Out predIt can be understood as the output result of the target crystal property prediction model; k can be understood as the number of target atomic paths; β can be understood as a parameter set according to actual prediction requirements; i can be understood as the i-th target atom; logits i is the target state feature of the i-th target atom.

[0133] S490: Determine the property prediction result of the target crystal according to the output result of the target crystal property prediction model.

[0134] According to the technical solution of an embodiment of the present invention, the target crystal property prediction model determines the output result through the following steps: for each target atomic path in the target atomic path, the path atom weight corresponding to each path atom on the target atomic path is determined based on the path atom distance corresponding to the target atomic path; the implicit state of the target atomic path is determined based on the target path characteristics corresponding to the target atomic path and the path atom weights corresponding to each path atom; and the output result is determined based on the implicit state corresponding to each target atomic path. In this embodiment of the present invention, by introducing distance-related parameters into the target crystal property prediction model, the accuracy of the output result of the target crystal property prediction model can be improved.

[0135] In order to better understand the technical solution of the above embodiment of the present invention, an optional example is provided here. Figure 5 , the embodiment of the present invention requires prediction of target crystal energy. Part (a) is an example of target atomic path generation (An Example of Atom Paths Generation). Obtain the atomic positions and atomic types of 15 initial atoms in three adjacent complete periods obtained after periodically expanding the target crystal, and obtain a trained target crystal property prediction model, wherein the atoms in each complete period include 5 initial atoms a, b, c, d and e; according to the atomic positions of the 15 initial atoms and the preset truncation radius, determine the 5 target atomic paths constructed based on the 15 initial atoms as well as And determine the path atom distance between at least one path atom on each of the five target atom paths.

[0136] Part (b) is an example of energy calculation by Path-LSTM and weighted Pooling. For example, you can Reverse the path atoms on , and update the reversed result to It should be noted that, in the embodiment of the present invention, reversal can also be understood as inversion. The input is to the embedding layer. The embedding layer determines the two features E1(c) and E2(C) corresponding to the path atom c, the two features E1(b) and E2(b) corresponding to the path atom b, the two features E1(e) and E2(e) corresponding to the path atom e, and the two features E1(a) and E2(a) corresponding to the path atom a based on the atomic position and atom type of at least one initial atom. The two features corresponding to each path atom are spliced ​​to obtain the target path features corresponding to at least one target atomic path. The target path features and the path atom distance corresponding to at least one target atomic path are input into the target crystal property prediction model. The encoding layer in the target crystal property prediction model determines the path atom weight corresponding to at least one path atom on the target atomic path according to the path atom distance corresponding to the target atomic path, realizes the distance weight calculation, and obtains the hidden state of each path atom. and And the hidden state corresponding to the last obtained path atom a After feature enhancement through the Feature Enhence Network (FEN) structure, the implicit state e1 of the target atomic path is obtained. Correspondingly, the target crystal property prediction model is adaptively adjusted according to the number of target atoms to obtain the five paths-LSTM to obtain the implicit states e1, e2, e3, e4 and e5 of the five target atomic paths, and the formula is:

[0137]

[0138] Perform energy calculation through weighted pooling (Weighted Pooling For Energy Computing) to obtain the output result Out of the target crystal property prediction model pred ; and predict the output of the model based on the target crystal properties Out pred , determine the property prediction results of the target crystal.

[0139] Example 5

[0140] Figure 6This is a block diagram of the structure of the crystal property prediction device provided in the fifth embodiment of the present invention. The device is used to execute the crystal property prediction method provided in any of the above embodiments. The device and the crystal property prediction method of the above embodiments belong to the same inventive concept. For details not fully described in the embodiment of the crystal property prediction device, please refer to the embodiment of the above crystal property prediction method. Figure 6 The device may specifically include: a target crystal property prediction model acquisition module 510, a path atomic distance determination module 520, a target path feature determination module 530 and a property prediction result determination module 540.

[0141] The target crystal property prediction model acquisition module 510 is used to obtain the atomic positions and atomic types of the initial atoms in the target crystal and to obtain the trained target crystal property prediction model;

[0142] a path atom distance determination module 520 for determining a target atom path constructed based on the initial atoms according to the atomic position of the initial atoms and a preset cutoff radius, and determining the path atom distance between the path atoms on each target atom path in the target atom path;

[0143] A target path feature determination module 530 is configured to determine target path features corresponding to target atomic paths based on the atomic positions and atomic types of the initial atoms;

[0144] The property prediction result determination module 540 is used to input the target path characteristics and path atom distances corresponding to the target atomic paths into the target crystal property prediction model, and determine the property prediction results of the target crystal based on the output results of the target crystal property prediction model.

[0145] Optionally, the path atom distance determination module 520 may include:

[0146] a target atom determination unit, configured to determine target atoms from initial atoms, wherein the number of target atoms is the same as the number of initial atoms in any complete period after periodic expansion of the target crystal, and each target atom in the target atoms is not an identical point to another;

[0147] a path atom determination unit, configured to determine, for each target atom in the target atoms, path atoms from the initial atoms according to a preset cutoff radius, an atomic position corresponding to the target atom, and an atomic position of the initial atom;

[0148] The target atom path obtaining unit is used to combine the path atoms into target atom paths corresponding to the target atoms, so as to obtain the target atom paths corresponding to the target atoms respectively.

[0149] Based on the above solution, optionally, the path atom determination unit may include:

[0150] A first neighbor atom determination subunit is configured to determine, from the initial atoms, a first neighbor atom that is within a preset cutoff radius of the target atom and is closest to the target atom based on the atomic position corresponding to the target atom and the atomic position of the initial atom;

[0151] A second neighbor atom determination subunit is configured to determine, from the initial atoms, a second neighbor atom that is within a preset cutoff radius of the first neighbor atom and is closest to the first neighbor atom based on the atomic position corresponding to the first neighbor atom and the atomic position of the initial atom;

[0152] The path atom obtaining subunit is used to convert at least the target atom, the first neighbor atom and the second neighbor atom into path atoms corresponding to the target atom, so as to obtain the path atom corresponding to the target atom.

[0153] Based on the above solution, optionally, the path atom distance determination module 520 may include:

[0154] The path atom distance determining unit is used to determine the path atom distance between the path atoms on the target atom path corresponding to the target atom based on at least the path atom distance between the target atom and the first neighbor atom, and the path atom distance between the first neighbor atom and the second neighbor atom.

[0155] Optionally, the target path feature determination module 530 may include:

[0156] an initial atom feature determination unit, configured to determine, for each target atom path in the target atom path and each path atom in the path atoms on the target atom path, a periodic atomic distance between the path atom and other atoms according to the atomic position of the initial atom, and determine the initial atomic features of the path atom according to the periodic atomic distance and a feature extraction function, wherein the other atoms include initial atoms other than the path atom in the initial atoms within the period in which the path atom is located;

[0157] a representation vector determining unit, configured to determine the atomic type of the path atom according to the atomic type of the initial atom, and determine the representation vector of the path atom from a target matrix according to the atomic type of the path atom, wherein the target matrix includes the representation vectors corresponding to the atomic types of the initial atoms respectively;

[0158] a target atom feature determination unit, configured to determine target atom features of path atoms based on initial atom features and representation vectors;

[0159] The target path feature determination unit is used to determine the target path feature of the target atom path according to the target atom features corresponding to the path atoms.

[0160] Optionally, the crystal property prediction device may further include:

[0161] The target path feature updating module is used to invert each target path feature before inputting the target path features and path atomic distances corresponding to the target atomic paths into the target crystal property prediction model, and update the inverted results as the target path features.

[0162] Based on the above solution, the crystal property prediction device may optionally further include the following modules to enable the target crystal property prediction model to determine the output result:

[0163] A path atom weight determination module is used to determine, for each target atom path in the target atom path, the path atom weights corresponding to the path atoms on the target atom path according to the path atom distances corresponding to the target atom paths;

[0164] An implicit state determination module is used to determine the implicit state of the target atomic path according to the target path characteristics corresponding to the target atomic path and the path atom weights corresponding to the path atoms respectively;

[0165] The output result determination module is used to determine the output result according to the implicit states corresponding to the target atomic paths.

[0166] The crystal property prediction device provided in the fifth embodiment of the present invention is used to obtain the atomic position and atomic type of the initial atoms in the target crystal through the target crystal property prediction model acquisition module, and obtain the trained target crystal property prediction model; determine the target atomic path constructed based on the initial atoms according to the atomic position of the initial atoms and the preset cutoff radius through the path atomic distance determination module, and determine the path atomic distance between the path atoms on each target atomic path in the target atomic path; determine the target path features corresponding to the target atomic paths based on the atomic position and atomic type of the initial atoms through the target path feature determination module; input the target path features and path atomic distance corresponding to the target atomic paths into the target crystal property prediction model through the property prediction result determination module, and determine the property prediction result of the target crystal based on the output result of the target crystal property prediction model. The above device takes into account that the distance between the atoms constituting the crystal will also affect the properties of the crystal, and thus determines the property prediction result of the target crystal based on the path atomic distance, so that the predicted property prediction result is more accurate, thereby improving the accuracy of the crystal property prediction.

[0167] The crystal property prediction device provided in the embodiment of the present invention can execute the crystal property prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0168] It is worth noting that in the embodiment of the above-mentioned crystal property prediction device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0169] Example 6

[0170] Figure 7 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0171] like Figure 7 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0172] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0173] The processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the crystal property prediction method.

[0174] In some embodiments, the crystal property prediction method can be implemented as a computer program that is tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the crystal property prediction method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the crystal property prediction method in any other appropriate manner (e.g., by means of firmware).

[0175] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0176] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0177] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0178] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0179] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0180] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0181] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0182] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for predicting crystal properties, characterized in that: include: Obtaining the atomic positions and atomic types of initial atoms in the target crystal, and obtaining a trained target crystal property prediction model; Determining a target atomic path constructed based on the initial atoms according to the atomic position of the initial atoms and a preset cutoff radius, and determining a path atomic distance between path atoms on each target atomic path in the target atomic path; Determining target path features corresponding to the target atomic paths based on the atomic positions and atomic types of the initial atoms; Inputting the target path characteristics and path atomic distances respectively corresponding to the target atomic paths into the target crystal property prediction model, and determining the property prediction result of the target crystal according to the output result of the target crystal property prediction model; The step of determining a target atomic path constructed based on the initial atoms according to the atomic positions of the initial atoms and a preset cutoff radius includes: Determining target atoms from the initial atoms, wherein the number of the target atoms is the same as the number of the initial atoms in any complete period after periodic expansion of the target crystal, and each target atom in the target atoms is not an equivalent point to another; For each target atom in the target atoms, determining a path atom from the initial atoms according to the preset cutoff radius, the atomic position corresponding to the target atom, and the atomic position of the initial atom; The path atoms are combined into target atom paths corresponding to the target atoms to obtain target atom paths corresponding to the target atoms respectively.

2. The method according to claim 1, characterized in that The determining of path atoms from the initial atoms according to the preset cutoff radius, the atomic position corresponding to the target atom, and the atomic position of the initial atom comprises: Determining, from the initial atoms, a first neighbor atom that is within the preset cutoff radius of the target atom and is closest to the target atom according to the atomic position corresponding to the target atom and the atomic position of the initial atom; Determining, from the initial atoms, a second neighbor atom that is within the preset cutoff radius of the first neighbor atom and is closest to the first neighbor atom, according to the atomic position corresponding to the first neighbor atom and the atomic position of the initial atom; At least the target atom, the first neighbor atom, and the second neighbor atom are respectively used as path atoms corresponding to the target atom, so as to obtain the path atom corresponding to the target atom.

3. The method according to claim 2, characterized in that Determining the path atom distance between path atoms on each target atom path in the target atom path includes: The path atom distances between the path atoms on the target atom path corresponding to the target atom are determined based at least on the path atom distances between the target atom and the first neighbor atom, and the path atom distances between the first neighbor atom and the second neighbor atom.

4. The method according to claim 1, wherein The determining of target path features corresponding to the target atomic paths based on the atomic positions and atomic types of the initial atoms includes: For each target atom path in the target atom path and each path atom in the path atoms on the target atom path, determining the periodic atomic distance between the path atom and other atoms according to the atomic position of the initial atom, and determining the initial atomic features of the path atom according to the periodic atomic distance and a feature extraction function, wherein the other atoms include initial atoms other than the path atom in the initial atoms within the period in which the path atom is located; Determining the atomic type of the path atom according to the atomic type of the initial atom, and determining the representation vector of the path atom from a target matrix according to the atomic type of the path atom, wherein the target matrix includes the representation vectors corresponding to the atomic types of the initial atoms respectively; determining target atom features of the path atom according to the initial atom features and the representation vector; The target path characteristics of the target atom path are determined according to the target atom characteristics corresponding to the path atoms respectively.

5. The method according to claim 1, wherein Before inputting the target path features and path atomic distances respectively corresponding to the target atomic paths into the target crystal property prediction model, the method further includes: Each target path feature in the target path features is reversed, and the reversed result is updated as the target path feature.

6. The method according to claim 5, characterized in that The target crystal property prediction model determines the output result through the following steps: For each target atom path in the target atom path, determining the path atom weights corresponding to the path atoms on the target atom path according to the path atom distance corresponding to the target atom path; Determining an implicit state of the target atomic path according to the target path features corresponding to the target atomic path and the path atom weights corresponding to the path atoms; The output result is determined according to the implicit states respectively corresponding to the target atomic paths.

7. A crystal property prediction device, characterized in that: include: A target crystal property prediction model acquisition module is used to obtain the atomic positions and atomic types of initial atoms in the target crystal and to obtain a trained target crystal property prediction model; a path atom distance determination module, configured to determine a target atom path constructed based on the initial atoms according to the atomic position of the initial atoms and a preset cutoff radius, and determine a path atom distance between path atoms on each target atom path in the target atom path; a target path feature determination module, configured to determine target path features corresponding to the target atomic paths based on the atomic positions and atomic types of the initial atoms; a property prediction result determination module, configured to input the target path characteristics and path atom distances respectively corresponding to the target atomic paths into the target crystal property prediction model, and determine the property prediction result of the target crystal according to the output result of the target crystal property prediction model; The path atomic distance determination module includes: a target atom determining unit, configured to determine target atoms from the initial atoms, wherein the number of the target atoms is the same as the number of initial atoms in any complete period after periodic expansion of the target crystal, and each target atom in the target atoms is not an identical point to another; a path atom determination unit, configured to determine, for each target atom in the target atoms, a path atom from the initial atoms according to the preset cutoff radius, the atomic position corresponding to the target atom, and the atomic position of the initial atom; The target atom path obtaining unit is used to combine the path atoms into target atom paths corresponding to the target atoms, so as to obtain target atom paths corresponding to the target atoms respectively.

8. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the crystal property prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the crystal property prediction method according to any one of claims 1 to 6 when executed.

Images