Perovskite crystal property prediction method, device, equipment and medium

Abstract

The invention discloses a perovskite crystal property prediction method and device, equipment and a medium. The method comprises the following steps: respectively determining element characteristic information of all elements in the perovskite crystal; wherein the element types of the elements comprise an atom type and a molecule type; respectively inputting the element feature information of each element into an encoder corresponding to the element type of the element to obtain a feature encoding result of each element; and obtaining a property prediction result of the perovskite crystal according to the feature coding result of each element and other attribute information of the perovskite crystal. According to the method, the atoms and the molecules in the perovskite crystal are coded respectively, so that the coding accuracy of the atoms and the molecules is improved, and the accuracy of crystal feature coding is further improved; and the crystal property is predicted by using an atom and molecule coding fusion mode, so that deep mining of the overall structure and physical information of the crystal is realized, and the accuracy of crystal property prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of solar photovoltaic materials, and in particular to a method, device, equipment and medium for predicting the properties of perovskite crystals. Background Art

[0002] The unique crystal structure of perovskite gives it excellent performance in optoelectronics, thermoelectricity, catalysis and other fields, making it a frontier hotspot in materials science and energy technology research. However, due to its insufficient chemical stability, poor environmental adaptability and limited large-scale application, the widespread promotion of perovskite materials still faces many challenges. Therefore, accurate prediction of the properties of perovskite crystals is particularly important, which can not only help to deeply understand the performance mechanism of materials, but also provide strong support for optimizing material design and accelerating the development of new high-performance materials.

[0003] Existing methods for predicting the properties of perovskite crystals include first-principles calculations based on quantum mechanics, which can accurately simulate the electronic structure and energy band characteristics of materials and reveal microscopic mechanisms; high-throughput computing methods that combine material databases and computing tools can efficiently screen perovskite materials that meet specific property requirements and are widely used in large-scale material exploration; and machine learning methods that predict material properties by training models and using experimental or simulation data for rapid predictions, thereby improving the efficiency of material design.

[0004] However, the computational cost of first-principles calculations is high, and it is difficult to handle complex crystal structures and conduct large-scale material screening. Although high-throughput computing can efficiently screen materials, its results are heavily dependent on the quality and coverage of the database, and the parameter optimization process is complex. Perovskite crystal materials have complex structures and multiple parameters, and traditional machine learning methods may not be able to accurately capture the subtle changes and nonlinear relationships therein, affecting the accuracy of crystal property predictions. Summary of the invention

[0005] The present invention provides a method, device, equipment and medium for predicting the properties of perovskite crystals, so as to improve the accuracy and efficiency of predicting the properties of perovskite crystals.

[0006] According to one aspect of the present invention, a method for predicting perovskite crystal properties is provided, comprising:

[0007] Determining the element characteristic information of all elements in the perovskite crystal respectively; wherein the element type of the element includes an atomic type and a molecular type;

[0008] Inputting the element feature information of each element into an encoder corresponding to the element type of the element to obtain a feature encoding result of each element;

[0009] According to the characteristic coding result of each element and other attribute information of the perovskite crystal, the property prediction result of the perovskite crystal is obtained.

[0010] According to another aspect of the present invention, there is provided a device for predicting properties of perovskite crystals, comprising:

[0011] An element characteristic determination module, used to respectively determine the element characteristic information of all elements in the perovskite crystal; wherein the element type of the element includes an atomic type and a molecular type;

[0012] An element feature encoding module, used to input the element feature information of each element into an encoder corresponding to the element type of the element, to obtain a feature encoding result of each element;

[0013] The crystal property prediction module is used to obtain the property prediction result of the perovskite crystal according to the characteristic coding result of each element and other attribute information of the perovskite crystal.

[0014] According to another aspect of the present invention, there is provided an electronic device, the electronic device comprising:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] 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 so that the at least one processor can execute the method for predicting the properties of the perovskite crystal as described in any embodiment of the present invention.

[0018] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for predicting the properties of perovskite crystals described in any embodiment of the present invention when executed.

[0019] The technical solution of the embodiment of the present invention improves the coding accuracy of atoms and molecules by encoding the atoms and molecules in the perovskite crystal separately, thereby improving the accuracy of the crystal feature coding; and predicts the crystal properties by fusion of atomic and molecular coding, thereby achieving in-depth mining of the overall structure and physical information of the crystal, and improving the accuracy of the crystal property prediction.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 is a flow chart of a method for predicting perovskite crystal properties provided according to an embodiment of the present invention;

[0023] Figure 2 is a flow chart of another method for predicting perovskite crystal properties provided according to an embodiment of the present invention;

[0024] Figure 3 is a flow chart of another method for predicting perovskite crystal properties provided according to an embodiment of the present invention;

[0025] Figure 4 is a schematic structural diagram of a device for predicting perovskite crystal properties provided according to an embodiment of the present invention;

[0026] Figure 5 It is a schematic diagram of the structure of an electronic device for implementing the method for predicting the properties of perovskite crystals according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme 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 described embodiments 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 creative work should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "candidate", "target", etc. in the specification 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 interchanged 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. 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 inherent to these processes, methods, products or devices.

[0029] Figure 1 A flowchart of a method for predicting the properties of a perovskite crystal is provided for an embodiment of the present invention. This embodiment is applicable to the case of predicting the properties of a perovskite crystal. The method can be performed by a device for predicting the properties of a perovskite crystal. The device for predicting the properties of a perovskite crystal can be implemented in the form of hardware and / or software. The device for predicting the properties of a perovskite crystal can be configured in a server with computing capabilities. Figure 1 As shown, the method includes:

[0030] S110, respectively determining element characteristic information of all elements in the perovskite crystal.

[0031] Among them, the element type of the element includes an atomic type and a molecular type.

[0032] Among them, all elements in the perovskite crystal refer to the components that make up the perovskite crystal. For example, the components of the perovskite crystal system are: ,in, For metal atoms, It is a composite molecule of non-metallic elements such as halogen or oxygen. Assume that the molecule , and is the type of atoms that make up the molecule, and The atomic ratio.

[0033] The element characteristic information refers to the atomic characteristic information and the molecular characteristic information in the perovskite crystal. The atomic characteristic information includes the property information related to the atomic components, and the molecular characteristic information includes the property information related to the molecular components.

[0034] Exemplarily, the atomic characteristic information includes data such as average atomic formation energy, energy, average atomic energy, Fermi energy, lattice constant of cubic unit cell, α angle of lattice, β angle of lattice, γ angle of lattice, volume, number of sites, number of space groups, etc., atomic equivalent radius, atomic ionization energy, Martynov-Batsanov electronegativity scale, covalent radius of element, highest oxidation state, lowest oxidation state, first ionization energy, second ionization energy, third ionization energy, electron affinity, electronegativity, resistivity, ionization potential, critical radius, boiling point, melting point, Mendeleev number, enthalpy of fusion at melting point, allowed factor, van der Waals radius, atomic radius, number of periodic table groups, number of periodic table rows, atomic number, average anion radius, average cation radius, average ion radius, and also includes spatial position under structural stability conditions, bond length and bond angle between atoms and other information, for example, including 1) spatial position and geometric relationship between atoms, 2) bond length and bond angle between atoms, and 3) electron orbital information of atoms. The molecular characteristic information includes the relevant characteristic information of atoms in the molecular components, and the relevant characteristic information of the atoms can refer to the above-mentioned atomic characteristic information.

[0035] S120. Input the element feature information of each element into an encoder corresponding to the element type of the element to obtain a feature encoding result of each element.

[0036] Different encoders are constructed according to different element types of the elements, so that different encoders can encode element feature information of different elements according to different element types, thereby improving the accuracy of feature encoding results of different elements.

[0037] Specifically, the atomic characteristic information of each atom is input into the atomic encoder to obtain the characteristic encoding result of each atom; the molecular characteristic information of each molecule is input into the molecular encoder to obtain the characteristic encoding result of each molecule.

[0038] Exemplarily, the atomic encoder and the molecular encoder can be pre-trained based on the sample atomic feature information and the sample molecular feature information in the sample crystal. For example, the atomic encoder architecture and the molecular encoder architecture are constructed based on a multi-layer perceptron, the sample atomic feature information is input into the atomic encoder for training, and the sample molecular feature information is input into the molecular encoder for training, to obtain the trained atomic encoder and molecular encoder.

[0039] S130, obtaining a property prediction result of the perovskite crystal according to the characteristic coding result of each element and other property information of the perovskite crystal.

[0040] According to the characteristic coding results of each atom in the perovskite crystal, the characteristic coding results of each molecule and other attribute information, the property prediction results of the perovskite crystal are obtained. Exemplarily, the property prediction model is pre-trained, and the characteristic coding results of each atom in the perovskite crystal, the characteristic coding results of each molecule and other attribute information are input into the property prediction model to obtain the property prediction results of the perovskite crystal. Among them, the other attribute information of the perovskite crystal is used to describe other characteristics between the components in the crystal. For example, the crystal is composed of a plurality of unit cells stacked, each unit cell includes a plurality of atoms and a plurality of molecules, and the other attribute information can be the attribute information of the unit cell, such as the size information of the unit cell or the relationship characteristics of the components in the unit cell.

[0041] The property prediction results can be determined according to the actual scene requirements, such as the formation energy results of perovskite crystals, etc. The property prediction model is trained based on the sample perovskite crystal. The specific structure of the property prediction model can select different network models according to the actual scene, which is not limited here. Specifically, the atomic feature information of each atom and the molecular feature information of each molecule in the sample perovskite crystal are determined, and the atomic feature information is input into the trained atomic encoder to obtain the atomic feature encoding result, and the molecular feature information is input into the trained molecular encoder to obtain the molecular feature encoding result, and the atomic feature encoding result, the molecular feature encoding result and other property information of the sample perovskite crystal are input into the pre-built property prediction model to obtain the property prediction result of the sample perovskite crystal. , based on the properties of the pre-determined sample perovskite crystals As labels, the mean squared error is used to evaluate the predicted values and the true value The square error between them, for example, is used to determine the loss using the following square error formula, and the parameters in the model are trained according to the calculated loss value.

[0042] .

[0043] In a feasible embodiment, S130 includes:

[0044] Determine the fusion feature information of each element according to the feature coding result of each element, the feature coding results of the adjacent node elements of the element, and the positional relationship between the element and each adjacent node element;

[0045] Based on the fusion characteristic information of each element and other property information of the perovskite crystal, the property prediction results of the perovskite crystal are obtained.

[0046] According to the positional relationship between each element in the perovskite crystal and its adjacent node element, the feature coding results of the adjacent node elements and the feature coding results of the element are fused as the new fused feature information of the element. Exemplarily, the distance parameter is determined according to the positional relationship between the element and the adjacent node element, and the weight of the adjacent node element is determined according to the distance parameter; or, it is determined whether the element is located in the same unit cell according to the positional relationship between the element and the adjacent node element, and the weight of the adjacent node element is determined according to the result. For example, if the element is located in the same unit cell as the first adjacent node element and in a different unit cell from the second adjacent node element, the weight of the first adjacent node element is greater than the weight of the first adjacent node element. Finally, according to the weight of the adjacent node element and the weighted result of the feature coding result, it is fused into the feature coding result of the element to obtain the fused feature information of the element.

[0047] According to the fusion feature information of each atom in the perovskite crystal, the fusion feature information of each molecule, and other property information, the property prediction result of the perovskite crystal is obtained. Exemplarily, a property prediction model is pre-trained, and the fusion feature information of each atom in the perovskite crystal, the fusion feature information of each molecule, and other property information are input into the property prediction model to obtain the property prediction result of the perovskite crystal.

[0048] Since the adjacent node elements of each element include atoms and / or molecules, this embodiment fully fuses the coding features of atoms and molecules in the perovskite crystal, so that the fused feature information can better reflect the structure and physical information of the perovskite, and then determine the crystal properties by fusing the feature information. In the prediction process, the structure and physical information of the perovskite are fully utilized to improve the accuracy of the crystal property prediction.

[0049] The technical solution of the embodiment of the present invention improves the coding accuracy of atoms and molecules by encoding the atoms and molecules in the perovskite crystal separately, thereby improving the accuracy of the crystal feature coding; and predicts the crystal properties by fusion of atomic and molecular coding, thereby achieving in-depth mining of the overall structure and physical information of the crystal, and improving the accuracy of the crystal property prediction.

[0050] Figure 2 This is a flow chart of another method for predicting the properties of perovskite crystals provided in an embodiment of the present invention. This embodiment further refines the crystal prediction process in the above embodiment. Figure 2 As shown, the method includes:

[0051] S210, respectively determining element characteristic information of all elements in the perovskite crystal.

[0052] Among them, the element type of the element includes an atomic type and a molecular type.

[0053] S220. Input the element feature information of each element into an encoder corresponding to the element type of the element to obtain a feature encoding result of each element.

[0054] S230: Classify the adjacent node elements of the element according to the positional relationship between each element and each adjacent node element of the element, and obtain the position type of each adjacent node element relative to the element.

[0055] Since crystals are made up of multiple unit cells stacked together, the interaction relationships between atoms and molecules in different unit cells are not the same. Therefore, according to whether the positional relationship between each element and each adjacent node element of the element is located in the same unit cell, the adjacent node elements of the element are classified, and the position type of each adjacent node element relative to the element includes within the unit cell unit and between unit cell units.

[0056] For example, first the perovskite crystal can be described as a multi-image ,in ,in is the atomic characteristic information of atom A, is the atomic characteristic information of atom B, is the molecular characteristic information of molecule X. is the set of edges representing related pairs of elements in the crystal, N is the total number of elements in the crystal, Indicates a connection relationship Elements and The edges between elements, In order to further determine the impact of different types of multi-edge graphs, the edges in the multi-edge graph are divided into two sub-graphs: the first sub-graph includes the edges within the unit cell, and the second sub-graph includes the edges between the unit cells.

[0057] S240: Determine the element relationship feature between each element and each adjacent node element of the element according to the feature coding result of each element and the feature coding results of the adjacent node elements of the element.

[0058] The adjacent node elements of each element are determined according to the position information and connection relationship of each element in the perovskite crystal, and the influence relationship between the adjacent node elements and the element is determined according to the characteristic coding result of each element and the characteristic coding result of each adjacent node element. The element relationship characteristics are determined according to the influence relationship judgment result.

[0059] Exemplarily, the element relationship characteristics between the first element and the first adjacent node element are determined according to the comparison result of the feature coding result of the first element and the feature coding result of the first adjacent node element of the first element, for example, the comparison result can be determined according to the difference. Wherein, the first element is any element in the perovskite crystal, which can be an atom or a molecule, and the first adjacent node element of the first element is any adjacent element of the first element, which can be an atom or a molecule. Alternatively, the feature coding result of the first element and the feature coding result of the first adjacent node element are input into a predetermined network model, and the network model is learned according to the feature coding results of the first element and the first adjacent node element to obtain the element relationship characteristics between the two elements, wherein the network model can be pre-trained using an unsupervised network model, and the type of the network model is not limited here.

[0060] In a feasible embodiment, S240 includes:

[0061] Based on the strategy of graph attention network, self-attention calculation is performed on the feature encoding results of each element and the feature encoding results of the adjacent node elements of the element to obtain the element relationship features between each element and each adjacent node element of the element.

[0062] Based on the strategy of graph attention network, self-attention calculation is performed on the feature encoding results of each element and the feature encoding results of the adjacent node elements of the element. For example, the calculation process of element relationship features is expressed by the following formula:

[0063]

[0064] ;

[0065] in, represents a single-layer feedforward neural network, is a shared linear transformation of the weight matrix, It represents the encoder. Indicates from The element feature information of any element selected in They represent the elemental characteristic information of A atom, B atom and X molecule in the perovskite system respectively. express The corresponding feature encoding result is When is an atom, is the atomic encoder, when When is a molecule, For molecular encoders, Characteristic of elemental relationships between different atoms and molecules.

[0066] Optional, S240, includes:

[0067] Based on the position type of each adjacent node element relative to the element, and according to the feature coding result of each element and the feature coding results of the adjacent node elements of the element, the element relationship feature between each element and each adjacent node element of the element is determined.

[0068] Exemplarily, the feature coding result of the first element, the feature coding result of the first adjacent node element, and the position type of the first adjacent node element relative to the first element are input into a predetermined network model, and the network model learns based on the feature coding results and position features of the first element and the first adjacent node element to obtain the element relationship feature between the two elements. The position features and coding features between different elements are integrated to determine the element relationship features between the elements, thereby improving the accuracy of the description of the relationship between different elements by the element relationship features.

[0069] S250: Determine a correlation coefficient between each element and each adjacent node element of the element according to an element relationship feature between each element and each adjacent node element of the element, and a position type of the adjacent node element relative to the element.

[0070] Specifically, the first element includes multiple adjacent node elements, and the correlation coefficient between the first element and the first adjacent node element is determined based on the element relationship characteristics between the first element and the first adjacent node element, the element relationship characteristics between the first element and other adjacent node elements, and whether the position type of other adjacent node elements relative to the first element is the same as the position type of the first adjacent node element and the first element.

[0071] Exemplarily, if the position type of the first other adjacent node element relative to the first element is different from the position type of the first adjacent node element and the first element, then the influence weight of the first other adjacent node element is determined to be the first weight; if the position type of the second other adjacent node element relative to the first element is the same as the position type of the first adjacent node element and the first element, then the influence weight of the second other adjacent node element is determined to be the second weight, and the second weight is greater than the first weight. The specific values ​​of the first weight and the second weight can be determined according to actual scenario requirements.

[0072] The total impact factor is determined according to the weighted value of the element relationship feature between the first element and each other adjacent node element, the first impact factor is determined according to the element relationship feature between the first element and the first adjacent node element, and the correlation coefficient between the first element and the first adjacent node element is determined according to the ratio of the first impact factor to the total impact factor. Optionally, the element relationship feature may be processed and then weighted to determine the total impact factor and the first impact factor.

[0073] For example, define an indicator function , used to determine the element and elements Are they located in the same unit cell? If so, Equal to 1, if not equal to Equal to 0, use The weights of adjacent node elements are represented. The value of can be determined according to the subgraph content in the above embodiment. The calculation formula of the correlation coefficient between each element and each adjacent node element of the element is as follows:

[0074] ;

[0075] ;

[0076] in is the intra-unit edge correlation coefficient normalized by the softmax function, that is, the element and elements Located in the same unit cell, it is determined based on the edge information in the first subgraph. is the inter-element edge correlation coefficient, and the element and elements Located in different unit cells, it is determined based on the edge information in the second subgraph. The correlation coefficient represents the neighbor node element For Elements The importance of. represents the LeakyRelu activation function, Representation elements The set of neighbor node elements.

[0077] S260: Determine fusion feature information of each element according to the correlation coefficient between each element and each adjacent node element of the element, and the feature encoding result of each adjacent node element.

[0078] The correlation coefficient between each element and each adjacent node element of the element is used as the influence weight of the adjacent node element, and the fused feature information of each element is determined according to the feature encoding result of each adjacent node element and the weighted result of the corresponding influence weight.

[0079] By combining the intra-unit cell attention and inter-unit cell attention mechanisms, we can effectively learn the structures between different elements of the crystal, improve the accuracy of fused feature information, and thus improve the accuracy of subsequent property predictions.

[0080] Exemplarily, based on the above example, an attention operation is performed on each element, and the weighted fusion feature information of the element is calculated. :

[0081] ;

[0082] in, represents a concatenation operation. Optionally, on this basis, the feature encoding result of element i itself can also be concatenated, that is, .

[0083] S270. Obtain a property prediction result of the perovskite crystal based on the fusion characteristic information of each element and other property information of the perovskite crystal.

[0084] According to the fusion feature information of each atom in the perovskite crystal, the fusion feature information of each molecule, and other property information, the property prediction result of the perovskite crystal is obtained. Exemplarily, a property prediction model is pre-trained, and the fusion feature information of each atom in the perovskite crystal, the fusion feature information of each molecule, and other property information are input into the property prediction model to obtain the property prediction result of the perovskite crystal.

[0085] In a feasible embodiment, the other property information includes at least one of the following: position information of each element and size information of a unit cell in a perovskite crystal;

[0086] The position information of an element is used to represent the position information of the atom or molecule in the crystal. For example, a coordinate system is established in the crystal, and the coordinate information of the atom or molecule in the coordinate system is used as the position information of the element according to the established coordinate system. Since the perovskite crystal is composed of stacked identical unit cells, the size information of the unit cell refers to the length, width and height information of a single unit cell, as well as the size information of the constituent atoms and molecules in the unit cell.

[0087] Accordingly, S270 includes:

[0088] Determine the interaction characteristics between elements in the perovskite crystal based on the fusion characteristic information of each element;

[0089] Determining a position coding feature based on position information of each element; and / or determining a size coding feature based on size information of a unit cell within a perovskite crystal;

[0090] According to the position coding features and / or size coding features, as well as the interaction features between elements in the perovskite crystal, they are input into a pre-trained network model to obtain the property prediction results of the perovskite crystal.

[0091] The fusion feature information of each element in the perovskite crystal is input into the pre-trained interaction capture network to obtain the interaction features between the elements in the perovskite crystal. For example, the DimeNet++ network is used to capture the interactions between different atoms and molecules in the perovskite crystal. Specifically, spherical functions are used to represent angles, and radial functions are used to represent the distances between elements. DimeNet++ refines the element representation through its network structure. , through which the interaction characteristics between elements in the perovskite crystal are captured as the crystal interaction .

[0092] Assume that the position information of each element in the crystal is expressed as , the size information of the unit cell in the perovskite crystal is expressed as , passing the unit cell size information to the Niggli algorithm to be encoded as a lattice representation Similarly, the position information of each element in the crystal is encoded through the fully connected layer as . is the position encoding feature, Represented as a size-coded feature.

[0093] The position coding features and the interaction features between elements in the perovskite crystal are input into a pre-trained network model, or the size coding features and the interaction features between elements in the perovskite crystal are input into a pre-trained network model, or the position coding features, the size coding features and the interaction features between elements in the perovskite crystal are input into a pre-trained network model to obtain the property prediction results of the perovskite crystal. Exemplarily, the position coding features and / or the size coding features, and the interaction features between elements in the perovskite crystal are passed through a multi-layer fully connected network. Post-mapping for predicted values ​​of properties of perovskite crystals , expressed as: .

[0094] The solution of the embodiment of the present invention, by introducing the positional relationship between crystal elements and adjacent node elements, fuses the feature coding results of each element and the adjacent node elements, thereby improving the accuracy of the fused feature information of the elements, and further improving the accuracy of predicting crystal properties by utilizing the fusion of atomic and molecular codes, thereby achieving in-depth mining of the overall structure and physical information of the crystal.

[0095] Figure 3 This is a flow chart of another method for predicting the properties of perovskite crystals provided by an embodiment of the present invention. This embodiment further refines the element encoding process in the above embodiment. Figure 3 As shown, the method includes:

[0096] S310, determining element characteristic information of all elements in the perovskite crystal respectively.

[0097] Among them, the element type of the element includes an atomic type and a molecular type.

[0098] S320, inputting the atomic characteristic information of each atom into the atomic encoder to obtain the characteristic encoding result of each atom.

[0099] The atom encoder is a pre-trained network structure for encoding the characteristics of atoms in the crystal. The atomic characteristic information of each atom in the perovskite crystal is processed, and the processed atomic characteristic information is input into the atom encoder, and the output of the atomic encoder is the characteristic encoding result of the atom. Exemplarily, the processing of the atomic characteristic information can be operations such as missing value filling, outlier identification, and magnitude normalization.

[0100] S330, inputting the molecular characteristic information of each molecule into a molecular encoder to obtain a characteristic encoding result of each molecule.

[0101] The molecular encoder is a pre-trained network structure for feature encoding of molecules in the crystal. The molecular feature information of each molecule in the perovskite crystal is processed, and the processed molecular feature information is input into the molecular encoder, and the output of the molecular encoder is the feature encoding result of the molecule. Exemplarily, the processing of the molecular feature information can be operations such as missing value filling, outlier identification, and magnitude normalization.

[0102] The atomic encoder and the molecular encoder are pre-trained according to at least two encoding methods.

[0103] The first encoding method is used to construct and train the atomic encoder and the molecular encoder, and the encoding result of the second encoding method is used to guide and correct the direction of parameter optimization during the training process of the first encoding method.

[0104] Exemplarily, the loss function of the training process of the atomic encoder and the molecular encoder corresponding to the first encoding method is determined according to the deviation between the encoding result of the second encoding method and the encoding result of the first encoding method. For example, the type of reference encoding method can be increased, for example, the reference encoding method includes at least one encoding method, and the loss function of the training process of the atomic encoder and the molecular encoder corresponding to the first encoding method is determined according to the deviation between the encoding result of each reference encoding method and the encoding result of the first encoding method.

[0105] By combining at least two encoding methods to train the atomic encoder and the molecular encoder, deviation in the training directions of the atomic encoder and the molecular encoder can be avoided, thereby improving the accuracy of the encoding results of the encoders.

[0106] In a feasible embodiment, the training process of the atomic encoder and the molecular encoder includes the following:

[0107] respectively obtaining sample atomic characteristic information of each atom and sample molecular characteristic information of each molecule in the sample crystal;

[0108] Inputting the sample atom characteristic information into an initial atom encoder constructed by using a first encoding method to obtain a first sample atom encoding result;

[0109] Inputting the sample molecular characteristic information into an initial molecular encoder constructed using a first encoding method to obtain a first sample molecular encoding result;

[0110] Inputting the sample atom characteristic information into a reference atom encoder constructed using a second encoding method to obtain a second sample atom encoding result;

[0111] Inputting the sample molecular characteristic information into a reference molecular encoder constructed using a second encoding method to obtain a second sample molecular encoding result;

[0112] Obtain a first sample fusion encoding result according to the first sample atom encoding result and the first sample molecule encoding result;

[0113] Obtain a second sample fusion encoding result according to the second sample atom encoding result and the second sample molecule encoding result;

[0114] Decoding the first sample fusion encoding result to obtain a first sample atom decoding result and a first sample molecule decoding result;

[0115] Determine the first encoding mode loss information according to the first sample atom decoding result and the sample atom feature information, and the first sample molecule decoding result and the sample molecule feature information;

[0116] Determine the second encoding mode loss information according to the first sample fusion encoding result and the second sample fusion encoding result;

[0117] The initial atomic encoder and the initial molecular encoder are trained according to the first encoding mode loss information and the second encoding mode loss information to obtain trained atomic encoder and molecular encoder.

[0118] The sample atomic feature information of each atom in the sample crystal includes property information related to the atomic composition; the sample molecular feature information of each molecule includes property information related to the molecular composition. The sample crystal can be a perovskite crystal or other types of crystals, and there is no restriction on the crystal type, so as to improve the generalization ability of the encoder.

[0119] Exemplarily, the sample atomic characteristic information includes data such as average atomic formation energy, energy, average atomic energy, Fermi energy, lattice constant of cubic unit cell, α angle of lattice, β angle of lattice, γ angle of lattice, volume, number of sites, number of space groups, etc., atomic equivalent radius, atomic ionization energy, Martynov-Batsanov electronegativity scale, covalent radius of element, highest oxidation state, lowest oxidation state, first ionization energy, second ionization energy, third ionization energy, electron affinity, electronegativity, resistivity, ionization potential, critical radius, boiling point, melting point, Mendeleev number, enthalpy of fusion at melting point, allowed factor, van der Waals radius, atomic radius, number of periodic table groups, number of periodic table rows, atomic number, average anion radius, average cation radius, average ion radius, and also includes spatial position under structural stability conditions, bond length and bond angle between atoms and other information, for example, including 1) spatial position and geometric relationship between atoms, 2) bond length and bond angle between atoms, and 3) electron orbit information of atoms. The sample molecular characteristic information includes relevant characteristic information of atoms in the molecular components. The relevant characteristic information of the atoms can refer to the above-mentioned sample atomic characteristic information.

[0120] For example, is the initial atomic encoder constructed using the first encoding method, is an initial molecular encoder constructed using the first encoding method, is a reference atom encoder constructed using the second encoding method, is a reference molecular encoder constructed using the second encoding method, For the decoder, we designed an atomic encoder based on a multi-layer perceptron. and molecular encoders and decoder The structure of Map the sample atomic feature information V and sample molecular feature information U into latent vectors and ,in, Encoding result for the first sample molecule, The result of encoding the first sample atom is , , , The encoding result is fused for the first sample; then, the decoder is used The fused latent vector Decoded into reconstructed molecule vector and atomic vector , The atomic decoding result for the first sample and Decode the result for the first sample molecule.

[0121] At the same time, through the one-hot encoder Map the sample atomic feature information V and sample molecular feature information U into latent vectors and ,in, Encoding result for the second sample molecule, The result of the second sample atom encoding is , , , is the second sample fusion encoding result, and It is the encoding result of the sample atomic feature information V and the sample molecular feature information U based on one-hot.

[0122] The goal of training the encoder is to make and and and The deviation between them is minimized, for example, the mean square error (MSE) can be minimized. The calculation method of the loss information of the first encoding method is as follows:

[0123] ;

[0124] At the same time, the direction of training the encoder is the encoder constructed using the first encoding method and the encoder constructed using the second encoding method The encoding results between them are close. Therefore, the training is performed by minimizing the loss function determined by the encoding results of the two encoding methods, that is, the calculation method of the loss information of the second encoding method is as follows:

[0125] ;

[0126] The objective loss function of the final atom encoder and molecule encoder is , and The specific expression is as follows:

[0127] .

[0128] S340, obtaining a property prediction result of the perovskite crystal according to the characteristic coding result of each atom, the characteristic coding result of each molecule and other property information of the perovskite crystal.

[0129] The solution of the embodiment of the present invention combines at least two encoding methods to train the atomic encoder and the molecular encoder, thereby avoiding deviation in the training direction of the atomic encoder and the molecular encoder and improving the accuracy of the encoder encoding result.

[0130] The present invention proposes a method for predicting the properties of perovskite crystals based on pre-training. The method first trains encoders of different atoms and molecules based on a large number of sample crystals, so that the trained encoders can accurately encode atoms and molecules in any perovskite crystal, thereby improving the generalization performance of the encoder's encoding ability. In the part of predicting the properties of perovskite crystals, the encoded representation of perovskite crystal molecules and atoms obtained by the pre-trained encoder is first obtained, and then the property prediction results are obtained by the property prediction model trained with small batch data, thereby reducing the dependence on a large amount of perovskite sample data while ensuring the calculation speed, and improving the efficiency and accuracy of the prediction.

[0131] Figure 4 A schematic diagram of the structure of a device for predicting the properties of perovskite crystals provided by an embodiment of the present invention. Figure 4 As shown, the device comprises:

[0132] The element characteristic determination module 410 is used to respectively determine the element characteristic information of all elements in the perovskite crystal; wherein the element type of the element includes an atomic type and a molecular type;

[0133] The element feature encoding module 420 is used to input the element feature information of each element into the encoder corresponding to the element type of the element to obtain the feature encoding result of each element;

[0134] The crystal property prediction module 430 is used to obtain the property prediction result of the perovskite crystal according to the characteristic coding result of each element and other attribute information of the perovskite crystal.

[0135] The technical solution of the embodiment of the present invention improves the coding accuracy of atoms and molecules by encoding the atoms and molecules in the perovskite crystal separately, thereby improving the accuracy of the crystal feature coding; and predicts the crystal properties by fusion of atomic and molecular coding, thereby achieving in-depth mining of the overall structure and physical information of the crystal, and improving the accuracy of the crystal property prediction.

[0136] Optional, crystal property prediction module, including:

[0137] An element fusion feature determination unit, used to determine the fusion feature information of each element according to the feature coding result of each element, the feature coding results of the adjacent node elements of the element, and the positional relationship between the element and each adjacent node element;

[0138] The property prediction unit is used to obtain the property prediction result of the perovskite crystal according to the fusion characteristic information of each element and other attribute information of the perovskite crystal.

[0139] Optionally, the element fusion feature determination unit includes:

[0140] An element position type determination subunit is used to classify the adjacent node elements of the element according to the position relationship between each of the elements and each of the adjacent node elements of the element, and obtain the position type of each of the adjacent node elements relative to the element;

[0141] An inter-element relationship determination subunit, used to determine the element relationship feature between each element and each adjacent node element of the element according to the feature coding result of each element and the feature coding result of the adjacent node elements of the element;

[0142] An inter-element correlation coefficient determination subunit, used to determine the correlation coefficient between each of the elements and each of the adjacent node elements of the element according to the element relationship characteristics between each of the elements and each of the adjacent node elements of the element, and the position type of the adjacent node elements relative to the element;

[0143] The element fusion feature determination subunit is used to determine the fusion feature information of each element according to the correlation coefficient between each element and each adjacent node element of the element, and the feature encoding result of each adjacent node element.

[0144] Optionally, the relationship between elements determines the subunit, specifically used for:

[0145] Based on the strategy of the graph attention network, self-attention calculation is performed on the feature encoding results of each element and the feature encoding results of the adjacent node elements of the element to obtain the element relationship features between each element and each adjacent node element of the element.

[0146] Optionally, the other property information includes at least one of the following: position information of each of the elements and size information of a unit cell in the perovskite crystal;

[0147] Accordingly, the property prediction unit is specifically used for:

[0148] Determining the interaction characteristics between the elements in the perovskite crystal according to the fusion characteristic information of each of the elements;

[0149] Determining a position coding feature based on the position information of each of the elements; and / or determining a size coding feature based on the size information of a unit cell in the perovskite crystal;

[0150] The position coding features and / or the size coding features, as well as the interaction features between elements in the perovskite crystal, are input into a pre-trained network model to obtain a prediction result of the properties of the perovskite crystal.

[0151] Optional, element feature encoding module, specifically used for:

[0152] Inputting the atomic characteristic information of each atom into an atomic encoder to obtain a characteristic encoding result of each atom;

[0153] Inputting the molecular characteristic information of each molecule into a molecular encoder to obtain a characteristic encoding result of each molecule;

[0154] The atomic encoder and the molecular encoder are pre-trained according to at least two encoding methods.

[0155] Optionally, the device further comprises an encoder training module, which is used to train the atomic encoder and the molecular encoder, specifically for:

[0156] respectively obtaining sample atomic characteristic information of each atom and sample molecular characteristic information of each molecule in the sample crystal;

[0157] Inputting the sample atom characteristic information into an initial atom encoder constructed by adopting a first encoding method to obtain a first sample atom encoding result;

[0158] Inputting the sample molecular characteristic information into an initial molecular encoder constructed using a first encoding method to obtain a first sample molecular encoding result;

[0159] Inputting the sample atom characteristic information into a reference atom encoder constructed using a second encoding method to obtain a second sample atom encoding result;

[0160] Inputting the sample molecule characteristic information into a reference molecule encoder constructed using a second encoding method to obtain a second sample molecule encoding result;

[0161] Obtaining a first sample fusion encoding result according to the first sample atom encoding result and the first sample molecule encoding result;

[0162] Obtaining a second sample fusion encoding result according to the second sample atom encoding result and the second sample molecule encoding result;

[0163] Decoding the first sample fusion encoding result to obtain a first sample atom decoding result and a first sample molecule decoding result;

[0164] Determine first encoding mode loss information according to the first sample atom decoding result and the sample atom feature information, and the first sample molecule decoding result and the sample molecule feature information;

[0165] Determining second encoding mode loss information according to the first sample fusion encoding result and the second sample fusion encoding result;

[0166] The initial atomic encoder and the initial molecular encoder are trained according to the first encoding mode loss information and the second encoding mode loss information to obtain trained atomic encoder and molecular encoder.

[0167] The device for predicting the properties of perovskite crystals provided in the embodiment of the present invention can execute the method for predicting the properties of perovskite crystals provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0168] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations and do not violate public order and good morals.

[0169] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0170] Figure 5 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 required herein.

[0171] like Figure 5 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and 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 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0172] A number of 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 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 through a computer network such as the Internet and / or various telecommunication networks.

[0173] The processor 11 may be a variety of general and / or dedicated processing components 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 dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a method for predicting the properties of perovskite crystals.

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

[0175] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific reference products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including 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, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0177] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, device, or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0178] To provide interaction with a user, the systems and techniques described herein may 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 a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0179] The systems and techniques described herein may be implemented in a computing system that includes a backend component (e.g., as a data server), or a computing system that includes a switch component (e.g., an application server), or a computing system that includes a frontend component (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, switch components, or frontend components. The components of the system may 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 a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0181] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0182] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting perovskite crystal properties, characterized in that: The method includes: Determining the element characteristic information of all elements in the perovskite crystal respectively; wherein the element type of the element includes an atomic type and a molecular type; Inputting the element feature information of each element into an encoder corresponding to the element type of the element to obtain a feature encoding result of each element; According to the characteristic coding result of each element and other attribute information of the perovskite crystal, the property prediction result of the perovskite crystal is obtained.

2. The method according to claim 1, characterized in that According to the characteristic coding result of each element and other property information of the perovskite crystal, the property prediction result of the perovskite crystal is obtained, including: Determine the fusion feature information of each element according to the feature coding result of each element, the feature coding results of the adjacent node elements of the element, and the positional relationship between the element and each adjacent node element; According to the fusion characteristic information of each element and other property information of the perovskite crystal, the property prediction result of the perovskite crystal is obtained.

3. The method according to claim 2, characterized in that Determining fusion feature information of each element according to the feature coding result of each element, the feature coding results of the adjacent node elements of the element, and the positional relationship between the element and each adjacent node element includes: Classify the adjacent node elements of the element according to the positional relationship between each of the elements and each of the adjacent node elements of the element, and obtain the position type of each of the adjacent node elements relative to the element; Determine the element relationship feature between each element and each adjacent node element of the element according to the feature coding result of each element and the feature coding results of the adjacent node elements of the element; Determine the correlation coefficient between each of the elements and each of the adjacent node elements of the element according to the element relationship characteristics between each of the elements and each of the adjacent node elements of the element, and the position type of the adjacent node elements relative to the element; The fused feature information of each element is determined according to the correlation coefficient between each element and each adjacent node element of the element, and the feature encoding result of each adjacent node element.

4. The method according to claim 3, characterized in that Determining the element relationship feature between each element and each adjacent node element of the element according to the feature coding result of each element and the feature coding results of the adjacent node elements of the element, including: Based on the strategy of the graph attention network, self-attention calculation is performed on the feature encoding results of each element and the feature encoding results of the adjacent node elements of the element to obtain the element relationship features between each element and each adjacent node element of the element.

5. The method according to any one of claims 2 to 4, characterized in that: The other property information includes at least one of the following: position information of each of the elements and size information of a unit cell in the perovskite crystal; Accordingly, according to the fusion characteristic information of each element and other property information of the perovskite crystal, the property prediction result of the perovskite crystal is obtained, including: Determining the interaction characteristics between the elements in the perovskite crystal according to the fusion characteristic information of each of the elements; Determining a position coding feature based on the position information of each of the elements; and / or determining a size coding feature based on the size information of a unit cell in the perovskite crystal; The position coding features and / or the size coding features, as well as the interaction features between elements in the perovskite crystal, are input into a pre-trained network model to obtain a prediction result of the properties of the perovskite crystal.

6. The method according to claim 1, characterized in that Inputting the element feature information of each element into the encoder corresponding to the element type of the element respectively, and obtaining the feature encoding result of each element, including: Inputting the atomic characteristic information of each atom into an atomic encoder to obtain a characteristic encoding result of each atom; Inputting the molecular characteristic information of each molecule into a molecular encoder to obtain a characteristic encoding result of each molecule; The atomic encoder and the molecular encoder are pre-trained according to at least two encoding methods.

7. The method according to claim 6, characterized in that in, The training process of the atomic encoder and the molecular encoder includes the following: respectively obtaining sample atomic characteristic information of each atom and sample molecular characteristic information of each molecule in the sample crystal; Inputting the sample atom characteristic information into an initial atom encoder constructed by adopting a first encoding method to obtain a first sample atom encoding result; Inputting the sample molecular characteristic information into an initial molecular encoder constructed using a first encoding method to obtain a first sample molecular encoding result; Inputting the sample atom characteristic information into a reference atom encoder constructed using a second encoding method to obtain a second sample atom encoding result; Inputting the sample molecule characteristic information into a reference molecule encoder constructed using a second encoding method to obtain a second sample molecule encoding result; Obtaining a first sample fusion encoding result according to the first sample atom encoding result and the first sample molecule encoding result; Obtaining a second sample fusion encoding result according to the second sample atom encoding result and the second sample molecule encoding result; Decoding the first sample fusion encoding result to obtain a first sample atom decoding result and a first sample molecule decoding result; Determine first encoding mode loss information according to the first sample atom decoding result and the sample atom feature information, and the first sample molecule decoding result and the sample molecule feature information; Determining second encoding mode loss information according to the first sample fusion encoding result and the second sample fusion encoding result; The initial atomic encoder and the initial molecular encoder are trained according to the first encoding mode loss information and the second encoding mode loss information to obtain trained atomic encoder and molecular encoder.

8. A device for predicting perovskite crystal properties, characterized in that: The device includes: An element characteristic determination module, used to respectively determine the element characteristic information of all elements in the perovskite crystal; wherein the element type of the element includes an atomic type and a molecular type; An element feature encoding module, used to input the element feature information of each element into an encoder corresponding to the element type of the element, to obtain a feature encoding result of each element; The crystal property prediction module is used to obtain the property prediction result of the perovskite crystal according to the characteristic coding result of each element and other attribute information of the perovskite crystal.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and 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 so that the at least one processor can perform the method for predicting the properties of the perovskite crystal according to any one of claims 1 to 7.

10. 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 method for predicting the properties of a perovskite crystal according to any one of claims 1 to 7 when executed.

Images