Reaction speed prediction device for catalytic reaction, reaction speed prediction method for catalytic reaction, and reaction speed prediction program for catalytic reaction

Abstract

To enable accurate prediction of reaction speed of catalytic reaction.SOLUTION: A reaction speed prediction device for a catalytic reaction according to the present invention comprises: a simulation unit configured to simulate a characteristic value on reaction speed of a catalytic reaction on the basis of energy of a three-dimensional structure of intermediates and transition states of a reactant generated in a process of generating a product from the reactant through the catalytic reaction consisting of repetition of multiple gradually progressing elementary reactions; a comparison unit for comparing the characteristic value to a preset reference value; and an optimization unit configured to use a result of the comparison to optimize the energy on the basis of a result of the comparison by changing an index having a correlation with the energy in such a way that the characteristic value approaches the reference value.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a catalytic reaction rate prediction device, a catalytic reaction rate prediction method, and a catalytic reaction rate prediction program. [Background technology]

[0002] Catalysts are used in many chemical reactions and are widely used in various fields such as hydrogen production, exhaust gas purification, etc. To understand the mechanism of catalytic reactions that produce the desired products from reactants on the catalyst surface, analysis and understanding based on the electronic state is necessary, and theoretical calculations are generally used to identify the electronic state.

[0003] Theoretical calculations utilize computational chemistry such as quantum chemical calculations, and molecular simulations using density functional theory (DFT) are performed to calculate the energies of intermediates and transition states of reactants, thereby predicting the reaction rates of catalytic reactions, etc. If a prediction method that can reproduce experiments can be established using computational chemistry, it will be possible to obtain information such as the state of intermediates that is difficult to observe experimentally, and this will be useful in elucidating mechanisms.

[0004] As a method for predicting the reaction rate of catalytic reactions using computational chemistry, for example, a method has been disclosed in which the intrinsic rates of basic reaction steps that control the catalyst turnover rate, product distribution, preferred mechanistic pathway, rate, and selectivity in complex reaction mechanisms such as the Fischer-Tropsch (FT) reaction are predicted using microkinetic modeling (see, for example, Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Bart Zijlstra, et al., "The Vital Role of Step-Edge Sites for Both CO Activation and Chain Growth on Cobalt Fischer-Tropsch Catalysts Revealed through First-Principles-Based Microkinetic Modeling Including Lateral Interactions",_ACS Catalysis, 2020, 10, p.9376-9400 Summary of the Invention [Problem to be solved by the invention]

[0006] However, when predicting the reaction rate of a catalytic reaction using computational chemistry, as in the method of Non-Patent Document 1, there is a problem in that the predicted value of the reaction rate of a catalytic reaction predicted using computational chemistry may deviate from the actual measured value obtained by experiment, etc., depending on factors such as the model of the reactant structure used for the prediction, the selection of the elementary reactions that make up the catalytic reaction, and the accuracy of the potential.

[0007] An object of the present invention is to make it possible to predict the reaction rate of a catalytic reaction with high accuracy. [Means for solving the problem]

[0008] One aspect of the present invention is a simulation unit that simulates a characteristic value related to the reaction rate of the catalytic reaction based on the energy of three-dimensional structures of intermediates and transition states of reactants that are generated in the process of generating products from reactants through a catalytic reaction in which a plurality of elementary reactions that proceed stepwise are repeated; and a comparison unit that compares the characteristic value with a set reference value; an optimization unit that optimizes the energy by changing an index correlated with the energy based on a result of the comparison so that the characteristic value approaches the reference value; The present invention provides a catalytic reaction rate prediction device comprising:

[0009] Another aspect of the present invention is The computer a simulation step of simulating a characteristic value related to the reaction rate of the catalytic reaction based on the energy of three-dimensional structures of intermediates and transition states of reactants that are generated in the process of generating products from reactants through a catalytic reaction in which a plurality of elementary reactions that proceed stepwise are repeated; a comparison step of comparing the characteristic value with a set reference value; an optimization step of optimizing the energy by changing an index correlated with the energy based on a result of the comparison so that the characteristic value approaches the reference value; The present invention provides a method for predicting the reaction rate of a catalytic reaction.

[0010] Another aspect of the present invention is On the computer, a simulation step of simulating a characteristic value related to the reaction rate of the catalytic reaction based on the energy of three-dimensional structures of intermediates and transition states of reactants that are generated in the process of generating products from reactants through a catalytic reaction in which a plurality of elementary reactions that proceed stepwise are repeated; a comparison step of comparing the characteristic value with a set reference value; an optimization step of optimizing the energy by changing an index correlated with the energy based on a result of the comparison so that the characteristic value approaches the reference value; This is a program for predicting the reaction rate of catalytic reactions. [Effects of the Invention]

[0011] The present invention makes it possible to predict the reaction rate of a catalytic reaction with high accuracy. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a diagram illustrating the concept of a catalytic reaction rate prediction device according to an embodiment of the present invention. FIG. [Figure 2] 1 is a block diagram showing the configuration of a catalytic reaction rate prediction device according to an embodiment of the present invention. [Figure 3] FIG. 10 is a diagram showing an example of the contents of listed elementary reactions. [Figure 4] FIG. 2 is an explanatory diagram showing gas molecules and surface molecules present on a catalyst surface. [Figure 5] An example of converting the three-dimensional structure of the surface molecules of a reactant into a descriptor is shown. [Figure 6] FIG. 1 is a diagram showing an example of the relationship between the number of terminal carbon atoms of an intermediate structure and the calculated value of relative energy. [Figure 7] FIG. 1 is a diagram showing an example of the relationship between the number of terminal carbon atoms of a transition state structure and the calculated value of the relative energy. [Figure 8] FIG. 1 is a diagram showing an example of the relationship between the number of terminal carbon atoms of a transition state structure and the predicted value of the relative energy. [Figure 9] FIG. 10 is a diagram showing an example of the relationship between the calculated values ​​of the relative energies of reactants C3 to C20 and the predicted values ​​predicted by the energy prediction unit. [Figure 10] FIG. 10 is an explanatory diagram showing prediction of values ​​outside the range of known database values. [Figure 11] FIG. 10 is a diagram showing an example of the results of fitting predicted values ​​of reactants C1 to C5 to actual measured values. [Figure 12] FIG. 2 is a block diagram showing the hardware configuration of a catalytic reaction rate prediction device and a catalytic reaction energy prediction device. [Figure 13] 1 is a flowchart illustrating a reaction rate prediction method according to an embodiment of the present invention. [Figure 14] 10 is a flowchart showing a regression model generation step (step S11). [Figure 15] 10 is a flowchart showing a reaction rate prediction step (step S20). DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, embodiments of the present invention will be described in detail. In this specification, unless otherwise specified, the symbol "to" indicating a numerical range means that the numerical values ​​before and after it are included as the lower and upper limits. Furthermore, when a unit is specified for only the upper limit of a numerical range expressed by "to," it means that the lower limit is also expressed in the same unit.

[0014] <Catalytic reaction rate prediction device> The catalytic reaction rate prediction device according to this embodiment (hereinafter simply referred to as the "reaction rate prediction device") will be described.

[0015] The reaction rate prediction device according to this embodiment includes a catalytic reaction energy prediction device (hereinafter simply referred to as the "energy prediction device"). As shown in FIG. 1, the energy prediction device lists a plurality of elementary reactions that proceed stepwise when a reactant reacts with a catalyst, calculates the energy E of the three-dimensional structure of an intermediate and a transition state that occurs during the catalytic reaction of the reactant, and generates a descriptor x of the three-dimensional structure of the intermediate and the transition state at the energy E. The energy prediction device performs learning using the descriptor x of the three-dimensional structure of the intermediate and the transition state as an explanatory variable and the energy E of the three-dimensional structure of the reactant intermediate and the transition state as a target variable, and generates a regression model that predicts the energy E of the three-dimensional structure of the reactant intermediate and the transition state from the descriptor x of the three-dimensional structure of the reactant intermediate and the transition state. The energy prediction device uses the generated regression model to accurately predict the energy E of the three-dimensional structure of the reactant intermediate and the transition state from the descriptors of the three-dimensional structure of the intermediate and the transition state.

[0016] As shown in Figure 1, the reaction rate prediction device according to this embodiment uses the regression model to simulate the reaction rate in the three-dimensional structure of the reactant intermediate and transition state from the predicted energy to obtain a predicted reaction rate, and then compares the predicted value with a reference value, such as an experimentally measured value. Based on the comparison results, such as a large difference between the predicted value and the reference value, the reaction rate prediction device according to this embodiment optimizes the regression model by, for example, changing the regression parameters of the regression model, and optimizing the energy predicted by the regression model, thereby improving the prediction accuracy of the reaction rate of the three-dimensional structure of the reactant intermediate and transition state. This improves the prediction accuracy of the reaction rate of a catalytic reaction.

[0017] A catalytic reaction is a type of chemical reaction in which multiple elementary reactions proceed stepwise, and examples of catalytic reactions include the Fischer-Tropsch (FT) reaction and the methanol synthesis reaction.

[0018] The catalyst may be a simple metal, an alloy containing multiple metals, or a compound containing a metal.

[0019] Reactants are substances used to produce products.

[0020] 2 is a block diagram showing the configuration of a reaction rate prediction device according to this embodiment. As shown in FIG. 2, the reaction rate prediction device 1 includes an energy prediction device 10, a reaction rate prediction unit 20, and an output unit 30.

[0021] [Energy prediction device] The energy prediction device 10 includes a regression model generation unit 11 and an energy prediction unit 12 .

[0022] (Regression model generation part) The regression model generation unit 11 generates a regression model M1 that predicts the energy of these three-dimensional structures based on the three-dimensional structure of a reactant intermediate (hereinafter also referred to as the "intermediate structure") and the three-dimensional structure of a transition state (hereinafter also referred to as the "transition state structure") that are generated in the process of generating a product from a reactant by a catalytic reaction, and the energies of these three-dimensional structures.

[0023] The regression model generation unit 11 includes a first acquisition unit 111 , an enumeration unit 112 , a structural optimization calculation unit 113 , a descriptor generation unit 114 , and a model generation unit 115 .

[0024] ((Acquisition Department)) The first acquisition unit 111 acquires information on the types of catalysts and reactants involved in the target catalytic reaction, intermediate structures and transition state structures of the reactants that occur in the process of generating products from the reactants through the catalytic reaction, and multiple elementary reactions involved in the catalytic reaction. The multiple elementary reactions involved in the catalytic reaction may be acquired from data stored in the storage unit 100 or an external database. The first acquisition unit 111 may acquire the multiple elementary reactions involved in the catalytic reaction sequentially, or may acquire the multiple elementary reactions involved in the multiple catalytic reactions used in the calculation all at once.

[0025] ((enumeration part)) The enumeration unit 112 enumerates the multiple elementary reactions acquired by the first acquisition unit 111. The elementary reactions to be enumerated may be appropriately selected by the user based on the type of catalytic reaction, information on catalytic reactions disclosed in publicly known documents, etc. The data enumerated by the enumeration unit 112 includes, for example, as shown in FIG. 3, the names and numbers of reactants 1 and 2 and products (intermediates) 1 and 2, and transition states occurring in the elementary reactions, for each elementary reaction.

[0026] ((Structural optimization calculation part)) The structural optimization calculation unit 113 calculates the energies of intermediate structures and transition state structures included in a plurality of elementary reactions. The structural optimization calculation unit 113 calculates the energies from the three-dimensional structures of the gas molecules, intermediates, and transition states included in the elementary reaction formulas.

[0027] 4, for example, on the surface of a catalyst, there are gas molecules that are not adsorbed to the catalyst and surface molecules that are adsorbed molecules that are adsorbed to the catalyst. The structure optimization calculation unit 113 calculates the three-dimensional structure of the gas molecules (hereinafter also referred to as "gas molecular structure") of the elementary reaction, and the energies of the intermediate structure and transition state structure of the surface molecules.

[0028] The gas molecules may be any reactant gas molecules that appear in elementary reactions, such as carbon monoxide (CO), hydrogen (H2), methane (CH4), water vapor (H2O), and carbon dioxide (CO2).

[0029] An intermediate is a product of a reactant that appears in an intermediate reaction among multiple elementary reactions. For example, CO * , H * , C.O.H. * Examples include:

[0030] The transition state is the transition state from the initial state (IS) before the reaction to the final state (FS) in an elementary reaction. For example, CO * and H * By reaction with COH * and * When CO is produced, the elementary reaction is * +H * →CO-H * + * →COH * + * " and the initial structure included in the elementary reaction equation is (CO * ,H * ) and the final structure is (COH * ) and the transition state structure is (CO-H * )

[0031] The structural optimization calculation unit 113 preferably performs structural optimization calculations of the three-dimensional structures of the gas molecules, intermediates, and transition states of the reactants in the listed elementary reactions. That is, the structural optimization calculation unit 113 preferably performs structural optimization of the three-dimensional structures of the gas molecules, intermediates, and transition states of the reactants to search for an optimized structure (also referred to as a stable structure), and calculates the energy of the stable structure.

[0032] When performing structural optimization calculations for gas molecule structures, gas molecules are prepared in a vacuum, and then a stable structure is searched for using a structural optimization method, and the energy at the time of the stable structure is obtained.

[0033] Examples of structural optimization methods that can be used include the BFGS (Broyden-Fletcher-Goldfarb-Shanno algorithm), the LBFGS (Limited Memory Broyden-Fletcher-Goldfarb-Shanno algorithm), BFGSLineSearch, and FIRE (fast inertial relaxation engine).

[0034] The convergence conditions may be selected appropriately. For example, convergence may be determined when the maximum force (fmax) acting between atoms is 0.01 or less.

[0035] When performing structural optimization calculations for the intermediate structure, surface molecules are placed on the surface of the catalyst, and then a stable structure of the surface molecules on the catalyst surface is searched for using a structural optimization method, and the energy at that stable structure is obtained.

[0036] The structural optimization method can be the same as that used in the structural optimization calculation of the gas molecule structure described above.

[0037] The convergence conditions can be the same as those used in the structural optimization calculation of the gas molecule structure.

[0038] When performing structural optimization calculations for the transition state structure, after structural optimization of the initial and final state structures, the transition state structure that occurs when generating the final state structure from the initial state structure is calculated using a method such as Nudged Elastic Band (NEB), and the transition state energy for that structure is calculated. The NEB method selects a path from the initial state structure to the final state structure, and performs optimization so that the activation energy of the transition state on the selected path is low. The energy of the optimized transition state structure at this time is taken as the transition state energy.

[0039] The convergence conditions can be the same as those used in the structural optimization calculation of the gas molecule structure.

[0040] The structural optimization calculation unit 113 preferably uses a trained model or machine learning potential that predicts the energies of the intermediate structure and transition state structure of the reactant from the descriptors of the intermediate structure and transition state structure of the reactant.

[0041] The trained model preferably applies a machine learning algorithm, which may be any of supervised learning, unsupervised learning, and reinforcement learning, or any combination thereof.

[0042] Examples of machine learning algorithms include regression algorithms, instance-based algorithms, decision tree algorithms, Bayesian algorithms, clustering algorithms, association rule learning algorithms, dimensionality reduction algorithms, ensemble learning algorithms, and deep learning algorithms.

[0043] Deep learning algorithms include, for example, neural networks (NNs), deep Boltzmann machines (DBMs), deep belief networks (DBNs), and stacked autoencoders.

[0044] The neural network can use deep learning, which is a neural network with more than three layers. Types of neural networks that can be used include, for example, convolutional neural networks (CNNs), recurrent neural networks (RNNs), and general regression neural networks.

[0045] The machine learning potential is an interatomic potential using a machine learning method that outputs energy from information about the atomic structure. Examples of machine learning potentials include neural network potential (NNP), Gaussian approximation potential (GAP), spectral neighbor analysis potential (SNAP), and moment tensor potential (MTP). Among these, NNP is preferred as the machine learning potential because of the high flexibility of neural networks. Preferred potential (PFP) may also be used as the NNP.

[0046] ((Descriptor Generation Part)) The descriptor generation unit 114 generates descriptors representing the intermediate structure and the transition state structure as structural descriptors based on the intermediate structure and the transition state structure. That is, the descriptor generation unit 114 acquires information such as intermolecular bonds and molecule-catalyst bonds of the intermediate structure and the transition state structure obtained by the structural optimization calculation unit 113, converts the information so as to express the characteristics of the intermediate structure and the transition state structure, and generates structural descriptors.

[0047] The structural descriptors generated by the descriptor generation unit 114 are descriptors based on information surrounding the active sites of the reactants with the catalyst, and may be generated based on peripheral information of the catalyst and the reactants in the three-dimensional structure.

[0048] The structural descriptor may be expressed, for example, using the numbers of C, H, O, and N in the surface molecules that are reactants, the numbers of C, H, O, and N in the surface molecules that are bonded to the catalyst, the number of metal atoms in the catalyst that are bonded to the surface molecules, and the average coordination number of the metal atoms in the catalyst that are bonded to the surface molecules.

[0049] The structural descriptors may include molecular descriptors that are not involved in binding with the catalyst, and interaction descriptors that are directly involved in binding with the catalyst.

[0050] 5 shows an example of converting the three-dimensional structure of the surface molecules of a reactant into a descriptor. In FIG. 5, the reactant is C4H8. As shown in FIG. 5, the three-dimensional structure of the intermediate or transition state of C4H8 obtained by the structure optimization calculation unit 113 (see FIG. 5(a)) is converted into a graph structure, thereby generating a structural descriptor (see FIG. 5(b)) having a molecular descriptor and an interaction descriptor.

[0051] When the structural descriptor is a transition state structure, the structural descriptor of the transition state structure of the reactant may be generated by converting the three-dimensional structures of the initial state structure and the final state structure based on the energy difference between the initial state structure and the final state structure of the catalytic reaction.

[0052] ((Model generation part)) The model generation unit 115 generates a regression model M1 using the structural descriptors generated by the descriptor generation unit 114 as explanatory variables and the energy obtained by the structural optimization calculation unit 113 as a response variable.

[0053] The regression model M1 may use, for example, a function such as the following formula (1): In formula (1), E is energy, X is a structural descriptor, and W is a constant (also called a regression parameter). E=W×X (1)

[0054] Examples of regression models that can be used include linear regression models such as Ridge regression, Lasso regression, Elastic Net regression, multiple linear regression, and polynomial regression, as well as nonlinear regression models such as Kernal Ridge regression, Random Forest, gradient boosting, support vector regression (SMR), and Gaussian process regression (GPR). Among these, Lasso regression, which is a linear model with excellent interpretability, and GPR, which is a nonlinear model with excellent expressive power, are preferred.

[0055] The model generation unit 115 may generate the regression model of the intermediate structure and the regression model of the transition state structure separately, or may generate them as a single regression model.

[0056] When the energies of the intermediate structure and the transition state structure are optimized in the optimization unit 27 of the reaction rate prediction unit 20, the regression model M1 may use the regression model changed in the optimization unit 27.

[0057] (Energy Prediction Department) The energy prediction unit 12 acquires the descriptors of the intermediate structure and the descriptors of the transition state structure generated by the descriptor generation unit 114. The energy prediction unit 12 predicts the energies of the intermediate structure and the transition state structure from the descriptors of the intermediate structure and the descriptors of the transition state structure generated by the descriptor generation unit 114, using the regression model M1 generated by the model generation unit 115.

[0058] Furthermore, when the energies of the intermediate structure and the transition state structure are optimized by correcting the regression parameters of the regression model in the optimization unit 27 of the reaction rate prediction unit 20, which will be described later, the energy prediction unit 12 may use the optimized energies. That is, the energy prediction unit 12 may predict the energies of the intermediate structure and the transition state structure using the regression model optimized in the optimization unit 27 as the regression model M1. In this way, the energy prediction unit 12 can newly predict the energies of the intermediate structure and the transition state structure using the regression model M1 optimized based on the energies of the intermediate structure and the transition state structure predicted before optimization.

[0059] The energy prediction unit 12 preferably uses a machine learning potential to predict the energies of the intermediate structure and the transition state structure from the descriptors of the intermediate structure and the transition state structure generated by the descriptor generation unit 114. This enables the energy prediction unit 12 to increase the calculation speed of the energies of the intermediate structure and the transition state structure and to predict the energies with high accuracy. Note that the machine learning potential is the same as described above, and therefore details thereof will be omitted.

[0060] The energy prediction unit 12 may predict the energy of the transition state (hereinafter referred to as "transition state energy") without using the structural descriptors of the intermediate structure and the transition state structure generated by the descriptor generation unit 114. The transition state energy of a catalytic reaction is related to the energy before and after the elementary reaction, as represented by the BEP (Bell-Evans-Polanyi) rule. For example, a regression model can be generated that predicts the transition state energy from the reaction energy of the elementary reaction obtained by the structural optimization calculation unit 113, using the energy difference before and after the elementary reaction (also referred to as reaction energy) as an explanatory variable and the transition state energy as a target variable. In this case, there is no need to calculate and prepare all transition state structures, so the energy prediction unit 12 can predict the transition state energy even when the structural optimization calculation unit 113 cannot calculate the transition state structure.

[0061] As described above, the energy prediction device 10 includes a regression model generation unit 11 and an energy prediction unit 12. In the energy prediction device 10, it is preferable that the regression model generation unit 11 generates a regression model M1, and the energy prediction unit 12 uses the regression model M1 generated by the regression model generation unit 11. This enables the energy prediction device 10 to calculate with high accuracy the energies of intermediate structures and transition state structures included in each elementary reaction involved in a catalytic reaction.

[0062] Here, when the intermediate structure and transition state structure are structurally optimized by the structural optimization calculation unit 113 and then the energies of the structurally optimized intermediate structure and transition state structure are calculated, outliers, which are values ​​that are significantly different from the energy values ​​of the intermediate structure and transition state structure, or missing values, which are values ​​where no energy is calculated, may occur. FIG. 6 shows an example of a calculation of the relative energy of a structurally optimized intermediate structure, and FIG. 7 shows an example of a calculation of the relative energy of a structurally optimized transition state structure. Note that FIGS. 6 and 7 show an example of the relationship between the type of terminal hydrocarbon of the reactant and the relative energy when the reactant has 3 to 20 carbon atoms (C3 to C20). The relative energy is a relative value when the energy of the gas molecules H2, CH4, and HO is set to 0 eV as a reference when calculating the energies of the intermediate structures and transition state structures in multiple elementary reactions in the structural optimization calculation unit 113. The calculated values ​​show the results of calculations using a general NNP. As shown in Figure 6, in the case of intermediate structures, the outliers in the relative energy of the intermediate structures tended to increase as the number of carbon atoms in the reactants increased, regardless of the terminal hydrocarbon of the reactant. As shown in Figure 7, in the case of transition state structures, many outliers and missing values ​​occurred at the stage where the number of carbon atoms in the reactants was small, regardless of the terminal hydrocarbon of the reactant. This is because, as hydrocarbons grow from the terminal hydrocarbon of the reactant, the number of carbon atoms in the reactant increases, resulting in a highly covered surface where a high proportion of reactants cover the surface of the catalyst, making it difficult for the structure optimization calculation unit 113 to search for the most stable transition state structure.

[0063] In contrast, in the energy prediction device 10, the energy prediction unit 12 uses the regression model M1 generated by the model generation unit 115 to predict these energies from the descriptors of the intermediate structure and the descriptors of the transition state structure generated by the descriptor generation unit 114, and the predicted energies have high accuracy. FIG. 8 shows an example of the relative energy of a structurally optimized transition state structure predicted using the regression model M1. Note that, like FIGS. 6 and 7, FIG. 8 also shows an example of the relationship between the type of terminal hydrocarbon and the relative energy of the reactants for C3 to C20, and the calculated values ​​show the results of calculations using NNP. As shown in FIG. 8, the predicted energies showed a relationship between the carbon number of the reactants and the relative energy depending on the carbon number, regardless of the type of terminal hydrocarbon of the reactants and the carbon number of the reactants. There were no outliers or missing values ​​in the relative energy, and a similar tendency was observed between the carbon number of the reactants and the relative energy. Therefore, even if the energy prediction unit 12 corrects the outliers and missing values ​​of the energy that occur when calculating the energies of the structurally optimized intermediate structure and transition state structure using the regression model M1, it can accurately calculate the energies of the structurally optimized intermediate structure and transition state structure.

[0064] FIG. 9 shows an example of the relationship between the calculated relative energy values ​​of the intermediate structures and transition state structures of the C3-C20 reactants and the predicted values ​​predicted by the energy prediction unit 12. The calculated values ​​in FIG. 9(a) are the calculated relative energy values ​​of the intermediate structures shown in FIG. 6, the calculated values ​​in FIG. 9(b) are the calculated relative energy values ​​of the transition state structures shown in FIG. 7, and the predicted values ​​are the predicted relative energy values ​​of the transition state structures shown in FIG. 8. The predicted values ​​in FIG. 9(a) are the relative energies of the intermediate structures predicted from the descriptors of the intermediate structures generated by the descriptor generation unit 114 using the regression model M1 generated by the model generation unit 115. As shown in FIG. 9, in both cases of the intermediate structures and transition state structures of the C3-C20 reactants, there is a relationship between the calculated relative energy values ​​and the predicted values ​​predicted by the energy prediction unit 12. The mean absolute error (MAE) of the predicted values ​​of the intermediate structures is 0.14, and the MAE of the predicted values ​​of the transition state structures is 0.31. Therefore, even if the structure optimization calculation unit 113 optimizes the intermediate structure and transition state structure, the energy prediction unit 12 can accurately calculate the energies of the optimized intermediate structure and transition state structure.

[0065] Furthermore, hydrocarbons grow chain-like from the terminal hydrocarbons of the reactant, and the greater the number of carbon atoms in the reactant, the greater the number of linked hydrocarbons in the resulting reactant. However, this increases the coverage of the catalyst surface with the reactant, which may affect the calculation of the energies of the intermediate structure and transition state structure of the reactant. The energy prediction unit 12 can accurately calculate the energies of the structurally optimized intermediate structure and transition state structure of the reactant, even when the number of carbon atoms in the reactant is, for example, 3 (C3) or more.

[0066] Therefore, the energy prediction device 10 uses the regression model M1 generated in the regression model generation unit 11 in the energy prediction unit 12, and corrects the outliers and missing values ​​of energy that occur when calculating the energies of the structurally optimized intermediate structure and transition state structure with the regression model M1, thereby making it possible to calculate the energies of the structurally optimized intermediate structure and transition state structure with high accuracy.

[0067] In the energy prediction device 10, the regression model generation unit 11 preferably acquires the energy using a trained model or a machine learning potential. This allows the regression model generation unit 11 to easily calculate the energies of the intermediate structure and the transition state structure from their descriptors. Therefore, the energy prediction device 10 can more easily generate the regression model M1.

[0068] Furthermore, it is preferable that the regression model generation unit 11 acquires the energy using a machine learning potential. When performing energy calculations using density functional theory (DFT), which has conventionally been generally used in quantum chemical calculations, a large amount of calculations and an enormous amount of calculation time are required. By using a machine learning potential in the structural optimization calculation unit 113, the energy prediction device 10 can speed up the calculations in the structural optimization calculation unit 113, thereby further shortening the time required for energy calculations. Therefore, the energy prediction device 10 can significantly reduce the time required to generate the regression model M1.

[0069] In the energy prediction device 10, the energy prediction unit 12 may predict energy using a machine learning potential. By using the machine learning potential to predict the energies of the intermediate structure and the transition state structure from the descriptors of the intermediate structure and the transition state structure, the energy prediction unit 12 can speed up calculations, thereby further shortening the time required for energy calculation. Therefore, the energy prediction device 10 can significantly reduce the time required to predict the energies of the intermediate structure and the transition state structure.

[0070] In the energy prediction device 10, the regression model generation unit 11 preferably includes a descriptor generation unit 114. This allows the descriptor generation unit 114 to generate structural descriptors that represent intermediate structures and transition state structures of reactants. By using the structural descriptors, the regression model generation unit 11 can generate a more accurate regression model M1, thereby allowing for more accurate calculation of the energies of the intermediate structures and transition state structures. Therefore, the energy prediction device 10, in the energy prediction unit 12, can more accurately calculate the energies of the intermediate structures and transition state structures included in each elementary reaction involved in the catalytic reaction.

[0071] In the energy prediction device 10, the descriptor generation unit 114 preferably includes an interaction descriptor directly related to bonding with the catalyst in the structural descriptor. By using the interaction descriptor, the regression model generation unit 11 can generate the regression model M1 based on information surrounding the active site with the catalyst, thereby enabling energy calculation of the intermediate structure and transition state structure in a state closer to reality. Therefore, the energy prediction device 10, in the energy prediction unit 12, can calculate the energy of the intermediate structure and transition state structure included in each elementary reaction involved in the catalytic reaction with even greater accuracy.

[0072] In the energy prediction device 10, the descriptor generation unit 114 preferably calculates the number of C, H, O, and N atoms in the surface molecules constituting the reactant, the number of C, H, O, and N atoms in the surface molecules that are bonded to the catalyst, and the number and average coordination number of metal atoms in the catalyst that are bonded to the surface molecules. The regression model generation unit 11 can generate a more accurate regression model M1 by limiting the descriptors used to specific descriptors that satisfy the above conditions, thereby enabling more accurate energy calculation of intermediate structures and transition state structures. Therefore, the energy prediction device 10, in the energy prediction unit 12, can more accurately calculate the energies of intermediate structures and transition state structures included in each elementary reaction involved in the catalytic reaction.

[0073] In the energy prediction device 10, the descriptor generation unit 114 preferably generates a descriptor representing the transition state structure of a reactant based on the energy difference between the initial state structure and the final state structure of the catalytic reaction. This allows the regression model generation unit 11 to appropriately generate a descriptor representing the transition state structure and more reliably generate a highly accurate regression model M1, thereby allowing appropriate energy calculation of the intermediate structure and the transition state structure. Therefore, the energy prediction device 10, in the energy prediction unit 12, can more reliably and accurately calculate the energy of the intermediate structure and the transition state structure included in each elementary reaction involved in the catalytic reaction.

[0074] In the energy prediction device 10, the regression model generation unit 11 preferably uses Lasso regression or Gaussian process regression as the regression model M1, which allows the regression model generation unit 11 to easily generate the regression model M1.

[0075] [Reaction rate prediction section] As shown in FIG. 2, the reaction rate prediction unit 20 includes a second acquisition unit 21, a gas molecule energy correction unit 22, a reaction rate simulation unit 23, a setting unit 24, a comparison unit 25, a determination unit 26, and an optimization unit 27.

[0076] (Second Acquisition Section) The second obtaining unit 21 obtains the energies of the intermediate structure and transition state structure of the reactant predicted by the energy predicting unit 12.

[0077] (Energy correction for gas molecules) The gas molecule energy correction unit 22 corrects the energy of the gas molecular structure calculated by the structure optimization calculation unit 113, which is a reactant that does not react with the catalyst. As described above, the structure optimization calculation unit 113 of the regression model generation unit 11 calculates the energy of each of the three-dimensional structures of the gas molecules, intermediates, and transition states of multiple elementary reactions. At this time, calculations are not performed for the gas molecular structure to generate a regression model M1 to improve the accuracy of the energy of the intermediate structures and transition state structures, as is the case with the intermediate structures and transition state structures. Similarly to the intermediate structures and transition state structures, the energy of the gas molecular structure calculated by the structure optimization calculation unit 113 may not necessarily be highly accurate. Furthermore, when simulating the reaction rate in the reaction rate simulation unit 23, it is preferable that the energy of the gas molecular structure be highly accurate in order to improve the accuracy of the simulated reaction rate. Therefore, the gas molecule energy correction unit 22 preferably corrects the energy of the gas molecular structure calculated by the structure optimization calculation unit 113 based on values ​​from a known experimental database, etc.

[0078] The known experimental database is not particularly limited, but examples thereof include the Active Thermochemical Tables (ATcT) and the like.

[0079] An example of the relationship between the number of hydrocarbons and the relative energy of the calculated values ​​of the energy of gas molecules calculated by the structural optimization calculation unit 113 and the numerical data of a known experimental database is shown in Fig. 10. As shown in Fig. 10, when there are values ​​not recorded in the known experimental database and there are values ​​calculated by the structural optimization calculation unit 113, the values ​​in the range not recorded in the known experimental database are calculated by extrapolation based on the calculated values ​​by the structural optimization calculation unit 113. Then, with reference to this calculated extrapolated value, the calculated value calculated by the structural optimization calculation unit 113 is corrected so that it matches the extrapolated value.

[0080] In Fig. 10, the vertical axis represents relative energy, but this may be changed as appropriate depending on the type of characteristic in the known experimental database used for correction. For example, if the characteristic stored in the known experimental database is formation enthalpy, the vertical axis may represent formation enthalpy. In this case, the formation enthalpy may be converted into potential energy or the like, and then finally converted into relative energy.

[0081] (Reaction Rate Simulation Section) The reaction rate simulation unit 23 simulates characteristic values ​​relating to the reaction rate of a catalytic reaction based on the energy of the intermediate structure and transition state structure of the reactants that are generated in the process of generating a product from the reactants through a catalytic reaction.

[0082] That is, the reaction rate prediction device 1 predicts the reaction rate of a catalytic reaction by performing microkinetics, which consists of enumerating elementary reactions in the enumeration unit 112 of the energy prediction device 10, calculating the energies of intermediate structures and transition state structures included in the elementary reactions in the energy prediction unit 12, and calculating the reaction rate in the reaction rate simulation unit 23.

[0083] The characteristic value relating to the reaction rate may be any characteristic value that can specify the reaction rate, such as the reaction rate and the selectivity. Either one of these may be used, or two or more of them may be used. In this embodiment, the characteristic value is the reaction rate.

[0084] The reaction rate simulation unit 23 converts the energy calculated by the regression model M1 into a reaction rate equation based on absolute reaction rate theory, and solves simultaneous ordinary differential equations to obtain the reaction rate of each elementary reaction listed in the enumeration unit 112 described above.

[0085] The reaction rate equation based on absolute reaction rate theory is not particularly limited, and may be any commonly used equation.

[0086] For example, if the product is methane and the reaction rate is the rate of methane formation, then ((CH4(g)+* )<-CH4 * ) and calculate the methane production rate y sim. get.

[0087] (Settings section) The setting unit 24 sets the reference value.

[0088] As the reference value, for example, a characteristic value already obtained, such as a characteristic value obtained by an experiment or a characteristic value disclosed in a document, may be used.

[0089] (Comparison section) The comparison unit 25 compares the characteristic value obtained by the reaction rate simulation unit 23 through the reaction rate simulation with the reference value set by the setting unit 24 .

[0090] The comparison unit 25 compares the production rate y of the product generated in each elementary reaction in the simulation obtained by the reaction rate simulation unit 23. sim. In addition, by calculating the ratio of the production rates, the selection rate S sim. It is also possible to calculate the actual rate of production of the product y exp. Or selection rate S expm. can be obtained.

[0091] The production rate of these simulated products, y sim. and selection rate S sim. and the actual product formation rate y exp. and selection rate S expm. Using the above, the degree of agreement between the simulation value obtained by simulation and the reference value can be compared using the following equations (11) and (12). Diff_rate = Σ(y sim.,i- y exp.,i ) 2 ···(11) Diff_selectivity = = Σ(S sim.,i- S exp.,i ) 2 ···(12)

[0092] In equation (1), Diff_rate is the error in the generation rate, Diff_selectivity is the error in the selectivity, and y sim.,i is the simulated production rate of product i, and y exp.,i is the production rate of product i determined by experiments, etc. In equation (2), S sim.,i is the selectivity of product i obtained by simulation, and S exp.,i is the selectivity of product i determined by experiments or other means.

[0093] The comparison unit 25 may compare the characteristic value acquired by the reaction rate simulation in the reaction rate simulation unit 23 with at least one of the reaction rate and the selectivity as the reference value set in the setting unit 24 by calculating the sum of square errors.

[0094] (Judgment Department) The determination unit 26 determines whether the difference in absolute value between the characteristic value and the reference value is equal to or less than a predetermined value. If the difference between the characteristic value and the reference value is equal to or less than the predetermined value, the determination unit 26 determines that the difference between the characteristic value and the reference value is sufficiently small, and therefore the accuracy of the predicted characteristic value is small enough not to affect use, and terminates the calculation. On the other hand, if the difference between the characteristic value and the reference value exceeds the predetermined value, the determination unit 26 determines that the difference between the characteristic value and the reference value is large, and therefore the accuracy of the predicted characteristic value is large enough to affect use, and determines that optimization is necessary.

[0095] The predetermined value is not particularly limited and may be set to any appropriate value.

[0096] (Optimization section) Based on the comparison result, the optimization unit 27 changes the index correlated with energy so that the characteristic value approaches the reference value, and optimizes the energy of the intermediate structure and transition state structure predicted by the energy prediction unit 12.

[0097] The index correlated with energy may be any index that affects the magnitude of the energy of the intermediate structure and the transition state structure, and an example of the index correlated with energy is the regression parameter (see W in formula (1)), which is a constant of the regression model M1.

[0098] That is, when the index correlated with energy is a regression parameter of the regression model M1, the optimization unit 27 can optimize the energies of the intermediate structure and the transition state structure by changing the regression parameter of the regression model M1 so that the characteristic value approaches the reference value. In this case, the energy prediction unit 12 predicts the energies of the intermediate structure and the transition state structure of the reactant again using the optimized regression model M1, and the reaction rate simulation unit 23 simulates the reaction rate of the catalytic reaction again using the predicted energies of the intermediate structure and the transition state structure of the reactant.

[0099] The parameter modification method can use global optimization methods such as random search and Bayesian optimization, and local optimization methods such as steepest descent method, etc. Among these, it is preferable to use at least one of random search and Bayesian optimization as the parameter modification method.

[0100] In the reaction rate prediction unit 20, in addition to changing the parameters of the regression model M1, the reaction rate simulation values ​​output by the reaction rate simulation unit 23 can also be changed by directly modifying the energy values ​​calculated by the gas molecule energy correction unit 22. However, when the catalytic reaction is a complex catalytic reaction consisting of many elementary reactions, such as an FT reaction, the number of parameters that must be optimized is enormous, making this unrealistic. Furthermore, the energies of the intermediate structure and the transition state structure on the catalyst surface are not independent but interrelated. Directly optimizing the energy values ​​is undesirable because it disrupts the interrelationship between the energies of the intermediate structure and the transition state structure on the catalyst surface, thereby losing their physical meaning. In the reaction rate prediction unit 20, optimizing the parameters of the regression model M1 using the optimization unit 27 reduces the number of parameters that need to be optimized and enables the search for parameters that reproduce experimental results in a realistic timeframe. Therefore, optimizing the parameters of the regression model M1 is preferable.

[0101] By changing the regression parameters of the regression model M1 so that the predicted values ​​of the reactants approach the actual measured values, the predicted values ​​can be fitted to the actual measured values.

[0102] FIG. 11 shows an example of the results of fitting the predicted values ​​of reactants with 1 to 5 carbon atoms (C1 to C5) to the measured values. In FIG. 11, predicted value 1 and measured value 1 represent the case where no trace metal is added to the catalyst, predicted value 2 and measured value 2 represent the case where a trace amount of Na is added to the catalyst as a promoter, and predicted value 3 and measured value 3 represent the case where a trace amount of Mg is added to the catalyst as a promoter. The reaction temperature was 320°C and the pressure was 0.8 MPa. As shown in FIG. 11, predicted values ​​1 to 3 of reactants C1 to C5 can be fitted to measured values ​​1 to 3 by changing the regression parameters of regression model M1 so that predicted values ​​1 to 3 approach measured values ​​1 to 3.

[0103] The optimization unit 27 repeatedly optimizes the energy of the intermediate structure and the transition state structure so that the absolute value of the difference between the characteristic value acquired by the reaction rate simulation and the predetermined value becomes equal to or less than the predetermined value.

[0104] [Output section] The output unit 30 outputs, by display or the like, the energy values ​​of the intermediate structure and transition state structure of the reactant or the energy values ​​of the optimized intermediate structure and transition state structure, the characteristic value, the difference in absolute value between the characteristic value and the reference value, the fact that the calculation has ended, etc. The output unit 30 may also output, by display or the like, information related to the regression model M1.

[0105] As described above, the reaction rate prediction device 1 includes an energy prediction device 10 and a reaction rate prediction unit 20, and the reaction rate prediction unit 20 includes a reaction rate simulation unit 23, a comparison unit 25, and an optimization unit 27. The reaction rate prediction unit 20 optimizes the energies of the intermediate structure and the transition state structure based on the reaction rate of the catalytic reaction simulated by the reaction rate simulation unit 23 so that the characteristic values ​​approach the reference values. This enables the reaction rate prediction device 1 to improve the accuracy of prediction of the energies of the intermediate structure and the transition state structure included in the elementary reactions of the catalytic reaction, thereby enabling highly accurate prediction of the reaction rate of the catalytic reaction.

[0106] In the reaction rate prediction device 1, the energy prediction device 10 preferably includes a regression model generation unit 11 as described above. As a result, the reaction rate prediction device 1 can use the energies of the intermediate structures and transition state structures included in each elementary reaction involved in the catalytic reaction, predicted using the regression model M1 generated by the regression model generation unit 11, in a simulation of the reaction rate of the catalytic reaction in the reaction rate prediction unit 20. By using the regression model M1, the reaction rate prediction device 1 can accurately predict the energies of the intermediate structures and transition state structures included in each elementary reaction involved in the catalytic reaction, and therefore can accurately simulate the reaction rate of the catalytic reaction in the reaction rate prediction unit 20. Therefore, the reaction rate prediction device 1 can predict the reaction rate of the catalytic reaction with high accuracy.

[0107] In the reaction rate prediction device 1, the optimization unit 27 preferably optimizes the energies of the intermediate structure and the transition state structure by changing the parameters of the regression model M1 so that the characteristic value approaches the reference value. Optimizing the regression parameters of the regression model M1 makes it easy to adjust the energies of the intermediate structure and the transition state structure of the reactant, making it easy to predict the energies of the intermediate structure and the transition state structure of the reactant. Therefore, by using the optimized regression model M1, the reaction rate prediction device 1 can improve the accuracy of the simulation of the reaction rate of the catalytic reaction and predict the reaction rate of the catalytic reaction with even higher accuracy.

[0108] In the reaction rate prediction device 1, the reaction rate prediction unit 20 preferably uses at least one of the reaction rate and the selectivity as the characteristic value. This allows the reaction rate prediction unit 20 to easily simulate the reaction rate of the catalytic reaction in the reaction rate simulation unit 23, making it easier to optimize the energies of the intermediate structure and the transition state structure. Therefore, the reaction rate prediction device 1 can improve the accuracy of the simulation of the reaction rate of the catalytic reaction and more easily and accurately predict the reaction rate of the catalytic reaction.

[0109] In the reaction rate prediction device 1, the comparison unit 25 preferably compares at least one of the reaction rate and the selectivity as the characteristic value and the reference value using the sum of squares error. This makes it easier to compare the characteristic value with the reference value, and the reaction rate prediction unit 20 can more appropriately determine whether the reaction rate of the catalytic reaction simulated by the reaction rate simulation unit 23 should be optimized by the optimization unit 27. Therefore, the reaction rate prediction device 1 can appropriately optimize the reaction rate of the catalytic reaction simulated by the reaction rate prediction unit 20, and can therefore appropriately simulate the reaction rate of the catalytic reaction, thereby more accurately predicting the reaction rate of the catalytic reaction.

[0110] In the reaction rate prediction device 1, it is preferable that the optimization unit 27 repeatedly optimizes the energies of the intermediate structure and the transition state structure so that the absolute value of the difference between the characteristic value and the reference value is equal to or less than a predetermined value. This allows the optimization unit 27 to optimize the energies of the intermediate structure and the transition state structure so that the characteristic value becomes closer to the reference value, thereby enabling the energies of the intermediate structure and the transition state structure to be adjusted with greater accuracy. This allows the reaction rate prediction device 1 to improve the prediction accuracy of the simulation of the reaction rate of the catalytic reaction, thereby enabling the reaction rate of the catalytic reaction to be predicted with even greater accuracy.

[0111] In the reaction rate prediction device 1, the reaction rate prediction unit 20 preferably uses at least one of random search and Bayesian optimization for optimization. This allows the optimization unit 27 to easily and appropriately optimize the energies of the intermediate structure and the transition state structure. Therefore, the reaction rate prediction device 1 can easily predict the simulation of the reaction rate of a catalytic reaction, and can easily predict the reaction rate of a catalytic reaction.

[0112] In the reaction rate prediction device 1, it is preferable that the regression model generation unit 11 acquires the energy using a trained model or a machine learning potential, as described above. This allows the reaction rate prediction device 1 to easily and accurately calculate the energies of the intermediate structure and the transition state structure from their descriptors in the regression model generation unit 11. This allows the reaction rate prediction device 1 to more easily generate the regression model M1.

[0113] Furthermore, in the reaction rate prediction device 1, it is preferable that the regression model generation unit 11 acquires the energy using machine learning potential as described above. This can speed up the calculation in the structural optimization calculation unit 113, thereby further shortening the time required for energy calculation. As a result, the reaction rate prediction device 1 can significantly reduce the time required to generate the regression model M1.

[0114] In the reaction rate prediction device 1, the reaction rate prediction unit 20 preferably includes a gas molecule energy correction unit 22. This allows the reaction rate prediction unit 20 to improve the accuracy of the simulation of the reaction rate of the catalytic reaction simulated by the reaction rate simulation unit 23. Therefore, the reaction rate prediction device 1 can predict the reaction rate of the catalytic reaction with even higher accuracy.

[0115] It is preferable that the reaction rate prediction device 1 uses an experimentally measured value as the reference value, which allows the reaction rate prediction unit 20 to easily compare the characteristic value with the reference value, and therefore the reaction rate prediction device 1 can easily predict the reaction rate of the catalytic reaction.

[0116] As described above, the energy prediction device 10 can calculate with high accuracy the energies of the intermediate structures and transition state structures contained in each elementary reaction involved in a catalytic reaction, and therefore can be effectively used to calculate the energies of the intermediate structures and transition state structures contained in each elementary reaction, such as an FT reaction, which has many reaction paths and a complex reaction mechanism, among catalytic reactions.

[0117] As described above, the reaction rate prediction device 1 can predict the reaction rate of a catalytic reaction with high accuracy, and therefore can be effectively used to calculate the reaction rate of an FT reaction or the like.

[0118] <Hardware configuration of reaction rate prediction device and energy prediction device> Next, an example of the hardware configuration of the reaction rate prediction device 1 and the energy prediction device 10 will be described. FIG. 12 is a block diagram showing the hardware configuration of the reaction rate prediction device 1 and the energy prediction device 10. As shown in FIG. 12, the reaction rate prediction device 1 and the energy prediction device 10 are configured as an information processing device (computer), and can be physically configured as a computer system including a CPU (Central Processing Unit: processor) 101, which is an arithmetic processing unit, a RAM (Random Access Memory) 102 and a ROM (Read Only Memory) 103, which are main storage devices, an input device 104, which is an input device, an output device 105, a communication module 106, and an auxiliary storage device 107 such as a hard disk. These are connected to each other by a bus 108. Note that the output device 105 and the auxiliary storage device 107 may be provided externally.

[0119] The CPU 101 controls the overall operation of the reaction rate prediction device 1 and the energy prediction device 10, and performs various information processing. The CPU 101 can execute, for example, a catalytic reaction reaction rate prediction method (hereinafter simply referred to as the "reaction rate prediction method"), a catalytic reaction energy prediction method (hereinafter simply referred to as the "energy prediction method"), a catalytic reaction reaction rate prediction program (hereinafter simply referred to as the "reaction rate prediction program"), and a catalytic reaction energy prediction program (hereinafter simply referred to as the "energy prediction program"), all of which are stored in the ROM 103 or the auxiliary storage device 107, to predict the reaction rate of the catalytic reaction and the energies of intermediate structures and transition state structures included in each elementary reaction.

[0120] The RAM 102 is used as a work area for the CPU 101 and may include a non-volatile RAM for storing main control parameters and information.

[0121] The ROM 103 stores a basic input / output program, etc. The reaction rate prediction program and the energy prediction program may be stored in the ROM 103.

[0122] The input device 104 is an input device such as a keyboard, a mouse, operation buttons, a touch panel, and a display screen, and receives information input by a user as an instruction signal and outputs the instruction signal to the CPU 101 .

[0123] The output device 105 is a display device such as a monitor display, an audio device such as a speaker, a printing device such as a printer, etc. In the output device 105, for example, information such as the catalyst selection result is displayed on a display device such as a monitor display, and the displayed screen is updated in response to input operations via the input device 104 or the communication module 106.

[0124] The communication module 106 is a data transmission / reception device such as a network card, and functions as a communication interface that receives information from an external data recording server or the like and outputs analysis information to other electronic devices.

[0125] The auxiliary storage device 107 is a storage device such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive), and stores, for example, various data and files required for the operation of the reaction rate prediction device 1.

[0126] Each function of the reaction rate prediction device 1 is realized by reading predetermined computer software (including a reaction rate prediction program and an energy prediction program) from a main memory device such as RAM 102 or an auxiliary memory device 107 and executing it with the CPU 101, thereby reading and writing data in a main memory device such as RAM 102 or an auxiliary memory device 107, and operating the input device 104, the output device 105, and the communication module 106.

[0127] Therefore, each part of the reaction rate prediction device 1 shown in Figure 2 is realized by software and hardware working together in a computer equipped with the reaction rate prediction device 1, where a processor executes predetermined computer software (including a reaction rate prediction program and an energy prediction program) that is pre-stored.

[0128] The reaction rate prediction program and the energy prediction program can be stored, for example, in a main storage device or an auxiliary storage device 107 of a computer. Alternatively, the reaction rate prediction program and the energy prediction program may be stored on a computer connected to a communication line such as the Internet, and a part or all of the reaction rate prediction program and the energy prediction program may be provided by being downloaded via the communication line. Furthermore, the reaction rate prediction program and the energy prediction program may be configured to be provided or distributed via the communication line.

[0129] The reaction rate prediction program and the energy prediction program may be recorded (including installed) into a computer from a state in which part or all of them are stored on a portable storage medium such as an optical disk such as a CD-ROM or DVD-ROM, or a semiconductor memory such as a flash memory.

[0130] <Method for predicting the reaction rate of a catalytic reaction> The reaction rate prediction method according to this embodiment will be described. The reaction rate prediction method according to this embodiment can be performed using the above-mentioned reaction rate prediction device 1, and the energy prediction method can be performed using the above-mentioned energy prediction device 10. Therefore, some of the content already explained will be omitted.

[0131] Fig. 13 is a flowchart showing a reaction rate prediction method according to this embodiment. As shown in Fig. 13, in the reaction rate prediction method, the energy prediction device 10 predicts the energy of a catalytic reaction in which a product is produced from a reactant by a catalytic reaction in which a plurality of elementary reactions that proceed stepwise are repeated (energy prediction step: step S10).

[0132] In the energy prediction step (step S10), as shown in FIG. 14, the regression model generation unit 11 generates a regression model M1 that predicts the energy of the intermediate structure and the transition state structure based on the intermediate structure and the transition state structure of the reactant that are generated in the process of generating a product from a reactant by a catalytic reaction and the energies of the intermediate structure and the transition state structure (regression model generation step: step S11).

[0133] In the regression model generation step (step S11), the first acquisition unit 111 acquires information on a plurality of elementary reactions related to a target catalytic reaction or the like (acquisition step: step S111).

[0134] Next, the enumeration unit 112 enumerates the plurality of elementary reactions acquired in the acquisition step (step S111) (enumeration step: step S112).

[0135] Next, the structure optimization calculation unit 113 calculates the energies of the intermediate structures and transition state structures involved in the plurality of elementary reactions (energy calculation step: step S113).

[0136] Next, the descriptor generating unit 114 generates, based on the intermediate structure and the transition state structure, descriptors representing the intermediate structure and the transition state structure as structural descriptors (descriptor generating step: step S114).

[0137] Next, the model generation unit 115 generates a regression model M1 using the structural descriptors generated in the descriptor generation step (step S114) as explanatory variables and the energy obtained in the energy calculation step (step S113) as a response variable (model generation step: step S115).

[0138] Next, the energy prediction unit 12 acquires the descriptors of the intermediate structure and the descriptors of the transition state structure generated in the descriptor generation step (step S114), and predicts the energies of the intermediate structure and the transition state structure from the descriptors of the intermediate structure and the descriptors of the transition state structure generated in the descriptor generation step (step S114) using the regression model M1 generated in the model generation step (step S115) (energy prediction step: step S12).

[0139] The energy prediction method generates a regression model M1 that predicts the energies of intermediate structures and transition state structures of reactants in a regression model generation step (step S11), and then predicts the energies of the intermediate structures and transition state structures of the reactants using the generated regression model M1 in an energy prediction step (step S12).The energy prediction method makes it possible to predict the energies of intermediate structures and transition state structures of reactants with high accuracy, and therefore makes it possible to calculate the energies of intermediate structures and transition state structures included in each elementary reaction involved in a catalytic reaction with high accuracy.

[0140] Next, as shown in FIG. 13, the reaction rate prediction unit 20 predicts the reaction rate of the catalytic reaction (reaction rate prediction step: step S20).

[0141] In the reaction rate prediction step (step S20), as shown in FIG. 15, the second acquisition unit 21 acquires the energies of the intermediate structure and transition state structure of the reactant predicted in the energy prediction step (step S12) (reaction rate prediction acquisition step: step S21).

[0142] Next, the gas molecule energy correction unit 22 corrects the energy of the gas molecule structure calculated in the energy calculation step (step S113), which exists among the reactants without reacting with the catalyst (gas molecule energy correction step: step S22).

[0143] Next, the reaction rate simulation unit 23 simulates characteristic values ​​related to the reaction rate of the catalytic reaction based on the energy of the intermediate structure and transition state structure of the reactant that are generated in the process of generating a product from the reactant through the catalytic reaction (reaction rate simulation process: step S23).

[0144] Next, the setting unit 24 sets a reference value (setting step: step S24).

[0145] Next, the comparison unit 25 compares the characteristic value acquired by simulating the reaction rate in the reaction rate simulation step (step S23) with the reference value set in the setting step (step S24) (comparison step: step S25).

[0146] Next, the determination unit 26 determines whether the difference in absolute value between the characteristic value and the reference value is equal to or less than a predetermined value (determination step: step S26).

[0147] If the difference in absolute value between the characteristic value and the reference value is less than or equal to a predetermined value (step S26: Yes), the judgment unit 26 determines that the characteristic value predicted by the regression model M1 is approximately the same as or close to the known value, and that there is no need to change the energies of the intermediate structure and transition state structure predicted by the regression model M1.

[0148] Next, as shown in FIG. 13, the output unit 30 outputs, by display or the like, the energy values ​​of the intermediate structure and transition state structure of the reactant, the characteristic values, the difference in absolute value between the characteristic values ​​and the reference values, and the fact that the calculation has been completed (output process: step S30).

[0149] On the other hand, as shown in FIG. 15, in the judgment process (step S26), if the difference in absolute value between the characteristic value and the reference value exceeds a predetermined value (step S26: No), the judgment unit 26 determines that the characteristic value predicted by the regression model M1 differs greatly from the known value, and that the energies of the intermediate structure and transition state structure predicted by the regression model M1 need to be changed.

[0150] Next, the optimization unit 27 changes the index correlated with energy based on the comparison result in the judgment step (step S26) so that the characteristic value predicted in the simulation step (step S23) approaches the reference value, and optimizes the energy of the intermediate structure and transition state structure predicted in the energy prediction step (step S12) (optimization step: step S28).

[0151] In the optimization step (step S27), the optimization unit 27 optimizes the regression parameters of the regression model M1, and then the process proceeds to the energy prediction step (step S12). Then, the energy prediction unit 12 performs the same steps as above using the energies of the optimized intermediate structure and transition state structure.

[0152] Then, in the determination step (step S26), the optimization of the energies of the intermediate structure and the transition state structure is repeatedly performed in the optimization step (step S27) until it is determined that the difference in absolute value between the characteristic value and the reference value is equal to or less than a predetermined value.

[0153] Finally, after it is determined in the determination step (step S26) that the difference in absolute value between the characteristic value and the reference value is equal to or less than a predetermined value, as described above, the output unit 30 outputs the energy values ​​of the intermediate structure and transition state structure of the reactant optimized in the optimization step (step S27) (output step: step S30).

[0154] The reaction rate prediction method according to this embodiment optimizes the energies of intermediate structures and transition state structures one or more times based on the results of the catalytic reaction rate simulated in the reaction rate prediction step (step S20) so that a characteristic value related to the reaction rate of the catalytic reaction approaches a set reference value. As a result, the reaction rate prediction method according to this embodiment can predict the reaction rate of the catalytic reaction with high accuracy by improving the prediction accuracy of the energies of intermediate structures and transition state structures included in each elementary reaction of the predicted catalytic reaction.

[0155] Although the embodiments have been described above, they are presented as examples and the present invention is not limited to the above embodiments. The above embodiments can be implemented in various other forms, and various combinations, omissions, substitutions, modifications, etc. can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents as set forth in the claims.

[0156] The embodiments of the present invention are as follows, for example. <1> a simulation unit that simulates a characteristic value related to the reaction rate of the catalytic reaction based on the energy of three-dimensional structures of intermediates and transition states of reactants that are generated in the process of generating products from reactants through a catalytic reaction in which a plurality of elementary reactions that proceed stepwise are repeated; and a comparison unit that compares the characteristic value with a set reference value; an optimization unit that optimizes the energy by changing an index correlated with the energy based on a result of the comparison so that the characteristic value approaches the reference value; A catalytic reaction rate prediction device comprising: <2> a regression model generating unit configured to generate a regression model that predicts the energy of the three-dimensional structure based on three-dimensional structures of the intermediate and transition state of the reactant and the energy of the three-dimensional structure; <1> The reaction rate prediction device for the catalytic reaction described in <3> the optimization unit changes parameters of the regression model so that the characteristic value approaches the reference value, thereby optimizing the energy of the three-dimensional structure of the intermediate and the transition state. <2> The reaction rate prediction device for the catalytic reaction described in <4> the characteristic value is at least one of the reaction rate and the selectivity; <1> ~ <3> 10. The reaction rate prediction device for a catalytic reaction according to claim 9. <5> the comparison unit compares at least one of the reaction rate and the selectivity as the characteristic value and the reference value by calculating a square sum error; <4> The reaction rate prediction device for the catalytic reaction described in <6> the optimization unit iteratively optimizes the energies of the intermediate and the transition state so that an absolute value of a difference between the characteristic value and the reference value is equal to or less than a predetermined value. <1> ~ <5> 10. The reaction rate prediction device for a catalytic reaction according to claim 9. <7> the optimization is at least one of a random search and a Bayesian optimization; <1> ~ <6> 10. The reaction rate prediction device for a catalytic reaction according to claim 9. <8> the regression model generation unit obtains the energies of the three-dimensional structures of the intermediate and the transition state of the reactant from descriptors of the three-dimensional structures of the intermediate and the transition state of the reactant using a trained model or a machine learning potential that predicts the energies of the three-dimensional structures of the intermediate and the transition state of the reactant. <2> or <3> The reaction rate prediction device for the catalytic reaction described in <9> an energy correction unit that corrects the energy of the three-dimensional structure of gas molecules that exist among the reactants without reacting with the catalyst; <1> ~ <8> 10. The reaction rate prediction device for a catalytic reaction according to claim 9. <10> The reference value is an actual measured value obtained by an experiment. <1> ~ <9> 10. The reaction rate prediction device for a catalytic reaction according to claim 9. <11> The computer a simulation step of simulating a characteristic value related to the reaction rate of the catalytic reaction based on the energy of three-dimensional structures of intermediates and transition states of reactants that are generated in the process of generating products from reactants through a catalytic reaction in which a plurality of elementary reactions that proceed stepwise are repeated; a comparison step of comparing the characteristic value with a set reference value; an optimization step of optimizing the energy by changing an index correlated with the energy based on a result of the comparison so that the characteristic value approaches the reference value; A method for predicting the reaction rate of a catalytic reaction. <12> On the computer, a simulation step of simulating a characteristic value related to the reaction rate of the catalytic reaction based on the energy of three-dimensional structures of intermediates and transition states of reactants that are generated in the process of generating products from reactants through a catalytic reaction in which a plurality of elementary reactions that proceed stepwise are repeated; a comparison step of comparing the characteristic value with a set reference value; an optimization step of optimizing the energy by changing an index correlated with the energy based on a result of the comparison so that the characteristic value approaches the reference value; A program for predicting the reaction rate of catalytic reactions. [Explanation of symbols]

[0157] 1. Catalytic reaction rate prediction device 10. Energy prediction device for catalytic reactions 11 Regression model generation unit 12 Energy Prediction Department 20 Reaction rate prediction section 21 Second Acquisition Section 22 Energy correction section for gas molecules 23 Reaction Rate Simulation Section 24 Setting section 25 Comparison section 26 Judgment section 27 Optimization Section 30 Output section 111 First Acquisition Section 112 Enumeration 113 Structural Optimization Calculation Unit 114 Descriptor Generator 115 Model Generation Unit M1 regression model

Claims

1. a simulation unit that simulates a characteristic value related to the reaction rate of the catalytic reaction based on the energy of three-dimensional structures of intermediates and transition states of reactants that are generated in the process of generating products from reactants through a catalytic reaction in which a plurality of elementary reactions that proceed stepwise are repeated; and a comparison unit that compares the characteristic value with a set reference value; an optimization unit that optimizes the energy by changing an index correlated with the energy based on a result of the comparison so that the characteristic value approaches the reference value; A catalytic reaction rate prediction device comprising:

2. 2. The reaction rate prediction device for a catalytic reaction according to claim 1, further comprising a regression model generation unit that generates a regression model that predicts the energy of the three-dimensional structure based on the three-dimensional structures of the intermediates and transition states of the reactants and the energies of the three-dimensional structures.

3. 3. The reaction rate prediction device for a catalytic reaction according to claim 2, wherein the optimization unit changes parameters of the regression model so that the characteristic value approaches the reference value, thereby optimizing the energy of the three-dimensional structure of the intermediate and the transition state.

4. 2. The apparatus for predicting the reaction rate of a catalytic reaction according to claim 1, wherein the characteristic value is at least one of the reaction rate and the selectivity.

5. 5. The apparatus for predicting the reaction rate of a catalytic reaction according to claim 4, wherein the comparison unit compares at least one of the reaction rate and the selectivity as the characteristic value and the reference value by calculating a square sum error.

6. 3. The reaction rate prediction device for a catalytic reaction according to claim 1, wherein the optimization unit repeatedly optimizes the energies of the intermediate and the transition state so that the absolute value of the difference between the characteristic value and the reference value is equal to or less than a predetermined value.

7. 3. The apparatus for predicting a reaction rate of a catalytic reaction according to claim 1, wherein the optimization is at least one of random search and Bayesian optimization.

8. 4. The reaction rate prediction device for a catalytic reaction according to claim 2, wherein the regression model generation unit obtains the energy of the three-dimensional structure of the intermediate and the transition state of the reactant from descriptors of the three-dimensional structure of the intermediate and the transition state of the reactant using a trained model or machine learning potential that predicts the energy of the three-dimensional structure of the intermediate and the transition state of the reactant.

9. 3. The reaction rate prediction device for a catalytic reaction according to claim 1, further comprising an energy correction unit that corrects the energy of the three-dimensional structure of gas molecules that exist among the reactants without reacting with the catalyst.

10. 3. The apparatus for predicting a reaction rate of a catalytic reaction according to claim 1, wherein the reference value is an actual measurement value obtained through an experiment.

11. The computer a simulation step of simulating a characteristic value related to the reaction rate of the catalytic reaction based on the energy of three-dimensional structures of intermediates and transition states of reactants that are generated in the process of generating products from reactants through a catalytic reaction in which a plurality of elementary reactions that proceed stepwise are repeated; a comparison step of comparing the characteristic value with a set reference value; an optimization step of optimizing the energy by changing an index correlated with the energy based on a result of the comparison so that the characteristic value approaches the reference value; A method for predicting the reaction rate of a catalytic reaction.

12. On the computer, a simulation step of simulating a characteristic value related to the reaction rate of the catalytic reaction based on the energy of three-dimensional structures of intermediates and transition states of reactants that are generated in the process of generating products from reactants through a catalytic reaction in which a plurality of elementary reactions that proceed stepwise are repeated; a comparison step of comparing the characteristic value with a set reference value; an optimization step of optimizing the energy by changing an index correlated with the energy based on a result of the comparison so that the characteristic value approaches the reference value; A program for predicting the reaction rate of catalytic reactions.

Images