Design support device, learning device, design support method, learning method, and program
Patent Information
- Application Number
- JP2025539328
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Priority Date
- 2023-08-10
- Filing Date
- 2024-07-31
- Publication Date
- 2025-02-13
Abstract
Description
Design support device, learning device, design support method, learning method, and program
[0001] The present disclosure relates to a design support device, a learning device, a design support method, a learning method, and a program.
[0002] Conventionally, metal-supported zeolite catalysts with desired properties have been designed by trial and error (see, for example, Patent Document 1). In recent years, a design method has been disclosed in which the Si / Al ratio, Cu loading amount, and zeolite framework shape of a Cu-supported zeolite catalyst are designed using machine learning (see, for example, Non-Patent Document 1).
[0003] Patent No. 6600082
[0004] Catalysts informatics: paradigm shift towards data-driven catalyst design K. Takahashi, et. al., Chem. Commun. 59, 2222-2238 (2023).
[0005] However, Patent Document 1 and Non-Patent Document 1 do not describe how to efficiently design a metal-supported zeolite catalyst that has desired properties and uses a novel supported metal.
[0006] An object of the present disclosure is to provide a design support device, a learning device, a design support method, a learning method, and a program that support the more efficient design of metal-supported zeolite catalysts having desired properties.
[0007] The present disclosure has the following configuration.
[0008] [1] A design support device that supports the design of a metal-supported zeolite catalyst by using a machine learning model that has been trained to determine the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and reaction yields of compounds, the design support device comprising: a parameter generation unit that generates comprehensive combinations of the zeolite parameters and parameters of the metal cations; a prediction unit that predicts the reaction yield of the compound corresponding to the combination using the trained machine learning model; and a candidate output unit that outputs data on candidate metal-supported zeolite catalysts based on the predicted reaction yield of the compound corresponding to the combination.
[0009] [2] The design support device according to [1], wherein the trained machine learning model has trained a correspondence relationship between the zeolite parameters, the metal cation parameters, and information on anion species and a reaction yield of the compound, and the parameter generation unit generates a comprehensive combination of the zeolite parameters, the metal cation parameters, and information on the anion species to be input to the trained machine learning model.
[0010] [3] The design support system according to [1] or [2], wherein the reaction yield of the compound is at least one of a propylene yield, an olefin / paraffin ratio, and a useful component yield.
[0011] [4] A learning device for training a machine learning model, comprising: a training dataset acquisition unit that acquires a training dataset that represents a correspondence between zeolite parameters and parameters of metal cations supported by ion exchange, and a reaction yield of a compound; and a learning unit that trains the machine learning model based on the training dataset, using the zeolite parameters and the metal cation parameters as explanatory variables and the reaction yield of the compound as a target variable.
[0012] [5] The learning device according to [4], wherein the training data set represents a correspondence relationship between the zeolite parameters, the metal cation parameters, and information on anion species and the reaction yield of the compound, and the learning unit trains the machine learning model based on the training data set using the zeolite parameters, the metal cation parameters, and information on anion species as explanatory variables and the compound reaction yield as a target variable.
[0013] [6] The learning device according to [4] or [5], wherein the reaction yield of the compound is at least one of a propylene yield, an olefin / paraffin ratio, and a useful component yield.
[0014] [7] A design support method in which a computer supports the design of a metal-supported zeolite catalyst by using a machine learning model that has learned the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and reaction yields of compounds, the design support method comprising: a parameter generation step of generating comprehensive combinations of the zeolite parameters and parameters of the metal cations; a prediction step of predicting the reaction yield of the compound corresponding to the combination using the trained machine learning model; and a candidate output step of outputting data on candidate metal-supported zeolite catalysts based on the predicted reaction yield of the compound corresponding to the combination.
[0015] [8] A learning method in which a computer trains a machine learning model, the learning method comprising: a learning dataset acquisition step of acquiring a learning dataset that represents a correspondence between zeolite parameters and parameters of metal cations supported by ion exchange, and a reaction yield of a compound; and a learning step of training the machine learning model based on the training dataset, using the zeolite parameters and the metal cation parameters as explanatory variables and the reaction yield of the compound as a target variable.
[0016] [9] A program for causing a computer that assists in the design of metal-supported zeolite catalysts, using a machine learning model that has learned the correspondence between zeolite parameters and parameters of metal cations supported by ion exchange, and the reaction yield of compounds, to execute the following steps: a parameter generation procedure that generates comprehensive combinations of the zeolite parameters and the metal cation parameters; a prediction procedure that predicts the reaction yield of the compound corresponding to the combination, using the trained machine learning model; and a candidate output procedure that outputs data on candidate metal-supported zeolite catalysts based on the predicted reaction yield of the compound corresponding to the combination.
[0017]
[10] A program for causing a computer that trains a machine learning model to execute: a training dataset acquisition procedure for acquiring a training dataset that represents the correspondence between zeolite parameters and parameters of metal cations supported by ion exchange, and the reaction yield of a compound; and a learning procedure for training the machine learning model based on the training dataset, using the zeolite parameters and the metal cation parameters as explanatory variables and the reaction yield of the compound as a target variable.
[0018] According to the present disclosure, it is possible to provide a design support device, a learning device, a design support method, a learning method, and a program that support the more efficient design of metal-supported zeolite catalysts having desired properties.
[0019] FIG. 1 is a configuration diagram of an example of an information processing system according to the present embodiment. FIG. 2 is a hardware configuration diagram of an example of a computer according to the present embodiment. FIG. 3 is a functional configuration diagram of an example of an information processing system according to the present embodiment. FIG. 4 is a flowchart of an example of a process for training a machine learning model in the information processing system according to the present embodiment. FIG. 5 is a configuration diagram of an example of a training dataset before parameter conversion. FIG. 6 is a configuration diagram of an example of a parameter table for metal cations. FIG. 7 is a configuration diagram of an example of a training dataset after parameter conversion. FIG. 8 is a flowchart of an example of a process for predicting candidates for metal-supported zeolite catalysts having desired properties in the information processing system according to the present embodiment. FIG. 9 is an explanatory diagram of an example of a search range for metal-supported zeolite catalysts. FIG. 10 is a configuration diagram of an example of a parameter table. FIG. 11 is an explanatory diagram of an example showing the relationship between predicted values and actual measured values of a trained machine learning model for propylene yield according to the present embodiment. FIG. 12 is an explanatory diagram of an example showing the relationship between predicted values and actual measured values of a trained machine learning model for useful component yield according to the present embodiment. FIG. 13 is an explanatory diagram showing data of candidates for metal-supported zeolite catalysts with high prediction results for propylene yield or useful component yield. FIG. 14 is a diagram showing an example of experimental results for propylene yield. FIG. 15 is a diagram showing an example of experimental results for C2-5 olefin / paraffin ratio. FIG. 16 is a diagram showing an example of experimental results for useful component yield. ＋ , Sr ２＋ , or Rb ＋ FIG. 10 is a diagram showing experimental results when using
[0020] Next, an embodiment of the present invention will be described in detail, but the present invention is not limited to the following embodiment.
[0021] [First Embodiment] <System Configuration> Fig. 1 is a configuration diagram of an example of an information processing system according to this embodiment. The information processing system 1 in Fig. 1 includes a design support device 10, a learning device 11, and a user terminal 12. The design support device 10, the learning device 11, and the user terminal 12 are connected to each other so as to be able to communicate data with each other via a communication network 18 such as a local area network (LAN) or the Internet.
[0022] The user terminal 12 is an information processing terminal operated by an operator, such as a PC, tablet terminal, or smartphone. The user terminal 12 displays a screen on a display device for accepting information input from the operator and accepts the information input from the operator. The user terminal 12 also transmits the information accepted from the operator to the design support device 10, which executes a process to support the more efficient design of a metal-supported zeolite catalyst having desired properties. The user terminal 12 receives information on the execution results of the process from the design support device 10, displays it on a display device, and allows the operator to check it. The user terminal 12 also transmits information accepted from the operator to the learning device 11, which executes a process to train a machine learning model, as described below. The user terminal 12 may also receive information on the execution results of the process from the learning device 11, and display it on a display device for the operator to check.
[0023] The design support device 10 is an information processing device such as a PC that supports the more efficient design of a metal-supported zeolite catalyst having desired properties, as described below. The design support device 10 supports the more efficient design of a metal-supported zeolite catalyst having desired properties by executing a process for outputting candidates for metal-supported zeolite catalysts having desired properties.
[0024] The design support device 10 may use a mathematical model such as a machine learning model (trained machine learning model) that has learned the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and the reaction yield of a compound. For example, a regression or classification method may be used as the machine learning method. The design support device 10 may use a mathematical model such as a machine learning model that has learned the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and information on anion species and the reaction yield of a compound. For example, a regression or classification method may be used as the machine learning method.
[0025] Specifically, examples of machine learning techniques that may be used include linear, generalized linear (such as lasso, ridge, elastic net, or logistic), partial least squares, kernel ridge, Gaussian process, k-nearest neighbor method, decision tree, random forest, Adaboost, bagging, gradient boosting, support vector machine, and supervised learning techniques such as neural networks.
[0026] The design support device 10 generates comprehensive combinations of zeolite parameters and parameters of metal cations supported by ion exchange, and predicts the reaction yield of compounds corresponding to the comprehensive combinations using a trained machine learning model. The design support device 10 may generate comprehensive combinations of zeolite parameters, parameters of metal cations supported by ion exchange, and information on anion species, and predict the reaction yield of compounds corresponding to the comprehensive combinations using a trained machine learning model.
[0027] By predicting the reaction yields of compounds corresponding to such comprehensive combinations using a trained machine learning model, the design support device 10 can output candidates for metal-supported zeolite catalysts having the desired properties.
[0028] For example, when predicting the reaction yield of a compound using a trained machine learning model, parameters of the zeolite and parameters of the metal cations supported by ion exchange are input to the trained machine learning model. Also, when predicting the reaction yield of a compound using a trained machine learning model, parameters of the zeolite, parameters of the metal cations supported by ion exchange, and information on anion species may be input to the trained machine learning model.
[0029] The design support device 10 receives information input by an operator to the user terminal 12 and executes processing to support the design of a metal-supported zeolite catalyst. The design support device 10 transmits processing result information to the user terminal 12 and causes the user terminal 12 to display the processing result information (e.g., data on one or more candidate metal-supported zeolite catalysts).
[0030] The learning device 11 causes a machine learning model to learn the correspondence relationships between the parameters of the zeolite and the parameters of the metal cations supported by ion exchange, and the reaction yield of the compound, as described below. The learning device 11 may also cause a machine learning model to learn the correspondence relationships between the parameters of the zeolite, the parameters of the metal cations supported by ion exchange, and the information on the anion species, and the reaction yield of the compound, as described below.
[0031] The information processing system 1 in Fig. 1 can be realized by a design support device 10 and a learning device 11 having a web server function, and a user terminal 12 that executes a web application using a web browser function. The information processing system 1 in Fig. 1 may also be realized by an application installed in the user terminal 12 performing processing in cooperation with programs installed in the design support device 10 and the learning device 11.
[0032] The information processing system 1 in Figure 1 is merely an example, and it goes without saying that there are various system configuration examples depending on the application and purpose. For example, the design support device 10 and the learning device 11 may be realized by multiple computers, may be realized as a cloud computing service, or may be realized by a single computer. Furthermore, the information processing system 1 in Figure 1 may be realized by a stand-alone computer.
[0033] <Hardware Configuration> The design support device 10, learning device 11, and user terminal 12 in FIG. 1 are realized by, for example, a computer 500 having the hardware configuration shown in FIG.
[0034] Fig. 2 is a diagram showing an example of the hardware configuration of a computer according to this embodiment. The computer 500 in Fig. 2 includes an input device 501, a display device 502, an external I / F 503, a RAM 504, a ROM 505, a CPU 506, a communication I / F 507, and an HDD 508, all of which are interconnected by a bus B. The input device 501 and the display device 502 may be connected to each other for use.
[0035] The input device 501 includes a touch panel, operation keys, buttons, a keyboard, a mouse, etc., which are used by the user to input various signals. The display device 502 includes a display such as a liquid crystal or organic EL display for displaying a screen, a speaker for outputting audio data such as voice and sound, etc. The communication I / F 507 is an interface for the computer 500 to perform data communication.
[0036] The HDD 508 is an example of a non-volatile storage device that stores programs and data. The stored programs and data include an OS, which is basic software that controls the entire computer 500, and applications that provide various functions on the OS. Note that the computer 500 may use a drive device that uses flash memory as a storage medium (e.g., a solid-state drive (SSD)) instead of the HDD 508.
[0037] The external I / F 503 is an interface with an external device. The external device may be a recording medium 503a. This allows the computer 500 to read and / or write data from and to the recording medium 503a via the external I / F 503. The recording medium 503a may be a flexible disk, a CD, a DVD, an SD memory card, a USB memory, or the like.
[0038] The ROM 505 is an example of a non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. The ROM 505 stores programs and data such as the BIOS, OS settings, and network settings that are executed when the computer 500 starts up. The RAM 504 is an example of a volatile semiconductor memory (storage device) that temporarily retains programs and data.
[0039] The CPU 506 is a computing device that reads programs and data from storage devices such as the ROM 505 and HDD 508 onto the RAM 504 and executes processing to realize overall control and functions of the computer 500. The computer 500 according to this embodiment can realize various functions of the design support device 10, the learning device 11, and the user terminal 12 by executing programs.
[0040] <Functional Configuration> The functional configuration of the information processing system 1 according to this embodiment will be described. Fig. 3 is a functional configuration diagram of an example of the information processing system according to this embodiment. Note that the configuration diagram in Fig. 3 appropriately omits parts that are not necessary for explaining this embodiment.
[0041] The design support device 10 has a request receiving unit 20, a response sending unit 22, a registration receiving unit 24, a search range receiving unit 26, a parameter generating unit 28, a prediction unit 30, a candidate output unit 32, a display control unit 34, a search range information storage unit 40, a parameter table storage unit 42, and a trained machine learning model storage unit 44.
[0042] The learning device 11 includes a request receiving unit 50 , a learning dataset acquisition unit 52 , a learning unit 54 , a registration request unit 56 , a parameter table storage unit 57 , and a machine learning model storage unit 58 .
[0043] The user terminal 12 includes an information display unit 60 , an operation reception unit 62 , a request transmission unit 64 , and a response reception unit 66 .
[0044] The information display unit 60 displays on the display device 502 a screen for accepting information input from an operator and a screen for displaying information on the execution results of processing by the design support device 10 or the learning device 11. The operation accepting unit 62 accepts operations by the operator, such as input of information. The request sending unit 64 sends a processing request corresponding to the information input from the operator to the design support device 10 or the learning device 11. In addition, the response receiving unit 66 receives a response to the processing request sent by the request sending unit 64 from the design support device 10 or the learning device 11.
[0045] The request receiving unit 20 receives a processing request from the user terminal 12. The response sending unit 22 responds with the execution result of the processing in response to the processing request. The registration accepting unit 24 accepts registration of search range information (described below) from, for example, an operator operating the user terminal 12, and stores the search range information in the search range information storage unit 40. The search range information storage unit 40 stores the search range information (described below). The registration accepting unit 24 also accepts registration of a parameter table (described below) from, for example, an operator operating the user terminal 12, and stores the parameter table in the parameter table storage unit 42. The parameter table storage unit 42 stores the parameter table (described below). The registration accepting unit 24 also accepts registration of a trained machine learning model from the learning device 11, and stores the trained machine learning model in the trained machine learning model storage unit 44. The trained machine learning model storage unit 44 stores the trained machine learning model. The search range accepting unit 26 accepts input of the range of comprehensive combinations generated by the parameter generating unit 28 from the operator.
[0046] The parameter generating unit 28 generates a comprehensive combination of the parameters of the zeolite and the parameters of the metal cations supported by ion exchange. The design support system 10 may also generate a comprehensive combination of the parameters of the zeolite, the parameters of the metal cations supported by ion exchange, and information on anion species.
[0047] The prediction unit 30 predicts the reaction yield of compounds corresponding to the exhaustive combinations generated by the parameter generation unit 28 using a trained machine learning model.
[0048] The candidate output unit 32 outputs data on candidate metal-supported zeolite catalysts based on the reaction yields of compounds corresponding to the exhaustive combinations predicted by the prediction unit 30. The display control unit 34 controls the display of the candidate metal-supported zeolite catalyst data output by the candidate output unit 32 on the user terminal 12.
[0049] The request receiving unit 50 receives a processing request from the user terminal 12. The training dataset acquiring unit 52 acquires a training dataset representing the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and the reaction yield of a compound. The training dataset acquiring unit 52 may acquire a training dataset representing the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and information on anion species, and the reaction yield of a compound. The training dataset acquiring unit 52 may also acquire a parameter table, which will be described later, and store it in the parameter table storage unit 57. The parameter table storage unit 57 stores the parameter table, which will be described later.
[0050] The learning unit 54 causes the machine learning model to learn the correspondence relationships between the parameters of the zeolite and the parameters of the metal cations supported by ion exchange and the reaction yield of the compound, based on the training dataset acquired by the training dataset acquisition unit 52. Furthermore, the learning unit 54 may cause the machine learning model to learn the correspondence relationships between the parameters of the zeolite, the parameters of the metal cations supported by ion exchange, and information on the anion species and the reaction yield of the compound, based on the training dataset acquired by the training dataset acquisition unit 52.
[0051] The registration request unit 56 requests the design support device 10 to register the machine learning model that has been trained by the learning unit 54, and stores the trained machine learning model in the trained machine learning model storage unit 44. Furthermore, the machine learning model storage unit 58 stores the machine learning model that the learning unit 54 trains.
[0052] 3 is an example. The information processing system 1 according to this embodiment can be configured in various ways. For example, the search range information storage unit 40, the parameter table storage unit 42, the trained machine learning model storage unit 44, the parameter table storage unit 57, and the machine learning model storage unit 58 may be included in a storage device, computer, cloud storage, or the like that can communicate data with the design support device 10 or the learning device 11.
[0053] <Processing> <Learning Process> The information processing system 1 according to this embodiment performs a learning process for a machine learning model, for example, according to the procedure shown in Fig. 4. Fig. 4 is a flowchart of an example of a process for learning a machine learning model in the information processing system according to this embodiment. For example, the flowchart in Fig. 4 represents a process that starts upon receiving a request for the learning process for a machine learning model from the user terminal 12.
[0054] In step S10, the training dataset acquisition unit 52 of the learning device 11 acquires a training dataset to be used for training the machine learning model. The training dataset is a training dataset that represents the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and the reaction yield of a compound, or a training dataset that represents the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and information on anion species, and the reaction yield of a compound. Here, an example of a training dataset 1000 shown in FIG. 5 will be described.
[0055] Fig. 5 is a diagram showing an example of a training dataset before parameter conversion. The training dataset 1000 shown in Fig. 5 has the following items: crystal system, pore size, cation species, BET specific surface area, Si / Al ratio, ion-exchange anion, propylene yield, C2-5 olefin / paraffin ratio, and useful component yield.
[0056] The crystal system represents the zeolite framework structure using a framework code. The pore size, BET specific surface area, and Si / Al ratio are examples of zeolite framework parameters. The cation species represents the cation element. The ion-exchange anion represents information on the anion species. The propylene yield, C2-5 olefin / paraffin ratio, and useful component yield represent the reaction yield of the compound.
[0057] In step S12, the training dataset acquisition unit 52 performs parameter conversion on items that need to be converted into parameters among the items of the training dataset acquired in step S10. If there are no items that need to be converted into parameters, the processing of step S12 may be skipped.
[0058] For example, in the case of the training data set 1000 in Fig. 5, the training data set acquisition unit 52 converts the data in the item "cation species" in Fig. 5 into metal cation parameters such as cation electronegativity, cation ion crystal radius, and cation charge, using the parameter table 1010 in Fig. 6. Fig. 6 is a diagram illustrating an example of a parameter table for metal cations.
[0059] By the processing of step S12, the training dataset before parameter conversion in FIG. 5 is converted into a training dataset 1020 in FIG. 7. FIG. 7 is a structural diagram of an example of a training dataset after parameter conversion. The training dataset 1020 shown in FIG. 7 represents the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, information on anion species, and the reaction yield of the compound. In FIG. 7, the item "information on anion species" is converted to the value "0" indicating that no anion is used as a dummy variable, or the value "1" indicating that an anion is used.
[0060] In the training data set 1020, the items "pore diameter," "BET specific surface area," and "Si / Al ratio" indicate parameters of the zeolite framework (examples of zeolite parameters). In addition, in the training data set 1020, the items "cation ion crystal radius," "cation electronegativity," and "cation charge" indicate parameters of metal cations. In addition, in the training data set 1020, the items "ion exchange anion_NO ３ － " and "Ion exchange anion _OH － " indicates information on anion species. In the learning dataset 1020, the items "propylene yield," "C2-5 olefin / paraffin ratio," and "useful component yield" indicate the reaction yield of the compound.
[0061] In step S14, the learning unit 54 causes the machine learning model to learn the correspondence between the zeolite parameters, the parameters of the metal cations supported by ion exchange, and the information on the anion species and the reaction yield of the compound, based on the training data set after parameter conversion, for example, as shown in Figure 7.
[0062] Specifically, the learning unit 54 trains the machine learning model using, for example, the zeolite parameters of the training data set after parameter conversion shown in Figure 7, the parameters of the metal cations supported by ion exchange, and information on the anion species as explanatory variables, and the reaction yield of the compound as a target variable.
[0063] The machine learning model learned by the learning device 11 is used for prediction processing in the design support device 10.
[0064] <<Prediction Process>> The information processing system 1 according to this embodiment predicts candidates for metal-supported zeolite catalysts having desired properties, for example, according to the procedure shown in Fig. 8. Fig. 8 is a flowchart of an example of a process for predicting candidates for metal-supported zeolite catalysts having desired properties in the information processing system according to this embodiment. For example, the flowchart in Fig. 8 represents a process that starts upon receiving a request for prediction process from the user terminal 12.
[0065] In step S20, the design support device 10 acquires a trained machine learning model trained by the learning device 11. In step S22, the search range receiving unit 26 receives input of a search range for the metal-supported zeolite catalyst. For example, the search range receiving unit 26 receives input of a search range 1100 for the metal-supported zeolite catalyst as shown in FIG. 9 from the search range information stored in the search range information storage unit 40.
[0066] 9 is an explanatory diagram of an example of a search range for a metal-supported zeolite catalyst. The search range in FIG. 9 is an example, and multiple values such as BET specific surface area or Si / Al ratio may be specified. For example, if the BET specific surface area is included in the search range for a metal-supported zeolite catalyst, 5, 100, 200, 300, 400, 500, 600, and 700 may be specified.
[0067] In step S24, the parameter generating unit 28 generates comprehensive combinations of metal cations, anions, and zeolites within the search range of the metal-supported zeolite catalysts input in step S22.
[0068] In step S26, the parameter generation unit 28 performs parameter conversion on items that require conversion into parameters for each of the comprehensive combinations of metal cations, anions, and zeolites generated in step S24. For example, the parameter generation unit 28 uses the parameter table 1010 shown in FIG. 6 to convert data in the "cation species" category into cation electronegativity, cation ion crystal radius, and cation charge, which are parameters of metal cations. Furthermore, the parameter generation unit 28 uses the parameter table 1120 shown in FIG. 10 to convert data in the "crystal system" category into pore size, which is a parameter of the zeolite framework. Note that the BET specific surface area and Si / Al ratio are not uniquely determined by the zeolite framework, and therefore require separate input of numerical values.
[0069] In step S28, the prediction unit 30 predicts the reaction yield of the compound by inputting the comprehensive combinations of metal cations, anions, and zeolites generated by the parameter generation unit 28 into a trained machine learning model. For example, if the reaction yield of the compound is the propylene yield, the C2-5 olefin / paraffin ratio, and the useful component yield, the comprehensive combinations of metal cations, anions, and zeolites are input into the trained machine learning model for the propylene yield, the trained machine learning model for the C2-5 olefin / paraffin ratio, and the trained machine learning model for the useful component yield.
[0070] In step S30, the candidate output unit 32 tallyes the reaction yields of compounds corresponding to the exhaustive combinations predicted by the prediction unit 30. For example, the candidate output unit 32 sorts the reaction yields of compounds corresponding to the exhaustive combinations predicted by the prediction unit 30, for example, in descending order of propylene yield or useful component yield.
[0071] In step S32, the candidate output unit 32 outputs data on candidate metal-supported zeolite catalysts based on the reaction yields of compounds corresponding to the comprehensive combinations predicted by the prediction unit 30. For example, the candidate output unit 32 can select and output data on candidate metal-supported zeolite catalysts from the comprehensive combinations predicted by the prediction unit 30 that have a high propylene yield or useful component yield.
[0072] Example: When decomposing plastics to produce low molecular weight compounds (methyl, ethyl, propyl, ethylene, propylene, etc.), metal-supported zeolite is used as a catalyst. There is a demand for using metal-supported zeolite as a catalyst to further increase the yield of desired components (propylene yield, useful component yield, etc.).
[0073] The yield of useful components is the sum of the yields of olefins up to C5 and useful aromatics (benzene, toluene, ethylbenzene, styrene, xylene (o-xylene, m-xylene, and p-xylene may be combined)).
[0074] 11A and 11B are explanatory diagrams showing an example of the relationship between the predicted values and actual measured values of the trained machine learning model for propylene yield and useful component yield according to this embodiment. FIG. 11A shows the relationship between the predicted values and actual measured values of the trained machine learning model for propylene yield, and the SHAP (Shapley Value) value of the explanatory variable item. FIG. 11B shows the relationship between the predicted values and actual measured values of the trained machine learning model for useful component yield, and the SHAP value of the explanatory variable item. The SHAP value represents the degree of contribution of each explanatory variable item to the predicted value of the trained machine learning model.
[0075] When searching for metal cations that further increase the propylene yield and the yield of useful components, it is clear that the lower the electronegativity from the SHAP value and the larger the ionic crystal radius of the metal cation, the more likely it is to improve the propylene yield. Also, when searching for metal cations that further increase the propylene yield and the yield of useful components, it is clear that the higher the periodicity of the cation, the better.
[0076] In the information processing system 1 according to this embodiment, after predicting the reaction yield of compounds for all the exhaustive combinations in the search range, when sorting in descending order of propylene yield, the metal cation "Rb ＋ The system was selected.
[0077] In addition, in the information processing system 1 according to this embodiment, after predicting the reaction yield of compounds for all the exhaustive combinations in the search range, when sorting them in descending order of useful component yield, "Sr ２＋ The system was selected.
[0078] 12 is an explanatory diagram showing data of candidate metal-supported zeolite catalysts with high predicted propylene yields or useful component yields. Experiments were conducted on data 1200 of candidate metal-supported zeolite catalysts with high predicted propylene yields or useful component yields, and actual measured values were obtained as experimental results.
[0079] 13A to 13C are diagrams showing examples of experimental results. Fig. 13A shows the experimental results for propylene yield, Fig. 13B shows the experimental results for C2-5 olefin / paraffin ratio, and Fig. 13C shows the experimental results for useful component yield.
[0080] As shown in Figures 13A to 13C, "Rb ＋ When the "Metal-supported zeolite catalyst" was used, the propylene yield, C2-5 olefin / paraffin ratio, and useful component yield were improved. In particular, the improvements in the propylene yield and C2-5 olefin / paraffin ratio were remarkable.
[0081] In addition, FIG. 14 shows the results of the analysis using Na as a metal cation. ＋ , Sr ２＋ , or Rb ＋ As shown in FIG. 14, when Rb was used as the metal cation, ＋ The experimental results when using Na as the metal cation ＋ , or Sr ２＋ The experimental results show higher performance than when using
[0082] According to the information processing system 1 of this embodiment, a machine learning model that has learned the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and the reaction yield of a compound is used to screen comprehensive combinations of zeolite parameters and metal cation parameters, and data on a metal-supported zeolite catalyst having desired properties is output. Furthermore, according to the information processing system 1 of this embodiment, a machine learning model that has learned the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and information on anion species, and the reaction yield of a compound is used to screen comprehensive combinations of zeolite parameters, metal cation parameters, and anion species, and data on a metal-supported zeolite catalyst having desired properties is output.
[0083] Therefore, the information processing system 1 according to this embodiment can assist in more efficient design of a metal-supported zeolite catalyst having desired properties.
[0084] [Other Embodiments] The metal-supported zeolite catalyst designed by the design support device 10 according to this embodiment may be produced by supplying data on the metal-supported zeolite catalyst to a manufacturing device that produces the metal-supported zeolite catalyst by blending a plurality of raw materials, for example, to produce a metal-supported zeolite catalyst having desired properties.
[0085] Although the present embodiment has been described above, it will be understood that various modifications in form and details are possible without departing from the spirit and scope of the claims. While the present invention has been described above based on examples, the present invention is not limited to the above examples and various modifications are possible within the scope of the claims. This application claims priority from basic application No. 2023-131006, filed with the Japan Patent Office on August 10, 2023, the entire contents of which are incorporated herein by reference.
[0086] REFERENCE SIGNS LIST 1 Information processing system 10 Design support device 11 Learning device 12 User terminal 18 Communication network 28 Parameter generation unit 30 Prediction unit 32 Candidate output unit 52 Learning dataset acquisition unit 54 Learning unit
Claims
1. A design support device that supports the design of metal-supported zeolite catalysts by using a machine learning model that has been trained to determine the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and reaction yields of compounds, the design support device comprising: a parameter generation unit that generates comprehensive combinations of the zeolite parameters and parameters of the metal cations; a prediction unit that uses the trained machine learning model to predict the reaction yield of the compound corresponding to the combination; and a candidate output unit that outputs data on candidate metal-supported zeolite catalysts based on the predicted reaction yield of the compound corresponding to the combination.
2. The design support device of claim 1, wherein the trained machine learning model has learned a correspondence between the zeolite parameters, the metal cation parameters, and information on anion species and the reaction yield of the compound, and the parameter generation unit generates a comprehensive combination of the zeolite parameters, the metal cation parameters, and information on the anion species to be input to the trained machine learning model.
3. The design support system according to claim 1 or 2, wherein the reaction yield of the compound is at least one of a propylene yield, an olefin / paraffin ratio, and a useful component yield.
4. A learning device for training a machine learning model, comprising: a training dataset acquisition unit for acquiring a training dataset representing a correspondence between zeolite parameters and parameters of metal cations supported by ion exchange, and a reaction yield of a compound; and a learning unit for training the machine learning model based on the training dataset, using the zeolite parameters and the metal cation parameters as explanatory variables and the reaction yield of the compound as a target variable.
5. The learning device according to claim 4, wherein the learning data set represents a correspondence between the zeolite parameters, the metal cation parameters, and the anion species information and the reaction yield of the compound, and the learning unit trains the machine learning model based on the learning data set using the zeolite parameters, the metal cation parameters, and the anion species information as explanatory variables and the compound reaction yield as a target variable.
6. The learning device according to claim 4 or 5, wherein the reaction yield of the compound is at least one of a propylene yield, an olefin / paraffin ratio, and a useful component yield.
7. A design support method in which a computer supports the design of a metal-supported zeolite catalyst using a machine learning model that has been trained on the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and reaction yields of compounds, the design support method comprising: a parameter generation step of generating comprehensive combinations of the zeolite parameters and parameters of the metal cations; a prediction step of predicting the reaction yield of the compound corresponding to the combination using the trained machine learning model; and a candidate output step of outputting data on candidate metal-supported zeolite catalysts based on the predicted reaction yield of the compound corresponding to the combination.
8. A learning method in which a computer trains a machine learning model, the learning method comprising: a learning dataset acquisition step of acquiring a learning dataset that represents a correspondence between zeolite parameters and parameters of metal cations supported by ion exchange, and a reaction yield of a compound; and a learning step of training the machine learning model based on the learning dataset, using the zeolite parameters and the metal cation parameters as explanatory variables and the reaction yield of the compound as a target variable.
9. A program for causing a computer that assists in the design of metal-supported zeolite catalysts using a machine learning model that has been trained on the correspondence between zeolite parameters, parameters of metal cations supported by ion exchange, and reaction yields of compounds to execute the following steps: a parameter generation step that generates comprehensive combinations of the zeolite parameters and parameters of the metal cations; a prediction step that predicts the reaction yield of the compound corresponding to the combination using the trained machine learning model; and a candidate output step that outputs data on candidate metal-supported zeolite catalysts based on the predicted reaction yield of the compound corresponding to the combination.
10. A program for causing a computer that trains a machine learning model to execute: a training dataset acquisition procedure for acquiring a training dataset that represents the correspondence between zeolite parameters and parameters of metal cations supported by ion exchange, and the reaction yield of a compound; and a learning procedure for training the machine learning model based on the training dataset, using the zeolite parameters and the metal cation parameters as explanatory variables and the reaction yield of the compound as a target variable.