System, method and medium for acquiring feature data of material composition information based on AI

Abstract

Systems, methods, and media for acquiring feature data of material composition information based on AI are disclosed. The system may include a memory configured to store a first AI model configured to output first feature data for composition information of a material and a second AI model configured to output second feature data for structural information of the material; and a processor configured to learn the first AI model and the second AI model. The processor may be configured to learn the first AI model based on second feature data for structural information of the material output by the second AI model and / or to learn the second AI model based on first feature data for composition information of the material output by the first AI model.

Description

Technical Field

[0001] The present disclosure generally relates to an apparatus, method and program for obtaining characteristic data of material composition information. More specifically, some exemplary embodiments of the present disclosure relate to an apparatus, method and program for obtaining characteristic data of material composition information based on artificial intelligence. Background Art

[0002] As lithium-ion battery technology develops, lithium-ion batteries are used in various fields such as electric vehicles and energy storage systems. As a result, positive electrode materials that can improve the energy storage capacity and energy density of lithium-ion batteries have been actively developed and researched.

[0003] The implementation method for analyzing various types of materials to develop such cathode materials requires a lot of time and cost. However, with the improvement of computer processing power and the development of databases for material research such as the Open Quantum Materials Database (OQMD), it has become possible to quickly study the molecular structure, crystal structure, etc. of materials by using artificial intelligence.

[0004] However, in the field of batteries, there has been no positive progress in the research of positive electrode materials using such artificial intelligence (especially deep learning artificial intelligence) because there is insufficient data related to the materials of the battery and it is difficult to obtain relevant data for structural information among the material information of the battery. This increases the need for methods that can analyze battery materials based on insufficient data. Summary of the invention

[0005] Some embodiments of the present disclosure may provide a device, method, and program for acquiring characteristic data of material composition information based on artificial intelligence.

[0006] The problems to be solved by the present disclosure are not limited to the above description, and other unmentioned problems can be clearly understood by those skilled in the art through the following description.

[0007] According to one aspect of the present disclosure, a device for obtaining characteristic data for composition information of a material for achieving the above-mentioned technical problem includes: a memory, the memory storing a first artificial intelligence (AI) model and a second AI model, the first AI model outputting first characteristic data for the composition information of the material, and the second AI model outputting second characteristic data for the structural information of the material; and a processor, the processor being configured to execute or learn the first AI model and the second AI model; wherein the processor can be configured to learn the first AI model based on the second characteristic data for the structural information of the material output by the second AI model, or to learn the second AI model based on the first characteristic data for the composition information of the material output by the first AI model.

[0008] And, the processor learning the first AI model based on second feature data related to structural information of the material output by the second AI model may involve: learning the first AI model to output feature data having an association with the second feature data.

[0009] Furthermore, the first feature data and the second feature data have multiple dimensions, and learning the first AI model by the processor to output feature data associated with the second feature data may involve: adjusting the first feature data to third feature data in such a way that they have a higher similarity when the first feature data and the second feature data have the same dimensions, thereby enabling learning the first AI model to output the adjusted third feature data.

[0010] Furthermore, the third feature data and the fourth feature data have multiple dimensions, and the processor can be configured to: adjust the second feature data to the fourth feature data so that if the second feature data and the third feature data have the same dimension, they have a higher similarity; learn the second AI model to output the adjusted fourth feature data, and adjust the third feature data to the fifth feature data so that if the third feature data and the fourth feature data have the same dimension, they have a higher similarity; and re-learn the first AI model to output the adjusted fifth feature data.

[0011] Furthermore, the first AI model is any one of a multilayer perceptron (MLP), a graph neural network (GNN), or a transformer encoder, and the second AI model can be a GNN.

[0012] Furthermore, the material is one of a plurality of materials including lithium oxide, and the processor may be configured to learn the first AI model and the second AI model based on the specific material, and then relearn the first AI model and the second AI model based on a material different from the specific material.

[0013] Furthermore, the first AI model learned based on the second feature data for the structural information of the material output by the second AI model can receive the composition information of the cathode material of the battery in a charged state and / or the composition information of the cathode material of the battery in a discharged state as input to obtain feature data for predicting the average voltage of the battery.

[0014] Furthermore, the feature data may be a feature vector.

[0015] Furthermore, according to another embodiment of the present disclosure, a method performed by a device for acquiring characteristic data of material composition information may include: a processor included in the device outputting first characteristic data by inputting the composition information of the material into a first AI model stored in a memory; the processor outputting second characteristic data by inputting the structural information of the material into a second AI model stored in the memory; and the processor learning the first AI model based on the second characteristic data related to the structural information of the material output by the second AI model or learning the second AI model based on the first characteristic data related to the composition information of the material output by the first AI model.

[0016] Also, learning, by the processor, the first AI model based on second feature data associated with structural information of the material output by the second AI model may involve learning the first AI model to output feature data associated with the second feature data.

[0017] Furthermore, the first feature data and the second feature data have multiple dimensions, and learning the first AI model by the processor to output feature data associated with the second feature data may involve: adjusting the first feature data to third feature data in such a way that they have a higher similarity when the first feature data and the second feature data have the same dimensions, thereby enabling learning the first AI model to output the adjusted third feature data.

[0018] Furthermore, the third feature data and the fourth feature data have multiple dimensions, and the present disclosure may also include: adjusting the second feature data to the fourth feature data by the processor, so that if the second feature data and the third feature data have the same dimension, they have a higher similarity, and learning the second AI model to output the adjusted fourth feature data; and adjusting the third feature data to the fifth feature data by the processor, so that if the third feature data and the fourth feature data have the same dimension, they have a higher similarity, and re-learning the first AI model to output the adjusted fifth feature data.

[0019] In addition, the present disclosure may also include: inputting the composition information of the cathode material of the battery in a charged state and / or the composition information of the cathode material of the battery in a discharged state into a first AI model learned based on second feature data for structural information of the material output by a second AI model; and obtaining feature data for predicting the average voltage of the battery by a processor.

[0020] Furthermore, a computer-readable recording medium storing a computer program for executing the method for implementing the present disclosure may also be provided.

[0021] Therefore, according to certain embodiments of the present disclosure, an apparatus, method, and program for acquiring characteristic data of material composition information based on artificial intelligence may be provided.

[0022] In addition, according to some embodiments of the present disclosure, characteristic data can be obtained only by composition information of the material, and there is no significant difference in the performance of the device, method and program for obtaining characteristic data by using structural information of the material. That is, since various embodiments of the present disclosure can obtain, extract and output characteristic data of the material with similar performance even by using composition information that is easy to obtain relative data instead of structural information that is difficult to obtain research and test data about the material, it can be further facilitated. In particular, some embodiments of the present disclosure can obtain characteristic data of positive electrode materials of batteries in which relevant data is insufficient.

[0023] The effects of the present disclosure are not limited to the above-mentioned effects, and other unmentioned effects may be clearly understood by those skilled in the art through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a schematic diagram of a system for acquiring characteristic data of material composition information based on artificial intelligence according to an embodiment of the present disclosure.

[0025] Figure 2 is a block diagram illustrating an apparatus for executing a method for acquiring feature data of material composition information based on artificial intelligence according to an embodiment of the present disclosure.

[0026] Figure 3 is a schematic block diagram illustrating a method for learning an AI model according to an embodiment of the present disclosure.

[0027] Figure 4 is a table for illustrating the performance of an AI model according to an embodiment of the present disclosure.

[0028] Figure 5 is a flowchart illustrating a method of learning an AI model according to an embodiment of the present disclosure.

[0029] Figure 6 is a flowchart illustrating a method of learning an AI model according to an embodiment of the present disclosure.

[0030] Figure 7 is a block diagram illustrating a method of predicting or calculating an average voltage of a battery using an AI model according to an embodiment of the present disclosure.

[0031] Figure 8 is a diagram for illustrating the performance of a method for extracting characteristic data of a battery positive electrode material using an AI model according to an embodiment of the present disclosure.

[0032] Reference numerals

[0033] 100: Installation

[0034] 110: Memory

[0035] 120: Communication module

[0036] 130: Display

[0037] 140: Input module

[0038] 150: Processor

[0039] 200: Database

[0040] 300: The first AI model

[0041] 340: Second AI Model

[0042] 700: AI Model

[0043] 740: Voltage prediction model DETAILED DESCRIPTION

[0044] Throughout this disclosure, the same reference numerals represent the same components. This disclosure does not describe all elements of the embodiments, and omits the general content in the technical field of this disclosure or the repetitive content among the embodiments. The terms "unit, module, component, block" used in this specification can be implemented in software or hardware, and according to the embodiment, multiple "units, models, components, blocks" can be implemented as a single component, or a single "unit, module, component, block" can also include multiple components.

[0045] Throughout the specification, when one part is described as being “connected” to another part, it includes not only the case where they are directly connected but also the case where they may be indirectly connected via a wireless communication network.

[0046] In addition, when a portion is described as “comprising” a certain component, it means that it may further include other components, and other components are not excluded unless explicitly stated otherwise.

[0047] Throughout the specification, when a member is described as being 'on' another member, it includes not only a case where the members are in contact but also a case where another member exists between them.

[0048] The terms “first” and “second” are used to distinguish one component from another component, and are not intended to limit the components to the scope described by the terms.

[0049] Unless the context clearly indicates otherwise, a singular expression includes a plural expression.

[0050] For ease of description, identification codes are provided for each step, and the identification codes do not specify the order of the steps, and unless a specific order is explicitly described, each step may be performed in a different order.

[0051] The operation principle and embodiments of the present disclosure will be described below with reference to the accompanying drawings.

[0052] The term device for acquiring characteristic data of material composition information in this specification includes various devices that can perform operations and provide results to users. For example, according to the present disclosure, the device for acquiring characteristic data of material composition information may include a computer, a server device, and a portable terminal, or it may take any of these forms.

[0053] For example, here, the computer may include a notebook computer, a desktop computer, a laptop computer, a tablet personal computer (PC), or a tablet computer equipped with a web browser.

[0054] The server device is a server that processes information by communicating with an external device, and may include an application server, a computing server, a database server, a file server, a game server, a mail server, a proxy server, and a network server.

[0055] For example, a portable terminal is a wireless communication device that ensures portability and mobility, and may include various handheld device-based wireless communication devices such as Personal Communication System (PCS), Global System for Mobile Communications (GSM), Personal Digital Cellular (PDC), Personal Handyphone System (PHS), Personal Digital Assistant (PDA), International Mobile Telecommunications (IMT)-2000, Code Division Multiple Access (CDMA)-2000, W Code Division Multiple Access (W-CDMA), Wireless Broadband Internet (WiBro) terminals, smart phones, and wearable devices such as watches, rings, bracelets, anklets, necklaces, glasses, contact lenses, or head mounted devices (HMD).

[0056] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

[0057] Figure 1 is a schematic diagram of a system for acquiring characteristic data for composition information of a material according to an embodiment of the present disclosure.

[0058] like Figure 1 As shown in , a system 1000 for acquiring characteristic data of composition information of a material may include a terminal or device 100 , a database 200 , and an artificial intelligence (AI) model 300 .

[0059] The device 100, the database 200, and the AI ​​model 300 included in the system 1000 may communicate via a network W. Here, the network W may include a wired network and a wireless network. For example, the network may include various types of networks, such as a local area network (LAN), a metropolitan area network (MAN), and a wide area network (WAN).

[0060] The network W may also include the World Wide Web (WWW). However, the network W according to an embodiment of the present disclosure is not limited to the networks listed above, and may at least partially include any type of network, such as a known wireless data network, a known telephone network, or a known wired or wireless television network.

[0061] The device 100 may obtain feature data about the composition information of the material based on or using the AI ​​model 300. Here, the AI ​​model 300 may output feature data for the composition information of the material, which may include a feature vector as data indicating each feature of the material, and the feature value may be a negative value or a positive value, and may include a value equal to or less than a decimal point in the composition information of the material.

[0062] For example, the composition information of the material includes a chemical formula, a composition ratio of elements or atoms, chemical information, etc., and the chemical information may include various types of molecular information (e.g., Simplified Molecular Input Line Entry System (SMILES), International Chemical Identifier (INCHI), or Self-Referencing Embedded String (SELFIES)). As an example, the material includes a cathode material of a battery.

[0063] In one embodiment, the apparatus 100 may calculate or predict various properties and characteristics of the material based on the acquired characteristic data, and these characteristic data may be used to calculate or predict the average voltage of the positive electrode material of the battery.

[0064] The database 200 may store various learning data for learning the AI ​​model 300. In addition, the database 200 may store chemical information such as the structure or composition of one or more materials and output data generated by the AI ​​model 300. If the learning of the AI ​​model 300 is completed, the database 200 may not be needed in the system 1000.

[0065] Figure 1 An exemplary embodiment is shown in which the database 200 is implemented as a device separate from the device 100. In this case, the database 200 may be connected to the device 100 in a wired and / or wireless manner. However, this is merely an exemplary embodiment, and the database 200 may be implemented as a component of the device 100 or included in the device 100.

[0066] Figure 1An exemplary embodiment is shown in which the AI ​​model 300 is connected to the device 100 via a network W (for example, implemented in a cloud-based manner), but the present disclosure is not limited thereto and may be implemented as a component in the device 100 or included in the device 100.

[0067] Figure 2 is a block diagram illustrating a device configured to perform a method of acquiring feature data of composition information of a material based on artificial intelligence according to an embodiment of the present disclosure.

[0068] like Figure 2 As shown in FIG. 1 , the device 100 may include a memory 110, a communicator or a communication module 120, a display 130, an input interface or an input module 140, and a processor 150. However, the device 100 is not limited thereto, and software and hardware configurations may be modified, added, or omitted within a range apparent from the perspective of a person skilled in the art according to necessary operations.

[0069] The memory 110 may store data supporting or associated with various functions of the device 100 and programs for the operation of the processor 150, may store input data and output data, and may store one or more application programs or applications running on the device 100, data for the operation of the device 100, one or more instructions, and an AI model. At least some of these applications may be downloaded from an external server via wireless communication.

[0070] The memory 110 may include, for example but not limited to, at least one type of storage medium selected from the group consisting of a flash memory type, a hard disk type, a solid state disk type (SSD), a silicon disk drive type (SDD), a multimedia card micro, a card-type memory (e.g., an SD or XD memory), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, and an optical disk.

[0071] In addition, the memory 110 may be separated from the device 100, or may not be included in the device 100, and the memory 110 may be included in a database connected by wire or wirelessly. That is, Figure 1 The database 200 shown in FIG. 1 may include the memory 110 or be implemented as a component of the memory 110 .

[0072] The communicator or communication module 120 may include one or more components configured to communicate with an external device, such as a broadcast receiving module, a wired communication module, a wireless communication module, a short-range communication module, and a location information module.

[0073] The wired communication module may include various wired communication modules, such as a local area network (LAN) module, a wide area network (WAN) module, or a value-added network (VAN) module, and various cable communication modules, such as a universal serial bus (USB), a high-definition multimedia interface (HDMI), a digital video interface (DVI), restricted standard 232 (RS-232), power line communication, or plain old telephone service (POTS).

[0074] In addition to the WiFi module and the wireless broadband module, the wireless communication module may include a wireless communication module configured to support various wireless communication methods, such as Global System for Mobile Communications (GSM), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Universal Mobile Telecommunications System (UMTS), Time Division Multiple Access (TDMA), Long Term Evolution (LTE), 4G, 5G and 6G.

[0075] The display 130 outputs (e.g., displays) information processed in the device 100 (e.g., data and physical property prediction information input or output by an AI model). In addition, the display 130 may display execution screen information of an application program (e.g., an application) driven by the device 100 or user interface (UI) and graphical user interface (GUI) information according to the execution screen information.

[0076] The input module 140 is configured to receive information from a user, and when the information is input through the user input unit, the processor 150 may control the operation of the device 100 to correspond to the input information.

[0077] In this way, the input module 140 may include hardware physical keys (e.g., buttons, dome switches, jog wheels, jog switches, touch panels, microphones, etc., located on at least one of the front surface, rear surface, and side surface of the device) and software touch keys. As an example, the touch keys may be virtual keys, soft keys, or visual keys displayed on the touch screen type display 130 through software processing, or may be touch keys provided in a portion other than the touch screen. On the other hand, the virtual keys or visual keys may be displayed on the touch screen in various forms (including, but not limited to, for example, graphics, text, icons, videos, or a combination thereof).

[0078] The processor 150 may be operably connected to the memory 110, which stores an algorithm for controlling the operation of the components in the device 100 (including the learning or execution of the AI ​​model) or data for reproducing the program of the algorithm. The processor 150 may use the data stored in the memory to perform or perform the above operations. In this case, the memory 110 and the processor 150 may be implemented as separate chips, or may be implemented as a single integrated chip.

[0079] In addition, the processor 150 may control one or more of the above-mentioned components of the device 100 so as to implement various embodiments according to the present disclosure described below on the device 100 .

[0080] Figure 3 is a schematic block diagram illustrating a method for learning an AI model according to an embodiment of the present disclosure.

[0081] When the composition information 310 of the material is input to the first AI model 300 , the first AI model 300 outputs feature data 320 for the input composition information 310 of the material.

[0082] Before performing learning according to an embodiment of the present disclosure, the first AI model 300 may be a predetermined or preconfigured algorithm to output a feature vector for composition information of a material, or may be a model learned based on various information of the material (e.g., composition enthalpy, refractive index, band gap, phonon frequency, bulk modulus, Debye temperature, thermal conductivity, thermal expansion coefficient, decomposition enthalpy, etc.). The algorithm of the first AI model 300 may be, for example, but not limited to, a multilayer perceptron (MLP), a graph neural network (GNN), a transformer encoder, etc.

[0083] In an embodiment, the composition information 310 input to the first AI model 300 has the same chemical formula as BaLi(B3O5)3, and the output feature data 320 is a feature vector and has N values ​​C1, C2, C3, ... C in N dimensions. N The first AI model 300 can extract the composition ratio information of each element from the input chemical formula, vectorize the information through Mat2Vec embedding, element embedding, fractional embedding, etc., and then output the extracted information to a feature vector through multi-head attention and residual network, and the output feature vector can be in matrix format, or can be multiple dimensions, and only some feature vectors can be selected and output from all feature vectors. When only some feature vectors are selected, the selected feature vectors can be values ​​representing the composition information of the material.

[0084] When the structural information 330 of the material is input, the second AI model 340 outputs feature data 350 for the input structural information 330 of the material. The information input to the first AI model 300 and the second AI model 340 is for the same material.

[0085] Before performing learning according to an embodiment of the present disclosure, the second AI model 340 may be a predetermined or preconfigured algorithm for outputting a feature vector for structural information of a material, or may be a model learned based on various structural information of the material. The algorithm of the second AI model 340 may be, for example, but not limited to, a graph neural network (GNN), a crystal graph convolutional neural network (CGCNN), a digital message passing neural network (DIMENET), or a material graph network (MEGNET).

[0086] In an embodiment, the structural information 330 input to the second AI model 340 is one or more of the information indicating the structure of BaLi(B3O5)3 (three-dimensional coordinates of each element, interatomic distances, molecular bond angles, structural images, etc.), and the output feature data 350 has N values ​​S1, S2, S3, ... S N Here, the feature vector 320 output by the first AI model 300 and the feature vector 350 output by the second AI model 340 preferably have the same dimension (ie, the same N) so as to correspond to each other 1:1, although this is not required.

[0087] After the first AI model 300 and the second AI model 340 output the first feature vector 320 and the second feature vector 350, respectively, the first AI model 300 is learned or updated based on the second feature vector 350 value as the output value of the second AI model 340, and the second AI model 340 is learned or updated based on the first feature vector 320 as the output value of the first AI model 300. That is, the first AI model 300 and the second AI model 340 learn from each other or compare and learn based on each other's output values, and such learning can be repeatedly performed. In some other embodiments, the first AI model 300 can be learned only based on the output value of the second AI model 340, and the second AI model 340 can be learned not based on the output value of the first AI model 300. In this embodiment, the second AI model 340 can be referred to as teaching the first AI model 300.

[0088] Learning the first AI model 300 based on the output value of the second AI model 340 may be performed by a processor of a device or computer, and the first AI model 300 may learn to output feature data having an association with feature data that is the output value of the second AI model 340. Here, the meaning of “having an association” (including similarity) includes a case where two data are associated in some way, such as being numerically similar or having a specific functional relationship, compared to other cases where there is no association, or two data are determined to be probabilistically related.

[0089] In an embodiment, the first vector C1 of the N first feature vectors 320 output by the first AI model 300 is compared and contrasted with the first vector S1 of the N second feature vectors 350 output by the second AI model 340, and C1 adjusts the C1 value so as to have similarity with S1 and has similarity with the remaining second feature vectors S2 to S3. N There is no similarity. Similarly, in S2 to S N When the dimensions are the same, the remaining second eigenvectors 350 (S2 to S N ) adjust the remaining first eigenvector 320 (C2 to C N Then, the first AI model 300 is learned or updated so that when the structural information of the same material is input, the adjusted feature vectors C1 to C N .

[0090] As a method of adjusting corresponding values ​​to be closer and more similar and non-corresponding values ​​to be farther and less similar, as in the following equation, the size of the corresponding value (positive value) can be larger and the size of the non-corresponding value (negative value) can be smaller. That is, the value on the left is a positive value and the value on the right is a negative value, so that the positive value is larger than the negative value. Cosine similarity can be used as the following score, and methods such as Euclidean distance, Manhattan distance, and Minkowski distance can be used as methods of measuring the similarity between two vectors.

[0091] Score(f(x),f(x + ))>Score(f(x),f(x - ))

[0092] Structural similarity can mean that two materials have similar crystal structures, which means that the arrangement of atoms, the symmetry of the crystal, the parameters of the unit cell, etc. are similar. Similar eigenvector values ​​for structural information lead to structural similarity. On the other hand, compositional similarity can mean that two materials have similar chemical compositions, which means that the types or proportions of atoms constituting the compositions are similar. Similar eigenvector values ​​for compositional information lead to compositional similarity. In some embodiments of the present disclosure, the eigenvectors of structural information and the eigenvectors of compositional information may be similar to each other.

[0093] As described above, learning the first AI model 300 may be repeatedly performed to output feature data having similarity to feature data that is an output value of the second AI model 340 .

[0094] In an embodiment, after learning the first AI model 300 to output adjusted feature data based on the output value of the second AI model 340, the second AI model 340 may also be learned based on the adjusted output data of the first AI model 300. That is, the second AI model 340 is trained to output feature data that is associated with the feature data that is the output value of the first AI model 300. In an embodiment, each of the N second feature vectors 350 output by the second AI model 340 is adjusted to have similarity with a corresponding one of the N first feature vectors 320 output by the first AI model 300, wherein the first feature vector is a value adjusted based on the second feature vector according to the above-described embodiment. Then, the second AI model 340 is learned or updated so that when the structural information of the same material is input, the adjusted feature vectors S1 to S2 are output. N .

[0095] On the other hand, in the above-mentioned embodiment, it is described that the first AI model 300 is first learned based on the feature data output by the second AI model 340. Alternatively, the second AI model 340 may be first learned based on the feature data output by the first AI model 300. Also, the first AI model 300 is learned based on the feature data output by the second AI model 340, and then, the second AI model 340 is learned based on the feature data output by the first AI model 300, and again, the first AI model 300 may be repeatedly relearned based on the feature data output by the second AI model 340, and after learning one material, another material may also be learned. In an embodiment, the first AI model 300 and the second AI model 340 may use only lithium oxide materials, or may use various materials including lithium oxide as learning target materials.

[0096] Generally, the structural information of a material includes more information than the composition information, so it is preferable to use the structural information rather than the composition information in order to extract the characteristic data of the material. However, since there is a problem that the structural information of a material may be difficult to know or obtain compared with the composition information, many studies and tests inevitably use the composition information to obtain the characteristic data, even though the composition information may show low performance.

[0097] However, the first AI model 300 according to an embodiment of the present disclosure not only receives the composition information of the material and outputs its characteristic data, but also learns based on the characteristic data of the structural information of the material output by the second AI model 340, so the output of the first AI model 300 can have a correlation with the structural information of the material. In addition, when the first AI model 300 and the second AI model 340 repeatedly perform learning by comparing learning with each other, even if only the composition information of the material is input, the first AI model 300 can extract characteristic data that is more closely related to the structural information of the material.

[0098] When the characteristic data of the material is extracted as described above, if the first AI model 300 according to an embodiment of the present disclosure is used (i.e., the second AI model is only used for learning the first AI model, and then only the first AI model is used without using the second AI model when extracting, obtaining and outputting the desired characteristic data of the material), a similar level of effect as the effect of extracting characteristic data using the structural information of the material can be obtained only by the composition information of the material.

[0099] Figure 4 is a table for illustrating the performance of an AI model according to an embodiment of the present disclosure.

[0100] exist Figure 4 In the present invention, the average voltage prediction performance was measured for four artificial intelligence models, for example, the structural encoder (S2P) using the known DimeNet, the reduced ER using the known Matminer and PCA (Principal Component Analysis), the transferred C2P based on the learning of lithium oxide materials in the artificial intelligence model according to the embodiment of the present disclosure (pre-training data: lithium oxide), and the transferred C2P based on the learning of all materials including lithium oxide in the artificial intelligence model according to the present disclosure (pre-training data: all materials).

[0101] First, in the case of S2P, since the structural information of the material is used as input data, it can be confirmed that the mean absolute error (MAE) score and the R2 score are the best (the lower the MAE score and the higher the R2 score, the better the performance can be determined). On the other hand, in the case of C2P and reduced ER, since the composition information of the material is used as input data, it can be confirmed that the MAE score and the R2 score are worse than the scores of S2P using the structural information as input data. By the way, in the case of C2P as a transfer of the artificial intelligence model according to an embodiment of the present disclosure, it can be confirmed that even when the composition information of the material is used as input data, the MAE score and the R2 score are not significantly different compared to S2P using the structural information as input data. In particular, it can be seen that the model learned from various materials has better performance than the model learned only from lithium oxide materials. As described above, it can be seen that even with the composition information of the material, the use of the AI ​​model according to some embodiments of the present disclosure can achieve similar effects as extracting feature data using the structural information of the material.

[0102] Figure 5 It is a flowchart for illustrating a method for learning an AI model according to an embodiment of the present disclosure.

[0103] In operation S510, the device inputs the composition information of the material into the first AI model. In operation S520, the first AI model outputs first feature data about the composition information of the material, and in operation S530, the device inputs the structure information of the material into the second AI model. In operation S540, the second AI model outputs second feature data about the structure information of the material.

[0104] Subsequently, in operation S550, the device adjusts the first feature data to the third feature data so that the first feature data for the composition information of the material output from the first AI model and the second feature data for the structural information of the material output from the second AI model can have similar values. In operation S560, the device learns for the first AI model to output the adjusted third feature data. According to an embodiment, the device can repeat operations S550 and S560 multiple times so that the adjusted third feature information and the adjusted second feature information can have similar values, and the first AI model can output the adjusted first feature information again.

[0105] Meanwhile, in an embodiment of the present disclosure, the device may use the first AI model learned after executing step S560 to obtain feature data on composition information of the target material to be analyzed, but may use the first AI model learned after performing step S560 according to the following Figure 6 The first AI model is re-learned after the additional learning process of the embodiment disclosed in the disclosure.

[0106] Figure 6 It is a flowchart for illustrating a method for learning an AI model according to an embodiment of the present disclosure.

[0107] In operation S610, the device adjusts the second feature data to the fourth feature data so that Figure 5 After learning the first AI model to output the adjusted third feature data from the composition information of the material, the third feature data and the second feature data output by the second AI model from the structural information of the material may have similar values. The similarity value may represent a value whose difference is within a predetermined allowable range. In operation S620, the second AI model is learned to output the fourth feature data adjusted in operation S610.

[0108] Subsequently, in operation S630, the third feature data is adjusted to fifth feature data so that the third feature data output by the first AI model and the fourth feature data output by the second AI model can have similar values. In operation S640, the first AI model is relearned to output the fifth feature data adjusted in operation S630.

[0109] exist Figure 6 In the embodiment disclosed in , since the first AI model for outputting characteristic data of composition information of the material and the second AI model for outputting characteristic data of structural information of the material repeatedly perform mutual learning or comparative learning with each other, the first AI model can be created or configured so that the structural characteristics of the material are further considered. Figure 4 As described above, even if only the composition information of the material is input, the AI ​​model according to the embodiment of the present disclosure can extract feature data that well shows the characteristics of the material at a level similar to that of the case where the structural information is input. This effect can be more clearly confirmed when the AI ​​model according to some embodiments of the present disclosure is used for the positive electrode material or material of the battery.

[0110] Figure 7 is a block diagram illustrating a method of predicting or calculating an average voltage of a battery using an AI model according to an embodiment of the present disclosure.

[0111] First, according to an embodiment of the present disclosure, the composition information of the positive electrode material in the battery charging state and / or the composition data of the positive electrode material in the battery discharging state (e.g., chemical formula, ratio or content of each element, etc.) are input to the AI ​​model 700. In an exemplary embodiment of the present disclosure, the battery may be a lithium-ion battery, and since the AI ​​model 700 is configured to output characteristic data according to the composition information of the material by adjusting the output characteristic data based on the characteristic data of the structural information according to an embodiment of the present disclosure, the performance difference can be reduced compared to a model that outputs characteristic data only based on the structural information of the material. Considering that the structural information of the positive electrode material of the battery is difficult to know or obtain, the technical effect of the model with the composition information based on the material according to certain embodiments of the present disclosure can obtain a similar effect to a model that uses only structural information to make related research easier and faster in a data-deficient environment.

[0112] The AI ​​model 700 receives the positive electrode material composition information of each of the battery charge and discharge states, and outputs the characteristic data 710 and 720 of the composition information of each of the battery charge or discharge states. The characteristic data 710 and 720 may be concatenated into the final characteristic data 730. For example, if the characteristic data of the composition information output by the AI ​​model 700 is 128-dimensional (i.e., 128 pieces of data), and the characteristic data of the structure information is 128-dimensional (128 pieces of data), the combined data may be 256-dimensional (256 pieces of data).

[0113] The combined feature data 730 can be input to the voltage prediction model 740 to predict the average voltage Vav of the positive electrode material. As described above, the combined feature data 730 is a feature vector output by the AI ​​model 700 based on the composition information, and the AI ​​model 700 can output data that can clearly indicate the characteristics of the material, because the output value is learned based on the feature data of the structural information of the material to adjust, and as a result, when the feature data is used as an input value of the voltage prediction model, a more accurate voltage (average voltage) can be predicted.

[0114] In addition, in the case of a battery, there is a problem in that the composition information is constantly changing (particularly the content of Li is changing) during the process of the battery changing from a charged state to a discharged state, so it is difficult to predict the average voltage, and the average voltage is not necessarily calculated according to the actual situation. Figure 7 The voltage prediction model 740 may be pre-learned based on the charging and discharging state data of the lithium-ion battery.

[0115] Figure 8 is a diagram for explaining the performance of a method for extracting characteristic data of a battery positive electrode material using an AI model according to an embodiment of the present disclosure.

[0116] Figure 8 The attention scores of C2P according to an embodiment of the present disclosure, an AI model learned from lithium oxide materials (pre-trained with LiO), and an AI model learned from various materials according to an embodiment of the present disclosure (re-trained with all) are shown. The attention scores are calculated in a lithium transition metal (TM)-oxygen structural system, where TM is Ti, V, Cr, Mn, Fe, Co, Ni, and Cu. Here, the attention score refers to how much each element is considered or referenced in the process of calculating the feature vector by the AI ​​model. If the attention score of the lithium (Li) element is high, it means that more lithium elements are considered or referenced than other elements when calculating the feature vector of the material containing lithium. When calculating the feature vector of a lithium-ion battery (especially the positive electrode material of the battery), it is important to reference the lithium element with a greater weight because the lithium-ion battery continuously changes the content of the lithium element as the charging and discharging states are repeated, which is an element that directly affects the battery voltage.

[0117] Figure 8 Four bars are shown for each model, the second bar is for lithium (Li), the third bar is for transition metals (TM), and the fourth bar is for oxygen (O) attention scores. As can be seen from the figure, the AI ​​model according to the embodiment of the present disclosure has a higher Li attention score than C2P, and in particular, it can be confirmed that the learning model based on various materials has a higher Li attention score. Therefore, when the AI ​​model according to the present disclosure is used to predict the average voltage of a lithium-ion battery, more accurate predictions and calculations can be performed.

[0118] At the same time, some embodiments of the present disclosure can be implemented in the form of a recording medium storing instructions that can be executed by a computer. The instructions can be stored in the form of program code, and when executed by a processor, a program module can be generated to perform the operations of the disclosed embodiments. The recording medium can be implemented as a computer-readable recording medium.

[0119] The computer-readable recording medium includes any type of recording medium in which instructions decipherable by a computer are stored, such as read-only memory (ROM), random access memory (RAM), magnetic tapes, magnetic disks, flash memory, optical data storage devices, etc.

[0120] As described above, the disclosed embodiments are described with reference to the accompanying drawings. It can be understood by those skilled in the art that the present disclosure can be implemented in a form different from the disclosed embodiments without changing the technical spirit or basic features of the present disclosure. The disclosed embodiments are illustrative and restrictive.

Claims

1. A system for obtaining characteristic data of material composition information based on AI, wherein the AI ​​is artificial intelligence, and the system comprises: a memory configured to store a first AI model configured to output first feature data for composition information of a material and a second AI model configured to output second feature data for structure information of a material; as well as a processor configured to execute or learn the first AI model and the second AI model, Wherein, the processor is configured to perform one or more of an operation of learning the first AI model based on the second feature data for structural information of the material output by the second AI model or an operation of learning the second AI model based on the first feature data for composition information of the material output by the first AI model.

2. The system for obtaining characteristic data of material composition information based on AI according to claim 1, wherein: The processor is configured to learn the first AI model based on the second feature data for structural information of a material output by the second AI model to output feature data associated with the second feature data for structural information of a material.

3. The system for obtaining characteristic data of material composition information based on AI according to claim 2, wherein: The first feature data and the second feature data have multiple dimensions, and The processor is configured to learn the first AI model to output feature data associated with the second feature data by adjusting the first feature data to third feature data so as to have a higher similarity between the first feature data and the second feature data, so that the first AI model is learned to output the third feature data adjusted from the first feature data.

4. The system for obtaining characteristic data of material composition information based on AI according to claim 3, wherein: The third feature data and the fourth feature data have multiple dimensions, and The processor is configured to: adjusting the second feature data to the fourth feature data to have a higher similarity between the second feature data and the third feature data, so that the second AI model is learned to output the fourth feature data adjusted from the second feature data, and The third feature data is adjusted to fifth feature data to have a higher similarity between the third feature data and the fourth feature data, so that the first AI model is relearned to output the fifth feature data adjusted from the third feature data.

5. The system for obtaining characteristic data of material composition information based on AI according to claim 1, wherein: The first AI model includes one or more of a multi-layer perceptron (MLP), a graph neural network (GNN), or a transformer encoder, and The second AI model includes GNN.

6. The system for obtaining characteristic data of material composition information based on AI according to claim 1, wherein: The material is one of a plurality of materials including lithium oxide, and The processor is configured to learn the first AI model and the second AI model based on one material among the plurality of materials, and relearn the first AI model and the second AI model based on another material among the plurality of materials.

7. The system for obtaining characteristic data of material composition information based on AI according to claim 1, wherein: The first AI model learned based on the second feature data for structural information of the material output by the second AI model is configured to receive composition information of the cathode material of the battery in a charged state and / or composition information of the cathode material of the battery in a discharged state as input to obtain feature data related to the average voltage of the battery.

8. The system for obtaining characteristic data of material composition information based on AI according to claim 1, wherein: The feature data is a feature vector.

9. A method for obtaining characteristic data of material composition information based on AI, the method being a computer-implemented method, the method comprising the following steps: Outputting first feature data by the processor by inputting composition information of the material into a first AI model stored in the memory; Outputting second feature data by the processor by inputting structural information of the material into a second AI model stored in the memory; as well as The processor performs one or more of an operation of learning the first AI model based on the second feature data for structural information of the material output by the second AI model or an operation of learning the second AI model based on the first feature data for composition information of the material output by the first AI model.

10. The method for obtaining characteristic data of material composition information based on AI according to claim 9, wherein: The operation of learning the first AI model based on the second feature data for structural information of a material output by the second AI model includes the step of learning the first AI model to output feature data associated with the second feature data.

11. The method for obtaining characteristic data of material composition information based on AI according to claim 10, wherein: The first feature data and the second feature data have multiple dimensions, and The step of learning the first AI model to output feature data associated with the second feature data includes the following steps: adjusting the first feature data to third feature data to have a higher similarity between the first feature data and the second feature data, so that the first AI model is learned to output the third feature data adjusted from the first feature data.

12. The method for obtaining characteristic data of material composition information based on AI according to claim 11, wherein: The third feature data and the fourth feature data have multiple dimensions, and, The computer-implemented method further comprises the following steps: adjusting, by the processor, the second feature data to the fourth feature data to have a higher similarity between the second feature data and the third feature data, so that the second AI model is learned to output the fourth feature data adjusted from the second feature data, and The processor adjusts the third feature data to fifth feature data to have a higher similarity between the third feature data and the fourth feature data, so that the first AI model is relearned to output the fifth feature data adjusted from the third feature data.

13. The method for obtaining characteristic data of material composition information based on AI according to claim 9, further comprising the following steps: inputting composition information of a cathode material of a battery in a charged state and / or composition information of a cathode material of the battery in a discharged state into the first AI model learned based on the second feature data for structural information of the material output by the second AI model; and The processor obtains characteristic data related to the average voltage of the battery.

14. A medium for acquiring characteristic data of material composition information based on AI, the medium being a non-transitory computer-readable storage medium having instructions that, when executed by one or more processors, cause the one or more processors to perform the following steps: Outputting first feature data by inputting composition information of the material into a first AI model stored in a memory; outputting second feature data by inputting structural information of the material into a second AI model stored in the memory; as well as Perform one or more of an operation of learning the first AI model based on the second feature data for structural information of a material output by the second AI model or an operation of learning the second AI model based on the first feature data for composition information of a material output by the first AI model.

Images