Structure screening method and device of positive electrode material, computer equipment and storage medium

Abstract

The invention relates to a structure screening method and device of a positive electrode material, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring unoptimized structural characteristics of a positive electrode material; based on a structure optimization network model, performing optimization processing on the unoptimized structure features to obtain optimized structure features; the structure optimization network model is used for realizing stability optimization of the unoptimized structure features; based on an energy value prediction model, performing stability analysis on the optimized structural features to obtain energy values of the optimized structural features; the energy value represents the stability of the optimized structural characteristics; according to the energy value, screening each optimized structural feature to obtain a target structural feature; and converting the target structure feature into a target structure. By adopting the method, the target structure characteristics with high stability can be efficiently and accurately screened out, and the target structure with high stability can be obtained.

Description

Technical Field

[0001] The present application relates to the technical field of electric energy, and particularly to a method and device for screening the structure of a cathode material, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] The cathode material is composed of alternating layers of lithium atoms and transition metal atoms, separated by oxygen atom layers in the middle. Finding the stable configuration of the cathode material is a fundamental problem in the computational study of cathode materials. The determination of the stable configuration will also affect the accurate calculation of many properties such as its subsequent electronic structure and electrochemical performance, so it is very important. Currently, the commonly used alternative search methods are the Ewald energy reference method and the cluster expansion fitting method. The Ewald energy reflects the electrostatic interaction in the structure and is usually correlated with the atomic arrangement and relative position of the structure. Only using the atomic coordinates and charge information of the structure, the electrostatic interaction energy can be quickly calculated through the Ewald summation method.

[0003] The energy calculated using the Ewald method does not always show a strictly linear correlation with the total energy of the structure, and for similar configurations with the same composition and only a small number of different atomic positions, their Ewald energies are often very close, making it difficult to accurately distinguish the relative stability between configurations only using this parameter. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method and device for screening the structure of a cathode material, a computer device, a computer-readable storage medium, and a computer program product, which can efficiently and accurately screen out the target structure with high stability.

[0005] In a first aspect, the present application provides a method for screening the structure of a cathode material, including:

[0006] Obtaining the unoptimized structural features of the cathode material;

[0007] Based on a structure optimization network model, performing optimization processing on the unoptimized structural features to obtain optimized structural features; the structure optimization network model is used to optimize the stability of the unoptimized structural features;

[0008] Based on an energy value prediction model, performing stability analysis on the optimized structural features to obtain the energy value of the optimized structural features; the energy value represents the stability of the optimized structural features;

[0009] According to the energy value, screening each of the optimized structural features to obtain target structural features; converting the target structural features into a target structure.

[0010] In one embodiment, the training steps of the structure optimization network model include:

[0011] Obtain the initial structure sample and the optimized structure sample of the positive electrode material;

[0012] Based on the initial generator network model, perform a structure transformation on the initial structure sample to obtain a transformed structure feature;

[0013] Through the discriminator network model, determine whether the optimized structure sample matches the transformed structure feature;

[0014] If they do not match, optimize the initial generator network model according to the difference between the optimized structure sample and the transformed structure feature to obtain an optimized generator network model; update the initial generator network model to the optimized generator network model, and return to the step of performing a structure transformation on the initial structure sample;

[0015] If they match, use the initial generator network model as the structure optimization network model.

[0016] In one embodiment, the obtaining of the initial structure sample and the optimized structure sample of the positive electrode material includes:

[0017] Obtain the initial structure sample of the positive electrode material;

[0018] Based on the density functional theory, determine the energy value and the inter-particle force of the initial structure sample;

[0019] According to the energy value and the inter-particle force, adjust the atomic positions of the initial structure sample until the adjusted structure sample meets the stable configuration condition, and use the adjusted structure sample as the optimized structure sample.

[0020] In one embodiment, the training steps of the energy value prediction model include:

[0021] Based on a set of lithium-containing inorganic compound structure samples, train a graph neural network model for energy value prediction to obtain an initial energy value prediction model;

[0022] Through the set of structure samples of the positive electrode material, train the initial energy value prediction model for energy value prediction to obtain the energy value prediction model;

[0023] Wherein, the number of samples in the feature sample set of the lithium-containing inorganic compound is more than the number of samples in the set of structure samples of the positive electrode material.

[0024] In one embodiment, the method further includes:

[0025] Determine the first transition metal atom in the optimized structural feature;

[0026] Based on the optimized structural feature and a preset quantity, determine the second transition metal atom closest to the first transition metal atom;

[0027] Based on the ratio of the quantity of the second transition metal atoms to the preset quantity, obtain the transition metal adjacency; there is a situation where the first transition metal atom is replaced by the second transition metal atom;

[0028] Wherein, the transition metal adjacency is used to represent the stability of the optimized structural feature.

[0029] In one embodiment, the first transition metal atom is a nickel atom, and the second transition metal atom is at least two transition metal atoms other than the nickel atom.

[0030] In a second aspect, the present application also provides a structural screening device for a cathode material, including:

[0031] An acquisition module, configured to acquire the unoptimized structural feature of the cathode material;

[0032] An optimization module, configured to perform optimization processing on the unoptimized structural feature based on a structural optimization network model to obtain an optimized structural feature; the structural optimization network model is used to achieve the stability optimization of the unoptimized structural feature;

[0033] A prediction module, configured to perform stability analysis on the optimized structural feature based on an energy value prediction model to obtain the energy value of the optimized structural feature; the energy value represents the stability of the optimized structural feature;

[0034] A screening module, configured to screen each of the optimized structural features according to the energy value to obtain a target structural feature; convert the target structural feature into a target structure.

[0035] In a third aspect, the present application also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the structural screening method for the cathode material in any of the above embodiments are implemented.

[0036] In a fourth aspect, the present application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the structural screening method for the cathode material in any of the above embodiments are implemented.

[0037] Fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program which, when executed by a processor, implements the steps of the method for screening the structure of the positive electrode material in any of the above embodiments.

[0038] For the above method, device, computer device, computer-readable storage medium and computer program product for screening the structure of the positive electrode material, the unoptimized structural features of the positive electrode material are obtained; based on the structure optimization network model, the unoptimized structural features are optimized to obtain optimized structural features; since the structure optimization network model is trained with the goal of optimizing the stability of the unoptimized structural features, it can accurately and efficiently adjust the unoptimized structural features to increase their stability; furthermore, based on the energy value prediction model, the optimized structural features are analyzed for stability to obtain the energy values of the optimized structural features; the energy values represent the stability of the optimized structural features; since the structure optimization network model and the energy value prediction model belong to two different machine learning models and their training processes do not interfere with each other, the energy values determined by the energy value prediction model are relatively accurate and the processing efficiency is relatively high; furthermore, according to the energy values, each of the optimized structural features is screened to obtain target structural features, so as to efficiently screen out target structural features with better stability; and then the target structural features are converted into a target structure to realize the analysis process of the stable configuration. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments of the present application or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other related drawings can be obtained without creative efforts based on these drawings.

[0040] Figure 1 It is an application environment diagram of the method for screening the structure of the positive electrode material in an embodiment;

[0041] Figure 2 It is a flowchart of the method for screening the structure of the positive electrode material in an embodiment;

[0042] Figure 3 It is a flowchart of the training steps of the structure optimization network model in an embodiment;

[0043] Figure 4 It is a flowchart of the method for screening the structure of the positive electrode material and its model training in an embodiment;

[0044] Figure 5 It is a schematic diagram of the effect of the energy value prediction model in an embodiment;

[0045] Figure 6 Schematic diagram of the stable configuration in an embodiment;

[0046] Figure 7 Schematic diagram of the voltage curve during charge and discharge in an embodiment;

[0047] Figure 8 Box plot of the energy value and the proximity of transition metals in an embodiment;

[0048] Figure 9 Structural block diagram of the structure screening device for the positive electrode material in an embodiment;

[0049] Figure 10 Internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0050] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0051] The structure screening method for the positive electrode material provided by the embodiments of the present application can be applied to an application environment as shown in Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or can be placed in the cloud or other network servers.

[0052] Among them, the terminal 102 can be, but is not limited to, various personal computers, laptop computers and other devices. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. This method can be executed based on the terminal 102, or can be implemented based on the server 104, or can also be implemented based on the interaction process between the terminal 102 and the server 104 and the two.

[0053] In an exemplary embodiment, as shown in Figure 2 A structure screening method for the positive electrode material is provided, and taking this method applied to the server 104 in Figure 1 as an example for illustration, it includes the following steps 202 to step 208. Among them:

[0054] Step 202, obtain the unoptimized structural features of the positive electrode material.

[0055] The positive electrode material is the material used at the positive electrode of the battery. The stability of the positive electrode material determines the service life of the battery and can effectively avoid the aging or structural changes of the material. The thermal stability and chemical stability of the positive electrode material are very important; for example, in high-power applications such as electric vehicles, the stability of the positive electrode material is used to prevent dangers during overheating, overcharging or short-circuiting.

[0056] The positive electrode material can be the positive electrode material of a lithium-ion battery; lithium-ion batteries mainly achieve charge and discharge based on the insertion and extraction process of lithium ions between the positive electrode and the negative electrode. A lithium-ion battery consists of parts such as a positive electrode, a negative electrode, an electrolyte, a separator and a current collector. The positive electrode material is usually a compound with a relatively high redox potential, such as NMC, NMX or other high-nickel positive electrode materials, whose function is to accept lithium ions and store energy during charging and release lithium ions during discharging; the negative electrode material generally has a relatively low potential and can accommodate lithium ions (such as graphite, etc.); the electrolyte is responsible for conducting lithium ions; the separator prevents the positive and negative electrodes from contacting and short-circuiting; the current collector is used to collect and conduct electrons. It has advantages such as high energy density, long cycle life, and low self-discharge rate, and thus is widely used in many fields. Among them, NMC refers to lithium nickel cobalt manganese oxide (LiNi x Co γ Mn 1-x-γ O 2 ), where x and y represent the molar ratios of nickel and cobalt. Different values of x and y form different NMC configurations, such as NMC811 (x = 0.8, y = 0.1), NMC622 (x = 0.6, y = 0.2), etc., thus forming configurations such as NMC811 and NMC622. These configurations represent the crystal structure characteristics at different atomic ratios and have important effects on battery performance such as energy density and cycle stability. NMX is a representation of lithium nickel manganese oxide (LiNi x Mn γ X 1-x-γ O 2 ), where x represents the molar ratio of nickel. Different values of x form different NMX configurations, which is also a structural form of the positive electrode material of a lithium-ion battery, and its performance characteristics are related to the ratios of nickel and manganese atoms.

[0057] Exemplarily, the positive electrode material can be the high-nickel positive electrode material of a lithium-ion battery. The nickel content of the high-nickel positive electrode material is relatively high; due to the high nickel content, the high-nickel positive electrode material can provide a relatively high voltage platform and energy density, which helps to improve the energy density and endurance of the lithium-ion battery and has important application potential in fields with high battery energy requirements such as electric vehicles and portable electronic devices.

[0058] The unoptimized structural features are potential crystal structural features of the cathode material. Due to the excessive number of unoptimized structural features, which may have low stability and poor balance, the optimized structure corresponding to the unoptimized structure is not calculated through density functional theory, nor is the stability analysis directly performed using the unoptimized structural features. The unoptimized structural features can be characterized in the form of graph vectors for processing by machine learning models that handle graph vectors; a graph vector is a representation form that maps the graph structure data obtained by transforming the crystal structure of the material into a vector space. Through specific algorithms or models, information such as the nodes, edges, and their mutual relationships of the graph is encoded into a vector, enabling the graph to be processed and analyzed in the vector space, such as being used as the input of a structure optimization network model. For example, the graph representation technology is used to transform the crystal structure into graph data, where atoms are used as nodes and chemical bonds between atoms are used as edges to form the graph vector form of the unoptimized structural features.

[0059] The possible atomic positions can be generated according to the crystal structure type of the material, and the simulated annealing is used to adjust the atomic positions to obtain the unoptimized structural features of the cathode material. The machine learning-assisted structure optimization method can also be used to optimize a large number of virtual configurations generated based on possible atomic arrangements. Exemplarily, first, the Li30Ni30O60 crystal structure containing 120 atoms is obtained; the unit cell size and the spatial positions of the atoms in the Li30Ni30O60 crystal structure are known; each Ni atom in the Li30Ni30O60 crystal structure is traversed and replaced with Mn or Co atoms in the NMC material to obtain the virtual configuration of NMC811, that is, the above-mentioned unoptimized structural features are obtained.

[0060] Step 204: Based on the structure optimization network model, perform optimization processing on the unoptimized structural features to obtain the optimized structural features; the structure optimization network model is used to optimize the stability of the unoptimized structural features.

[0061] The structure optimization network model is used to adjust the structure of the unoptimized structural features to achieve the stability optimization of the unoptimized structural features. The structure optimization network model is a machine learning model, that is, a model in the field of Machine Learning (ML). A machine learning model enables a computer to learn patterns and rules from data rather than performing tasks through explicit programming. The structure optimization network model is established using the crystal structure data, performance data, etc. of the material, and is used to optimize material design. It can handle large-scale and complex unoptimized structural features, automatically discover hidden patterns in the unoptimized structural features, and improve the research efficiency. Therefore, when processing a large amount of cathode material data with different configurations, it can quickly screen out material configurations with specific properties, reduce experimental and computational costs, and accelerate the material research and development process.

[0062] The optimized structural features are the features obtained by adjusting the atomic positions of the unoptimized structural features. The optimized structural features can also exist in the form of a graph vector, and this graph vector can be converted into the optimized structure, that is, the optimized crystal structure. Since the optimized structural features are obtained by processing with the structural optimization network model, the stability of the optimized structural features is relatively high and closer to the equilibrium state. The structural features of the equilibrium state can also be called the features of the equilibrium state configuration. The equilibrium state configuration is the configuration formed by the arrangement of atoms or molecules in the material system in the state of the lowest energy and the most stable structure.

[0063] Exemplarily, using the Crystalor machine learning model, these virtual configurations are converted into the optimized structural features after structure optimization. The Crystalor machine learning model is a neural network model developed based on the generative adversarial neural network and is used for material structure optimization. Its basic framework is similar to the image conversion software package pix2pix. It uses the generative adversarial neural network method to train the structural optimization network model. Taking the unoptimized material configuration as the input, through the adversarial training of the initial generator network model and the discriminator network model, the structural optimization network model is obtained, and then the optimized structural features are obtained by using the structural optimization network model; and then the optimized structural features are converted into the corresponding optimized structure.

[0064] Step 206, based on the energy value prediction model, perform a stability analysis on the optimized structural features to obtain the energy value of the optimized structural features; the energy value represents the stability of the optimized structural features.

[0065] The energy value prediction model and the structural optimization network model are different machine learning models, and the two are trained with different samples respectively. Therefore, the energy value of the optimized structure is determined through the energy value prediction model to more carefully judge the stability of the optimized structural features through this energy value.

[0066] The energy value prediction model is a machine learning model. The energy value prediction model is established by using the crystal structure data, performance data, etc. of the material and is used to predict the stability of the material. It can process large-scale and complex data, automatically discover the hidden patterns in the data, and improve the research efficiency. Therefore, when processing a large amount of data of cathode materials with different configurations, it can quickly screen out the material configurations with specific performances, reduce the experimental and calculation costs, and accelerate the material R & D process.

[0067] When the optimized structural features are represented by graph vectors, the energy value prediction model is trained based on a graph neural network model. A graph neural network model is a type of neural network specialized for processing graph-structured data. It forms a graph vector with atoms as nodes and chemical bonds between atoms as edges, takes the graph vector as input, and learns the representations of nodes and the graph by propagating information on the graph. Each node updates its own state according to its own features and the information of neighboring nodes, thereby realizing the feature extraction and analysis of the entire optimized structural features. The energy value prediction model can effectively capture the complex relationships between atoms in the crystal structure, predict the energy value of the material, and can also be used for structure classification or discovering the relationship between structure and properties. The energy value is the result output by the energy value prediction model and is used to represent the stability of the optimized structural features; the lower the energy value, the more stable the optimized structural features; the higher the energy value, the more unstable the optimized structural features.

[0068] Optionally, the energy value prediction model can be trained based on the MEGNet network model (MatErials GraphNetworks); the MEGNet model uses strategies such as graph message passing and can effectively process the crystal structure data of materials, and predicts the properties of materials by learning the features of atoms and chemical bonds; it has a high prediction accuracy. The energy value prediction model can also be trained based on the CGCNN network model (Crystal Graph Convolutional Neural Network), and the CGCNN network model extracts the features of the crystal structure through operations such as convolutional layers, and then predicts the properties of materials.

[0069] Exemplarily, based on the energy value prediction model, the energy value of the optimized structural features of the high-nickel cathode material is predicted to obtain the energy value. This energy value is used to obtain its thermodynamic stability index, that is, through corresponding prediction means, to accurately evaluate the stable degree of this crystal structure thermodynamically; according to the crystal structure and the required target performance, the stable configuration of the high-nickel cathode is predicted by high-throughput.

[0070] Step 208, according to the energy value, screen each optimized structural feature to obtain the target structural feature; convert the target structural feature into the target structure.

[0071] The target structural feature is one of the optimized structural features, and its energy value meets certain conditions. Exemplarily, multiple optimized structural features with energy values greater than the threshold can be used as the target structural features, and then the target structural features are screened and analyzed.

[0072] The target structural feature is the corresponding feature of the target structure. Converting the target structural feature into the target structure can more accurately screen out the corresponding cathode material and facilitate subsequent use.

[0073] In one example, to obtain an accurate charge-discharge curve, it is first necessary to determine the stable configurations at different lithium ion contents and calculate the corresponding average electric potential based on the energies of these stable configurations. Using the method proposed in this embodiment, the energy values of the stable configurations can be predicted by a combined machine learning method, and the charge-discharge curve can be calculated to help accelerate the computationally assisted design of high-performance cathode materials for lithium ion batteries.

[0074] The charge-discharge curve is used to describe the curve of the voltage change of a lithium ion battery over time or capacity during charging and discharging. The charging curve shows the upward trend of the voltage during the charging process of the battery, while the discharging curve represents the downward trend of the voltage during the discharging process of the battery. The area under the curve is related to the capacity of the battery, and the charge-discharge capacity of the battery can be calculated by integrating the charge-discharge curve. The shape and variation trend of the curve can also reflect the characteristics of the battery's internal resistance, polarization, etc.; the charge-discharge curve is an important basis for evaluating the performance of lithium ion batteries.

[0075] In the above method for screening the structure of the cathode material, the unoptimized structural features of the cathode material are obtained; based on the structure optimization network model, the unoptimized structural features are optimized to obtain the optimized structural features; since the structure optimization network model is trained with the goal of optimizing the stability of the unoptimized structural features, it can accurately and efficiently adjust the unoptimized structural features to increase their stability; furthermore, based on the energy value prediction model, the stability analysis of the optimized structural features is carried out to obtain the energy values of the optimized structural features; the energy value represents the stability of the optimized structural features; since the structure optimization network model and the energy value prediction model belong to two different machine learning models and their training processes do not interfere with each other, the energy values determined by the energy value prediction model are relatively accurate and the processing efficiency is relatively high; furthermore, according to the energy values, each optimized structural feature is screened to obtain the target structural features, so as to efficiently screen out the target structural features with better stability; then the target structural features are converted into the target structure to realize the analysis process of the stable configuration.

[0076] In some embodiments, as Figure 3 shown, the training steps of the structure optimization network model include:

[0077] Step 302, obtain the initial structure samples and optimized structure samples of the cathode material.

[0078] The initial structure samples are unoptimized crystal structures for training. The initial structure samples and the unoptimized structural features may have the same structure or form, and there are corresponding optimized structure samples for the initial structure samples to obtain the structure optimization network model, while the unoptimized structural features need to be input into the trained structure optimization network model to obtain the optimized structural features.

[0079] The optimized structure sample is the optimized crystal structure for training. The optimized structure sample and the optimized structure feature may have the same structure or form, and there is a corresponding initial structure sample for the optimized structure sample to obtain a structure optimization network model, while the optimized structure feature is the output result of the structure optimization network model.

[0080] The initial structure sample and the optimized structure sample can be determined based on the same sample set, and the initial structure sample can also be optimized based on the density functional theory to obtain the optimized structure sample.

[0081] In some embodiments, obtaining the initial structure sample and the optimized structure sample of the cathode material includes: obtaining the initial structure sample of the cathode material; determining the energy value and the inter-particle force of the initial structure sample based on the density functional theory; adjusting the atomic positions of the initial structure sample according to the energy value and the inter-particle force until the adjusted structure sample meets the stable configuration condition, and then taking the adjusted structure sample as the optimized structure sample.

[0082] The density functional theory (DFT) is a computational method in quantum mechanics. The electron density and energy value of the initial structure sample can be determined by the Kohn-Sham algorithm of the density functional theory.

[0083] The energy value of the initial structure sample is the overall energy value of the initial structure sample, which is used to represent the stability of the initial structure sample. The inter-particle force represents the interaction situation of the particles in the initial structure sample, and the inter-particle force includes but is not limited to: Coulomb force, exchange-correlation force, and van der Waals force.

[0084] Adjusting the atomic positions of the initial structure sample according to the energy value and the inter-particle force can make the adjusted structure sample approach the true equilibrium state through the energy value and the inter-particle force to ensure its stability.

[0085] The stable configuration condition is a condition set for the energy value and the inter-particle force. When the energy value and the inter-particle force meet the stable configuration condition, it can be determined that the adjustment of the initial structure sample is completed and approaches the true equilibrium state.

[0086] Exemplarily, the energy value of the initial structure sample is determined by calculating the electron density and the effective potential; the inter-particle force can be calculated by the gradient of the total energy with respect to the atomic coordinates or the particle positions.

[0087] In this embodiment, since the optimized structural samples are obtained by adjusting based on the density functional theory, the transformed structural features generated by the initial generator network model gradually approach the stable configuration under the density functional theory. Therefore, from the perspective of the dimensionality-reduced structural features and energy prediction values, the optimized structural features output by the structural optimization network model are similar to the optimized structural samples of DFT. In this case, without performing additional DFT structural optimization calculations, it is helpful to obtain the electronic structure properties of the optimized structural features with high-precision hierarchy and accuracy, and can effectively save calculation time and improve the overall research efficiency.

[0088] Step 304: Based on the initial generator network model, perform a structural transformation on the initial structural samples to obtain the transformed structural features.

[0089] The initial generator network model is a generator network model used to improve the stability of crystal structures. The initial generator network model is used to generate new transformed structural features according to the input initial structural samples, such as generating optimized transformed structural features in the Crystalor toolkit. The initial generator network model converts the input of the initial structural samples into transformed structural features similar to the optimized structural samples by learning the distribution characteristics of the optimized structural samples. For example, converting the unoptimized NMC configuration vector into a configuration vector close to the DFT-optimized one.

[0090] The parameters of the initial generator network model are gradually updated based on the discriminator network model. After each update, the transformed structural samples output by the initial generator network model are closer to the optimized structural samples. Through iterative training, the initial generator network model gradually learns to generate realistic transformed structural samples, providing an effective solution for material structure optimization.

[0091] Exemplarily, based on the initial generator network model, at least one of the atomic positions and chemical bonds between atoms in the initial structural samples can be adjusted to obtain the transformed structural features.

[0092] Step 306: Use the discriminator network model to determine whether the optimized structural samples match the transformed structural features.

[0093] The discriminator network model can determine the training completion status of the initial generator network model. The main task of the discriminator network model is to determine whether the transformed structural features approach the optimized structural samples to make adaptive adjustments to the training completion status.

[0094] When it is determined through the discriminator network model that the optimized structural samples match the transformed structural features, the parameters of the discriminator network model can be adjusted to improve the judgment ability of the discriminator network model, forming adversarial training.

[0095] Exemplarily, through multiple layers in the discriminator network model, feature extraction is performed on the transformed structural features to obtain the structural features to be classified; the probability value that the structural features to be classified belong to the optimized structural sample is determined; when the probability value and the corresponding threshold are greater than the threshold, it is determined that the optimized structural sample matches the transformed structural features; when the probability value and the corresponding threshold are less than the threshold, it is determined that the optimized structural sample does not match the transformed structural features.

[0096] Step 308, if they do not match, then optimize the initial generator network model according to the difference between the optimized structural sample and the transformed structural features to obtain an optimized generator network model; update the initial generator network model to the optimized generator network model, and return to the step of performing structural transformation on the initial structural sample.

[0097] The optimized generator network model is obtained by adjusting the parameters of the initial generator network model according to the feedback of the discriminator network model. In each round of training, if the optimized structural sample and the transformed structural features do not match, then optimize the initial generator network model to obtain an optimized generator network model through this round of training. Furthermore, use the optimized generator network model as the initial generator network model in the next round of training process, and perform the next round of training process, and so on for multiple rounds of iterative training of the initial generator network model.

[0098] The difference between the optimized structural sample and the transformed structural features can be determined based on a corresponding loss function. For example, the difference between the optimized structural sample and the transformed structural features can be determined based on the mean absolute error (MAE), the mean square error (MSE), or other loss functions.

[0099] Returning to the step of performing structural transformation on the initial structural sample means using the optimized generator network model obtained in the current round of training as the initial generator network model in the next round, and then executing step 304 and step 306 in the next round.

[0100] Step 310, if they match, then use the initial generator network model as the structural optimization network model.

[0101] In some embodiments, using the initial generator network model as the structural optimization network model includes: using the initial generator network model obtained in the current round of training as the structural optimization network model.

[0102] The initial generator network model and the discriminator network model form a Generative Adversarial Network (GAN); the task of the initial generator network model is to generate optimized structural features similar to the optimized structural samples, and the discriminator network model is responsible for distinguishing between the optimized structural samples and the optimized structural features generated by the generator network model. During the training process, the initial generator network model and the discriminator network model confront each other. The initial generator network model tries to generate more balanced transformed structural features for the discriminator network model to detect; the discriminator network model detects them and continuously improves its discrimination ability until it is satisfied, and the initial generator network model can generate high-quality transformed structural features similar to the optimized structural samples. The structure optimization network model is the trained generator network model.

[0103] In this embodiment, through the adversarial training between the initial generator network model and the discriminator network model, and by the initial generator network model generating corresponding transformed structural features until the transformed structural features gradually approach the optimized structural samples. When the optimization ability of the initial generator network model reaches a relatively high level, the structure optimization network model is obtained to ensure that the structure optimization network model has a relatively high structure optimization ability.

[0104] In some embodiments, the training steps of the energy value prediction model include: training a graph neural network for energy value prediction based on a set of lithium-containing inorganic substance structure samples to obtain an initial energy value prediction model; training the initial energy value prediction model for energy value prediction through a set of structural samples of the cathode material to obtain the energy value prediction model; wherein, the number of samples in the feature sample set of the lithium-containing inorganic substances is more than the number of samples in the set of structural samples of the cathode material.

[0105] The set of lithium-containing inorganic substance structure samples is a sample set composed of inorganic substances containing lithium atoms, and some of the lithium-containing inorganic substances in the set of lithium-containing inorganic substance structure samples can serve as the cathode material of a lithium battery.

[0106] The initial energy value prediction model trained by the graph neural network model only needs to be trained with a small number of structural samples of the cathode material and their energy samples in the set, so that the predicted energy value error is small. And as the amount of data used in the fine-tuning stage of transfer learning increases, the prediction accuracy of this initial energy value prediction model continuously increases until the model convergence is judged by the corresponding loss function, and the energy value prediction model is obtained. The graph neural network model therein can be MEGNet or CGCNN.

[0107] First, train based on the lithium-containing inorganic substance structure sample set, and then train through the structure sample set of the cathode material to form a transfer learning strategy. Utilize the training process of predicting energy values in the lithium-containing inorganic substance structure sample set to accelerate the training process of predicting energy values in the structure sample set of the cathode material, so as to obtain good performance even on the structure sample set of the cathode material with relatively small data volume.

[0108] The training process carried out on the lithium-containing inorganic substance structure sample set is called pre-training. In this stage, the model learns the general feature representations and patterns of the lithium-containing inorganic substance structure, perhaps learning the general patterns of atomic interactions, common features of crystal structures, etc. from the structural and property data of a large number of different types of materials; it can provide good initial model parameters for the subsequent fine-tuning on the structure sample set of the cathode material, enabling the model to converge faster and improve the energy value prediction performance of the cathode material. The pre-trained model can capture the general laws in the data, and these laws may have a certain degree of generality in different related tasks, thus reducing the difficulty and computational cost of training from scratch on the target task.

[0109] The training process carried out on the structure sample set of the cathode material is called fine-tuning. In the fine-tuning stage, the parameters of the initial energy value prediction model will be adjusted according to the characteristics of the structure sample set of the cathode material to make it more suitable for the energy value prediction requirements of the cathode material and improve the accuracy.

[0110] Exemplarily, first obtain a lithium-containing inorganic substance structure sample set of 50,000 data points, train the graph neural network model according to the lithium-containing inorganic substance structure sample set to obtain an initial energy value prediction model; furthermore, determine a structure sample set of the cathode material containing approximately 200 data points; train the initial energy value prediction model according to the structure sample set of the cathode material to achieve transfer learning, so that the model can obtain relatively accurate prediction capabilities when training on a relatively small data set.

[0111] The structural sample set of lithium-containing inorganic substances can be obtained through the Materials Project database. The graph neural network model can be MEGNet. The structural sample set of the cathode material can be the structural sample sets of NMC and NMX cathode materials. The structural sample sets of NMC and NMX cathode materials can also be obtained through the Materials Project database. The Materials Project database contains rich material information, including crystal structure data, calculated energies, physical property data such as electronic structures of many materials. The database can be accessed through a web interface or software tools such as pymatgen to obtain the structural sample set of lithium-containing inorganic substances. The pymatgen software is a software package for materials science calculations and analysis, capable of reading and writing crystal structure files in CIF format or other multiple formats, facilitating the operation and conversion of material crystal structures to form the structural sample sets of NMC and NMX cathode materials.

[0112] In one example, the crystal file in the database is read and transformed into a structural sample set of lithium-containing inorganic substances. The structural sample set of lithium-containing inorganic substances is a data object that is easy for a computer to process. Atoms, bonds, and their spatial features are extracted from this data object, and these features extracted from the data object are stored in a matrix. Subsequently, the matrix is transformed to form a graph vector corresponding to the material. The graph vector is input into the graph neural network model for training of energy value prediction. Specifically, the graph vector is sequentially mapped through multiple neural networks in the graph neural network model to obtain the energy value during the training process. According to this energy value and the corresponding loss function, the parameters of the graph neural network are adjusted until the model converges or meets the corresponding conditions, and an initial energy value prediction model is obtained. It can be understood that by replacing the structural sample set of lithium-containing inorganic substances in this example with the structural sample set of the cathode material and replacing the graph neural network with the initial energy value prediction model, and training again, an energy value prediction model can be obtained.

[0113] In this embodiment, a transfer learning strategy is adopted. First, a training process for energy value prediction is carried out based on the structural sample set of lithium-containing inorganic substances to obtain an initial energy value prediction model. Then, the initial energy value prediction model is trained using the structural sample set of the cathode material. Since the number of samples in the structural sample set of the cathode material is relatively small, the training process of the initial energy value prediction model belongs to the process of model fine-tuning, and the training efficiency is relatively high. At the same time, since the cathode material belongs to lithium-containing inorganic substances, the effectiveness of this transfer strategy is relatively high. Therefore, the energy value prediction model can more accurately obtain key features from the structure of the cathode material, and then implement corresponding energy value prediction and analysis operations.

[0114] In some embodiments, the method further includes: determining a first transition metal atom in the optimized structural feature; determining a second transition metal atom closest to the first transition metal atom based on the optimized structural feature and a preset quantity; obtaining a transition metal adjacency based on the ratio of the quantity of the second transition metal atoms to the preset quantity; and there is a case where the first transition metal atom is replaced by the second transition metal atom; wherein, the transition metal adjacency is used to represent the distribution of transition metal atoms in the optimized structural feature.

[0115] The transition metal adjacency is used to represent the distribution characteristics of multiple transition metal ions, and the transition metal adjacency is determined based on the order of the second transition metal atoms that are in a short range with the first transition metal atom. The more dispersed the second transition metal atoms are, the lower the transition metal adjacency, the lower the total configuration energy, and the more stable the configuration; the more concentrated the second transition metal atoms are, the greater the transition metal adjacency, the higher the total configuration energy, and the more unstable the configuration. The transition metal adjacency is mutually verified with the energy value output by the energy prediction model to perform a stability analysis of the optimized structural feature from multiple dimensions.

[0116] The preset quantity represents the calculation range of the transition metal adjacency. The quantity of the second transition metal atoms to be counted can be determined through the preset quantity, and then the adjacent range with the first transition metal atom in the transition metal adjacency can be controlled. The larger the preset quantity, the larger range each first transition metal atom uses to count the second transition metal atoms; the smaller the preset quantity, the smaller range each first transition metal atom uses to count the second transition metal atoms.

[0117] In this embodiment, the transition metal adjacency is mutually verified with the energy value output by the energy prediction model to perform a stability analysis of the optimized structural feature from multiple dimensions. Moreover, representing the distribution of transition metal atoms based on the transition metal adjacency helps to more deeply analyze the stability of the optimized structural feature.

[0118] In some embodiments, the above-mentioned first transition metal atom is a nickel atom, and the above-mentioned second transition metal atoms are at least two transition metal atoms other than the nickel atom.

[0119] Transition metal atoms include metal atoms in the d-block or f-block of the periodic table of elements. Transition metal atoms are usually the redox centers during the process of lithium ion insertion and extraction. The type, content, and distribution pattern of transition metal atoms will affect the crystal structure stability of the cathode material. Different transition metal atoms have different bonding strengths with oxygen atoms, and their arrangement patterns in the crystal structure will affect structural characteristics such as the lattice parameters and symmetry of the material, and further affect the diffusion kinetics of lithium ions in the material and the cycle stability of the battery.

[0120] Taking high-nickel cathode materials NMC and NMX as examples, Ni, Mn, Co, and X therein are transition metal atoms. In the NMC cathode material, the oxidation states of transition metal atoms such as nickel, cobalt, and manganese change with the insertion and extraction of lithium ions, thereby realizing the charge and discharge process of the battery. Their redox potentials and electrochemical activities have important effects on battery performance such as voltage platform and energy density.

[0121] Exemplarily, at least two transition metal atoms are manganese atoms and cobalt atoms; the specific calculation method is to start from nickel that is often doped and replaced in the high-nickel cathode configuration, calculate the sum of the number of manganese and the number of cobalt among the 6 nearest-neighbor transition metal atoms of each nickel, obtain the transition metal adjacency of each nickel, and take the sum of the transition metal adjacencies of each nickel as the total transition metal adjacency; the total transition metal adjacency is used to represent the stability of the optimized structural characteristics.

[0122] In this embodiment, when the first transition metal atom is a nickel atom, the above-mentioned second transition metal atom is the sum of different transition metal atoms other than the nickel atom, so that the stability of the nickel-containing cathode material can be analyzed more objectively.

[0123] In a specific embodiment, as Figure 4 shown in (a) of

[0124] Step 402, obtain a virtual configuration; the virtual configuration includes an unoptimized structure;

[0125] Step 404, obtain the unoptimized structural characteristics of the cathode material based on the virtual configuration; accurately predict the equilibrium state configuration based on the structure optimization network model to obtain the optimized structural characteristics.

[0126] Step 406, input the optimized structural characteristics into the energy value prediction model for energy value prediction.

[0127] Step 408, determine the stable target structural characteristics according to the energy value, and the target structural characteristics correspond to the stable configuration.

[0128] As Figure 4As shown in (b), the training process of the structure optimization network model includes: in the first branch, input the virtual configuration into the graph encoder for encoding to obtain the initial structure sample; input the initial structure sample into the initial generator model for structure transformation to obtain the transformed structure features; input the transformed structure features into the discriminator network model for detection; in the second branch, after obtaining the DFT-optimized structure sample based on the density functional theory, use the DFT-optimized structure sample as the real data and input it into the discriminator network model to determine whether the optimized structure sample matches the transformed structure features through the discriminator network model; when there is no match, obtain the optimized generator network model, update the initial generator network model to the optimized generator network model, and continue training until they match, and then use the initial generator network model as the structure optimization network model.

[0129] As Figure 4 As shown in (c), the training process of the energy value prediction model includes a pre-training stage, transfer learning on a small dataset, and a fine-tuning stage. The pre-training stage is the training stage for obtaining the initial energy value prediction model; the transfer learning and fine-tuning stage on the small dataset is the training stage for obtaining the energy value prediction model.

[0130] As Figure 5 shown, Figure 5 The ordinate of (a) is the mean absolute error (MAE), and its unit is eV / atom. Figure 5 The abscissa of (a) is the size of the training set; it can be seen that as the data in the training set increases, the mean absolute error gradually decreases, indicating that the prediction accuracy of the three graph neural network models such as MEGNet improves with more training data; correspondingly, as Figure 5 shown in (b), the performance of the untrained model, CGCNN, AtomSets, and MEGNet in predicting the energy calculated by density functional theory (DFT). The mean absolute error of the untrained dark squares is the largest, and the mean absolute errors of the other models are relatively small; the horizontal axis is the real energy value obtained by DFT calculation, and the vertical axis is the energy value predicted by the model. The dashed line in the figure represents the perfect prediction in the ideal case.

[0131] According to the above embodiments, the method for screening the stable configuration of the material crystal structure in this embodiment can be realized. Based on the new screening method in this embodiment, this embodiment further provides application examples of the material crystal structure. Specifically, these application examples include but are not limited to: using the method for screening the material crystal structure in this embodiment for screening the stable configuration of high-nickel cathode NMC811, accelerating the calculation of charge-discharge curves, and quantitatively analyzing the relationship between the arrangement of transition metal atoms and the configuration stability in high-nickel layered cathode materials. The detailed introduction of the application method is as follows:

[0132] For the application example of screening stable configurations, with the aid of the machine learning workflow constructed in steps 202-208, this embodiment performed a prediction operation on the configurational stability of the NMC811 cathode material. Through in-depth analysis and processing of relevant data, from the perspective of thermodynamic energy, 10 of the most stable configurations of NMC811 were determined by energy values. The specific crystal structures of these stable configurations are shown in the form of schematic diagrams in Figure 6 to more comprehensively and deeply reflect the specific characteristics of these stable configurations, providing solid basic data support for further research on the performance of the NMC811 cathode material in different application scenarios; among them, Figure 6 (a) represents the names of different atoms, Figure 6 (b)- Figure 6 (k) represent different stable configurations, that is, crystal structures with target structural characteristics, namely the target structure.

[0133] For the example of accelerating the calculation of charge-discharge voltage curves, it is as shown in Figure 7 ; based on the stable configurations in Figure 6 , this embodiment deeply calculated the configurational characteristics concerned by the actual battery. As one of the crucial characteristics of the lithium-ion battery cathode material, the calculated charge-discharge curves are often used to evaluate the charge-discharge performance of this material in experiments. To obtain accurate charge-discharge curves, it is first necessary to determine the stable configurations at different lithium-ion contents and calculate the corresponding average potential based on the energies of these stable configurations. As shown in Figure 7 , using the method proposed in this embodiment, the energy values of stable configurations can be predicted by a combinatorial machine learning method without performing additional DFT structure optimization calculations, and the charge-discharge curves can be calculated to help accelerate the design of high-performance cathode materials for lithium-ion batteries with computational assistance. Figure 7 (a) shows the voltage comparison of this embodiment using machine learning (ML) and the embodiment of Ewald with or without using DFT under different battery capacities; Figure 7 (b) The units of both the horizontal axis and the vertical axis are electron volts per atom (eV / atom), and a prediction comparison between this embodiment and Ewald is formed; electron volts per atom (eV / atom) is the energy value per atom, and the unit of the energy value is electron volt.

[0134] For the application of analyzing the rules of stable configurations, using the index of the adjacent degree of transition metal atoms, this embodiment can achieve a quantitative analysis of the structural characteristics of the configurations of the high-nickel cathode NMC811 and mutually confirm them with its configurational stability. From the perspective of the overall configurational energy, as shown in Figure 8As shown, the relationship between the dispersion degree, adjacency, and configurational energy of non-Ni atoms in NMC is as follows: the more dispersed the non-Ni atoms are, the lower the adjacency, the lower the total configurational energy, and the more stable the configuration. This result is consistent with the established patterns of previous computational and experimental studies. Figure 8 In it, the vertical axis represents the energy value, with the unit of electron volts per atom (eV / atom). Electron volts per atom (eV / atom) is the energy value per atom, and the unit of the energy value is electron volts; the horizontal axis represents the adjacent degree of transition metal elements; among them, the box represents the middle 50% of the data (i.e., the 25th percentile to the 75th percentile); the median line is the horizontal line within the box, representing the median of the data; the average value is marked with a dot, representing the average value of the data; the outlier is marked with a star, representing the data point outside the normal range; the whisker line is the vertical line extending out of the box, representing the range of the data, usually extending within the range of 1.5 times the interquartile range (IQR) from the edge of the box.

[0135] As Figure 8 (a) shows that when the energy in the NMC811 configuration is the lowest, the distribution characteristics of Mn and Co ions are: almost evenly distributed in different transition metal layers, while in the configurations with higher energy, there are multiple Mn ions and Co ions doped in the same layer to form a cluster structure, which is significantly different from the characteristics of the stable state structures found in existing experimental and theoretical studies; for the NMC622 and NMC333 configurations, as Figure 8 (b) and Figure 8 (c) show, fewer and evenly distributed Mn / Co atoms can stabilize the entire structure, which is consistent with the phenomena observed in previous experimental studies. Generally speaking, the combined machine learning method demonstrates the ability to search for the stable configurations of NMC811 materials through high-throughput without excessive DFT calculations. This provides a convenient and effective method for exploring the stability of various high-nickel layered cathode materials, and helps to design and optimize high-performance battery materials. This embodiment not only accelerates the search for the stable configurations of NMC811, but also helps to more deeply understand the potential mechanism of the structural stability of NMC811.

[0136] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0137] Based on the same inventive concept, an embodiment of the present application also provides a device for screening the structure of a cathode material for implementing the above-mentioned method for screening the structure of a cathode material. The solution provided by this device for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the device for screening the structure of a cathode material provided below can refer to the limitations on the method for screening the structure of a cathode material in the above text, and will not be elaborated here.

[0138] In an exemplary embodiment, as Figure 9 shown, a device for screening the structure of a cathode material is provided, including:

[0139] An acquisition module 902, configured to acquire the unoptimized structural features of the cathode material;

[0140] An optimization module 904, configured to perform optimization processing on the unoptimized structural features based on a structure optimization network model to obtain optimized structural features; the structure optimization network model is used to realize the stability optimization of the unoptimized structural features;

[0141] A prediction module 906, configured to perform stability analysis on the optimized structural features based on an energy value prediction model to obtain the energy value of the optimized structural features; the energy value represents the stability of the optimized structural features;

[0142] A screening module 908, configured to screen each of the optimized structural features according to the energy value to obtain target structural features; and convert the target structural features into a target structure.

[0143] In one of the embodiments, the optimization module 904 is configured to:

[0144] Acquire the initial structural sample and the optimized structural sample of the cathode material;

[0145] Based on the initial generator network model, perform a structural transformation on the initial structural sample to obtain the transformed structural features;

[0146] Through the discriminator network model, determine whether the optimized structural sample matches the transformed structural features;

[0147] If they do not match, optimize the initial generator network model according to the difference between the optimized structural sample and the transformed structural features to obtain an optimized generator network model; update the initial generator network model to the optimized generator network model, and return to the step of performing a structural transformation on the initial structural sample;

[0148] If they match, use the initial generator network model as the structural optimization network model.

[0149] In one embodiment, the optimization module 904 is configured to:

[0150] Obtain the initial structural sample of the positive electrode material;

[0151] Based on the density functional theory, determine the energy value and the inter-particle force of the initial structural sample;

[0152] According to the energy value and the inter-particle force, adjust the atomic positions of the initial structural sample until the adjusted structural sample meets the stable configuration condition, and use the adjusted structural sample as the optimized structural sample.

[0153] In one embodiment, the prediction module 906 is configured to:

[0154] Based on the set of lithium-containing inorganic substance structural samples, train the graph neural network model for energy value prediction to obtain an initial energy value prediction model;

[0155] Through the set of structural samples of the positive electrode material, train the initial energy value prediction model for energy value prediction to obtain the energy value prediction model;

[0156] Wherein, the number of samples in the feature sample set of the lithium-containing inorganic substance is more than the number of samples in the set of structural samples of the positive electrode material.

[0157] In one embodiment, the prediction module 906 is configured to:

[0158] Determine the first transition metal atom in the optimized structural features;

[0159] Based on the optimized structural features and a preset number, determine the second transition metal atoms closest to the first transition metal atom;

[0160] Based on the ratio of the number of the second transition metal atoms to the preset number, the transition metal adjacency is obtained; there is a situation where the first transition metal atoms are replaced by the second transition metal atoms;

[0161] wherein, the transition metal adjacency is used to represent the stability of the optimized structural features.

[0162] In one embodiment, the first transition metal atoms are nickel atoms, and the second transition metal atoms are at least two transition metal atoms other than the nickel atoms.

[0163] Each module in the above structure screening device for the cathode material can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to each of the above modules.

[0164] In an exemplary embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 10 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with external terminals through a network connection. When the computer program is executed by the processor, it implements a method for screening the structure of a cathode material.

[0165] Those skilled in the art can understand that Figure 10 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0166] In one embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0167] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0168] In one embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0169] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0170] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., without limitation.

[0171] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in the present application.

[0172] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for structural screening of positive electrode materials, characterized in that: The method comprises: Obtaining non-optimized structural features of cathode materials; Based on the structural optimization network model, the non-optimized structural features are optimized to obtain optimized structural features; the structural optimization network model is used to achieve stability optimization of the non-optimized structural features; Based on the energy value prediction model, stability analysis is performed on the optimized structural features to obtain the energy value of the optimized structural features; the energy value represents the stability of the optimized structural features; According to the energy value, each of the optimized structural features is screened to obtain a target structural feature; and the target structural feature is converted into a target structure.

2. The method according to claim 1, characterized in that The training steps of the structure optimization network model include: Obtaining an initial structure sample and an optimized structure sample of the positive electrode material; Based on the initial generator network model, the initial structure sample is structurally transformed to obtain transformed structural features; Determining whether the optimized structure sample matches the transformed structure feature through a discriminator network model; If there is no match, the initial generator network model is optimized according to the difference between the optimized structure sample and the transformed structure feature to obtain the optimized generator network model; the initial generator network model is updated to the optimized generator network model, and the step of performing structural transformation on the initial structure sample is returned; If they match, the initial generator network model is used as the structure optimization network model.

3. The method according to claim 2, characterized in that The obtaining of the initial structure sample and the optimized structure sample of the positive electrode material comprises: Obtaining an initial structure sample of the positive electrode material; Based on the density functional theory, determining the energy value and the inter-particle force of the initial structure sample; According to the energy value and the inter-particle force, the atomic positions of the initial structure sample are adjusted until the adjusted structure sample meets a stable configuration condition, and the adjusted structure sample is used as an optimized structure sample.

4. The method according to claim 1, characterized in that: The training steps of the energy value prediction model include: Based on the sample set of lithium-containing inorganic structures, the graph neural network model is trained to predict energy values, and an initial energy value prediction model is obtained; The initial energy value prediction model is trained for energy value prediction by using the structure sample set of the positive electrode material to obtain the energy value prediction model; Among them, the number of samples in the characteristic sample set of the lithium-containing inorganic substance is greater than the number of samples in the structural sample set of the positive electrode material.

5. The method according to claim 1, characterized in that The method further comprises: determining the first transition metal atom in the optimized structural feature; Determining a second transition metal atom closest to the first transition metal atom based on the optimized structural features and a preset number; Based on the ratio of the number of the second transition metal atoms to the preset number, a transition metal proximity is obtained; the first transition metal atom is replaced by the second transition metal atom; The transition metal proximity is used to represent the transition metal atomic distribution of the optimized structural features.

6. The method according to claim 5, characterized in that The first transition metal atom is a nickel atom, and the second transition metal atom is at least two transition metal atoms except the nickel atom.

7. A structure screening device for positive electrode materials, characterized in that: The device comprises: An acquisition module, used for acquiring unoptimized structural features of the positive electrode material; An optimization module, used for optimizing the non-optimized structural features based on a structural optimization network model to obtain optimized structural features; the structural optimization network model is used to achieve stability optimization of the non-optimized structural features; A prediction module, used to perform stability analysis on the optimized structural features based on an energy value prediction model to obtain an energy value of the optimized structural features; the energy value represents the stability of the optimized structural features; A screening module is used to screen each of the optimized structural features according to the energy value to obtain a target structural feature; and convert the target structural feature into a target structure.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images