Machine learning models accelerate the screening and optimization of fast ion conductor materials

Abstract

The present invention relates to a method and material for accelerating the screening and optimization of fast ion conductor materials using a machine learning model. The method comprises: inputting a data set of materials to be screened into an ion migration barrier model, screening out materials with lithium ion migration barriers less than a preset barrier value to form a first material set; the data set of materials to be screened includes a plurality of Li-A-O-S compounds having two anions; determining a second material set that meets the availability evaluation requirements of the A element, and selecting a material with the lowest lithium ion migration barrier from the second material set as the material to be optimized; doping and modifying the material to be optimized to obtain a plurality of derivative structures with different doping amounts and doping site distributions; performing structural relaxation on each derivative structure according to crystal energy and atomic force information corresponding to each derivative structure to obtain a converged derivative structure corresponding to each doping amount, and then screening out the derivative structure with the smallest lithium ion migration barrier as a screening optimization output result through the ion migration barrier model.

Description

Technical Field

[0001] The present invention relates to the field of new energy material data mining, and in particular to a method and material for accelerating the screening and optimization of fast ion conductor materials using a machine learning model. Background Art

[0002] In today's world, not only are gases released from the combustion of fossil fuels and biomass seriously polluting the air, but global warming, driven by a series of environmental damages in modern society, is also sounding the alarm. Given these concerns, the industry is focused on finding new, environmentally friendly energy sources to replace fossil fuels.

[0003] In recent years, the industry has made significant progress in the development of environmentally friendly energy sources, such as wind turbines, solar thermal receivers, and solar cells, which has placed higher demands on the development of energy storage devices. Therefore, secondary batteries can store sustainable energy stably for a long time and have a long cycle life, which conforms to the concept of environmental friendliness and has become an important breakthrough in the development of modern electrochemistry. In addition, all-solid-state batteries have gained increasing attention as the next generation of environmentally friendly batteries due to their higher chemical stability, higher power density, and higher energy density compared to traditional liquid electrolyte batteries. The exploration of inorganic solid electrolyte materials with high ionic conductivity is one of the key tasks in the development of all-solid-state batteries.

[0004] Most widely used solid-state electrolyte materials were discovered experimentally. However, the vast space of candidate materials complicates the task of synthesizing and measuring their properties, making experimental exploration of new solid-state electrolyte materials challenging. In addition to experimental methods, theoretical calculations have gradually improved their accuracy in simulating material properties. As an auxiliary tool, they have successfully predicted many material properties, making material screening and design possible.

[0005] The screening of fast ion conductor materials based on high-throughput computing involves a large number of computational tasks. The screening process usually includes steps such as structure optimization, energy calculation, charge analysis, and ion migration barrier estimation. The first step involves finding equilibrium atomic positions in a large number of candidate compounds or derivative structures. Since a dense material structure space needs to be explored, the computational cost is very high. The last step involves the study of the ion dynamics properties in the material, where transition state theory and molecular dynamics are the main simulation methods, but both are very time-consuming computational tools. Therefore, to this day, the screening and optimization of fast ion conductor materials is still a costly, time-consuming, and labor-intensive task. Summary of the Invention

[0006] Embodiments of the present invention provide a method and material for screening and optimizing fast ion conductor materials using a machine learning model. This method uses a screening optimization method to explore the optimal doping scheme, including the optimal doping ratio and doping site distribution, to obtain fast ion conductor materials with improved ion mobility.

[0007] In a first aspect, an embodiment of the present invention provides a method for accelerating the screening and optimization of fast ion conductor materials using a machine learning model, the screening and optimization method comprising:

[0008] Inputting a data set of materials to be screened into an ion migration barrier model to screen out materials having a lithium ion migration barrier less than a preset barrier value to form a first material set; the data set of materials to be screened includes a plurality of Li-AOS compounds having two anions, wherein A is a trivalent metal element;

[0009] Determining, from the first material set, a second material set that meets the A element usability evaluation requirements based on predetermined A element usability evaluation parameters;

[0010] selecting the material with the lowest lithium ion migration barrier from the second material set as the material to be optimized;

[0011] Doping and modifying the material to be optimized based on a preselected doping material to obtain a plurality of derivative structures with different doping amounts and doping site distributions;

[0012] According to the crystal energy and atomic force information corresponding to each derivative structure, structural relaxation is performed on the derivative structures with different doping amounts and doping site distributions, and a converged derivative structure corresponding to each doping amount is obtained as the third material set;

[0013] The third material set is input into the ion migration barrier model, and the derivative structure with the smallest lithium ion migration barrier is screened out as the screening optimization output result.

[0014] Preferably, the preset potential barrier value is specifically 1.265 eV.

[0015] Preferably, the element A availability evaluation parameter is determined by evaluating the price and abundance of element A.

[0016] Preferably, performing structural relaxation on the derivative structures with different doping amounts and doping site distributions according to the crystal energy and atomic force information corresponding to each derivative structure to obtain a converged derivative structure corresponding to each doping amount specifically includes:

[0017] Determining the crystal energy and atomic force information corresponding to each derived structure; wherein the atomic force information includes the atomic force information of each element in the derived structure;

[0018] The velocity-Verlet molecular dynamics algorithm is used to update the atomic positions in the derived structure, and the most stable derived structure under each doping amount is determined according to the crystal energy and atomic force information.

[0019] Further preferably, the crystal energy and atomic force information corresponding to each derivative structure is determined as follows:

[0020] The crystal energy and atomic force information corresponding to each derived structure are determined according to the crystal energy model and the atomic force model; wherein the atomic force model is used to respectively determine the atomic force information of each element in the derived structure.

[0021] Preferably, the ion migration barrier model uses a random forest algorithm; the construction of the ion migration barrier model includes:

[0022] The inorganic crystal database was used to obtain all the n+ The basic data set is constructed by compound, where M n+ Includes: Ag + ,Al 3+ ,Ca 2+ ,Li + ,Mg 2+ ,Na + , or Zn 2+ ;

[0023] The M is calculated based on the bond valence theory. n+ The migration barrier of the compound is less than 2eV. n+ The compounds were used as model datasets;

[0024] The Matminer package is used to extract the features of the model data, where the model data includes M n+ The mobile ions, framework ions and all ions in the compound; the mobile ions are M n+ , the framework ion is M n+ In addition to M n + Other ions;

[0025] Based on the extracted features, the ion migration barrier model is trained.

[0026] Preferably, the preselected doping material is Mg;

[0027] The derivative structure with the smallest lithium ion migration barrier is Li 1-2x Mg x BiOS; x = 0.1875; the stoichiometric ratio of Li, Mg, Bi, O, and S in the chemical formula is 10:3:16:16:16; unit cell parameters α=β=γ=90°, and the space group is Pca21.

[0028] Preferably, the screening optimization method is used for optimizing the screening of fast ion conductor materials.

[0029] In the second aspect, the embodiment of the present invention provides a fast lithium ion conductor material, the material is Li 1- 2x Mg x BiOS; x<0.375; space group is Pca21.

[0030] Preferably, the fast lithium ion conductor material is used as a solid electrolyte material in lithium ion batteries or lithium metal batteries.

[0031] The present invention uses a machine learning model to accelerate the screening and optimization of fast ion conductor materials. This method develops an accelerated fast ion conductor material screening process. This method utilizes a machine learning model to accelerate time-consuming steps in the screening process, accelerating doping structure relaxation and migration barrier estimation. This significantly shortens the material screening cycle and enables more accurate screening and optimization of fast ion conductor materials. By exploring the optimal doping scheme, including the optimal doping ratio and doping site distribution, the fast ion conductor materials obtained through the present invention's screening and optimization method exhibit improved ion migration performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The technical solutions of the embodiments of the present invention are further described in detail below through the accompanying drawings and examples.

[0033] Figure 1 A flow chart of a method for accelerating the screening and optimization of fast ion conductor materials using a machine learning model provided in an embodiment of the present invention;

[0034] Figure 2 a is a schematic diagram of the performance prediction of the ion migration barrier model proposed in an embodiment of the present invention;

[0035] Figure 2 b is an important descriptive factor of the ion migration barrier model according to the embodiment of the present invention;

[0036] Figure 2 c is a schematic diagram of quantitative parameters of important descriptive factors of the ion migration barrier model according to an embodiment of the present invention;

[0037] Figure 3 The lithium ion migration barriers of 49 materials in the original material data set based on which the screening and optimization of the embodiments of the present invention are performed;

[0038] Figure 4 4a and 4b are schematic diagrams of the crystal structure of the preselected doping material Li-Bi-OS proposed in an embodiment of the present invention;

[0039] Figure 4 c and 4d are schematic diagrams of lithium ion migration performance of the preselected doping material Li-Bi-OS proposed in an embodiment of the present invention;

[0040] Figure 5 5a and 5b are schematic diagrams of the prediction performance of the crystal energy model proposed in the embodiment of the present invention;

[0041] Figure 6 af are schematic diagrams of the predicted performance of the atomic force model based on five different elements: Li, Mg, Bi, O, and S according to an embodiment of the present invention;

[0042] Figure 7 a, 7b are Li proposed in the embodiment of the present invention 1-2x Mg x Schematic diagram of the crystal structure of BiOS (x = 0.1875);

[0043] Figure 7 c is Li proposed in the embodiment of the present invention 1-2x Mg x Schematic diagram of lithium ion migration performance of BiOS (x=0.1875);

[0044] Figure 8 Schematic diagram of the process of the method for accelerating the screening and optimization of fast ion conductor materials using a machine learning model proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The present invention will be further described below through the accompanying drawings and specific embodiments, but it should be understood that these embodiments are only used for more detailed description and should not be understood as limiting the present invention in any form, that is, they are not intended to limit the scope of protection of the present invention.

[0046] The present invention proposes a method for accelerating the screening and optimization of fast ion conductor materials using a machine learning model. The method is used to accelerate the screening and optimization of fast ion conductor materials. The main steps of the method are as follows: Figure 1 Shown, including:

[0047] Step 110: Input the material data set to be screened into the ion migration barrier model to screen out materials with lithium ion migration barriers less than a preset barrier value to form a first material set;

[0048] The construction and training of the ion migration barrier model includes:

[0049] S1, using the inorganic crystal database to obtain all the crystals containing a specific cation M n+ The basic data set is constructed by compound, where M n+ Includes: Ag + ,Al 3+,Ca 2+ ,Li + ,Mg 2+ ,Na + , or Zn 2+ ;

[0050] S2, calculate the M based on bond valence theory n+ The migration barrier of the compound is less than 2eV. n+ The compounds were used as model datasets;

[0051] The ion migration barrier of the compound is used as a label for the model output. The model data set composed of compounds with a migration barrier less than 2 eV selected for model training in the present invention contains a total of 3136 training sample structures.

[0052] S3, using Matminer package to extract the features of model data, where the model data includes M n+ The mobile ions, framework ions and all ions in the compound; wherein the mobile ion is M n+ , the framework ion is M n+ In addition to M n+ Other ions;

[0053] Specifically, the Matminer package is used to extract the features of the model data and the material features used as the model input: based on the basic properties of the material of composition and structure, the features can be divided into those that focus only on mobile ions (i.e., ions that can migrate in the structure, here refers to M n+ ), only focusing on the framework ions other than the migrating ions (i.e., ions that hardly migrate in the structure, here refers to ions other than M n+ All features were extracted using the Matminer package. Detailed feature information is shown in Table 1.

[0054]

[0055] Table 1

[0056] In addition to these compositional and structural features extracted from the material's chemical formula and structure file, features describing the migrating ions can also be added, including the migrating ion species, migrating ion valence, migrating ion radius, and migrating ion electronegativity. In the specific implementation of the present invention, 296 features were ultimately extracted for each material.

[0057] S4, based on the extracted features, the ion migration barrier model is trained.

[0058] Because the ion migration barriers of the compounds in the inorganic crystal database are known, the ion migration barrier model can be obtained by training and converging the model training output results based on the known ion migration barriers.

[0059] The ion migration barrier model of the present invention uses a random forest algorithm. The hyperparameter settings of the algorithm are shown in Table 2.

[0060]

[0061] Table 2

[0062] The trained ion migration barrier model achieved a mean absolute error of 0.265 eV on the test set. This error is less than the 1 eV commonly used screening criteria for fast ion conductor materials, so the model's predictions can be used for fast ion conductor screening.

[0063] The relationship between the ion migration barrier predicted by the model and the ion migration barrier calculated by bond valence theory is as follows: Figure 2 As shown in Figure 2a, most data points are distributed around the reference line (dashed line), indicating that the model has good predictive performance. The random forest algorithm is an interpretable algorithm that can automatically identify the most important features in the model. Figure 2 b shows the ten most important features of the model. Figure 2 c shows the Spearman correlation between the ten most important characteristics and the ion migration barrier. Variables with positively correlated parameters are positively correlated, and vice versa. Properties numbered 1-11 in the figure represent: silver content fraction, average bond angle of migrating ions, lithium content fraction, average bond length of migrating ions, average p-shell valence electron content, number of migrating ion sites, order parameter of migrating ions in a tetrahedral coordination environment, energy band center, average distance offset between framework ions and surrounding ions, Young's modulus, and activation energy of ion migration calculated using bond valence theory.

[0064] The trained ion migration barrier model is used to screen a material dataset, wherein the material dataset includes multiple Li-AOS compounds with two anions, where A is a trivalent metal element.

[0065] In the implementation process of the present invention, first, based on a series of virtual structures obtained by element substitution in the article "Computational Discovery of Stable Heteroanionic Oxychalcogenides ABXO (A, B = Metals; X = S, Se, and Te) and Their Potential Applications" by He J et al., 49 lithium-ion-containing compounds ABXO (A is Li, B is a trivalent metal element; X = S, Se, Te) in stable and metastable states were selected. Their chemical formulas are shown in Figure 2. Figure 3 The horizontal axis of .

[0066] In order to obtain potential fast lithium ion conductor materials, the screening condition preset barrier value is specifically 1.265eV. Among the above 49 compounds, the vertical coordinate is located at Figure 3 All materials below the center dotted line meet the requirements.

[0067] Because sulfur has a smaller atomic mass than selenium and tellurium, materials containing oxysulfide ions have a greater mass energy density advantage. The present invention selects compounds where X=S to form the material data set to be screened.

[0068] The resulting data set of materials to be screened includes 9 types of Li-AOS, whose chemical formula components are Li-Bi-OS, Li-VOS, Li-Dy-OS, Li-Sc-OS, Li-Gr-OS, Li-As-OS, Li-Al-OS, Li-Ga-Os, and Li-In-OS.

[0069] Step 120 , determining a second material set that meets the A element usability evaluation requirements from the first material set based on the predetermined A element usability evaluation parameters;

[0070] Specifically, the evaluation parameter of element A availability is determined by evaluating the price and abundance of element A.

[0071] In a specific implementation, corresponding benchmark reference data are set according to the price and abundance of element A, and the prices and abundances of different elements A are quantitatively evaluated using the benchmark reference data.

[0072] In a specific example, the availability evaluation parameter of element A = α × (price of element A / benchmark reference price data) + β × (abundance of element A / benchmark reference abundance data). α is a price factor, and β is an abundance factor. α, β, the benchmark reference price data, and the benchmark reference abundance data can be set based on empirical data or experimental requirements.

[0073] This step can screen out expensive and / or rare A elements.

[0074] In this implementation, Li-Dy-OS, Li-As-OS, Li-In-OS, Li-Sc-OS, and Li-Ga-OS were eliminated. This left four materials with different compositions forming the second material set: Li-Bi-OS, Li-VOS, Li-Gr-OS, and Li-Al-OS, with lithium-ion migration barriers of 1.171 eV, 1.214 eV, 1.177 eV, and 1.175 eV, respectively.

[0075] Step 130 , selecting a material with the lowest lithium ion migration barrier from the second material set as the material to be optimized;

[0076] Among the four materials screened in the previous step, Li-Bi-OS, the material with the lowest lithium ion migration barrier, was selected for further physical property research.

[0077] The crystal structure of Li-Bi-OS is shown in Figure 4 a, 4b, the connected area in the structure represents the lithium ion migration path calculated by bond valence theory. It can be seen that the lithium ions of Li-Bi-OS are transported two-dimensionally in the ac plane. Figure 4 c It can be seen that the ion migration barrier calculated by the expansion elastic band theory is 0.047eV, indicating that the lithium ion migration barrier in Li-Bi-OS is indeed very low. Further first-principles molecular dynamics simulations are performed, such as Figure 4 As shown in Figure d, very few diffusion events occur in the structure after 120 ps simulation at 1200 K, indicating poor ion migration performance due to the high defect formation energy.

[0078] Step 140 , modifying the material to be optimized by doping with a preselected doping material to obtain a plurality of derivative structures with different doping amounts and doping site distributions;

[0079] In the present invention, the preselected doping material is divalent Mg ions, which is mainly used to solve the problem of poor ion migration performance caused by high defects. By replacing some lithium ions in Li-Bi-OS with divalent magnesium ions, Li-Bi-OS is doped and modified to generate 615 derivative structures Li with different doping amounts and doping site distributions. 1-2x Mg x BiOS (x = 0.0625, 0.125, 0.1875, 0.25, 0.3125, 0.375, 0.4375), at this time x = 0 corresponding to the LiBiOS has a total of 64 atoms.

[0080] If it is for other purposes or there is a specific material selection, different pre-selected doping materials can also be used to modify the material. The method of the present invention can also be applied to doping of other materials besides Mg doping.

[0081] Step 150 , performing structural relaxation on the derivative structures with different doping levels and doping site distributions based on the crystal energy and atomic force information corresponding to each derivative structure, and obtaining a converged derivative structure corresponding to each doping level as the third material set;

[0082] Specifically, the crystal energy and atomic force information corresponding to each derived structure are determined; wherein the atomic force information includes the atomic force information of each element in the derived structure; further in detail, the crystal energy and atomic force information corresponding to each derived structure can be determined based on the crystal energy model and the atomic force model; wherein the atomic force model is used to separately determine the atomic force information of each element in the derived structure.

[0083] The velocity-Verlet molecular dynamics algorithm is used to update the atomic positions in the derived structure, and the most stable derived structure under each doping amount is determined based on the crystal energy and atomic force information.

[0084] The crystal energy model and atomic force model training of the present invention adopts 66 randomly generated initial doping structures Li under different doping amounts. 1-2x Mg x BiOS (x = 0, 0.0625, 0.125, 0.1875, 0.25, 0.3125, 0.375, 0.4375), the 66 initial structures were subjected to density functional theory-based structural relaxation, and the structural snapshots during structural relaxation were obtained, and a total of 9858 structures were obtained as the training sample data set.

[0085] Crystal energy and atomic force information for all structures was collected as data tags for the two models. A total of 9,858 crystal energy data points were obtained. Regarding atomic forces, each structure contained multiple atoms, and each atom had information about the force components in the x, y, and z directions. From the 9,858 structures, every tenth one was selected and the force information for all atoms in the selected structure was collected, resulting in a total of 58,889 atomic force data points. This data was used as material tags for the model output.

[0086] Two material characteristics based on the local environment of atoms in the structure are used as model inputs.

[0087] For the crystal energy model, the smoothed expansion of atomic orbitals (SOAP) introduced by Himanen L et al. in the article “DScribe: Library of descriptors for machine learning in materials science” is used as the material feature. SOAP can be used to encode the local environment in the atomic structure. The atomic structure of an atom is converted into an atomic density field. In the atomic density field function, the spherical harmonic Y lm The maximum angular quantum number l in (θ,φ) max , radial basis function g n The maximum number of series n in (r) max and the cutoff radius r cut Need to be specified manually. Local features cannot be used directly to describe the material properties related to the whole. Therefore, we use the average SOAP value of all atoms in the structure as the characteristic of the total energy model of the crystal. Use parameter setting , l max =5,n max =5, and 1950 features were generated for each structure.

[0088] For the atomic force model, AGNI, introduced by Botu V et al. in the article "Machine Learning Force Fields: Construction, Validation, and Outlook", is used as the material feature. In AGNI, the local environmental influence on atom i in the k direction (k = x, y, z) is given by a function. The parameters of the function that need to be specified manually are the Gaussian function width η and the truncation radius r. cut . Use parameter settings , η ranges from 1.6 to 16 with an interval of 0.4, and 540 features are generated for each atom.

[0089] The crystal energy model uses the random forest model algorithm, and the atomic force model uses the ridge regression algorithm. The hyperparameter settings of the algorithms are shown in Tables 3 and 4.

[0090]

[0091] Table 3

[0092] As shown in Table 3, the average error of the crystal energy model on the test set is 1 meV / atom, which is comparable to the energy accuracy calculated based on density functional theory. The relationship between the energy predicted by the model and the energy calculated by density functional theory is shown in Figure 5 As shown in (a), most of the data points are distributed around the reference line (dashed line in the figure), indicating that the model has good prediction performance. Figure 5b is the absolute value of the error between the predicted energy and the density functional calculated energy. The prediction error of the model is very small.

[0093] For the atomic force model, the atomic force model is constructed separately according to the element type of the atom of concern. Because the system contains five different elements, lithium, bismuth, oxygen, sulfur, and magnesium, a total of five atomic force models are constructed.

[0094]

[0095] Table 4

[0096] As shown in Table 4, the average error of the atomic force model on the test set is The relationship between the forces predicted by the model and those calculated by density functional theory is shown in Figure 6 As shown in Figures ae, most of the data points are distributed around the reference line (dashed line in the figure), indicating that the model has good prediction performance. Figure 6 f is the absolute value of the error between the predicted force and the force calculated by density functional theory. The prediction error of the model is very small.

[0097] During structural relaxation of the initial doped structure, the two machine learning models are used to predict the total crystal energy and atomic forces, respectively. A modified velocity-Verlet molecular dynamics (MDMin) method is used to update the atomic positions in each relaxation step. The force used for relaxation has a convergence accuracy of

[0098] Using this structural relaxation model, the above 615 derived structures Li with different doping amounts and doping site distributions were analyzed. 1- 2x Mg x Structural relaxation was performed using BiOS (x = 0.0625, 0.125, 0.1875, 0.25, 0.3125, 0.375, and 0.4375). The most stable structures at each doping level were selected based on the predicted energies, yielding a total of seven structures for each of the seven doping levels.

[0099]

[0100] Table 5

[0101] Step 160 : Input the third material set into the ion migration barrier model, and screen out the derivative structure with the smallest lithium ion migration barrier as the screening optimization output result.

[0102] The ion migration barrier model was again used to predict the lithium ion migration barriers of the seven stable structures obtained in step 150. The energy prediction values of these seven structures are shown in Table 5. Among them, the structure corresponding to x = 0.1875 has the lowest ion migration barrier and is considered to be the optimal doping scheme, which is the output result.

[0103] In addition, we can see the derivative structure Li with x < 0.375 1-2x Mg x The lithium ion migration activation energy corresponding to BiOS is less than 1.1eV, so they all have good practical application prospects.

[0104] Li under the optimal doping scheme 1-2x Mg x The crystal structure of BiOS (x = 0.1875) is as follows Figure 7 As shown in a. The first principle molecular dynamics simulation is performed on it, as shown in Figure 7 c shows the mean square displacement of lithium ions during the simulation of 100ps at 525K. Compared with the results of the undoped LiBiOS structure ( Figure 4 d) Comparison, Li 1-2x Mg x BiOS (x = 0.1875) has significantly more migration events at lower temperatures and shorter simulation times, indicating that the lithium ion migration performance is significantly improved after doping. Figure 7 b shows Li 1-2x Mg x Figure 3. The migration path of a lithium ion in BiOS (x=0.1875). The lithium ions mainly migrate in the ac plane, and the two-dimensional ion migration characteristics in the original material are retained.

[0105] The present invention screens and optimizes the above materials to find the derivative structure with the smallest lithium ion migration barrier, Li 1-2x Mg x BiOS; x = 0.1875; the stoichiometric ratio of Li, Mg, Bi, O, and S in the chemical formula is 10:3:16:16:16; unit cell parameters α=β=γ=90°, and the space group is Pca21.

[0106] The machine learning model proposed in this invention accelerates the screening and optimization method of fast ion conductor materials, which mainly involves processing modules and their execution algorithms. Figure 8 As shown, they have been described in detail in the above method flow description and will not be repeated here.

[0107] The present invention applies this screening process to screen fast lithium ion conductors from 49 compounds with two anions, ABXO (A, B are Li and other trivalent metal elements; X = S, Se, Te), and screens LiBiOS as a potential fast ion conductor material. Although LiBiOS exhibits a relatively low migration barrier value, further analysis shows that its high carrier formation energy limits ion transport. Therefore, we use Mg in LiBiOS. 2+ Replace part of Li + By introducing vacancy defects into the structure to reduce the carrier formation energy, and using an accelerated screening process to explore the best doping scheme, including the optimal Li-Mg ratio and doping site distribution, the optimal doping scheme Li 1-2x Mg x BiOS (x=0.1875) has better ion migration performance than the undoped structure.

[0108] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for accelerating the screening and optimization of fast ion conductor materials using a machine learning model, characterized in that: The screening optimization method comprises: Inputting a data set of materials to be screened into an ion migration barrier model to screen out materials having a lithium ion migration barrier less than a preset barrier value to form a first material set; the data set of materials to be screened includes a plurality of Li-AOS compounds having two anions, wherein A is a trivalent metal element; Determining, from the first material set, a second material set that meets the A element usability evaluation requirements based on predetermined A element usability evaluation parameters; selecting the material with the lowest lithium ion migration barrier from the second material set as the material to be optimized; Doping and modifying the material to be optimized based on a preselected doping material to obtain a plurality of derivative structures with different doping amounts and doping site distributions; According to the crystal energy and atomic force information corresponding to each derivative structure, structural relaxation is performed on the derivative structures with different doping amounts and doping site distributions, and a converged derivative structure corresponding to each doping amount is obtained as the third material set; The third material set is input into the ion migration barrier model, and the derivative structure with the smallest lithium ion migration barrier is screened out as the screening optimization output result.

2. The screening optimization method according to claim 1, characterized in that The preset potential barrier value is specifically 1.265 eV.

3. The screening optimization method according to claim 1, characterized in that The element A availability evaluation parameter is determined by evaluating the price and abundance of element A.

4. The screening optimization method according to claim 1, characterized in that The method of performing structural relaxation on the derivative structures with different doping amounts and doping site distributions according to the crystal energy and atomic force information corresponding to each derivative structure to obtain a converged derivative structure corresponding to each doping amount specifically includes: Determining the crystal energy and atomic force information corresponding to each derived structure; wherein the atomic force information includes the atomic force information of each element in the derived structure; The velocity-Verlet molecular dynamics algorithm is used to update the atomic positions in the derived structure, and the most stable derived structure under each doping amount is determined according to the crystal energy and atomic force information.

5. The screening optimization method according to claim 4, characterized in that: The crystal energy and atomic force information corresponding to each derivative structure are determined as follows: The crystal energy and atomic force information corresponding to each derived structure are determined according to the crystal energy model and the atomic force model; wherein the atomic force model is used to respectively determine the atomic force information of each element in the derived structure.

6. The screening optimization method according to claim 1, characterized in that: The ion migration barrier model uses a random forest algorithm; the construction of the ion migration barrier model includes: The inorganic crystal database was used to obtain all the n+ The basic data set is constructed by compound, where M n+ Includes: Ag + ,Al 3+ ,Ca 2+ ,Li + ,Mg 2+ ,Na + , or Zn 2+ ; The M is calculated based on the bond valence theory. n+ The migration barrier of the compound is less than 2eV. n+ The compounds were used as model datasets; The Matminer package is used to extract the features of the model data, where the model data includes M n+ The mobile ions, framework ions and all ions in the compound; the mobile ions are M n+ , the framework ion is M n+ In addition to M n+ Other ions; Based on the extracted features, the ion migration barrier model is trained.

7. The screening optimization method according to claim 1, characterized in that: The preselected doping material is Mg; The derivative structure with the smallest lithium ion migration barrier is Li 1-2x Mg x BiOS; x = 0.1875; the stoichiometric ratio of Li, Mg, Bi, O, and S in the chemical formula is 10:3:16:16:16; unit cell parameters α=β=γ=90°, and the space group is Pca21.

8. The screening optimization method according to claim 1, characterized in that: The screening optimization method is used for optimizing the screening of fast ion conductor materials.

9. A fast lithium ion conductor material, characterized in that The material is Li 1-2x Mg x BiOS; x<0.375; space group is Pca21.

10. The fast lithium ion conductor material according to claim 9, characterized in that The fast lithium ion conductor material is used as a solid electrolyte material in lithium ion batteries or lithium metal batteries.

Images