Artificial intelligence-based device, method, and program for predicting physical properties of mixtures

Abstract

An apparatus for predicting physical properties of a mixture composed of multiple substances is disclosed. The apparatus includes a memory storing a first AI model trained to output first feature data for substance information and a second AI model trained to output physical property prediction information for the first feature data, and a processor for executing the first AI model and the second AI model, wherein the processor inputs substance information for each of the multiple substances into the first AI model to obtain first feature data for each of the multiple substances, and inputs the first feature data into the second AI model to obtain physical property prediction information for the mixture.

Description

[Technical Field]

[0001] The present disclosure relates to an apparatus, method, and program for analyzing a mixture, and more particularly to an apparatus, method, and program for predicting the physical properties of a mixture based on artificial intelligence. [Background technology]

[0002] In recent years, as consumer needs and demand trends have diversified and product development methods have become more diverse, the development of new mixtures that can be used in product manufacturing has been actively pursued. In order to more efficiently develop and mass-produce new mixtures that can be used in product manufacturing, it is essential to predict and analyze the physical properties of the mixtures.

[0003] Previously, techniques for predicting or analyzing the physical properties of a single material or alloy were developed and utilized. However, as the number of materials constituting the mixture increased and the compounding ratios of each material became more complex, the conventional techniques faced limitations in that they made it difficult to efficiently predict the physical properties of the mixture. As a result, there is a growing need for a more efficient method for predicting the physical properties of mixtures. Summary of the Invention [Problem to be solved by the invention]

[0004] The embodiments disclosed in the present disclosure aim to provide an apparatus, method, and program for predicting the physical properties of a mixture based on artificial intelligence.

[0005] The problems to be solved by the present disclosure are not limited to those described above, and other problems not described will become apparent to those skilled in the art from the following description. [Means for solving the problem]

[0006] In order to solve the above technical problems, an apparatus for predicting the physical properties of a mixture composed of multiple substances according to one embodiment of the present disclosure includes: a memory storing a first AI (artificial intelligence) model trained to output first feature data for substance information and a second AI model trained to output physical property prediction information for the first feature data; and a processor that executes the first AI model and the second AI model, wherein the processor inputs substance information for each of the multiple substances into the first AI model to obtain first feature data for each of the multiple substances, and inputs the first feature data into the second AI model to obtain physical property prediction information for the mixture.

[0007] The first AI model may include an attention-based model that may be trained to extract feature data of a specific substance based on at least one of polarity information and hydrophobicity information of the specific substance.

[0008] The first AI model may also include a molecular contrastive learning-based model that may be trained to align molecules with similar structures in a latent space.

[0009] Furthermore, the first AI model may be trained to update second feature data extracted from two-dimensional graph information for a specific substance based on third feature data extracted from three-dimensional structural characteristics for the specific substance, and to output the first feature data based on the updated data.

[0010] Additionally, the three-dimensional structural characteristics for the particular substance may include multiple conformers for the particular substance.

[0011] Additionally, the second AI model may include a multi-layer perceptron-based model.

[0012] Additionally, the second AI model may include a transformer encoder model.

[0013] The processor can also input the first feature data and component ratio information of multiple substances in the mixture into a second AI model to obtain property prediction information for the mixture.

[0014] In addition, the first AI model and the second AI model may be trained in an end-to-end training manner.

[0015] The substance information may also include chemical information, which may include molecular information (e.g., SMILES (simplified molecular-input line-entry system), INCHI (international chemical identifier), or SELFIES (Self-Referencing Embedded String)) for each of the plurality of substances.

[0016] The chemical information may include at least one of atomic properties or bonding properties associated with the mixture, and as another example, at least one of atomic properties or bonding properties associated with the mixture may be obtained based on the molecular information.

[0017] Furthermore, the first feature data can include molecular feature data for each of the plurality of substances.

[0018] The physical property prediction information may also include Gibbs free energy prediction information.

[0019] In another embodiment of the present disclosure, a method for predicting physical properties of a mixture composed of multiple substances, executed by an apparatus, may include the steps of inputting substance information for each of the multiple substances into a first AI model that has learned first feature data for the substance information, thereby obtaining first feature data for each of the multiple substances, and inputting the first feature data into a second AI model that has learned physical property prediction information for the first feature data, thereby obtaining physical property prediction information for the mixture.

[0020] Additionally, a computer program stored on a computer-readable recording medium may be provided that can be executed to implement the method of the present disclosure.

[0021] Additionally, a computer-readable recording medium may be provided that records a computer program executed to realize the method of the present disclosure. [Effects of the Invention]

[0022] According to the above-described means for solving the problems of the present disclosure, an apparatus, method, and program for predicting the physical properties of a mixture based on artificial intelligence can be provided.

[0023] Furthermore, according to the above-described means for solving the problems of the present disclosure, the properties of a mixture can be predicted more efficiently by using an artificial intelligence model that can take into account the interactions between the components of the mixture.

[0024] The effects of the present disclosure are not limited to the effects described above, and other effects not described will become apparent to those skilled in the art from the following description. [Brief explanation of the drawings]

[0025] [Figure 1] FIG. 1 is a schematic diagram of a system for implementing an artificial intelligence-based method for predicting physical properties of a mixture, according to one embodiment of the present disclosure. [Figure 2]FIG. 2 is a block diagram illustrating the configuration of an apparatus for executing an artificial intelligence-based method for predicting physical properties of a mixture according to one embodiment of the present disclosure. [Figure 3] FIG. 3 is a block diagram illustrating an artificial intelligence-based method for predicting physical properties of a mixture according to one embodiment of the present disclosure. [Figure 4] FIG. 4 is a block diagram illustrating the configuration and operation of an AI model for predicting the properties of a mixture according to one embodiment of the present disclosure. [Figure 5] FIG. 5 is a diagram illustrating a learning / inference method of an AI model for predicting the properties of a mixture according to one embodiment of the present disclosure. [Figure 6a] FIG. 6a is a diagram illustrating a learning method of an AI model that extracts feature data of substances that constitute a mixture, according to one embodiment of the present disclosure. [Figure 6b] FIG. 6b is a diagram illustrating a learning method of an AI model that extracts feature data of substances that constitute a mixture, according to one embodiment of the present disclosure. [Figure 7a] FIG. 7a is a diagram illustrating the performance of each type of AI model according to one embodiment of the present disclosure. [Figure 7b] FIG. 7b is a diagram illustrating the performance of each type of AI model according to one embodiment of the present disclosure. [Figure 7c] FIG. 7c is a diagram illustrating the performance of each type of AI model according to one embodiment of the present disclosure. [Figure 8a] FIG. 8a is a diagram illustrating the structure of an AI model for predicting the properties of a mixture according to one embodiment of the present disclosure. [Figure 8b] FIG. 8b is a diagram illustrating the structure of an AI model for predicting the properties of a mixture according to one embodiment of the present disclosure. [Figure 8c] FIG. 8c is a diagram illustrating the structure of an AI model for predicting the properties of a mixture according to one embodiment of the present disclosure. [Figure 9a]FIG. 9a is a diagram illustrating a learning / inference method of an AI model for predicting properties of a mixture according to one embodiment of the present disclosure. [Figure 9b] FIG. 9b is a diagram illustrating a learning / inference method of an AI model for predicting properties of a mixture according to one embodiment of the present disclosure. [Figure 9c] FIG. 9c is a diagram illustrating a learning / inference method of an AI model for predicting properties of a mixture according to one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0026] Throughout this disclosure, the same reference numerals refer to the same elements. This disclosure does not describe all elements of the embodiments, and matters common in the technical field to which the disclosure belongs and content that overlaps between embodiments will be omitted. Terms such as "unit," "module," "component," and "block" used in this specification may be realized by software or hardware, and depending on the embodiment, multiple "units," "modules," "components," and "blocks" may be realized as a single component, and one "unit," "module," "component," or "block" may include multiple components.

[0027] Throughout this specification, when a part is described as being "connected" to another part, this includes not only cases where the parts are directly connected, but also cases where the parts are indirectly connected, and such indirect connection also includes cases where the parts are connected via a wireless communication network.

[0028] Furthermore, when a part is described as "comprising" a certain component, unless otherwise specified, it does not exclude other components, but means that other components may also be included.

[0029] Throughout this specification, when an element is described as being "on" another element, this includes not only when the element is in contact with the other element, but also when there is another element between them.

[0030] Terms such as "first" and "second" are used to distinguish one component from another, and the components are not limited by these terms.

[0031] Any reference to the singular shall include the plural unless the context clearly indicates otherwise.

[0032] In each step, the identification numbers are used for convenience of explanation, and the identification numbers do not indicate the order of the steps, and the steps may be performed in an order different from the order described unless the context clearly dictates a specific order.

[0033] The working principle and embodiments of the present disclosure will be described below with reference to the accompanying drawings.

[0034] In this specification, the "device for predicting the physical properties of a mixture composed of two or more substances according to the present disclosure" includes all of various devices that can perform calculations and provide results to a user. For example, the device for predicting the physical properties of a mixture composed of two or more substances according to the present disclosure may include all of a computer, a server device, and a mobile terminal, or may be in the form of any one of them.

[0035] Here, the computer may include, for example, a notebook computer, a desktop computer, a laptop computer, a tablet PC, a slate PC, etc., equipped with a web browser.

[0036] The server device is a server that communicates with external devices and processes information, and may include an application server, a computing server, a database server, a file server, a game server, a mail server, a proxy server, a web server, and the like.

[0037] The mobile terminal is, for example, a wireless communication device that ensures portability and mobility, and can include all types of handheld-based wireless communication devices such as PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), WiBro (Wireless Broadband Internet) terminals, and smartphones, and can further include wearable devices such as watches, rings, bracelets, anklets, necklaces, glasses, contact lenses, or head-mounted devices (HMDs).

[0038] Furthermore, for purposes of describing this disclosure, the mixture may include, but is not limited to, a battery electrolyte, and may refer to various types of mixtures made up of multiple materials.

[0039] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0040] FIG. 1 is a schematic diagram of a system 1000 for implementing a method for predicting physical properties of a mixture composed of multiple substances, according to one embodiment of the present disclosure.

[0041] As shown in FIG. 1, a system 1000 for implementing a method for predicting the physical properties of a mixture composed of multiple substances can include an apparatus 100, a database 200, and an AI model 300.

[0042] The device 100, the database 200, and the AI ​​model 300 included in the system 1000 can communicate via a network W. Here, the network W can include a wired network and a wireless network. For example, the network can include various networks such as a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), and the like.

[0043] The network W may also include the well-known World Wide Web (WWW). However, the network W according to the embodiments of the present disclosure is not limited to the networks listed above, and may also include, at least in part, a well-known wireless data network, a well-known telephone network, or a well-known wired / wireless television network.

[0044] The device 100 can predict the physical properties of a mixture composed of multiple substances based on the AI ​​model 300. That is, the device 100 can predict the physical properties of a mixture based on an AI model that can take into account the interactions between the substances that compose the mixture.

[0045] Specifically, the device can input substance information for each of the multiple substances that make up the mixture into a first AI model and obtain first feature data for each of the substances.

[0046] Here, the first feature data may include molecular feature data (e.g., molecular feature vector) for each of the substances. The molecular feature data may be based on information acquired from an external device, but is not limited thereto, and may also be acquired by calculation (e.g., Python program, RdKit, Mordred, etc.).

[0047] The substance information also includes chemical information, which may include various types of molecular information (e.g., SMILES (simplified molecular-input line-entry system), INCHI (international chemical identifier), or SELFIES (Self-Referencing Embedded String)). For example, the device may input molecular information for each of a plurality of substances constituting a mixture into a first AI model to obtain first feature data for each of the substances.

[0048] As yet another example, the chemical information can include at least one of atomic properties or bond properties associated with the mixture.

[0049] In addition, the device may input the first feature data into a second AI model to obtain property prediction information for the mixture. As another example, the device may input the first feature data and component ratio information for multiple substances in the mixture into a second AI model to obtain property prediction information for the mixture. Here, the property prediction information may include, but is not limited to, Gibbs free energy prediction information.

[0050] The device can predict the physical properties of the mixture based on the acquired physical property prediction information (for example, Gibbs free energy prediction information).

[0051] As an example, the difference (e.g., root mean square error (RMSE)) between the predicted information of the Gibbs free energy obtained by the device and the actual Gibbs free energy may be less than 0.02. However, this is not limited thereto, and the RMSE value may vary depending on the experimental environment, the number of times the model is trained, etc.

[0052] Database 200 can record various types of learning data for training AI model 300. Database 200 can also record chemical information for two or more substances that make up a mixture. Database 200 can also record information on the component ratios of multiple substances in the mixture.

[0053] 1 illustrates a case where database 200 is implemented externally to device 100. In this case, database 200 may be connected to device 100 via a wired or wireless connection. However, this is merely an example, and database 200 may also be implemented as a component of device 100.

[0054] The AI ​​model 300 may include a model trained to predict the physical properties of a mixture composed of multiple substances. Specifically, the AI ​​model 300 may include a first AI model trained to extract feature data for each of the substances constituting the mixture and a second AI model trained to extract physical property prediction information for the mixture.

[0055] As an example, the first AI model and the second AI model may be trained in an end-to-end manner. The end-to-end training method refers to a training method that processes input to output in one go without a pipeline network. Here, a pipeline network refers to a partial network that makes up the entire network. In other words, when trained in an end-to-end manner, the first AI model and the second AI model may be realized as a single AI model.

[0056] Accordingly, the device can obtain prediction information for the Gibbs free energy of a mixture by inputting feature data for each substance that makes up the mixture into an AI model trained in an end-to-end manner.

[0057] However, this is just one example, and the first AI model and the second AI model may be realized as separate models (or separate pipeline networks), and each model may be trained separately.

[0058] FIG. 1 illustrates a case where the AI ​​model 300 is implemented externally to the device 100 (e.g., cloud-based), but is not limited thereto; the AI ​​model 300 may be implemented as a component of the device 100.

[0059] FIG. 2 is a block diagram illustrating the configuration of an apparatus for executing an artificial intelligence-based method for predicting physical properties of a mixture according to one embodiment of the present disclosure.

[0060] 2, the device 100 may include a memory 110, a communication module 120, a display 130, an input module 140, and a processor 150. However, the device 100 is not limited thereto, and the software and hardware configuration may be changed, added, or omitted within a scope obvious to a person skilled in the art depending on the required operation.

[0061] Memory 110 can store data supporting various functions of device 100 and programs for the operation of processor 150. It can also store input / output data (e.g., music files, still images, videos, etc.), as well as a plurality of application programs executed in the device and data and instructions for the operation of device 100. At least some of these application programs can be downloaded from an external server via wireless communication.

[0062] Such memory 110 may include at least one type of recording medium among flash memory type, hard disk type, solid state disk type (SSD type), silicon disk drive type (SDD type), multimedia card micro type, card type memory (e.g., SD or XD memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, and optical disk.

[0063] The memory 110 may also include a database that is separate from the device but connected to it via wired or wireless connections, i.e., the database 200 shown in Figure 1 may be implemented as a component of the memory 110.

[0064] The communication module 120 may include one or more components that enable communication with external devices. For example, the communication module 120 may include at least one of a broadcast communication module, a wired communication module, a wireless communication module, a local area communication module, and a location information module.

[0065] The wired communication module may include various wired communication modules such as a local area network (LAN) module, a wide area network (WAN) module, a value added network (VAN) module, etc., as well as various cable communication modules such as a Universal Serial Bus (USB) (registered trademark), a High Definition Multimedia Interface (HDMI) (registered trademark), a Digital Visual Interface (DVI), recommended standard 232 (RS-232), power line communication, or plain old telephone service (POTS).

[0066] The wireless communication module may include a Wi-Fi module, a WiBro (Wireless Broadband) module, as well as wireless communication modules that support various wireless communication methods such as GSM (global System for Mobile Communication), CDMA (Code Division Multiple Access), WCDMA (Wideband Code Division Multiple Access) (registered trademark), UMTS (universal mobile telecommunications system), TDMA (Time Division Multiple Access), LTE (Long Term Evolution), 4th generation (4G), 5th generation (5G), and 6th generation (6G).

[0067] The display 130 displays (outputs) information processed in the device 100, such as feature data and property prediction information output via each AI model (e.g., Gibbs free energy prediction information, mixture property prediction result information, etc.).

[0068] For example, display 130 can display execution screen information of an application program (e.g., an application) executed on device 100, or UI (User Interface) and GUI (Graphical User Interface) information based on the execution screen information. Types of UIs output to display 130 will be described later.

[0069] The input module 140 allows the user to input information, and when information is input via the user interface unit, the processor 150 can control the operation of the device 100 in accordance with the input information.

[0070] The input module 140 may include hardware-based physical keys (e.g., buttons, dome switches, jog wheels, jog switches, etc., located on at least one of the front, back, and side of the device) and software-based touch keys. For example, the touch keys may be virtual keys, soft keys, or visual keys displayed on the touchscreen display 130 through software processing, or may be touch keys located outside the touchscreen. The virtual keys or visual keys may be displayed on the touchscreen in various forms, and may be, for example, graphics, text, icons, videos, or a combination thereof.

[0071] Processor 150 may be configured with a memory for storing data related to an algorithm or a program for executing the algorithm for controlling the operation of components within device 100, and at least one processor (not shown) for performing the aforementioned operations using the data stored in the memory. In this case, the memory and processor may be implemented as separate chips, or the memory and processor may be implemented as a single chip.

[0072] In addition, the processor 150 can control one or more of the above components in combination to realize various embodiments of the disclosure described in Figures 3 to 8 below on the device 100.

[0073] FIG. 3 is a block diagram illustrating an artificial intelligence-based method for predicting physical properties of a mixture according to one embodiment of the present disclosure.

[0074] The device can input substance information for each of a plurality of substances into a first AI model to obtain first feature data for each of the plurality of substances (S310).

[0075] Here, the substance information may include chemical information, and the chemical information may include molecular information for each of the multiple substances (e.g., SMILES (simplified molecular-input line-entry system), INCHI (international chemical identifier), or SELFIES (Self-Referencing Embedded String), etc.).

[0076] The first AI model may include a model trained to output first feature data for each of the substances (e.g., molecular feature data for each of the substances). As shown in Fig. 4, input data for the first AI model 400 includes substance information for each of the multiple substances that make up the mixture, and output data from the first AI model 400 may include the first feature data.

[0077] For example, as shown in FIG. 5, the first AI model 500 may include at least one of an attention-based model, a molecular contrastive learning-based model, or a representative learning-based model.

[0078] Here, the attention-based model can be trained to extract feature data of chemical information for a specific substance based on the physicochemical properties of the specific substance (e.g., at least one of polarizability information and hydrophobicity information (e.g., sLogP)). That is, the attention-based model can be trained to extract first feature data in the final layer after being trained based on a specific physical property.

[0079] 6A shows an example of training result data of an attention-based model. The attention-based model can be trained to extract feature data of chemical information for a specific substance based on information highly related to Gibbs free energy (e.g., Wildman-Crippen LogP characteristic (sLogP), Wildman-Crippen MR characteristic (sMR), bond polarizability characteristic (bpol), or atomic polarizability characteristic (apol)).

[0080] A molecular contrast learning-based model can be trained to align structurally similar molecules in a latent space. That is, the molecular contrast learning-based model calculates the extent to which structurally similar molecules exist in the vicinity, and based on the calculation results, can be trained to align structurally similar molecules. Here, the latent space refers to the space of feature data (or molecular feature vectors) of chemical information for a specific substance.

[0081] For example, Figure 6b shows an example of learning result data of a molecular contrast learning-based model. In the graph shown in Figure 6b, the X axis represents the similarity between molecules, and the Y axis represents the distance. According to the graph shown in Figure 6b, as the distance between molecules increases, the similarity between molecules tends to decrease.

[0082] As another example, the first AI model may be trained to update second feature data extracted from two-dimensional graph information for a specific substance based on third feature data extracted from three-dimensional structural characteristics for the specific substance (e.g., multiple conformers for the specific substance). Furthermore, the first AI model may be trained to output first feature data based on the updated data. In this case, the first AI model may be trained to take into account multiple conformers for the specific substance.

[0083] Specifically, when chemical information for a specific substance represented as two-dimensional graph information is input, the pre-trained first AI model can process the second feature data extracted from the chemical information to be updated to reflect third feature data extracted from three-dimensional structural characteristics of the specific substance. Accordingly, the second feature data can be updated to include the three-dimensional structural characteristics. Furthermore, the pre-trained first AI model can output first feature data based on the updated data.

[0084] In the present disclosure, as another example, chemical information, which is an example of substance information, can include at least one of atomic properties or bonding properties associated with a mixture. For example, the apparatus can acquire first feature data by inputting molecular information (e.g., SMILES, INCHI (international chemical identifier), or SELFIES (Self-Referencing Embedded String)) for each of a plurality of substances into a first AI model. That is, the first AI model can be trained to output first feature data that takes into account both atomic properties and bonding properties based on the molecular information associated with the mixture.

[0085] The first AI model may be input with only molecular information, but is not limited thereto, and may selectively input at least one of atomic properties or bond properties.

[0086] The device can input the acquired first feature data into a second AI model to acquire property prediction information for the mixture (S320).

[0087] Specifically, the device can input the first feature data and component ratio information of the plurality of substances in the mixture into the second AI model to obtain property prediction information for the mixture, where the component ratio information (or fraction information) of the plurality of substances in the mixture can be information encoded in a predefined manner.

[0088] 4, the input data of the second AI model 410 may include at least one of the first feature data or the component ratio information of the plurality of substances, and the output data of the second AI model 410 may include the component ratio information of the plurality of substances in the mixture.

[0089] In addition, the second AI model can be trained to output property prediction information for the mixture (e.g., Gibbs free energy prediction information) based on component ratio information of multiple substances in the mixture and the first feature data.

[0090] For example, the second AI model may include at least one of a multi-layer perceptron (MLP)-based model, a ResNet, or a transformer encoder model. The apparatus may obtain prediction information for the physical properties of the mixture based on the physical property prediction information (e.g., prediction information for Gibbs free energy).

[0091] 7a, 7b, and 7c are diagrams illustrating the performance of each type of AI model according to one embodiment of the present disclosure.

[0092] FIG. 7a is a diagram illustrating an example of feature data extracted by a first AI model when the first AI model includes an attention-based model.

[0093] FIG. 7b is a diagram showing an example of feature data extracted by a first AI model when the first AI model is trained to update second feature data extracted from two-dimensional graph information for a specific substance based on third feature data extracted from three-dimensional structural characteristics for the specific substance.

[0094] FIG. 7c is a diagram showing an example of feature data extracted by the first AI model when the first AI model includes a molecular contrast learning-based model.

[0095] The upper images in Figures 7a, 7b, and 7c show the results of feature data extraction by the first AI model for mixtures that were not subjects of learning in the learning process. The lower images in Figures 7a, 7b, and 7c show the results of feature data extraction by the first AI model for mixtures that were subjects of learning in the learning process.

[0096] Comparing the top and bottom images in Figures 7a, 7b, and 7c, it can be seen that when feature data is extracted from the mixture that was the learning target during the learning process using the first AI model, the RMSE value is reduced compared to when it is not.

[0097] Figures 8a, 8b, and 8c are diagrams illustrating the structure of an AI model for predicting properties of a mixture according to an embodiment of the present disclosure, and Figures 9a, 9b, and 9c are diagrams illustrating the learning / inference scheme of an AI model for predicting properties of a mixture according to an embodiment of the present disclosure.

[0098] 8b and 8c show a case where the first AI model is realized as an attentive FP model (i.e., an example of an attention-based model), but is not limited to this. The first AI model may also be realized as an AI model capable of extracting feature data of a substance, such as a molecular contrast learning-based model.

[0099] 8b and 8c show the operation of inputting molecular information (e.g., SMILES (simplified molecular-input line-entry system), INCHI (international chemical identifier), or SELFIES (Self-Referencing Embedded String)) as substance information of the substances that make up the mixture into the first AI model, but this is not limited to this. The input data to the first AI model may consist of substance information only.

[0100] As another example, the first AI model may selectively input at least one of atomic properties or bonding properties for the mixture, and as another example, the at least one of atomic properties or bonding properties for the mixture may be obtained based on molecular information.

[0101] Also, as shown in Figures 8a, 8b, and 8c, the second AI model (e.g., a Transformer Encoder) shows the operation of outputting prediction information for Gibbs free energy, but is not limited to this and can output various types of physical property prediction information.

[0102] 8a shows a process of extracting predicted physical property information for a mixture using a one-hot encoding model and an MLP-based model. In the case of FIG. 8b, when substance information for each of the substances constituting the mixture is input, the one-hot encoding model can be trained to output molecular feature data mapped to the input substance information. In this case, the one-hot encoding model can be trained to output information that has been pre-mapped to the chemical information, rather than extracting feature data from the substance information (e.g., chemical information) of the substances constituting the mixture.

[0103] Figure 8b shows a process of extracting mixture property prediction information (e.g., Gibbs free energy prediction information) via a first AI model including an attention-based model and a second MLP-based AI model. Figure 8c shows a process of extracting mixture property prediction information (e.g., Gibbs free energy prediction information) via a first AI model including an attention-based model and a Transformer Encoder model.

[0104] As described above, FIG. 8a differs from FIG. 8b and FIG. 8c in that it does not include a process of extracting first feature data for each of the multiple substances that make up the mixture via a first AI model.

[0105] 9a, 9b, and 9c show the resulting data based on FIGS. 8a, 8b, and 8c.

[0106] Comparing Figures 9a and 9b, we can see the difference in performance that arises from utilizing feature data extracted from the molecular information (e.g., SMILES, INCHI, or SELFIES) of the substances that make up the mixture. Utilizing feature data extracted from the substance information of the substances can potentially reduce the difference between the actual physical property information (e.g., Gibbs free energy) and the predicted physical property information (e.g., predicted Gibbs free energy) obtained through the AI ​​model.

[0107] Comparing Figures 9a and 9b, we can see the performance difference between the MLP-based model and the Transformer Encoder model. The Transformer Encoder model may produce more accurate predictions than the MLP-based model.

[0108] Meanwhile, the embodiments of the present disclosure may be realized in the form of a recording medium having computer-executable instructions recorded thereon. These instructions are recorded in the form of program code and, when executed by a processor, generate program modules to perform the operations of the embodiments of the present disclosure. The recording medium may be realized as a computer-readable recording medium.

[0109] A computer-readable storage medium includes any type of storage medium that stores computer-readable instructions, such as a read-only memory (ROM), a random access memory (RAM), a magnetic tape, a magnetic disk, a flash memory, or an optical data storage device.

[0110] As described above, the embodiments of the present disclosure have been described with reference to the accompanying drawings. Those skilled in the art will understand that the present disclosure can be implemented in forms different from the embodiments of the present disclosure without departing from the technical spirit and essential features of the present disclosure. Therefore, the embodiments of the present disclosure are illustrative and should not be interpreted as limiting.

Claims

1. An apparatus for predicting the physical properties of a mixture composed of a plurality of substances, a memory storing a first AI (artificial intelligence) model trained to output first feature data for substance information and a second AI model trained to output physical property prediction information for the first feature data; a processor that executes the first AI model and the second AI model; The processor: inputting substance information for each of the plurality of substances into the first AI model to obtain first feature data for each of the plurality of substances; A physical property prediction device that inputs the first feature data into the second AI model to obtain physical property prediction information for the mixture.

2. the first AI model includes an attention-based model; The physical property prediction device according to claim 1 , wherein the attention-based model is trained to extract feature data of a specific substance based on at least one of polarizability information and hydrophobicity information of the specific substance.

3. the first AI model comprises a molecular contrastive learning based model; The property prediction device according to claim 1 , wherein the molecular contrast learning-based model is learned to align molecules having similar structures in a latent space.

4. The first AI model is learning to update second feature data extracted from two-dimensional graph information for a specific substance based on third feature data extracted from three-dimensional structural characteristics for the specific substance; The physical property prediction device according to claim 1 , which is trained to output the first feature data based on the updated data.

5. The three-dimensional structural characteristics for the specific substance are: The physical property prediction device according to claim 4 , further comprising a plurality of spatial conformers for the specific substance.

6. The second AI model is 10. The property prediction apparatus of claim 1, comprising a multi-layer perceptron-based model.

7. The second AI model is The physical property prediction device of claim 1 , comprising a transformer encoder model.

8. The processor: The property prediction device according to claim 1 , wherein the first feature data and component ratio information of the plurality of substances in the mixture are input to the second AI model to obtain property prediction information for the mixture.

9. The physical property prediction device according to claim 1 , wherein the first AI model and the second AI model are trained using an end-to-end training method.

10. the substance information includes chemical information; the chemical information includes molecular information for each of the plurality of substances; 2. The physical property prediction device according to claim 1, wherein the molecular information includes at least one of SMILES (Simplified Molecular-Input Line-Entry System), INCHI (International Chemical Identifier), and SELFIES (Self-Referencing Embedded String).

11. The property prediction apparatus of claim 10 , wherein the chemical information includes at least one of atomic properties or bonding properties associated with the mixture.

12. The physical property prediction apparatus according to claim 1 , wherein the first feature data includes molecular feature data for each of the plurality of substances.

13. The physical property prediction device according to claim 1 , wherein the physical property prediction information includes prediction information of Gibbs free energy.

14. 1. A method for predicting physical properties of a mixture of multiple substances, performed by an apparatus, comprising: inputting substance information for each of the plurality of substances into a first AI model that has learned first feature data for substance information, and acquiring first feature data for each of the plurality of substances; and inputting the first feature data into a second AI model that has learned property prediction information for the first feature data, thereby obtaining property prediction information for the mixture.

15. A computer program stored on a computer-readable recording medium, which is combined with a hardware device and causes the computer to execute the method for predicting physical properties of a mixture composed of a plurality of substances according to claim 14.

Images