A method for modeling performance of inorganic solid-state electrolytes using graph theory and machine learning
By using graph theory and machine learning methods, the crystal structure of inorganic solid electrolytes is modeled as a crystal network graph. Multi-layer convolution operations are performed using a crystal graph convolutional neural network, which solves the problem of performance prediction and doping design of inorganic solid electrolyte materials, and achieves efficient and accurate performance prediction and doping scheme design.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI XUANYI NEW ENERGY DEV CO LTD
- Filing Date
- 2026-03-11
- Publication Date
- 2026-06-16
AI Technical Summary
Existing technologies make it difficult to quickly and accurately predict the performance of inorganic solid electrolyte materials and design doping schemes, resulting in lengthy research and development cycles and high costs, which makes it difficult to meet the needs of rapid industrialization of new materials.
Using graph theory and machine learning methods, the crystal structure of inorganic solid electrolytes is abstracted into a crystal network graph. Multi-layer convolutional operations are performed using a crystal graph convolutional neural network to aggregate the local chemical environment features of atoms, and multi-dimensional performance indicators are obtained through mapping through fully connected layers.
This technology enables high-precision performance prediction of inorganic solid electrolytes, shortens the screening cycle, reduces computational resource consumption, provides a scientific theoretical basis, and offers guidance for doping design and industrial applications.
Smart Images
Figure CN122224366A_ABST
Abstract
Description
Technical Field
[0001] This article relates to the interdisciplinary field of materials science and artificial intelligence, specifically to a method for predicting the performance of inorganic solid electrolytes based on crystal graph convolutional neural networks. Background Technology
[0002] Solid-state batteries, due to their high energy density, high safety, high thermal stability, and wide operating temperature range, have gradually become an important development direction for next-generation energy storage devices. Among them, solid electrolytes, as the core component of solid-state batteries, have shown high ionic conductivity through structural and compositional design optimization, especially inorganic solid electrolytes (such as sulfide and oxide systems). The performance of some materials can even reach or exceed the level of traditional liquid electrolytes.
[0003] However, solid electrolytes still face numerous technical bottlenecks in industrial applications. For example, solid electrolyte materials are difficult to process and shape, the preparation process is complex, and production costs are high; the interfacial contact performance between solid electrolytes and electrode materials is poor, resulting in high interfacial resistance; and it is difficult to simultaneously achieve high ionic conductivity and water stability. Furthermore, while doping strategies can effectively improve material properties during the development of solid electrolyte materials, their specific mechanisms of action are still unclear, leading to a lack of theoretical guidance for doping design, lengthy research and development cycles, and high experimental costs, making it difficult to meet the demand for rapid development of new materials.
[0004] Therefore, there is an urgent need to establish an efficient R&D method that can quickly and accurately predict the performance of a large number of candidate materials and analyze the mechanism of doping schemes, so as to guide the design, screening and industrial application of new solid electrolyte materials. Summary of the Invention
[0005] Based on this, this application provides a method for modeling the performance of inorganic solid electrolytes using graph theory and machine learning, including: S1. Obtain the crystal structure data of the inorganic solid electrolyte to be predicted, extract the crystal structure features, and abstract the crystal structure into a crystal network graph based on graph theory; wherein, the crystal network graph consists of nodes representing atoms and edges representing interactions between atoms; S2. Perform multi-dimensional feature embedding operations on nodes and edges respectively to construct a network model of inorganic solid electrolyte; wherein, the node feature vectors generated by the feature embedding operation have a preset atomic feature dimension; S3. Using Crystal Graph Convolutional Neural Network (CGCNN), perform multi-layer convolution operations with a preset number of graph convolutional layers on the network model. By aggregating neighborhood environment features, update and obtain the local chemical environment feature vector of each atomic node. S4. A pooling layer is used to globally aggregate the local chemical environment feature vectors of each atomic node obtained after multi-layer convolution, thus obtaining a global feature vector characterizing the entire crystal structure; and S5. Input the global feature vector into a fully connected layer with a preset hidden layer dimension to obtain the prediction result of the inorganic solid electrolyte to be predicted; wherein, the prediction result includes thermodynamic performance index, mechanical performance index and electrochemical performance index.
[0006] On the other hand, this application also provides an inorganic solid electrolyte performance modeling apparatus, including a memory; and a processor connected to the memory, the memory being used to store instructions, and the processor being configured to perform steps of the inorganic solid electrolyte performance modeling method using graph theory and machine learning as described herein, based on the instructions stored in the memory.
[0007] On the other hand, this application also provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the inorganic solid electrolyte performance modeling method using graph theory and machine learning as described herein.
[0008] On the other hand, this application also provides a computer program product including instructions that, when executed by a computer, execute the inorganic solid electrolyte performance modeling method using graph theory and machine learning as described herein.
[0009] This application involves abstracting the microstructure of inorganic solid electrolytes into a crystal network graph containing node and edge features using graph theory. It then utilizes multi-layer convolutional operations of a crystal graph convolutional neural network to deeply capture the local chemical environment of atoms, aggregating this information into a global structural representation vector via a pooling layer. Finally, nonlinear mapping enables simultaneous prediction of multi-dimensional material performance indicators. This application, by aggregating atomic neighborhood environment features through a crystal graph convolutional neural network, can completely preserve crystal structure information, achieving a fine characterization of the microenvironment of inorganic solid electrolytes and significantly improving the prediction accuracy of the model in complex doped systems. This application can simultaneously obtain multi-dimensional prediction indicators such as thermodynamic, mechanical, and electrochemical performance indicators, effectively solving the traditional research and development challenge of balancing material stability, interfacial strain resistance, and ionic conductivity. Compared with traditional first-principles calculations (DFT), the machine learning prediction model established in this application maintains high accuracy while greatly reducing computational resource consumption, shortening the screening cycle of high-performance solid electrolytes, and reducing experimental development costs. By establishing a precise correlation between crystal structure and macroscopic performance, it provides a scientific and efficient numerical reference and theoretical basis for the elemental doping design, mechanism analysis, and industrial application of solid electrolytes.
[0010] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. Other advantages of this application can be realized and obtained by means of the embodiments described in the description and the accompanying drawings. Attached Figure Description
[0011] The accompanying drawings are used to provide an understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.
[0012] Figure 1 A schematic diagram of the architecture of an inorganic solid electrolyte performance modeling method using graph theory and machine learning provided in an embodiment of this application; Figure 2 This is a scatter plot showing the comparison between predicted and experimental values of the ionic conductivity of inorganic solid electrolytes provided in the embodiments of this application. Figure 3 This is a schematic diagram of the convergence curve of the mean absolute error (MAE) of the crystal graph convolutional neural network provided in the embodiments of this application as a function of the training epochs. Detailed Implementation
[0013] Unless otherwise stated, the technical and scientific terms used herein have the same meanings as commonly understood by one of ordinary skill in the art to which this application pertains. When a quantity, concentration, or other value or parameter is expressed as a range, preferred range, or preferred upper and lower numerical limits, it should be understood that this is equivalent to specifically disclosing any range by combining any pair of upper or preferred values with any lower or preferred value, regardless of whether the range is specifically disclosed. Unless otherwise stated, the numerical ranges listed herein are intended to include the endpoints of the range and all integers and fractions (decimals) within that range.
[0014] When used with a numerical variable, the terms "about" or "approximately" usually mean that the value of the variable and all values of the variable are within the experimental error (e.g., within the 95% confidence interval of the mean) or within ±10% of the specified value, or a wider range.
[0015] The expression "comprising," or similar expressions such as "including," "containing," and "having," is open-ended and does not exclude additional unlisted elements, steps, or components. The expression "consisting of," excludes any unspecified elements, steps, or components. The expression "substantially consisting of," limits the scope to the specified elements, steps, or components, plus optional elements, steps, or components that do not materially affect the essential and novel features of the claimed subject matter. It should be understood that the expression "comprising" encompasses both the expressions "substantially consisting of" and "consisting of."
[0016] The expression "at least one" or "one or more" indicates 1, 2, 3, 4, 5, 6, 7, 8, 9 or more kinds.
[0017] Traditional first-principles calculations (DFT) are resource-intensive and time-consuming, making it difficult to rapidly screen massive amounts of candidate materials. Existing modeling often focuses on chemical composition, neglecting the three-dimensional topological structure of crystal atoms and the local chemical environment, resulting in limited prediction accuracy. There is a lack of comprehensive prediction models that can simultaneously correlate thermodynamic, mechanical, and electrochemical performance indicators.
[0018] Based on this, this application provides a method for modeling the performance of inorganic solid electrolytes using graph theory and machine learning, including: S1. Obtain the crystal structure data of the inorganic solid electrolyte to be predicted, extract the crystal structure features, and abstract the crystal structure into a crystal network graph based on graph theory; wherein, the crystal network graph consists of nodes representing atoms and edges representing interactions between atoms; S2. Perform multi-dimensional feature embedding operations on nodes and edges respectively to construct a network model of inorganic solid electrolyte; wherein, the node feature vectors generated by the feature embedding operation have a preset atomic feature dimension; S3. Using a crystal graph convolutional neural network, perform multi-layer convolution operations with a preset number of graph convolutional layers on the network model. By aggregating neighborhood environment features, update and obtain the local chemical environment feature vector of each atomic node. S4. A pooling layer is used to globally aggregate the local chemical environment feature vectors of each atomic node obtained after multi-layer convolution, thus obtaining a global feature vector characterizing the entire crystal structure; and S5. Input the global feature vector into a fully connected layer with a preset hidden layer dimension to obtain the prediction result of the inorganic solid electrolyte to be predicted; wherein, the prediction result includes thermodynamic performance index, mechanical performance index and electrochemical performance index.
[0019] In some implementations, extracting crystal structure features in step S1 includes: extracting unit cell constants, space group information, atom types, atom position coordinates, and the type, bond energy, and bond length of interatomic interactions.
[0020] In some implementations, the central atomic node in the multi-layer convolution operation in step S3 i The feature vectors are updated according to the following formula: in, Indicates the first t In layer convolution, the central atomic node ieigenvectors; As the central atomic node i After the first t The updated feature vector obtained after +1 convolution operation; Represents the relationship between the central atomic node i All neighboring atoms that have direct interactions j and connecting the central atomic nodes i With neighboring atoms j All chemical bonds k Perform a traversal and summation; The combined feature vector is formed by concatenating the node feature vector and the edge feature vector; W f , W s This is the weight matrix. b f , b s It is the bias vector; and g For activation function, This represents element-wise multiplication.
[0021] In some implementations, the edge feature vector carries characteristic information about the interactions between atomic nodes, including at least one of bond length, bond energy, and bond type; the combined feature vector... The splicing formula is: in, and These are the central atomic nodes. i With neighboring atoms j The node feature vectors, For connecting two atoms k The edge feature vector of the edge, This indicates a vector concatenation operation.
[0022] In some implementations, the activation function and g Both are Sigmoid functions.
[0023] In some implementations, the global aggregation in step S4 uses mean pooling, and the formula for calculating mean pooling is as follows: in, v G This is the global feature vector obtained after aggregation; Atomic nodes output by the last convolutional layer i The final local chemical environment feature vector; VThis represents the total number of atomic nodes in the crystal network diagram.
[0024] In some implementations, the global aggregation in step S4 uses max pooling, which extracts the atomic nodes output by the last convolutional layer. i The global feature vector is obtained by finding the maximum value of each dimension in the final local chemical environment feature vector. v G The formula for max pooling is: in, V This represents the total number of atoms in the crystal network diagram. Atomic nodes output by the last convolutional layer i The final local chemical environment feature vector, v G This is the global feature vector obtained through aggregation.
[0025] In some implementations, the thermodynamic performance indicators, mechanical performance indicators, and electrochemical performance indicators include one or more of the following: total energy, formation energy, Young's modulus, shear modulus, Poisson's ratio, and ionic conductivity.
[0026] Some implementations also include steps for model training and performance evaluation of the modeling method: S6. Obtain inorganic solid electrolyte crystal structure data with known performance indicators as a dataset, and divide the dataset into training, validation, and test sets; based on the preset learning rate and batch size, iteratively train the network model using the training set; and use the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination to train the network model. R 2 At least one of the following, evaluate the accuracy of the modeling method in predicting the crystal structure of inorganic solid electrolytes with known performance indices.
[0027] Some implementations also include a modeling method optimization step: S7. By adjusting at least one hyperparameter among the atomic feature dimension in step S2, the number of graph convolutional layers in step S3, the hidden layer dimension in step S5, and the learning rate and batch size in step S6, the prediction error of the modeling method on the validation set converges with the training epochs, and the model hyperparameters at which the validation set performance is optimal are saved; wherein, the prediction error includes the mean absolute error (MAE), and the performance includes the coefficient of determination. R 2 .
[0028] Some implementation schemes also include steps for material design and screening based on the prediction results: S8. Construct a crystal structure model to be predicted that includes different element types or doping concentrations, construct or optimize the obtained crystal graph convolutional neural network to predict the various performance indicators of the model, and perform mechanism analysis and material recommendation for doping schemes based on the prediction results.
[0029] On the other hand, this application also provides an inorganic solid electrolyte performance modeling apparatus, including a memory; and a processor connected to the memory, the memory being used to store instructions, and the processor being configured to perform steps of the inorganic solid electrolyte performance modeling method using graph theory and machine learning as described herein, based on the instructions stored in the memory.
[0030] On the other hand, this application also provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the inorganic solid electrolyte performance modeling method using graph theory and machine learning as described herein.
[0031] On the other hand, this application also provides a computer program product including instructions that, when executed by a computer, execute the inorganic solid electrolyte performance modeling method using graph theory and machine learning as described herein.
[0032] This application describes several embodiments, but these descriptions are exemplary and not limiting, and it will be apparent to those skilled in the art that many more embodiments and implementations are possible within the scope of the embodiments described herein. Although many possible combinations of features are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with, or may replace, any feature or element of any other embodiment.
[0033] This application includes and contemplates combinations of features and elements known to those skilled in the art. The embodiments, features, and elements disclosed in this application can also be combined with any conventional features or elements to form unique inventive solutions. Any feature or element of any embodiment can also be combined with features or elements from other inventive solutions to form another unique inventive solution. Therefore, it should be understood that any feature shown and / or discussed in this application can be implemented individually or in any suitable combination. Therefore, the embodiments are not limited except by the limitations imposed by the appended claims and their equivalents. Furthermore, various modifications and changes can be made within the scope of the appended claims.
[0034] Furthermore, in describing representative embodiments, the specification may have presented methods and / or processes as a specific sequence of steps. However, the method or process should not be limited to the specific order of steps described herein, to the extent that it does not depend on such a specific order. As will be understood by those skilled in the art, other sequences of steps are also possible. Therefore, the specific order of steps set forth in the specification should not be construed as a limitation of the claims. Moreover, the claims concerning the method and / or process should not be limited to the steps performed in the written order, and those skilled in the art will readily understand that these orders can be varied and still remain within the spirit and scope of the embodiments of this application.
[0035] The equipment used in each step of the following embodiments is conventional equipment. Where no corresponding national standard exists, general international standards, conventional conditions, or conditions recommended by the manufacturer shall be followed. Unless otherwise defined or stated, all technical and scientific terms used in this application have the same meaning as those skilled in the art. Furthermore, any methods and materials similar to or equivalent to those described herein may be applied to the methods of this application.
[0036] Example Example 1. Preliminary Construction of a Performance Prediction Model for Inorganic Solid Electrolytes This embodiment provides a method for modeling the performance of inorganic solid electrolytes using graph theory and machine learning.
[0037] First, 599 crystal structure data points of inorganic solid electrolytes with known performance indicators were extracted from a crystal structure database to form a dataset. Feature information, including cell constants, space group information, atom types, atomic position coordinates, and the type, bond energy, and bond length of interatomic interactions, was extracted. A graph theory algorithm was used to abstract the crystal structure into a crystal network graph, where nodes represent atoms and edges represent interatomic interactions. Multidimensional feature embedding was performed on nodes and edges. The node embedding vector contained atomic features such as atom type, electronegativity, and ionic radius, while the edge embedding vector contained connection information such as chemical bond type, bond order, and bond length.
[0038] Then, a crystal graph convolutional neural network architecture is adopted, in which the central atomic node in the graph convolutional layer... i The feature vector update logic is as follows: Feature splicing: central atomic node i eigenvectors The splicing formula is: in, and These are the central atomic nodes. i With neighboring atomsj The node feature vectors, For connecting two atoms k The edge feature vector of the edge, This indicates a vector concatenation operation.
[0039] Convolutional update: Update according to the following formula: in, Indicates the first t In layer convolution, the central atomic node i eigenvectors; As the central atomic node i After the first t The updated feature vector obtained after +1 convolution operation; Represents the relationship between the central atomic node i All neighboring atoms that have direct interactions j and connecting the central atomic nodes i With neighboring atoms j All chemical bonds k Perform a traversal and summation; The combined feature vector is formed by concatenating the node feature vector and the edge feature vector; W f , W s This is the weight matrix. b f , b s It is the bias vector; and g For activation function, This represents element-wise multiplication.
[0040] After multiple convolutional layers, a pooling layer is used to globally aggregate the node features. This embodiment employs max-pooling to extract the maximum value of each dimension, obtaining the global feature vector. v G Finally, predictions for thermodynamic, mechanical, and electrochemical parameters are obtained through mapping of the fully connected layer.
[0041] The dataset was divided into a training set (480 records, used for iterative updates of model parameters), a validation set (60 records, used for hyperparameter selection and early stopping monitoring), and a test set (59 records, used to evaluate the final prediction performance of the model) in an 8:1:1 ratio.
[0042] In a basic test scenario of this embodiment, a set of initial hyperparameters (Table 2, Configuration A) were used for modeling verification: the atomic feature dimension was set to 32, the hidden layer dimension was set to 32, two graph convolutional layers were set, the initial learning rate was fixed at 0.01, and the batch size was 32. Experiments show that this basic configuration can initially establish a nonlinear mapping relationship between crystal structure features and material properties.
[0043] Example 2: Multidimensional performance evaluation experiment based on optimized configuration In this embodiment, performance prediction is performed using the hyperparameter optimization scheme of this application (Table 2, Configuration B). The specific hyperparameter configuration of this scheme is as follows: the atomic feature dimension is increased to 64, the hidden layer dimension is expanded to 128, a 3-layer graph convolutional layer is used for deep feature extraction, the initial learning rate is set to 0.001 and a segmented decay strategy is adopted, and the batch size is set to 64.
[0044] Under this optimized configuration, the model's multidimensional performance prediction results for the test set data are shown in Table 1.
[0045] Table 1: Evaluation Table of Predictive Performance of the Model on Different Target Properties Table 1 shows the coefficients of determination for the optimized model in terms of total energy and formation energy. R 2 All exceed 0.95. And as... Figure 2 As shown, taking ionic conductivity as an example, the predicted value and the experimental value are highly consistent ( R 2 = 0.831), with the scatter points closely distributed on both sides of the ideal fitting line, demonstrating the ability of the optimized model to capture the correlation between complex atomic local chemical environments and electrochemical performance in crystal network diagrams.
[0046] Example 3: Comparative Experiment on Model Hyperparameter Optimization To verify the technical effect of the hyperparameter optimization scheme of this application, this embodiment compares the performance of the initial settings (Table 2, Configuration A) and the optimized optimal settings (Table 2, Configuration B) in terms of formation energy and overall performance. The results are shown in Table 2.
[0047] Table 2: Comparison of Model Hyperparameter Optimization The results show that the accuracy of the prediction method is significantly improved after the model hyperparameter optimization: by configuring the optimization scheme B, the mean absolute error (MAE) of the model in predicting formation energy decreased from 0.082 to 0.041, and the prediction accuracy was improved by 50%. Another set of experimental data shows that, compared with the baseline model, the optimized model reduced the MAE of ionic conductivity prediction by 38.2% (baseline model MAE = 0.68, optimized model MAE = 0.42). R 2 Improved by 27.7% (baseline model) R 2 = 0.65, optimized model R 2 = 0.83). This indicates that through the hyperparameter co-optimization described in this application (such as increasing the feature dimension and network depth), the model can more accurately capture the nonlinear correlation between the complex atomic local chemical environment and ionic conductivity in the crystal network diagram.
[0048] Other examples Figure 3 As shown, under configuration B, the MAE of both the training and validation sets converges smoothly with increasing training epochs. The model reaches the minimum validation set error at epoch 86 and preserves the optimal parameters, effectively ensuring the model's generalization performance and avoiding overfitting.
[0049] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term "computer storage medium" includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.
[0050] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.
Claims
1. A method for modeling the performance of inorganic solid electrolytes using graph theory and machine learning, characterized in that, Includes the following steps: S1. Obtain the crystal structure data of the inorganic solid electrolyte to be predicted, extract the crystal structure features, and abstract the crystal structure into a crystal network graph based on graph theory; wherein, the crystal network graph consists of nodes representing atoms and edges representing interactions between atoms; S2. Perform multi-dimensional feature embedding operations on the nodes and edges respectively to construct the network model of the inorganic solid electrolyte; wherein the node feature vectors generated by the feature embedding operation have a preset atomic feature dimension; S3. The network model is subjected to a multi-layer convolution operation with a preset number of convolutional layers using a crystal graph convolutional neural network. By aggregating neighborhood environment features, the local chemical environment feature vectors of each atomic node are updated and obtained. S4. A pooling layer is used to globally aggregate the local chemical environment feature vectors of each atomic node obtained after multi-layer convolution, thus obtaining a global feature vector characterizing the entire crystal structure; and S5. Input the global feature vector into a fully connected layer with a preset hidden layer dimension to map and obtain the prediction result of the inorganic solid electrolyte to be predicted; wherein, the prediction result includes thermodynamic performance indicators, mechanical performance indicators and electrochemical performance indicators.
2. The method according to claim 1, characterized in that, The extraction of crystal structure features in step S1 includes: extracting unit cell constants, space group information, atom types, atom position coordinates, and the type, bond energy, and bond length of interatomic interactions.
3. The method according to claim 1, characterized in that, The central atomic node in the multi-layer convolution operation described in step S3 i The feature vectors are updated according to the following formula: in, Indicates the first t In layer convolution, the central atomic node i eigenvectors; The central atomic node i After the first t The updated feature vector obtained after +1 convolution operation; This indicates the relationship with the central atomic node. i All neighboring atoms that have direct interactions j and connecting the central atomic nodes i With the neighboring atoms j All chemical bonds k Perform a traversal and summation; The combined feature vector is formed by concatenating the node feature vector and the edge feature vector; W f , W s This is the weight matrix. b f , b s It is the bias vector; and g For activation function, This represents element-wise multiplication.
4. The method according to claim 3, characterized in that, The edge feature vector carries feature information about the interactions between atomic nodes, including at least one of bond length, bond energy, and bond type; the combined feature vector The splicing formula is: in, and These are the central atomic nodes. i With the neighboring atoms j The node feature vectors, For connecting two atoms k The edge feature vector of the edge, This indicates a vector concatenation operation.
5. The method according to claim 3, characterized in that, The activation function and g Both are Sigmoid functions.
6. The method according to claim 1, characterized in that, The global aggregation in step S4 uses max pooling, which extracts the atomic nodes output by the last convolutional layer. i The global feature vector is obtained by finding the maximum value of each dimension in the final local chemical environment feature vector. v G The maximum pooling calculation formula is as follows: in, V This represents the total number of atoms in the crystal network diagram. Atomic nodes output by the last convolutional layer i The final local chemical environment feature vector, v G The global feature vector obtained by aggregation.
7. The method according to claim 1, characterized in that, The global aggregation in step S4 uses mean pooling, and the calculation formula for mean pooling is as follows: in, v G This refers to the global feature vector obtained after aggregation; Atomic nodes output by the last convolutional layer i The final local chemical environment feature vector; V This represents the total number of atomic nodes in the crystal network diagram.
8. The method according to claim 1, characterized in that, The thermodynamic, mechanical, and electrochemical performance indicators include one or more of the following: total energy, formation energy, Young's modulus, shear modulus, Poisson's ratio, and ionic conductivity.
9. The method according to any one of claims 1 to 8, characterized in that, The method further includes the steps of model training and performance evaluation of the modeling method: S6. Obtain inorganic solid electrolyte crystal structure data with known performance indicators as a dataset, and divide the dataset into a training set, a validation set, and a test set; based on a preset learning rate and batch size, iteratively train the network model using the training set; and use the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination. R 2 At least one of the following, evaluate the accuracy of the modeling method in predicting the crystal structure of an inorganic solid electrolyte with the known performance indices.
10. The method according to claim 9, characterized in that, It also includes the modeling method optimization step: S7. By adjusting at least one of the following hyperparameters—the atomic feature dimension mentioned in step S2, the number of graph convolutional layers mentioned in step S3, the hidden layer dimension mentioned in step S5, and the learning rate and batch size mentioned in step S6—the prediction error of the modeling method on the validation set converges with each training epoch, and the model hyperparameters at which the validation set performance is optimal are saved; wherein, the prediction error includes the mean absolute error (MAE), and the performance includes the coefficient of determination. R 2 .
11. The method according to claim 10, characterized in that, The method also includes a step of material design and screening based on the prediction results: S8. Construct a crystal structure model to be predicted that includes different element types or doping concentrations, construct or optimize the obtained crystal graph convolutional neural network to predict the performance indicators of the model, and perform mechanism analysis and material recommendation for doping schemes based on the prediction results.
12. An inorganic solid electrolyte performance modeling apparatus, comprising a memory; and a processor connected to the memory, the memory for storing instructions, the processor being configured to perform the steps of the method as claimed in any one of claims 1 to 11 based on the instructions stored in the memory.
13. A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method as described in any one of claims 1 to 11.
14. A computer program product comprising instructions that, when executed by a computer, perform the method as claimed in any one of claims 1 to 11.