JP7487155B2 - Information processing device, program, and information processing method - Google Patents

Description

The present invention relates to an information processing device, a program, and an information processing method.

For example, Patent Literature 1 discloses that a user selects a compound space from a database that collects material data related to the analysis objective, and generates effective material features that are more suited to the analysis objective by learning an autoencoder. Patent Literature 2 discloses predicting the presence or absence of an effect of enhancing electron transfer between an enzyme and an electrode by nanocarbon using a prediction model expressed as a function of a descriptor related to the effect. Non-Patent Literatures 1 to 3 disclose the use of materials informatics to search for substances or processes.
[Prior Art Literature]
[Patent Documents]
[Patent Document 1] JP 2021-39534 A [Patent Document 2] JP 2021-40633 A [Non-Patent Document]
[Non-Patent Document 1] G. Nakada, Y. Igarashi, H. Imai, Y. Oaki, Adv. Theory Simul. 2019, 2, 1800180.
[Non-Patent Document 2] H. Numazawa, Y. Igarashi, K. Sato, H. Imai, Y. Oaki, Adv. Theory Simul. 2019, 2, 1900130.
[Non-Patent Document 3] R. Mizuguchi, Y. Igarashi, H. Imai, Y. Oaki, Nanoscale 2021, 13, 3853.

According to a first aspect of the present invention, an information processing device is provided. The information processing device predicts the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery, for example. The information processing device includes, for example, a material determination unit that determines a target material, which is an electrode material to be predicted. The information processing device includes, for example, a characteristic determination unit that determines a target characteristic, which is a characteristic to be predicted. For example, when the target characteristic is a reaction potential of a positive electrode when the target material is used as a positive electrode active material of a secondary battery, the information processing device includes an input value acquisition unit that acquires (i) an energy value of the LUMO level of a molecule of the target material, and (ii) an energy value of a level one level higher than the LUMO level of the molecule of the target material. The information processing device includes, for example, a prediction unit that inputs the value acquired by the input value acquisition unit into a prediction model and outputs a predicted value of the target characteristic of the target material. In the above information processing device, the prediction model is, for example, a regression model in which the objective variable is the potential of the positive electrode when the material is used as a positive electrode active material of a secondary battery, and the descriptors include (a) the energy of the LUMO level of the material's molecules and (b) the energy of the level one level higher than the LUMO level of the material's molecules.

According to a second aspect of the present invention, an information processing device is provided. The information processing device predicts the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery, for example. The information processing device includes, for example, a material determination unit that determines a target material, which is an electrode material to be predicted. The information processing device includes, for example, a characteristic determination unit that determines a target characteristic, which is a characteristic to be predicted. When the target characteristic is, for example, a capacity when the target material is used as a positive electrode active material of a secondary battery, the information processing device includes an input value acquisition unit that acquires (iii) a value per molecule of the number of reactions of the target material and the metal ions of the secondary battery derived from the molecular structure of the target material, and (iv) a value per molecule of the number of orbitals of the molecules of the target material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery. The information processing device includes, for example, a prediction unit that inputs the value acquired by the input value acquisition unit into a prediction model and outputs a predicted value of the target characteristic of the target material. In the above information processing device, the prediction model is a regression model that uses, for example, the positive electrode capacity when the material is used as a positive electrode active material of a secondary battery as the objective variable, (c) the number of reactions between the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule, as a descriptor.

According to a third aspect of the present invention, an information processing device is provided. The information processing device predicts the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery, for example. The information processing device includes, for example, a material determination unit that determines a target material, which is an electrode material to be predicted. The information processing device includes, for example, a characteristic determination unit that determines a target characteristic, which is a characteristic to be predicted. When the target characteristic is, for example, an energy density when the target material is used as a positive electrode active material of a secondary battery, the information processing device includes an input value acquisition unit that acquires (i) an energy value of a LUMO level of a molecule of the target material, (ii) an energy value of a level one level higher than the LUMO level of the molecule of the target material, (iii) a value per molecule of the number of reactions of the target material and the metal ion of the secondary battery derived from the molecular structure of the target material, and (iv) a value per molecule of the number of orbitals of the molecule of the target material having an energy level equal to or lower than the work function of the metal ion of the secondary battery. The information processing device includes, for example, a prediction unit that inputs the value acquired by the input value acquisition unit into a prediction model and outputs a predicted value of the target characteristic of the target material. In the above information processing device, the prediction model is a regression model that uses, as an objective variable, the energy density when the material is used as a positive electrode active material of a secondary battery, and includes, as a descriptor, (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule.

In the information processing device according to the first to third aspects described above (which may be simply referred to as the information processing device described above), the regression model may be represented by a linear regression equation or a nonlinear regression equation.

The information processing device may include a model construction unit that constructs a predictive model by machine learning. In the information processing device, the model construction unit may have a first variable selection unit that selects M explanatory variables (M is a positive integer between 1 and N) from N explanatory variables (N is a positive integer between 2 and N) each of which represents a characteristic of a material, as descriptors or candidates for the descriptors of a regression model. In the information processing device, the first variable selection unit may select the M explanatory variables by a sparse modeling method. In the information processing device, the first variable selection unit may determine regression coefficients or partial regression coefficients corresponding to each of the M explanatory variables.

In the above information processing device, the material may be an organic molecule or a salt thereof. In the above information processing device, the N explanatory variables may include (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. In the above information processing device, the N explanatory variables may further include at least one selected from the group consisting of (e) the dispersion term of the Hansen solubility parameters of the material and the electrolyte of the secondary battery, (f) the polarization term of the Hansen solubility parameters, (g) the hydrogen bond term of the Hansen solubility parameters, (h) the dipole moment of the molecules of the material, (i) the minimum value of the partial charge density of the molecules of the material, and (j) the number of elements other than carbon, hydrogen, and oxygen contained in the molecules of the material, which is the number of elements per molecule.

In the above information processing device, the N explanatory variables may include (k) the energy of the level two above the LUMO level of the molecules of the material, (l) the energy of the level three above the LUMO level of the molecules of the material, (m) the energy of the level four above the LUMO level of the molecules of the material, (n) the molecular weight of the molecules of the material, (o) the number of carbon atoms conjugated by π bonds contained in the molecules of the material, which is the number of carbon atoms per molecule, (p) the number of orbitals of the molecules of the material having an energy level of 0 or less, which is the number of orbitals per molecule, (q) the absolute value of the sum of the energy levels of 0 or less per molecule of the material, and (r) the maximum value of the partial charge density of the molecules of the material.

In the above information processing device, the material may be an organic molecule or a salt thereof. In the above information processing device, the M explanatory variables may include at least two selected from (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule.

In the above information processing device, the model construction unit may have a second variable selection unit that selects K explanatory variables (K is a positive integer between 1 and M) as descriptors from among M explanatory variables based on a user's instruction. In the above information processing device, the model construction unit may have a third variable selection unit that selects N explanatory variables as candidates for descriptors from among L explanatory variables (L is a positive integer between N and N), each of which represents a characteristic of a material. In the above information processing device, the third variable selection unit may determine the degree of correlation with other explanatory variables for each of the L explanatory variables. The third variable selection unit may select N explanatory variables from among the L explanatory variables by leaving one explanatory variable out of multiple explanatory variables whose degree of correlation is stronger than a predetermined standard and excluding the other explanatory variables.

According to a fourth aspect of the present invention, an information processing device is provided. The information processing device constructs a prediction model for predicting the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery, for example. In the information processing device, the prediction model is, for example, a regression model in which the characteristics of the secondary battery are set as a response variable and one or more characteristics related to the electrode material are each a descriptor. The information processing device includes, for example, a first variable selection unit that selects M explanatory variables (M is a positive integer between 1 and N) as descriptors or candidates for the regression model from N explanatory variables (N is a positive integer between 2 and 1) each of which represents a characteristic of a material. In the information processing device, the first variable selection unit receives, for example, a group of teacher data in which, for each of a plurality of materials, a value of one characteristic related to a secondary battery when the material is used as an electrode material is associated with values of N explanatory variables related to the characteristics of the material. The first variable selection unit selects M explanatory variables from the N explanatory variables by, for example, executing a regression analysis based on machine learning.

In the above information processing device, the N explanatory variables include, for example, (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. In the above information processing device, the N explanatory variables include, for example, at least one selected from the group consisting of (e) the dispersion term of the Hansen solubility parameters of the material and the electrolyte of the secondary battery, (f) the polarization term of the Hansen solubility parameters, (g) the hydrogen bond term of the Hansen solubility parameters, (h) the dipole moment of the molecules of the material, (i) the minimum value of the partial charge density of the molecules of the material, and (j) the number of elements other than carbon, hydrogen, and oxygen contained in the molecules of the material, which is the number of elements per molecule.

According to a fifth aspect of the present invention, an information processing method is provided. The information processing method is, for example, a method for predicting the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery. The information processing method has, for example, a material determination step in which a computer determines a target material, which is an electrode material to be predicted. The information processing method has, for example, a characteristic determination step in which a computer determines a target characteristic, which is a characteristic to be predicted. For example, when the target characteristic is the potential of the positive electrode when the target material is used as a positive electrode active material of a secondary battery, the information processing method has an input value acquisition step in which the computer acquires (i) the energy value of the LUMO level of the molecule of the target material, and (ii) the energy value of the level one level higher than the LUMO level of the molecule of the target material. The information processing method has, for example, a prediction step in which the computer inputs the values acquired in the input value acquisition step into a prediction model and outputs a predicted value of the target characteristic of the target material. In the above information processing method, the prediction model is, for example, a regression model in which the objective variable is the potential of the positive electrode when the material is used as a positive electrode active material of a secondary battery, and the descriptors include (a) the energy of the LUMO level of the material's molecules and (b) the energy of the level one level higher than the LUMO level of the material's molecules.

According to a sixth aspect of the present invention, an information processing method is provided. The information processing method is, for example, a method for predicting the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery. The information processing method has, for example, a material determination step in which a computer determines a target material, which is an electrode material to be predicted. The information processing method has, for example, a characteristic determination step in which a computer determines a target characteristic, which is a characteristic to be predicted. For example, when the target characteristic is a positive electrode capacity when the target material is used as a positive electrode active material of a secondary battery, the information processing method has an input value acquisition step in which the computer acquires (iii) a value per molecule of the number of reactions of the target material and the metal ions of the secondary battery derived from the molecular structure of the target material, and (iv) a value per molecule of the number of orbitals of the molecules of the target material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery. The information processing method has, for example, a prediction step in which the computer inputs the values acquired in the input value acquisition step into a prediction model and outputs a predicted value of the target characteristic of the target material. In the above information processing method, the prediction model is, for example, a regression model in which the objective variable is the positive electrode capacity when the material is used as a positive electrode active material of a secondary battery, (c) the number of reactions between the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule, are included as descriptors.

According to a seventh aspect of the present invention, an information processing method is provided. The information processing method is, for example, a method for predicting the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery. The information processing method has, for example, a material determination step in which a computer determines a target material, which is an electrode material to be predicted. The information processing method has, for example, a characteristic determination step in which a computer determines a target characteristic, which is a characteristic to be predicted. For example, when the target characteristic is an energy density when the target material is used as a positive electrode active material of a secondary battery, the information processing method has an input value acquisition step in which the computer acquires (i) the energy value of the LUMO level of the molecule of the target material, (ii) the energy value of the level one level above the LUMO level of the molecule of the target material, (iii) the value per molecule of the number of reactions of the target material and the metal ion of the secondary battery derived from the molecular structure of the target material, and (iv) the value per molecule of the number of orbitals having an energy level equal to or lower than the work function of the metal ion of the secondary battery among the orbitals of the molecule of the target material. The above information processing method includes, for example, a prediction step in which a computer inputs the values acquired in the input value acquisition step into a prediction model and outputs a predicted value of the target property of the target material. In the above information processing method, the prediction model is, for example, a regression model in which the energy density when the material is used as a positive electrode active material of a secondary battery is used as an objective variable, and includes (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule, as a descriptor.

According to an eighth aspect of the present invention, an information processing method is provided. The information processing method is, for example, a method for constructing a prediction model for predicting the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery. In the information processing method, the prediction model is, for example, a regression model in which the characteristics of the secondary battery are set as the objective variable and one or more characteristics related to the electrode material are each set as a descriptor. The information processing method has, for example, a first variable selection step in which a computer selects M explanatory variables (M is a positive integer between 1 and N) from N explanatory variables (N is a positive integer between 2 and 1) each representing a characteristic of a material as a descriptor or a candidate for a descriptor of the regression model. In the information processing method, the first variable selection step includes, for example, a step in which a computer acquires a group of teacher data in which, for each of a plurality of materials, a value of one characteristic related to a secondary battery when the material is used as an electrode material is associated with values of N explanatory variables related to the characteristics of the material. The first variable selection step includes, for example, a step in which a computer selects M explanatory variables from N explanatory variables by performing a regression analysis based on machine learning.

In the above information processing method, the N explanatory variables include, for example, (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. In the above information processing method, the N explanatory variables include, for example, at least one selected from the group consisting of (e) the dispersion term of the Hansen solubility parameters of the material and the electrolyte of the secondary battery, (f) the polarization term of the Hansen solubility parameters, (g) the hydrogen bond term of the Hansen solubility parameters, (h) the dipole moment of the molecules of the material, (i) the minimum value of the partial charge density of the molecules of the material, and (j) the number of elements other than carbon, hydrogen, and oxygen contained in the molecules of the material, which is the number of elements per molecule.

In a ninth aspect of the present invention, a program is provided. A computer-readable medium storing the above program may be provided. The computer-readable medium may be a non-transitory computer-readable medium or a computer-readable recording medium. The above program may be a program for causing a computer to function as an information processing device according to the first to fourth aspects above. The above program may be a program for causing a computer to execute an information processing method according to the fifth to eighth aspects above.

Note that the above summary of the invention does not list all of the necessary features of the present invention. Also, subcombinations of these features may also be inventions.

1 shows a schematic diagram of an example of a system configuration of a materials exploration system 100. 2 shows a schematic diagram of an example of the internal structure of the storage section 118. 1 illustrates an example of information processing in a database construction unit 120. 13 illustrates an example of information processing in a data set generating unit 142. 13 illustrates an example of a process for generating a reconstruction data set 460. 13 illustrates an example of information processing in the prediction model construction unit 140. 13 illustrates an outline of another example of information processing in the prediction model construction unit 140. Schematically illustrating an example of information processing in the characteristic prediction unit 160. 1 shows an outline of an example of an information processing method in the material exploration system 100. 3 shows an example of a system configuration of a computer 3000. 1 shows the relationship between the predicted and measured average reaction potentials in Example 1. 1 shows the relationship between predicted and actually measured values of the positive electrode capacity per unit mass in Example 1. 1 shows the relationship between predicted and measured mass energy density values in Example 1. 1 shows the RMSE of the average reaction potentials of Examples 1 to 6. The RMSE of the positive electrode capacity per unit mass of Examples 1 to 6 is shown. RMSE of mass energy density in Examples 1 to 6 is shown.

In recent years, there has been a demand for further improvements in the battery performance of storage batteries. For example, the potential, capacity, and energy density of a storage battery are highly dependent on the electrode material (e.g., the positive electrode material). For example, when developing a storage battery with a mass energy density of more than 400 to 500 [Wh/kg-storage battery], it is desirable to develop a positive electrode material that greatly exceeds the positive electrode capacity of existing positive electrode materials.

In addition, since organic materials are lightweight and relatively inexpensive, it is expected that the use of organic materials as positive electrode materials will lead to the production of inexpensive storage batteries with high mass energy density. However, it is said that there are more than 10 ⁶⁰ types of organic materials that can be used as positive electrode materials. It is not easy to find new materials that can be used as electrode materials from such a wide variety of organic materials.

In this regard, materials informatics (sometimes referred to as MI), which uses data science techniques to explore materials, processes, and the like, has been attracting attention in recent years. In general, MI requires large amounts of data for learning. However, when applying MI to the search for new electrode materials as described above, it is difficult to prepare large amounts of data.

According to one embodiment of the present technology, by utilizing MI that appropriately combines machine learning with the experience and consideration of a researcher, a prediction model for predicting the characteristics of a secondary battery based on the characteristics of the electrode material of the secondary battery is constructed using relatively small-scale data. According to this embodiment, for example, as the characteristics of the secondary battery, the teacher data used in the machine learning for constructing a prediction model for predicting the potential of the positive electrode, the capacity of the secondary battery, or the energy density of the secondary battery includes a new explanatory variable. This allows a new prediction model to be constructed. Also, according to this embodiment, for example, an explanatory variable to be used as a descriptor of the prediction model is selected from explanatory variables representing multiple features listed based on the experience and consideration of a researcher by a sparse modeling method. This allows a prediction model with good prediction accuracy to be constructed using relatively small-scale data.

According to another embodiment of the present technology, a prediction model is used to search for candidate materials for the electrode material of a secondary battery. Specifically, a regression model is used to search for candidate materials for the positive electrode active material of a secondary battery, with the characteristics of the secondary battery as the objective variable and with one or more features related to the electrode material as the descriptor. For example, the above regression model is used to search for candidate materials for the positive electrode active material of a secondary battery.

The present invention will be described below through embodiments of the invention, but the following embodiments do not limit the invention according to the claims. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[Outline of the material exploration system 100]
1 is a schematic diagram showing an example of a system configuration of a material exploration system 100. In this embodiment, the details of the material exploration system 100 are described using an example in which the material exploration system 100 predicts the characteristics of a secondary battery based on the characteristics of the electrode material of the secondary battery. In this embodiment, the details of the material exploration system 100 are described using an example in which the electrode material of the secondary battery is a material used as a positive electrode active material.

In this embodiment, the material exploration system 100 constructs a prediction model by machine learning using data representing the characteristics of the secondary battery and data representing the characteristics of the electrode material of the secondary battery as training data. For example, the material exploration system 100 constructs a regression model by regression analysis based on machine learning using the training data. The material exploration system 100 may construct a prediction model for each of one or more characteristics.

In this embodiment, the material exploration system 100 acquires input data representing one or more characteristics of the material to be predicted (sometimes referred to as the target material). The material exploration system 100 also acquires information indicating the characteristics of the secondary battery to be predicted (sometimes referred to as the target characteristics). The material exploration system 100 inputs the above input data into a prediction model for predicting the target characteristics, and outputs a predicted value of the target characteristics when the target material is used as an electrode material for a secondary battery.

The type of secondary battery is not particularly limited, but examples of the types of secondary batteries include lithium ion batteries, sodium ion batteries, and potassium ion batteries. The electrode material may be an inorganic material, an organic material, or a combination of an inorganic material and an organic material. When an organic material is used as the electrode material, a high-potential and/or high-capacity electrode can be produced inexpensively. Examples of organic materials include organic molecules or salts thereof.

The material to be searched for is not particularly limited, but the organic material to be searched for as the positive electrode active material of the secondary battery may be a conjugated organic molecule having a π bond or a salt thereof. The above organic material may be a conjugated organic molecule having two or more π bonds or a salt thereof. Examples of the above organic material include quinones and their derivatives and their salts (sometimes referred to as quinone-based organic compounds), disulfides and their derivatives and their salts (sometimes referred to as disulfide-based organic compounds), and aromatic imines and their derivatives and their salts (sometimes referred to as imine-based organic compounds).

Examples of the characteristics of the secondary battery to be predicted include the potential [V] of the positive electrode of the secondary battery, the capacity [Ah] of the positive electrode of the secondary battery, and the energy density [Wh/kg or Wh/L] of the secondary battery. The potential [V] may be a potential at which an oxidation-reduction reaction of the target compound occurs, determined from electrochemical measurements such as a charge/discharge curve. The potential may be a standard electrode potential. The potential may be an average reaction potential. When the secondary battery is a lithium ion battery, the potential [V] may be a potential [V vs Li/Li ⁺ ] based on the potential at which an oxidation-reduction reaction of lithium occurs. The capacity of the positive electrode (sometimes referred to as positive electrode capacity, capacity, etc.) may be the positive electrode capacity per unit mass [mAh/g or Ah/kg]. The energy density may be a mass energy density [Wh/kg or Wh/g] or a volumetric energy density [Wh/L].

The characteristics of the electrode material used in the prediction may be any that can be obtained as literature values, catalog values, experimental values, or calculated values, and the type is not particularly limited. For example, when the electrode material is an organic material, various properties related to the molecules that make up the organic material (sometimes referred to as constituent molecules) may be used as the characteristics of the electrode. Examples of the above characteristics include physical properties, chemical properties, and quantum properties. Examples of physical properties include mechanical properties, thermal properties, electrical properties, magnetic properties, and optical properties. Examples of chemical properties include molecular weight, dipole moment, properties of molecular structure, properties of constituent elements, properties of functional groups, solubility, number of functional groups, and number of heteroelements. Examples of quantum properties include properties of molecular orbitals and partial charge density.

[First feature group]
In this embodiment, when the electrode material is an organic material, the characteristics of the electrode material include (A) the melting point of the constituent molecule, (B) the molecular weight of the constituent molecule, (C) the dipole moment of the constituent molecule, (D) the number of carbons conjugated by π bonds among the carbons contained in the structural molecule, (E) the number of elements other than carbon, hydrogen, and oxygen contained in the structural molecule, (F) the number of carboxyl groups contained in the structural molecule, (G) the number of reactions between the structural molecule and the metal ions of the secondary battery, which is derived from the molecular structure of the structural molecule, (H) the maximum and minimum values of an index indicating the charge bias within the structural molecule (for example, a partial charge density indicating the bias of the electron density calculated by the density functional theory), and (I) at least one entropy related to the orbital of the structural molecule. (J) the energy level of the orbital of the constituent molecule, (K) the number of orbitals that meet the first condition among the orbitals of the constituent molecule, (L) the absolute value of the sum of the energy levels of the orbitals that meet the second condition among the orbitals of the constituent molecule, (M) the dispersion term of the Hansen solubility parameter between the structural molecule and the electrolyte of the secondary battery, (N) the polarization term of the Hansen solubility parameter between the structural molecule and the electrolyte of the secondary battery, (O) the hydrogen bond term of the Hansen solubility parameter between the structural molecule and the electrolyte of the secondary battery, and (P) the difference in HSP value between the structural molecule and the carbonate-based electrolyte (sometimes referred to as the first characteristic group). The value of the above characteristic may be a value per molecule of the structural molecule. The characteristic included in the first characteristic group may be an example of a characteristic listed based on the experience and consideration of the inventors.

In this embodiment, (C) the dipole moment of the constituent molecule, (H) the maximum and minimum values of the index showing the charge bias within the constituent molecule, (I) the energy of at least one energy level related to the orbital of the constituent molecule, (J) the energy difference between two energy levels related to the orbital of the constituent molecule, (K) the number of orbitals of the constituent molecule that meet the first condition, and (L) the absolute value of the sum of the energy levels of the orbitals of the constituent molecule that meet the second condition are derived, for example, by known DFT calculations. As an example of DFT calculations, B3LYP is used as the exchange-correlation functional and 6-311G is used as the basis function.

When searching for positive electrode materials using MI, it is conceivable to generate a descriptor based on structural information of candidate compounds. The structural information of the compound is described, for example, based on the SMILES notation. On the other hand, in this embodiment, a DFT calculation is performed based on the structural information of the compound, and the execution result is used as the above-mentioned features that become descriptor candidates. This allows researchers to eliminate from the candidates molecules that are clearly unable to demonstrate performance or that may pose a risk such as an explosion, although they cannot be determined from SMILES alone.

In this embodiment, the dispersion term of the Hansen solubility parameter between the (M) structural molecule and the electrolyte of the secondary battery, the polarization term of the Hansen solubility parameter between the (N) structural molecule and the electrolyte of the secondary battery, the hydrogen bond term of the Hansen solubility parameter between the (O) structural molecule and the electrolyte of the secondary battery, and the difference in HSP value between the (P) structural molecule and the carbonate-based electrolyte are derived, for example, using a known calculation method for the Hansen solubility parameter. As the electrolyte of the secondary battery, for example, any electrolyte is used that is determined by the ion mobility and/or conductivity of the electrolyte itself, the type of material used for the positive electrode and/or negative electrode, and the conditions related to charging and discharging. As the carbonate-based electrolyte, for example, a known carbonate-based electrolyte is used.

(I) Examples of the energy of at least one energy level related to the orbital of the constituent molecules include the energy of the lowest unoccupied molecular orbital (LUMO) energy level (sometimes referred to as the LUMO level), the energy of the level one level above the LUMO level, the energy of the level two levels above the LUMO level, the energy of the level three levels above the LUMO level, the energy of the level four levels above the LUMO level, the energy of the highest occupied molecular orbital (sometimes referred to as the HOMO level), the energy of the level one level below the HOMO level, the energy of the level two levels below the HOMO level, the energy of the level three levels below the HOMO level, and the energy of the level four levels below the HOMO level. (J) Examples of the energy difference between two energy levels related to the orbital of the constituent molecules include the energy difference between any two levels selected from the multiple levels described above.

(K) Examples of the number of orbitals of the constituent molecules that meet the first condition include the number of orbitals that have an energy level equal to or lower than the work function of the metal (e.g., lithium in the case of a lithium ion battery) obtained by reducing the ions that serve as the carrier of the secondary battery, and the number of orbitals that have an energy level equal to or lower than 0. (L) Examples of the absolute value of the sum of the energy levels of the orbitals of the constituent molecules that meet the second condition include the absolute value of the sum of the energy levels of orbitals that have an energy level equal to or lower than 0.

This allows the construction of a prediction model that includes one or more descriptors that represent each of the one or more features that make up the first feature group. As a result, the target properties of the target substance are accurately predicted, as shown in the examples described below.

[Second feature group]
As described above, the features constituting the first feature group are features selected based on the experience and consideration of the inventors. Therefore, the first feature group may include a set of features that are correlated with each other. Therefore, the inventors considered determining features of the electrode material that are candidates for the descriptor, taking into account the correlations of the features included in the first feature group.

In this embodiment, when the electrode material is an organic material, the characteristics of the electrode material include (B) the molecular weight of the constituent molecule, (C) the dipole moment of the constituent molecule, (D) the number of carbons conjugated by π bonds among the carbons contained in the structural molecule, (E) the number of elements other than carbon, hydrogen, and oxygen contained in the structural molecule, (G) the number of reactions between the structural molecule and the metal ions of the secondary battery, derived from the molecular structure of the structural molecule, (H) the maximum and minimum values of an index indicating the charge bias within the structural molecule (e.g., partial charge density), and (I) at least one of the characteristics related to the orbitals of the structural molecule. At least one characteristic selected from the group consisting of (M) the energy of the other energy level, (K) the number of orbitals of the constituent molecules that meet the first condition, (L) the absolute value of the sum of the energy levels of the orbitals of the constituent molecules that meet the second condition, (N) the polarization term of the Hansen solubility parameter between the structural molecule and the electrolyte of the secondary battery, and (O) the hydrogen bond term of the Hansen solubility parameter between the structural molecule and the electrolyte of the secondary battery (sometimes referred to as the second characteristic group). The value of the above characteristic may be a value per molecule of the structural molecule.

In the second feature group, (I) the energy of at least one energy level related to the orbital of the constituent molecule may be exemplified by the energy level of the lowest unoccupied molecular orbital (LUMO) (sometimes referred to as the LUMO level), the energy of the level one level above the LUMO level, the energy of the level two levels above the LUMO level, the energy of the level three levels above the LUMO level, and the energy of the level four levels above the LUMO level. (K) The number of orbitals of the constituent molecules that meet the first condition in the second feature group, and (L) the absolute value of the sum of the energy levels of the orbitals of the constituent molecules that meet the second condition in the second feature group may be the same as those in the first feature group.

This allows the construction of a prediction model that includes one or more descriptors that represent each of the one or more features that make up the second feature group. As a result, the target properties of the target substance are accurately predicted, as shown in the examples described below.

[Third feature group]
In this embodiment, when constructing a prediction model for searching for an organic material to be used as a positive electrode active material of a secondary battery, the characteristics of the above-mentioned electrode material include: (a) the energy of the LUMO level of the constituent molecule; (b) the energy of the level one level higher than the LUMO level of the constituent molecule; (c) the number of reactions between the constituent molecule and the metal ion of the secondary battery derived from the molecular structure of the constituent molecule, which is the number of reactions per molecule; (d) the number of orbitals of the constituent molecule having an energy level equal to or lower than the work function of the metal ion of the secondary battery, which is the number of orbitals per molecule; (e) dispersion term of the Hansen solubility parameters of the constituent molecules and the electrolyte of the secondary battery, (f) polarization term of the Hansen solubility parameters of the constituent molecules and the electrolyte of the secondary battery, (g) hydrogen bond term of the Hansen solubility parameters of the constituent molecules and the electrolyte of the secondary battery, (h) dipole moment of the constituent molecules, (i) minimum value of partial charge density of the constituent molecules, and (j) at least one characteristic selected from the group consisting of the number of elements other than carbon, hydrogen and oxygen contained in the constituent molecules, the number of elements per molecule (sometimes referred to as the third characteristic group). The value of the above characteristic may be a value per molecule of the constituent molecules.

The characteristics of the electrode material may include (a) the energy of the LUMO level of the constituent molecule, (b) the energy of the level one level higher than the LUMO level of the constituent molecule, (c) the number of reactions between the constituent molecule and the metal ion of the secondary battery derived from the molecular structure of the constituent molecule, which is the number of reactions per molecule, and (d) the number of orbitals of the constituent molecule that have an energy level equal to or lower than the work function of the metal ion of the secondary battery, which is the number of orbitals per molecule. The characteristics of the electrode material may include at least two of (a) to (d).

For example, when constructing a predictive model for predicting the potential of the positive electrode, the characteristics of the electrode material may include (a) the energy of the LUMO level of the constituent molecule and (b) the energy of the level one level higher than the LUMO level of the constituent molecule. For example, when constructing a predictive model for predicting the capacity of the positive electrode, the characteristics of the electrode material may include (c) the number of reactions of the constituent molecule and the metal ion of the secondary battery derived from the molecular structure of the constituent molecule, which is the number of reactions per molecule, and (d) the number of orbitals of the constituent molecule having an energy level equal to or lower than the work function of the metal ion of the secondary battery, which is the number of orbitals per molecule. For example, when constructing a prediction model for predicting the energy density of a secondary battery, the characteristics of the electrode material may include: (a) the energy of the LUMO level of the constituent molecule; (b) the energy of the level one level higher than the LUMO level of the constituent molecule; (c) the number of reactions between the constituent molecule and the metal ion of the secondary battery derived from the molecular structure of the constituent molecule, which is the number of reactions per molecule; and (d) the number of orbitals of the constituent molecule that have an energy level equal to or lower than the work function of the metal ion of the secondary battery, which is the number of orbitals per molecule.

The above-mentioned electrode material characteristics may further include other explanatory variables in addition to the combination of the explanatory variables described above. For example, the above-mentioned electrode material characteristics further include at least one selected from the group consisting of (e) the dispersion term of the Hansen solubility parameters of the constituent molecules and the electrolyte of the secondary battery, (f) the polarization term of the Hansen solubility parameters of the constituent molecules and the electrolyte of the secondary battery, (g) the hydrogen bond term of the Hansen solubility parameters of the constituent molecules and the electrolyte of the secondary battery, (h) the dipole moment of the constituent molecules, (i) the minimum value of the partial charge density of the constituent molecules, and (j) the number of elements other than carbon, hydrogen, and oxygen contained in the constituent molecules, which is the number of elements per molecule.

This allows the construction of a prediction model that includes, for example, one or more descriptors that represent each of the one or more features that make up the above feature group. As a result, the target properties of the target substance are predicted with high accuracy, as shown in the examples described below.

[Overview of each part of the material exploration system 100]
In this embodiment, the material exploration system 100 includes, for example, a control unit 112, an input unit 114, an output unit 116, and a storage unit 118. In this embodiment, the material exploration system 100 includes, for example, a database construction unit 120, a prediction model construction unit 140, and a property prediction unit 160.

In this embodiment, the predictive model construction unit 140 has, for example, a dataset generation unit 142, a sparse modeling unit 144, a descriptor adjustment unit 146, and a reconstruction unit 148. In this embodiment, the property prediction unit 160 has, for example, a target material determination unit 162, a target property determination unit 164, and a predicted value output unit 166. Each unit of the material exploration system 100 is configured, for example, to be able to send and receive information to and from each other.

In this embodiment, the control unit 112 controls the operation of the material exploration system 100. The control unit 112 may control various information processing in the material exploration system 100. Examples of the above information processing include input/output processing, database construction processing, prediction model construction processing, prediction processing, etc.

In this embodiment, the input unit 114 accepts inputs or requests from a user or other information processing devices. For example, the input unit 114 accepts user instructions for the material exploration system 100. The input unit 114 may accept data input from the user in the database construction process.

Examples of the input unit 114 include an input device and a communication device. Examples of the input device include a keyboard, a pointing device, a touch panel, a microphone, and a camera.

In this embodiment, the output unit 116 outputs information to a user or another information processing device. For example, the output unit 116 outputs a screen to show the user the results of information processing in the material exploration system 100. The output unit 116 may also output a screen for the user to input instructions, etc., to the material exploration system 100.

Examples of the output unit 116 include an output device and a communication device. Examples of the output device include a display device, an audio output device, and a vibration output device. Examples of the display device include a display and a projector.

In this embodiment, the storage unit 118 stores various types of information. In one embodiment, the storage unit 118 stores various types of information used in the information processing of the material exploration system 100. In another embodiment, the storage unit 118 stores various types of information generated in the information processing of the material exploration system 100. Details of the storage unit 118 will be described later.

[Database construction]
In this embodiment, the database construction unit 120 constructs various databases. In one embodiment, the database construction unit 120 constructs a database (sometimes referred to as a first DB) that stores, for each of a plurality of materials, at least one characteristic value related to the secondary battery when the material is used as an electrode material of the secondary battery and the values of the various features described above in association with each other. In another embodiment, a database (sometimes referred to as a second DB) that stores the values of the various features described above for each of a plurality of materials to be searched is constructed. Details of the database construction unit 120 will be described later.

[Construction of predictive model]
In this embodiment, the prediction model construction unit 140 constructs a prediction model for predicting characteristics of a secondary battery based on the characteristics of the electrode material of the secondary battery. The prediction model construction unit 140 constructs the prediction model by, for example, machine learning.

More specifically, the prediction model construction unit 140 constructs a prediction model by machine learning using data representing the characteristics of the secondary battery and data representing the characteristics of the electrode material of the secondary battery as training data. For example, the material exploration system 100 constructs a prediction model by regression analysis based on machine learning using the training data. The prediction model may be a regression model represented by a linear regression equation or a nonlinear regression equation. The prediction model may be a regression model in which the characteristics of the secondary battery are the objective variable and one or more characteristics related to the electrode material are each a descriptor. The prediction model may be a linear multiple regression model in which the characteristics of the secondary battery are the objective variable and multiple characteristics related to the electrode material are each a descriptor.

[Prediction model of positive electrode potential]
In one embodiment, the prediction model may be a regression model in which the potential of the positive electrode (e.g., reaction potential) when the material is used as a positive electrode active material of a secondary battery is used as a response variable, and (a) the energy of the LUMO level of the molecule of the material and (b) the energy of the level one level higher than the LUMO level of the molecule of the material are included as descriptors. According to this embodiment, the characteristic prediction unit 160 uses the above prediction model, so that the potential of the positive electrode when the target material is used as a positive electrode active material of a secondary battery can be accurately predicted.

[Positive electrode capacity prediction model]
In another embodiment, the prediction model may be a regression model including, as a response variable, the capacity of the positive electrode (e.g., capacity per unit mass) when the material is used as a positive electrode active material of a secondary battery, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule, as a descriptor. According to this embodiment, the characteristic prediction unit 160 uses the above prediction model, so that the capacity of the positive electrode of the secondary battery when the target material is used as a positive electrode active material of the secondary battery can be accurately predicted.

[Energy density prediction model]
In yet another embodiment, the prediction model may be a regression model that uses the energy density (for example, mass energy density) of the secondary battery when the material is used as a positive electrode active material of the secondary battery as a response variable, and includes (a) the energy of the LUMO level of the molecule of the material, (b) the energy of the level one level higher than the LUMO level of the molecule of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecule of the material having an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule, as a descriptor. According to this embodiment, the characteristic prediction unit 160 uses the above prediction model, so that the energy density of the secondary battery when the target material is used as a positive electrode active material of the secondary battery can be accurately predicted.

In this embodiment, the dataset generation unit 142 generates, for example, various datasets used for machine learning in the sparse modeling unit 144. For example, for each of a plurality of materials, the dataset generation unit 142 generates a dataset in which the value of one characteristic related to a secondary battery when the material is used as an electrode material is associated with the values of a plurality of explanatory variables (sometimes referred to as explanatory variables related to materials) that represent each of a plurality of characteristics of the material, as a group of teacher data for machine learning in the sparse modeling unit 144. The dataset generation unit 142 generates various datasets used to build a predictive model in the reconstruction unit 148. For example, the dataset generation unit 142 generates various datasets used for machine learning by referring to the first DB described above.

The dataset generation unit 142 generates, for example, various datasets used to construct a predictive model in the reconstruction unit 148. The dataset generation unit 142 generates, for example, various datasets used for machine learning in the reconstruction unit 148. The dataset generation unit 142 may generate, for example, a dataset for machine learning in the reconstruction unit 148 by a procedure similar to that for the dataset for machine learning in the sparse modeling unit 144. Details of the dataset generation unit 142 will be described later.

In this embodiment, the sparse modeling unit 144 generates candidates for a predictive model. The sparse modeling unit 144 may generate candidates for a predictive model for each of one or more characteristics of the secondary battery.

As described above, the candidate prediction model may be a regression model represented by a linear regression equation or a nonlinear regression equation. The candidate prediction model may be a regression model in which the characteristics of the secondary battery are the response variables and each of the one or more features related to the electrode material is a descriptor. The candidate prediction model may be a linear multiple regression model in which the characteristics of the secondary battery are the response variables and each of the multiple features related to the electrode material is a descriptor.

For example, the sparse modeling unit 144 generates candidate prediction models by machine learning using the dataset generated by the dataset generation unit 142 as training data. For example, the sparse modeling unit 144 generates candidate prediction models for predicting the potential of a positive electrode. For example, the sparse modeling unit 144 generates candidate prediction models for predicting the positive electrode capacity of a secondary battery. For example, the sparse modeling unit 144 generates candidate prediction models for predicting the energy density of a secondary battery.

At this time, the sparse modeling unit 144 may select explanatory variables to be used as descriptors or candidate descriptors of the prediction model from among multiple explanatory variables that represent the characteristics of the material. As described above, in this embodiment, a regression model is used as the prediction model. For example, the sparse modeling unit 144 selects M explanatory variables (M is a positive integer between 1 and N, inclusive; M may also be a positive integer between 1 and N, inclusive) from among N explanatory variables (N is a positive integer between 2 and 1, inclusive) that each represent the characteristics of the material, as descriptors or candidate descriptors of the regression model.

More specifically, for example, the sparse modeling unit 144 receives, as a teacher data group, a data set in which, for each of a plurality of materials, the value of one characteristic related to a secondary battery when the material is used as an electrode material is associated with the values of N explanatory variables related to the characteristics of the material. Examples of one characteristic related to a secondary battery include the potential of the positive electrode, the positive electrode capacity of the secondary battery, or the energy density of the secondary battery. Each of the N explanatory variables may be an explanatory variable representing each of the N characteristics among the multiple characteristics included in the first feature group, the second feature group, or the third feature group described above. Each of the N explanatory variables may be an explanatory variable representing the N characteristics included in the second feature group. Each of the N explanatory variables may be an explanatory variable representing the N characteristics included in the third feature group.

The sparse modeling unit 144 selects M explanatory variables from the N explanatory variables by performing a regression analysis based on machine learning. The sparse modeling unit 144 may further determine a regression coefficient or partial regression coefficient corresponding to each of the M explanatory variables. The sparse modeling unit 144 may also determine a regression constant of the regression model. This generates candidates for a prediction model.

The sparse modeling unit 144 selects M explanatory variables from the N explanatory variables, for example, by a sparse modeling method. Each of the M explanatory variables may be an explanatory variable representing the M features included in the second feature group. Each of the M explanatory variables may be an explanatory variable representing the M features included in the third feature group. The M explanatory variables may include at least two of the combinations of explanatory variables described in relation to the third feature group. The sparse modeling unit 144 may further determine a regression coefficient or partial regression coefficient corresponding to each of the M explanatory variables by the sparse modeling method. The sparse modeling unit 144 may further determine a regression constant by the sparse modeling method.

As the sparse modeling method, a known regression analysis method for selecting variables can be adopted. Examples of regression analysis methods using the sparse modeling method include (i) an approach that searches all combinations of explanatory variables, such as the exhaustive search for linear regression method (ES-LiR, sometimes called a brute force method, etc.), a relaxed approximation approach, such as lasso regression (sometimes called least absolute shrinkage and selection operator, LASSO, etc.), and a sampling approach using a sampling method.

The sparse modeling unit 144 selects, as the M explanatory variables, at least two selected from, for example, (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level above the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. This allows for the construction of a prediction model with excellent prediction accuracy when searching for a positive electrode active material.

In one embodiment, when the objective variable of the regression model is the potential of the positive electrode, the sparse modeling unit 144 may select M explanatory variables including (a) the energy of the LUMO level of the molecules of the material and (b) the energy of the level one level higher than the LUMO level of the molecules of the material. This allows a new prediction model with excellent prediction accuracy to be constructed when searching for a positive electrode active material. In particular, a new prediction model is constructed that predicts the potential of the positive electrode based on the energy of the level one level higher than the LUMO level of the molecules of the material.

In another embodiment, when the objective variable of the regression model is the positive electrode capacity of the secondary battery, the sparse modeling unit 144 may select M explanatory variables, including (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. This allows the construction of a new prediction model with excellent prediction accuracy when searching for a positive electrode active material.

In yet another embodiment, when the objective variable of the regression model is the energy density of the secondary battery, the sparse modeling unit 144 may select M explanatory variables, including (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. This allows the construction of a new prediction model with excellent prediction accuracy when searching for a positive electrode active material.

In this embodiment, the descriptor adjustment unit 146 selects K explanatory variables (K is a positive integer between 1 and M inclusive. K may be a positive integer between 1 and M inclusive) from the M explanatory variables selected by the sparse modeling unit 144 as descriptors for the regression model. The descriptor adjustment unit 146 selects K explanatory variables from the M explanatory variables based on, for example, a user instruction. The descriptor adjustment unit 146 outputs information indicating the selected K explanatory variables to the reconstruction unit 148.

The descriptor adjustment unit 146 selects, as the K explanatory variables, at least two selected from, for example, (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level above the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. This allows the construction of a new prediction model with excellent prediction accuracy when searching for a positive electrode active material.

In one embodiment, when the objective variable of the regression model is the potential of the positive electrode, the descriptor adjustment unit 146 may select K explanatory variables including (a) the energy of the LUMO level of the molecules of the material and (b) the energy of the level one level higher than the LUMO level of the molecules of the material. This allows the construction of a new prediction model with excellent prediction accuracy when searching for a positive electrode active material.

In another embodiment, when the objective variable of the regression model is the positive electrode capacity of the secondary battery, the descriptor adjustment unit 146 may select K explanatory variables, including (c) the number of reactions between the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. This allows the construction of a new prediction model with excellent prediction accuracy when searching for a positive electrode active material.

In yet another embodiment, when the objective variable of the regression model is the energy density of the secondary battery, the descriptor adjustment unit 146 may select K explanatory variables, including (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. This allows the construction of a new prediction model with excellent prediction accuracy when searching for a positive electrode active material.

In this embodiment, the reconstruction unit 148 constructs a prediction model based on the K explanatory variables output by the descriptor adjustment unit 146. The reconstruction unit 148 may construct a prediction model for each of one or more characteristics of the secondary battery. For example, the reconstruction unit 148 constructs a prediction model by machine learning using a dataset generated by the dataset generation unit 142 as training data. The reconstruction unit 148 may construct a regression model that uses one characteristic of the secondary battery as a response variable and includes the K explanatory variables output by the descriptor adjustment unit 146 for the one characteristic as descriptors.

[Prediction of battery characteristics]
In this embodiment, the characteristic prediction unit 160 predicts the characteristics of the secondary battery based on the characteristics of the electrode material of the secondary battery. For example, the characteristic prediction unit 160 predicts the target characteristics of the secondary battery when the target material is used as the electrode material of the secondary battery by using the prediction model constructed by the prediction model construction unit 140. This allows the user to search for new electrode materials.

In this embodiment, the target material determination unit 162 determines a target material, which is an electrode material that is the target of prediction. The target material determination unit 162 determines the target material based on, for example, an instruction from a user received by the input unit 114. The target material determination unit 162 may output information indicating the target material to the prediction value output unit 166.

In this embodiment, the target characteristic determination unit 164 determines a target characteristic, which is a characteristic to be predicted. The target characteristic determination unit 164 determines the target characteristic based on, for example, an instruction from a user received by the input unit 114. The target characteristic determination unit 164 may output information indicating the target characteristic to the predicted value output unit 166.

In this embodiment, the predicted value output unit 166 outputs a predicted value of the target characteristic of the target material. For example, the predicted value output unit 166 acquires input data representing one or more characteristics of the target material. The predicted value output unit 166 inputs the input data into a prediction model for predicting the target characteristic, thereby outputting a predicted value of the target characteristic when the target material is used as an electrode material for a secondary battery. Details of the predicted value output unit 166 will be described later.

[Specific configuration of each part of the material exploration system 100]
Each part of the material exploration system 100 may be realized by hardware, software, or a combination of hardware and software. Each part of the material exploration system 100 may be realized, at least in part, by a single server or by multiple servers. Each part of the material exploration system 100 may be realized, at least in part, on a virtual machine or a cloud system. Each part of the material exploration system 100 may be realized, at least in part, by a personal computer or a mobile terminal. Examples of mobile terminals include mobile phones, smartphones, PDAs, tablets, notebook computers or laptop computers, and wearable computers. Each part of the material exploration system 100 may store information using a distributed ledger technology such as a blockchain or a distributed network.

When at least some of the components constituting the material exploration system 100 are realized by software, the components realized by the software may be realized by starting software or a program that defines the operation of the components in an information processing device of a general configuration. The information processing device of the general configuration may include (i) a data processing device having a processor such as a CPU or GPU, a ROM, a RAM, a communication interface, etc., (ii) input devices such as a keyboard, a pointing device, a touch panel, a camera, a voice input device, a gesture input device, various sensors, a GPS receiver, etc., (iii) output devices such as a display device, a voice output device, a vibration device, etc., and (iv) storage devices such as memory, HDD, SSD, etc. (including external storage devices).

In the information processing device of the above general configuration, the above data processing device or storage device may store the above software or program. The above software or program, when executed by a processor, causes the above information processing device to execute the operation defined by the software or program. The above software or program may be stored in a non-transitory computer-readable recording medium. The above software or program may be a program for causing a computer to function as the material exploration system 100 or a part thereof. The above software or program may be a program for causing a computer to execute an information processing method in the material exploration system 100 or a part thereof.

In one embodiment, the information processing method may be an information processing method for predicting the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery. The information processing method has, for example, a material determination step for determining a target material, which is an electrode material to be predicted. The information processing method has, for example, a characteristic determination step for determining a target characteristic, which is a characteristic to be predicted. For example, when the target characteristic is the potential of the positive electrode when the target material is used as a positive electrode active material of a secondary battery, the information processing method has an input value acquisition step for acquiring (i) the energy value of the LUMO level of the molecule of the target material, and (ii) the energy value of the level one level higher than the LUMO level of the molecule of the target material. The information processing method has, for example, a prediction step for inputting the value acquired in the input value acquisition step into a prediction model and outputting a predicted value of the target characteristic of the target material. In the above information processing method, the prediction model is, for example, a regression model in which the objective variable is the potential of the positive electrode when the material is used as a positive electrode active material of a secondary battery, and the descriptors include (a) the energy of the LUMO level of the material's molecules and (b) the energy of the level one level higher than the LUMO level of the material's molecules.

In another embodiment, the information processing method may be an information processing method for predicting the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery. The information processing method may, for example, have a material determination step for determining a target material, which is an electrode material to be predicted. The information processing method may, for example, have a characteristic determination step for determining a target characteristic, which is a characteristic to be predicted. For example, when the target characteristic is a positive electrode capacity when the target material is used as a positive electrode active material of a secondary battery, the information processing method may have an input value acquisition step for acquiring (iii) a value per molecule of the number of reactions of the target material and the metal ions of the secondary battery derived from the molecular structure of the target material, and (iv) a value per molecule of the number of orbitals of the molecules of the target material having an energy level equal to or lower than the work function of the metal ions of the secondary battery. The information processing method may, for example, have a prediction step for inputting the values acquired in the input value acquisition step into a prediction model and outputting a predicted value of the target characteristic of the target material. In the above information processing method, the prediction model is, for example, a regression model in which the objective variable is the positive electrode capacity when the material is used as a positive electrode active material of a secondary battery, (c) the number of reactions between the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule, as a descriptor.

In yet another embodiment, the information processing method may be an information processing method for predicting the characteristics of a secondary battery based on the characteristics of an electrode material of the secondary battery. The information processing method may, for example, have a material determination step for determining a target material, which is an electrode material to be predicted. The information processing method may, for example, have a characteristic determination step for determining a target characteristic, which is a characteristic to be predicted. For example, when the target characteristic is an energy density when the target material is used as a positive electrode active material of a secondary battery, the information processing method may have an input value acquisition step for acquiring (i) an energy value of the LUMO level of the molecule of the target material, (ii) an energy value of the level one level higher than the LUMO level of the molecule of the target material, (iii) a value per molecule of the number of reactions of the target material and the metal ion of the secondary battery derived from the molecular structure of the target material, and (iv) a value per molecule of the number of orbitals of the molecule of the target material having an energy level equal to or lower than the work function of the metal ion of the secondary battery. The information processing method may, for example, have a prediction step for inputting the value acquired in the input value acquisition step into a prediction model and outputting a predicted value of the target characteristic of the target material. In the above information processing method, the prediction model is a regression model that uses, for example, the energy density when the material is used as a positive electrode active material of a secondary battery as the objective variable, and includes, as descriptors, (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule.

In yet another embodiment, the above information processing method may be an information processing method for constructing a prediction model for predicting the characteristics of a secondary battery based on the characteristics of the electrode material of the secondary battery. In the above information processing method, the prediction model is, for example, a regression model in which the characteristics of the secondary battery are the objective variable and each of one or more characteristics related to the electrode material is a descriptor.

The above information processing method includes, for example, a first variable selection step of selecting M explanatory variables (M is a positive integer between 1 and N) from among N explanatory variables (N is a positive integer between 2 and 1) each of which represents a characteristic of a material as a descriptor or candidate for a descriptor of a regression model. In the above information processing method, the first variable selection step includes, for example, a step of acquiring a group of teacher data in which, for each of a plurality of materials, a value of one characteristic related to a secondary battery when the material is used as an electrode material is associated with values of N explanatory variables related to the characteristics of the material. The first variable selection step includes, for example, a step of selecting M explanatory variables from among N explanatory variables by performing a regression analysis based on machine learning. In the above information processing method, the N explanatory variables include, for example, (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. The N explanatory variables further include, for example, at least one selected from the group consisting of (e) the dispersion term of the Hansen solubility parameters of the material and the electrolyte of the secondary battery, (f) the polarization term of the Hansen solubility parameters, (g) the hydrogen bond term of the Hansen solubility parameters, (h) the dipole moment of the molecules of the material, (i) the minimum value of the partial charge density of the molecules of the material, and (j) the number of elements other than carbon, hydrogen, and oxygen contained in the molecules of the material, which is the number of elements per molecule.

The material exploration system 100 may be an example of an information processing device. The predictive model construction unit 140 may be an example of a model construction unit or an information processing device. The sparse modeling unit 144 may be an example of a first variable selection unit, a model construction unit, or an information processing device. The descriptor adjustment unit 146 may be an example of a second variable selection unit. The property prediction unit 160 may be an example of an information processing device. The target material determination unit 162 may be an example of a material determination unit. The target property determination unit 164 may be an example of a property determination unit. The predicted value output unit 166 may be an example of an information processing device. The constituent molecules may be an example of molecules of the material. Various data sets used in machine learning may be an example of a teacher data group.

[An example of another embodiment]
In the present embodiment, an example of the material exploration system 100 has been described in which the sparse modeling unit 144 generates prediction model candidates by selecting M explanatory variables from N explanatory variables by the sparse modeling method. However, the material exploration system 100 is not limited to the present embodiment.

In other embodiments, the sparse modeling unit 144 may generate candidate predictive models using any linear or nonlinear regression analysis method. Examples of linear regression methods include the above-mentioned all-state search method (sometimes referred to as the ES-LiR method), lasso regression (sometimes referred to as LASSO), multiple regression (sometimes referred to as MLR), MCP regression, etc. Examples of nonlinear regression methods include random forest regression (sometimes referred to as RF), neural net regression (sometimes referred to as NN), support vector regression (sometimes referred to as SVR), logistic regression, etc.

In this embodiment, an example of the material exploration system 100 has been described using a case in which the sparse modeling unit 144 generates candidates for a predictive model, and then a predictive model is constructed through information processing in the descriptor adjustment unit 146 and the reconstruction unit 148. However, the material exploration system 100 is not limited to this embodiment.

In other embodiments, at least a portion of the information processing in the descriptor adjustment unit 146 and the reconstruction unit 148 may be omitted. In this case, for example, the candidate predictive model generated by the sparse modeling unit 144 is output as the predictive model.

Figure 2 shows a schematic diagram of an example of the internal structure of the storage unit 118. In this embodiment, the storage unit 118 includes a material database 220 and a model database 240. In this embodiment, the material database 220 stores, for each of one or more materials, a material ID 222, a substance ID 224, information 226 representing the characteristics of the secondary battery, and information 228 representing the characteristics of the material, in association with each other. In this embodiment, the model database 240 stores, for each of one or more prediction models, a model ID 242, a response variable ID 244, and information 246 representing the prediction model, in association with each other.

In this embodiment, identification information for identifying each of one or more materials is input to material ID 222. The above identification information may be a symbol assigned to each material, or may be the name of each material.

Each of the one or more materials may be a known electrode material or a material to be searched for. If the material identified by material ID 222 is a known electrode material, the record of the material is used, for example, as teacher data for machine learning processing in the prediction model construction unit 140. If the material identified by material ID 222 is a material to be searched for, the record of the material is used, for example, as input data for prediction processing in the characteristic prediction unit 160.

In this embodiment, identification information for identifying the substances contained in each material or the substances constituting each material is input to the substance ID 224. The above identification information may be a symbol assigned to each substance or may be the name of each substance. The identification information may include structural information of each substance.

In this embodiment, the information 226 representing the characteristics of the secondary battery includes values of the characteristics of the secondary battery when each material is used as an electrode material of the secondary battery. For example, the information 226 representing the characteristics of the secondary battery includes values of the characteristics of the secondary battery when each material is used as a positive electrode active material of the secondary battery. As described above, examples of the characteristics of the secondary battery include the potential of the positive electrode, the positive electrode capacity of the secondary battery, and the energy density of the secondary battery.

When the material identified by material ID 222 is a known electrode material, experimental values, literature values, etc. of the characteristics are input into information 226 representing the characteristics of the secondary battery. On the other hand, when the material identified by material ID 222 is a material to be searched for, a symbol indicating that the information is unknown may be input into information 226 representing the characteristics of the secondary battery, or information 226 representing the characteristics of the secondary battery may be left blank.

In this embodiment, the values of one or more characteristics of each material are input to information 228 representing the material characteristics. As explained in relation to the characteristics of the electrode material in FIG. 1, literature values, catalog values, experimental values, calculated values, etc. representing the characteristics of each material are input to information 228 representing the material characteristics. Examples of calculated values include values calculated by DFT calculation, calculated values of each term of the Hansen solubility parameter, and calculated values of the Hansen solubility parameter.

In this embodiment, identification information for identifying each of the one or more prediction models is input to model ID 242. The above identification information may be a symbol assigned to each prediction model, or may be the name of each prediction model.

In this embodiment, identification information of the objective variable of each prediction model is input to the objective variable ID 244. The above identification information may be a symbol assigned to each objective variable, or may be the name of each objective variable.

In this embodiment, various information for identifying a prediction model is input to the information 246 representing the prediction model. For example, identification information (sometimes referred to as a descriptor ID) for identifying each of one or more descriptors included in each prediction model is input to the information 246 representing the prediction model in association with information indicating the value of a regression coefficient corresponding to each descriptor. Information indicating the value of the regression constant of each prediction model may also be input to the information 246 representing the prediction model.

Figure 3 shows an example of information processing in the database construction unit 120. In this embodiment, the database construction unit 120 includes a material data registration unit 322 and a material data calculation unit 324.

In this embodiment, the material data registration unit 322 starts the material data registration process according to instructions from the control unit 112. The control unit 112 decides to start the material data registration process, for example, based on an instruction or request from the input unit 114. When the material data registration process starts, the material data registration unit 322 accepts various data inputs from the user, for example, via a data input screen for constructing the material database 220 displayed on the output unit 116. The material data registration unit 322 acquires input values from the user and registers the input values in the material database 220.

The material data registration unit 322 also requests the material data calculation unit 324 to execute a calculation process to derive a predetermined type of characteristic from among the characteristics input to the information 228 representing the material characteristics. The material data registration unit 322 acquires the calculated value output by the material data calculation unit 324 and registers the calculated value in the material database 220.

In this embodiment, the material data calculation unit 324 executes various calculation processes. For example, the material data calculation unit 324 acquires a substance ID of a material to be subjected to the calculation process from the material data registration unit 322. The material data calculation unit 324 executes a DFT calculation for a substance identified by the substance ID. The material data calculation unit 324 outputs the calculation result of the DFT calculation to the material data registration unit 322. The material data calculation unit 324 may calculate the value of each term of the Hansen solubility parameter, or may calculate the value of the Hansen solubility parameter.

Figure 4 shows an example of information processing in the dataset generation unit 142. In this embodiment, the details of the dataset generation unit 142 are described using an example in which the dataset generation unit 142 generates a training dataset 420, a validation dataset 440, and a reconstruction dataset 460.

In this embodiment, the dataset generation unit 142 generates a training dataset 420 and a validation dataset 440 in response to, for example, the sparse modeling unit 144 starting the process of constructing a predictive model. The dataset generation unit 142 generates an appropriate dataset in response to the objective variable of the predictive model generated by the sparse modeling unit 144, and outputs the generated dataset to the sparse modeling unit 144.

The training dataset 420 includes, for example, a dataset 422 used to train a predictive model for predicting the potential of the positive electrode, a dataset 424 used to train a predictive model for predicting the positive electrode capacity of the secondary battery, and/or a dataset 426 used to train a predictive model for predicting the energy density of the secondary battery. The validation dataset 440 includes, for example, a dataset 442 used to validate a predictive model for predicting the potential of the positive electrode, a dataset 444 used to validate a predictive model for predicting the positive electrode capacity of the secondary battery, and/or a dataset 446 used to validate a predictive model for predicting the energy density of the secondary battery.

For example, the dataset generation unit 142 first refers to the information 226 representing the characteristics of the secondary battery in the material database 220, and extracts, for each of the multiple materials, a value corresponding to the objective variable of the prediction model generated by the sparse modeling unit 144. In addition, the dataset generation unit 142 generates a dataset that serves as a teacher data group by, for example, associating the value corresponding to the objective variable of the extracted prediction model with the values of multiple explanatory variables related to each material included in the information 228 representing the characteristics of the material in the material database 220. Next, the dataset generation unit 142 divides the dataset that serves as the teacher data group to generate a training dataset 420 and a validation dataset 440.

In addition, in this embodiment, the dataset generation unit 142 generates a reconstruction dataset 460, for example, in response to the reconstruction unit 148 starting the construction process of the prediction model. The reconstruction dataset 460 includes, for example, a dataset 462 used to train a prediction model for predicting the potential of the positive electrode, a dataset 464 used to train a prediction model for predicting the positive electrode capacity of the secondary battery, and/or a dataset 466 used to train a prediction model for predicting the energy density of the secondary battery. The reconstruction dataset 460 may include a dataset (not shown) for verifying the prediction model corresponding to each of the dataset 462, the dataset 464, and the dataset 466.

In one embodiment, the dataset generator 142 generates the reconstruction dataset 460 by a procedure similar to that for the training dataset 420 and/or the validation dataset 440. The training dataset 420 and/or the validation dataset 440 may be reused as the reconstruction dataset 460.

In another embodiment, the dataset generator 142 generates the reconstruction dataset 460 by extracting data in which the value corresponding to the objective variable of the prediction model meets a predetermined condition from the training dataset 420 and/or the validation dataset 440. An example of the above condition is that the value corresponding to the objective variable of the prediction model is within a predetermined numerical range. The above numerical range may have only an upper limit, only a lower limit, or both an upper limit and a lower limit.

This allows, for example, outliers to be removed and data that will produce desirable results to be extracted. As a result, a predictive model with excellent prediction accuracy can be constructed.

Training dataset 420 may be an example of a teacher data group. Dataset 422 may be an example of a teacher data group. Dataset 424 may be an example of a teacher data group. Dataset 426 may be an example of a teacher data group.

Figure 5 shows an example of a process for generating the reconstruction dataset 460. In this embodiment, the details of the process for generating the reconstruction dataset 460 are described using as an example a case in which the dataset generator 142 generates a dataset 462 including Q pieces of data (Q is a positive integer between 1 and P) based on a dataset 422 including P pieces of data (P is a positive integer between 1 and P).

In this embodiment, the dataset generation unit 142 generates the dataset 462 by extracting data in which the value corresponding to the objective variable of the prediction model meets a predetermined condition from the dataset 422. The above condition is set, for example, based on an instruction from a user received by the input unit 114. As described above, an example of the above condition is that the value corresponding to the objective variable of the prediction model is within a predetermined numerical range.

For example, the dataset generation unit 142 generates the dataset 462 by extracting data in which the positive electrode potential value is greater than a predetermined value from the dataset 422. The dataset generation unit 142 may generate the dataset 462 by deleting data related to descriptors other than the descriptor selected by the descriptor adjustment unit 146 from the dataset 422.

Figure 6 shows an example of information processing in the predictive model construction unit 140. According to this embodiment, first, the control unit 112 causes the sparse modeling unit 144 to start the construction process of a predictive model, for example, in response to a request from a user. At this time, the control unit 112 may output information indicating the contents of various settings related to the construction process of a predictive model (sometimes referred to as setting information) to the sparse modeling unit 144. Examples of the above settings include the type of regression method, the type of predictive model, various settings related to cross-validation, and the selection of a descriptor.

The control unit 112 also causes the dataset generation unit 142 to start the process of generating the training dataset 420 and the validation dataset 440. At this time, the control unit 112 may output information indicating the type of the objective variable of the prediction model to the dataset generation unit 142.

The dataset generation unit 142 generates a training dataset 420 and a validation dataset 440 in response to instructions from the control unit 112. As described above, each of the training dataset 420 and the validation dataset 440 includes a plurality of records corresponding to each of a plurality of materials, and each of the plurality of records stores the value of a single objective variable in association with the values of N explanatory variables.

The dataset generation unit 142 outputs the training dataset 420 to the construction unit 642. The dataset generation unit 142 outputs the validation dataset 440 to the validation unit 644.

Next, the sparse modeling unit 144 generates candidates for a prediction model with the characteristics of the secondary battery specified by the user as the objective variable by machine learning using the input training dataset 420 as a teacher data group. At this time, the sparse modeling unit 144 may select M explanatory variables from the N explanatory variables included in the training dataset 420 by the sparse modeling method.

In one embodiment, the sparse modeling unit 144 optimizes the combination of descriptors by changing the combination of descriptors included in the model and repeating the construction of the model by the construction unit 642 and the verification of the model by the verification unit 644. When the construction of the model by the construction unit 642 and the verification of the model by the verification unit 644 are completed for all combinations of the N explanatory variables included in the training dataset 420, the output unit 646 outputs the optimized combination of descriptors. The output unit 646 may output information indicating the selected M explanatory variables and information indicating the values of the regression coefficients corresponding to the M explanatory variables. The output unit 646 may output information indicating the value of the regression constant. In this way, a candidate prediction model is generated. At this time, the value of the regression coefficient corresponding to the explanatory variable that was not selected may be 0. In this way, M explanatory variables are selected from the N explanatory variables included in the training dataset 420.

In another embodiment, the sparse modeling unit 144 optimizes the combination of descriptors by repeatedly constructing and verifying a model by changing the combination of descriptors included in the model using machine learning techniques such as LASSO regression that uses L1 regularization to reduce the number of descriptors, random forest that combines multiple decision trees to perform ensemble learning, and support vector regression that uses the idea of margin maximization to minimize the ε-insensitive loss function. The output unit outputs the optimized combination of descriptors. When a linear regression technique is used, the output unit may output information indicating the selected M explanatory variables and information indicating the values of the regression coefficients corresponding to the M explanatory variables. When a nonlinear regression technique is used, the output unit outputs information indicating the selected M explanatory variables and hyperparameters of the model using the M explanatory variables. As a result, M explanatory variables are selected from the N explanatory variables included in the training dataset 420.

Next, the descriptor adjustment unit 146 selects K explanatory variables as descriptors of the prediction model from the M explanatory variables selected by the sparse modeling unit 144. For example, the descriptor adjustment unit 146 causes the output unit 116 to display a screen for showing the output result of the sparse modeling unit 144. The descriptor adjustment unit 146 also causes the output unit 116 to display an input screen for accepting an instruction to select explanatory variables by the user. The descriptor adjustment unit 146 accepts the selection instruction input by the user to the input unit 114, and selects K explanatory variables as descriptors of the prediction model based on the selection instruction. The descriptor adjustment unit 146 outputs information indicating the selection result to the reconstruction unit 148.

Next, the reconstruction unit 148 constructs a prediction model based on the K explanatory variables output by the descriptor adjustment unit 146. The dataset generation unit 142 generates a reconstruction dataset 460 in response to an instruction from the control unit 112. The reconstruction unit 148 constructs a prediction model including K descriptors that represent the characteristics of the material explained by the K explanatory variables output by the descriptor adjustment unit 146, using, for example, machine learning with the input reconstruction dataset 460 as a training data group, the characteristics of the secondary battery specified by the user as the objective variable. The reconstruction unit 148 stores various information related to the constructed prediction model in the model database 240.

FIG. 7 is a schematic diagram showing another example of information processing in the prediction model construction unit 140. In this embodiment, the prediction model construction unit 140 is different from the embodiment described in relation to FIG. 6 in that the prediction model construction unit 140 includes a descriptor adjustment unit 746 instead of the descriptor adjustment unit 146 and does not include the reconstruction unit 148. In addition, this embodiment is different from the embodiment described in relation to FIG. 6 in that the descriptor adjustment unit 746 adjusts explanatory variables constituting the training dataset 420 and the validation dataset 440 before the dataset generation unit 142 generates the training dataset 420 and the validation dataset 440. With respect to features other than the above differences, the embodiment described in relation to FIG. 7 may have a configuration similar to that of the embodiment described in relation to FIG. 6.

As described above, in this embodiment, before the dataset generation unit 142 generates the training dataset 420 and the validation dataset 440, the descriptor adjustment unit 746 adjusts the explanatory variables constituting the training dataset 420 and the validation dataset 440. More specifically, for example, when the dataset generation unit 142 receives an instruction from the control unit 112, the dataset generation unit 142 first refers to the material database 220 and generates the adjustment dataset 720 in a similar procedure to the training dataset 420, etc.

In this embodiment, the adjustment dataset 720 includes a plurality of records corresponding to each of the plurality of materials, and each of the plurality of records stores the value of the objective variable indicated by the control unit 112 in correspondence with the values of L explanatory variables (L is a positive integer equal to or greater than N. L may be a positive integer greater than N). Each of the L explanatory variables represents a plurality of characteristics related to the material. Each of the L explanatory variables may be explanatory variables representing a respective one of the L characteristics among the plurality of characteristics included in the first feature group, the second feature group, or the third feature group described above. Each of the L explanatory variables may be explanatory variables representing the N characteristics included in the first feature group. The dataset generation unit 142 outputs the generated adjustment dataset 720 to the descriptor adjustment unit 746.

Next, the descriptor adjustment unit 746 selects N explanatory variables from the L explanatory variables included in the adjustment dataset 720 as candidates for the descriptor of the prediction model. For example, the descriptor adjustment unit 746 determines the degree of correlation with the other explanatory variables for each of the L explanatory variables. The descriptor adjustment unit 746 selects N explanatory variables from the L explanatory variables by leaving one explanatory variable out of multiple explanatory variables whose degree of correlation is stronger than a predetermined standard and excluding the other explanatory variables.

This adjusts the number of explanatory variables. The descriptor adjustment unit 746 outputs information indicating the adjustment result to the dataset generation unit 142. The information indicating the adjustment result includes, for example, identification information for each of the N explanatory variables selected by the descriptor adjustment unit 746.

Next, the dataset generation unit 142 refers to the material database 220 to obtain the values of the N explanatory variables selected by the descriptor adjustment unit 746, and generates a training dataset 420 and a validation dataset 440. As described above, each of the training dataset 420 and the validation dataset 440 includes a plurality of records corresponding to each of the multiple materials, and each of the multiple records stores the value of a single objective variable and the values of the N explanatory variables in correspondence with each other. The dataset generation unit 142 outputs the generated training dataset 420 and validation dataset 440 to the sparse modeling unit 144.

Next, the sparse modeling unit 144 generates candidates for prediction models using a procedure similar to that described in relation to FIG. 6. In this embodiment, the information processing in the descriptor adjustment unit 146 and the reconstruction unit 148 is omitted. Therefore, the above-mentioned candidates for prediction models are output as prediction models. The sparse modeling unit 144 stores various information related to the constructed prediction models in the model database 240.

The descriptor adjustment unit 746 may be an example of a third variable selection unit. The multiple explanatory variables that make up the adjustment dataset 720 may be an example of L explanatory variables.

[An example of another embodiment]
In this embodiment, an example of the material exploration system 100 has been described by taking as an example a case in which the predictive model construction unit 140 does not include the descriptor adjustment unit 146 and the reconstruction unit 148. However, the material exploration system 100 is not limited to this embodiment. In other embodiments, the predictive model construction unit 140 may include the descriptor adjustment unit 146 and the reconstruction unit 148 in addition to the descriptor adjustment unit 746.

Figure 8 shows an example of information processing in the property prediction unit 160. According to this embodiment, first, the target material determination unit 162 determines the target material. The target material determination unit 162 determines the target material, for example, based on instructions from the user received by the input unit 114. In addition, the target property determination unit 164 determines the target property. The target property determination unit 164 determines the target property, for example, based on instructions from the user received by the input unit 114.

Next, the prediction value output unit 166 identifies a prediction model for predicting the target characteristic. The prediction value output unit 166, for example, refers to the model database 240 to identify a prediction model that uses the target characteristic determined by the target characteristic determination unit 164 as a target variable. The prediction value output unit 166 also refers to the model database 240 to identify the descriptors included in the identified prediction model.

The predicted value output unit 166 then accesses, for example, the material database 220 to obtain information indicating the values of explanatory variables corresponding to the descriptors included in the identified prediction model. The predicted value output unit 166 inputs the obtained values into the prediction model and outputs the predicted values of the target properties of the target material. The output unit 116 presents the predicted values output by the predicted value output unit 166 to the user.

In one embodiment, when the target characteristic is the potential of the positive electrode when the target material is used as a positive electrode active material of a secondary battery, the predicted value output unit 166 acquires (i) the energy value of the LUMO level of the molecules of the target material, and (ii) the energy value of the level one level higher than the LUMO level of the molecules of the target material. The predicted value output unit 166 inputs the acquired values into the above-described prediction model of the positive electrode potential, and outputs the predicted value of the positive electrode potential when the target material is used as a positive electrode active material of a secondary battery.

In another embodiment, when the target characteristic is the positive electrode capacity when the target material is used as a positive electrode active material of a secondary battery, the predicted value output unit 166 acquires (iii) a value per molecule of the number of reactions between the target material and the metal ions of the secondary battery derived from the molecular structure of the target material, and (iv) a value per molecule of the number of orbitals of the molecules of the target material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery. The predicted value output unit 166 inputs the acquired values into the above-described positive electrode capacity prediction model, and outputs a predicted value of the positive electrode capacity of the secondary battery when the target material is used as a positive electrode active material of the secondary battery.

In yet another embodiment, when the target characteristic is the energy density when the target material is used as a positive electrode active material of a secondary battery, the predicted value output unit 166 acquires (i) the energy value of the LUMO level of the molecule of the target material, (ii) the energy value of the level one level higher than the LUMO level of the molecule of the target material, (iii) the value per molecule of the number of reactions between the target material and the metal ions of the secondary battery derived from the molecular structure of the target material, and (iv) the value per molecule of the number of orbitals of the molecule of the target material having an energy level equal to or lower than the work function of the metal ions of the secondary battery. The predicted value output unit 166 inputs the acquired values into the energy density prediction model described above, and outputs a predicted value of the energy density of the secondary battery when the target material is used as a positive electrode active material of the secondary battery.

Figure 9 shows an example of an information processing method in the material exploration system 100. According to this embodiment, first, in step 912 (step may be abbreviated as S), the database construction unit 120 constructs the material database 220.

Next, in S922, S924, and S926, a prediction model is constructed with a specific characteristic as the objective variable. First, in S922, explanatory variables that are candidates for the descriptors included in the prediction model are determined. For example, the descriptor adjustment unit 746 selects N explanatory variables from the L explanatory variables. Also, the sparse modeling unit 144 selects M explanatory variables from the N explanatory variables. In S924, a descriptor to be used in the prediction model is determined from the explanatory variables that are candidates for the descriptors. For example, the descriptor adjustment unit 146 selects M explanatory variables from the N explanatory variables selected by the sparse modeling unit 144. Then, in S926, a prediction model is constructed using the selected descriptors. For example, the reconstruction unit 148 constructs a prediction model that includes the K explanatory variables selected by the descriptor adjustment unit 146 as descriptors.

Next, in S932, the characteristic prediction unit 160 calculates a predicted value of the target characteristic using the prediction model constructed by the prediction model construction unit 140. This ends the process.

Figure 10 shows an example of a computer 3000 in which multiple aspects of the present invention may be embodied in whole or in part. At least a part of the material exploration system 100 may be realized by the computer 3000. For example, at least a part of the database construction unit 120 is realized by the computer 3000. For example, at least a part of the predictive model construction unit 140 is realized by the computer 3000. For example, at least a part of the property prediction unit 160 is realized by the computer 3000.

A program installed on the computer 3000 may cause the computer 3000 to function as or perform operations associated with an apparatus according to an embodiment of the present invention or one or more "parts" of the apparatus, and/or to perform a process or steps of the process according to an embodiment of the present invention. Such a program may be executed by the CPU 3012 to cause the computer 3000 to perform certain operations associated with some or all of the blocks of the flowcharts and block diagrams described herein.

The computer 3000 according to this embodiment includes a CPU 3012, a RAM 3014, a graphics controller 3016, and a display device 3018, which are interconnected by a host controller 3010. The computer 3000 also includes input/output units such as a communication interface 3022, a hard disk drive 3024, a DVD-ROM drive 3026, and an IC card drive, which are connected to the host controller 3010 via an input/output controller 3020. The computer also includes legacy input/output units such as a ROM 3030 and a keyboard 3042, which are connected to the input/output controller 3020 via an input/output chip 3040.

The CPU 3012 operates according to the programs stored in the ROM 3030 and the RAM 3014, thereby controlling each unit. The graphics controller 3016 acquires image data generated by the CPU 3012 into a frame buffer or the like provided in the RAM 3014 or into itself, and causes the image data to be displayed on the display device 3018.

The communication interface 3022 communicates with other electronic devices via a network. The hard disk drive 3024 stores programs and data used by the CPU 3012 in the computer 3000. The DVD-ROM drive 3026 reads programs or data from the DVD-ROM 3001 and provides the programs or data to the hard disk drive 3024 via the RAM 3014. The IC card drive reads programs and data from an IC card and/or writes programs and data to an IC card.

The ROM 3030 stores therein a boot program or the like that is executed by the computer 3000 upon activation, and/or a program that depends on the hardware of the computer 3000. The input/output chip 3040 may also connect various input/output units to the input/output controller 3020 via a parallel port, a serial port, a keyboard port, a mouse port, etc.

The programs are provided by a computer-readable storage medium such as a DVD-ROM 3001 or an IC card. The programs are read from the computer-readable storage medium, installed in the hard disk drive 3024, RAM 3014, or ROM 3030, which are also examples of computer-readable storage media, and executed by the CPU 3012. The information processing described in these programs is read by the computer 3000, and brings about cooperation between the programs and the various types of hardware resources described above. An apparatus or method may be constructed by realizing the operation or processing of information according to the use of the computer 3000.

For example, when communication is performed between computer 3000 and an external device, CPU 3012 may execute a communication program loaded into RAM 3014 and instruct communication interface 3022 to perform communication processing based on the processing described in the communication program. Under the control of CPU 3012, communication interface 3022 reads transmission data stored in a transmission buffer area provided in RAM 3014, hard disk drive 3024, DVD-ROM 3001, or a recording medium such as an IC card, and transmits the read transmission data to the network, or writes received data received from the network to a reception buffer area or the like provided on the recording medium.

The CPU 3012 may also cause all or a necessary portion of a file or database stored on an external recording medium such as the hard disk drive 3024, the DVD-ROM drive 3026 (DVD-ROM 3001), an IC card, etc. to be read into the RAM 3014, and perform various types of processing on the data on the RAM 3014. The CPU 3012 may then write back the processed data to the external recording medium.

Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and may undergo information processing. The CPU 3012 may execute various types of processing on the data read from the RAM 3014, including various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, information search/replacement, etc., as described throughout this disclosure and specified by the instruction sequence of the program, and write back the results to the RAM 3014. The CPU 3012 may also search for information in a file, database, etc. in the recording medium. For example, when a plurality of entries each having an attribute value of a first attribute associated with an attribute value of a second attribute are stored in the recording medium, the CPU 3012 may search for an entry whose attribute value of the first attribute matches a specified condition from among the plurality of entries, read the attribute value of the second attribute stored in the entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

The above-described programs or software modules may be stored in a computer-readable storage medium on the computer 3000 or in the vicinity of the computer 3000. In addition, a recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can be used as a computer-readable storage medium, thereby providing the above-described programs to the computer 3000 via the network.

The blocks in the flowcharts and block diagrams in the above embodiments may represent stages of a process in which an operation is performed or "parts" of a device responsible for performing the operation. Particular stages and "parts" may be implemented by dedicated circuitry, programmable circuitry provided with computer-readable instructions stored on a computer-readable storage medium, and/or a processor provided with computer-readable instructions stored on a computer-readable storage medium. Dedicated circuitry may include digital and/or analog hardware circuitry, and may include integrated circuits (ICs) and/or discrete circuits. Programmable circuitry may include reconfigurable hardware circuitry including AND, OR, XOR, NAND, NOR, and other logical operations, flip-flops, registers, and memory elements, such as, for example, field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), and the like.

A computer-readable storage medium may include any tangible device capable of storing instructions that are executed by a suitable device, such that a computer-readable storage medium having instructions stored thereon comprises an article of manufacture that includes instructions that can be executed to create means for performing the operations specified in the flowchart or block diagram. Examples of computer-readable storage media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, and the like. More specific examples of computer-readable storage media may include floppy disks, diskettes, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), electrically erasable programmable read-only memories (EEPROMs), static random access memories (SRAMs), compact disk read-only memories (CD-ROMs), digital versatile disks (DVDs), Blu-ray disks, memory sticks, integrated circuit cards, and the like.

The computer readable instructions may include either assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk (registered trademark), JAVA (registered trademark), C++, etc., and conventional procedural programming languages such as the "C" programming language or similar programming languages.

The computer-readable instructions may be provided to a processor of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, or a programmable circuit, either locally or over a local area network (LAN), a wide area network (WAN), such as the Internet, to cause the processor of the general-purpose computer, special-purpose computer, or other programmable data processing apparatus, or a programmable circuit, to execute the computer-readable instructions to generate means for performing the operations specified in the flowcharts or block diagrams. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc.

The following provides examples to explain this technology in detail. Note that this technology is not limited to the following examples.

[Example 1]
(Creating training data)
First, a material database was constructed according to the following procedure. Based on 17 previous studies of organic positive electrode active materials (referred to as References 1 to 17), literature values for the average reaction potential [V], positive electrode capacity per unit mass [mAh/g], and mass energy density [Wh/g] were collected for 38 types of low molecular weight compounds, including 26 types of quinone-based organic compounds, 2 types of disulfide-based organic compounds, and 5 types of imine-based organic compounds. In addition, for each of the above 38 types of low molecular weight compounds, values for 46 types of explanatory variables explained in relation to the above-mentioned first characteristic group were prepared by DFT calculation, Hansen solubility parameter calculation, etc. The structural formulas of the 38 types of low molecular weight compounds (referred to as Compounds 1 to 38) were as follows.

Furthermore, References 1 to 17 were as follows.
[Reference 1] MM Doeff, SJ Visco, LC De Jonghe, J. Appl. Electrochem, 1992, 22, 307.
[Reference 2] SR Deng, LB Kong, GQ Hu, T. Wu, D. Li, YH Zhou, ZY Li, Electrochemica Acta, 2006, 51, 2589.
[Reference 3] M. Lee, J. Hong, DH Seo, DH Nam, KT Nam, K. Kang, CB Park, Angew. Chem. Int. Ed, 2013, 52, 8322.
[Reference 4] S. Nishide, Y. Yamamoto, T. Takui, Y. Morita, ChemSusChem, 2013, 6, 794.
[Reference 5] T. Yokoji, Y. Kameyama, S. Sakaida, N. Maruyama, M. Satoh, H. Matsubara, Chem. Lett, 2015, 44, 1726.
[Reference 6] M. Yao, H. Senoh, S. Yamazaki, Z. Siroma, T. Sakai, K. Yasuda, Journal of Power Sources, 2010, 195, 8336.
[Reference 7] T. Yokoji, Y. Kameyama, N. Maruyama, H. Matsubara, J. Mater. Chem. A, 2016, 4, 5457.
[Reference 8] Y. Liang, P. Zhang, J. Chen, Chem. Sci, 2013, 4, 1330.
[Reference 9] Y. Liang, P. Zhang, S. Yang, Z. Tao, J. Chen, Adv. Energy Mater, 2013, 3, 600.
[Reference 10] Z. Luo, L. Liu, Q. Zhao, F. Li, J. Chen, Angew. Chem. Int. Ed, 2017, 56, 12561.
[Reference 11] A. Shimizu, H. Kuramoto, Y. Tsujii, T. Nokami, Y. Inatomi, N. Hojo, H. Suzuki, J. Yoshida, Journal of Power Sources, 2014, 260, 211.
[Reference 12] W. Wan, H. Lee, X. Yu, C. Wang, KW Nam, XQ Yang, H. Zhou, RSC Adv, 2014, 4, 19878.
[Reference 13] T. Yokoji, H. Matsubara, M. Satoh, J. Mater. Chem. A, 2014, 2, 19347.
[Reference 14] S. Gottis, AL Barres, F. Dolhem, P. Poizot, ACS Appl. Mater. Interfaces, 2014, 6, 10870.
[Reference 15] R. Zeng, L. Xing, Y. Qiu, Y. Wang, W. Huang, W. Li, S. Yang, Electrochemica Acta, 2014, 146, 447.
[Reference 16] M. Yao, S. Yamazaki, H. Senoh, T. Sakai, T. Kiyobayashi, Mater. Sci. Engineering B, 2012, 177, 483.
[Reference 17] B. Tian, Z. Ding, GH Ning, W. Tang, C. Peng, B. Liu, J. Su, C. Su, KP Loh, Chem. Commun, 2017, 53, 2914.

Next, based on the correlations among the 46 explanatory variables, the values of 28 explanatory variables were deleted. As a result, a data set was constructed in which, for each of the 38 low molecular weight compounds mentioned above, (i) literature values for the average reaction potential [V], positive electrode capacity per unit mass [mAh/g], and mass energy density [Wh/g] were associated with (ii) the values of the 18 explanatory variables described in relation to the second group of features mentioned above.

(Building a predictive model)
Next, the explanatory variables were reduced by performing sparse modeling based on the full state search method using the above dataset as training data. In the sparse modeling, data on about 30 types of low molecular weight compounds out of the data on 38 types of low molecular weight compounds were used as training data, and data on about 10 types of low molecular weight compounds were used as validation data.

Next, taking into consideration chemical considerations and ease of data acquisition, the explanatory variables extracted by sparse modeling were further reduced. As a result, (a) the energy of the LUMO level of the molecules of the material, and (b) the energy of the level one level higher than the LUMO level of the molecules of the material were extracted as descriptors of the prediction model of the average reaction potential [V]. In addition, (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule, were extracted as descriptors of the prediction model of the positive electrode capacity per unit mass [mAh/g]. Furthermore, as descriptors of the prediction model for mass energy density [Wh/g], (a) the energy of the LUMO level of the molecules of the material, (b) the energy of the level one level higher than the LUMO level of the molecules of the material, (c) the number of reactions between the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material that have an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule.

Next, machine learning of a linear regression model consisting of the extracted descriptors was performed using the above-mentioned dataset for each prediction model of the average reaction potential [V], the positive electrode capacity per unit mass [mAh/g], and the mass energy density [Wh/g]. At this time, data on 38 types of low molecular weight compounds was used as training data.

[Example 2]
When constructing the prediction model, instead of reducing the explanatory variables by performing sparse modeling based on the full state search method, sparse modeling based on Lasso regression (LASSO) was performed to reduce the explanatory variables, and the same procedures as in Example 1 were used to construct prediction models for the average reaction potential [V], the positive electrode capacity per unit mass [mAh / g], and the mass energy density [Wh / g].

[Example 3]
When constructing the prediction models, instead of reducing the explanatory variables by performing sparse modeling based on the full state search method, sparse modeling based on multiple linear regression (MLR) was performed to reduce the explanatory variables. Except for this, a prediction model was constructed for each of the average reaction potential [V], the positive electrode capacity per unit mass [mAh/g], and the mass energy density [Wh/g] by the same procedure as in Example 1.

[Example 4]
When constructing the prediction models, instead of reducing the explanatory variables by performing sparse modeling based on the full state search method, sparse modeling based on random forest regression (RF) was performed to reduce the explanatory variables. Except for this, a prediction model was constructed for each of the average reaction potential [V], the positive electrode capacity per unit mass [mAh/g], and the mass energy density [Wh/g] by the same procedure as in Example 1.

[Example 5]
When constructing the prediction models, instead of reducing the explanatory variables by performing sparse modeling based on the full state search method, sparse modeling based on neural network regression (NN) was performed to reduce the explanatory variables. Except for this, a prediction model was constructed for each of the average reaction potential [V], the positive electrode capacity per unit mass [mAh/g], and the mass energy density [Wh/g] by the same procedure as in Example 1.

[Example 6]
When constructing the prediction models, instead of reducing the explanatory variables by performing sparse modeling based on the full state search method, sparse modeling based on support vector regression (SVR) was performed to reduce the explanatory variables. Except for this, a prediction model was constructed for each of the average reaction potential [V], the positive electrode capacity per unit mass [mAh/g], and the mass energy density [Wh/g] by the same procedure as in Example 1.

[Evaluation of predictive models]
(Creating an evaluation dataset)
A data set for evaluating the predictive model was created according to the following procedure. First, based on 21 previous studies different from the 17 previous studies used to create the teacher data, literature values for the average reaction potential [V], positive electrode capacity per unit mass [mAh/g], and mass energy density [Wh/g] were collected for 22 types of low molecular weight compounds, including 15 types of quinone organic compounds and 6 types of imine organic compounds. The 22 types of low molecular weight compounds are different from the 38 types of low molecular weight compounds described above.

For each of the 22 types of low molecular weight compounds, values for four explanatory variables were prepared by DFT calculation, Hansen solubility parameter calculation, etc.: (a) the energy of the LUMO level of the molecules of the material; (b) the energy of the level one level higher than the LUMO level of the molecules of the material; (c) the number of reactions of the material and the metal ions of the secondary battery derived from the molecular structure of the material, which is the number of reactions per molecule; and (d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ions of the secondary battery, which is the number of orbitals per molecule. As a result, an evaluation dataset was constructed in which literature values for (i) the average reaction potential [V], the positive electrode capacity per unit mass [mAh/g], and the mass energy density [Wh/g] were associated with (ii) the values of the four explanatory variables for each of the 22 types of low molecular weight compounds.

(Evaluation of the prediction model of Example 1)
For each of the 22 types of low molecular weight compounds included in the evaluation dataset, the predicted values of the average reaction potential, the positive electrode capacity per unit mass, and the mass energy density were calculated by inputting the values of the explanatory variables included in the evaluation dataset into each prediction model constructed in Example 1. Also, for reference, for each of the 38 types of low molecular weight compounds included in the teacher data, the predicted values of the average reaction potential, the positive electrode capacity per unit mass, and the mass energy density were calculated by inputting the values of the explanatory variables included in the teacher data into each prediction model constructed in Example 1.

Figure 11 shows the relationship between the predicted and measured average reaction potential. Figure 12 shows the relationship between the predicted and measured positive electrode capacity per unit mass. Figure 13 shows the relationship between the predicted and measured mass energy density. In Figures 11 to 13, the vertical axis shows the predicted values, and the horizontal axis shows the measured values (literature values). In Figures 11 to 13, circular markers show the relationship between the predicted and measured values when teacher data is used, and triangular markers show the relationship between the predicted and measured values when evaluation data is used.

When the predicted values of the average reaction potential, the positive electrode capacity per unit mass, and the mass energy density were compared with the actual measured values included in the evaluation dataset, the root mean square error (RMSE) of the average reaction potential was 0.240 [V]. The root mean square error (RMSE) of the positive electrode capacity per unit mass was 52.2 [mAh/g]. The root mean square error (RMSE) of the mass energy density was 0.160 [Wh/g].

The RMSE for the evaluation data set (and the test data) was 0.293 [V] 100 [mAh/g] and 0.263 [Wh/g], respectively. This shows that the prediction model of Example 1 has sufficient prediction accuracy.

In addition, as shown in Figures 11 to 13, it can be seen that accurate predictions can be made even for evaluation data that exists in areas where no training data exists. This demonstrates that the prediction model of Example 1 has extrapolation capabilities.

(Evaluation of Prediction Models of Examples 2 to 6)
For the prediction models of Examples 2 to 6, the predicted values of the average reaction potential, the positive electrode capacity per unit mass, and the mass energy density were calculated for each of the 22 types of low molecular weight compounds included in the evaluation dataset using the same procedure as for the prediction model of Example 1. In addition, for reference, the predicted values of the average reaction potential, the positive electrode capacity per unit mass, and the mass energy density were calculated for each of the 38 types of low molecular weight compounds included in the teacher data.

For the results of each prediction model in Examples 2 to 6, the root mean square error (RMSE) of the average reaction potential, the positive electrode capacity per unit mass, and the mass energy density was calculated using the same procedure as for the prediction model in Example 1 when the evaluation data set was used. In addition, the root mean square error (RMSE) of the average reaction potential, the positive electrode capacity per unit mass, and the mass energy density was calculated when the training data was used.

Figure 14 shows the RMSE of the average reaction potential in Examples 1 to 6. Figure 15 shows the RMSE of the positive electrode capacity per unit mass in Examples 1 to 6. Figure 16 shows the RMSE of the mass energy density in Examples 1 to 6. In Figures 14 to 16, the vertical axis shows the magnitude of the RMSE of each characteristic. In Figures 14 to 16, the bar graphs on the left side of each Example show the RMSE when the training data was used, and the bar graphs on the right side of each Example show the RMSE when the evaluation dataset was used.

As shown in Figures 14 to 16, it can be seen that the prediction models of Examples 2 to 6 also have sufficient prediction accuracy. Furthermore, a person skilled in the art who has read the description of this specification can understand that, taking into account Examples 1 to 3, a relatively accurate prediction model can also be obtained using other linear regression methods such as MCP regression. Similarly, a person skilled in the art who has read the description of this specification can understand that, taking into account Examples 4 to 6, a relatively accurate prediction model can also be obtained using other non-linear regression methods such as logistic regression.

In particular, it can be seen that the method of Example 1 has a smaller RMSE for all of the average reaction potential, positive electrode capacity per unit mass, and mass energy density, and the difference between the RMSE for the training data and the RMSE for the evaluation data set is smaller, compared to the methods of the other examples. Thus, the method of Example 1 can construct a prediction model with excellent generalization performance for all of the average reaction potential, capacity per unit mass, and mass energy density.

Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It is clear to those skilled in the art that various modifications or improvements can be made to the above embodiments. Furthermore, the matters described for a specific embodiment can be applied to other embodiments to the extent that there is no technical contradiction. Furthermore, each component may have the same characteristics as other components with the same name but different reference numbers. It is clear from the claims that forms with such modifications or improvements can also be included in the technical scope of the present invention.

The order of execution of each process, such as operations, procedures, steps, and stages, in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not specifically stated as "before" or "prior to," and it should be noted that the processes may be performed in any order, unless the output of a previous process is used in a later process. Even if the operational flow in the claims, specifications, and drawings is explained using "first," "next," etc. for convenience, it does not mean that it is necessary to perform the processes in that order.

100 Material exploration system, 112 Control unit, 114 Input unit, 116 Output unit, 118 Storage unit, 120 Database construction unit, 140 Prediction model construction unit, 142 Data set generation unit, 144 Sparse modeling unit, 146 Descriptor adjustment unit, 148 Reconstruction unit, 160 Property prediction unit, 162 Target material determination unit, 164 Target property determination unit, 166 Prediction value output unit, 220 Material database, 222 Material ID, 224 Material ID, 226 Information, 228 Information, 240 Model database, 242 Model ID, 244 Object variable ID, 246 Information, 288 Information, 322 Material data registration unit, 324 Material data calculation unit, 420 Training data set, 422 Data set, 424 Data set, 426 Data set, 440 Verification data set, 442 Data set, 444 Data set, 446 Data set, 460 Reconstruction data set, 462 Data set, 464 Data set, 466 Data set, 642 Construction unit, 644 Verification unit, 646 Output unit, 720 Adjustment data set, 746 Descriptor adjustment unit, 3000 Computer, 3001 DVD-ROM, 3010 Host controller, 3012 CPU, 3014 RAM, 3016 Graphic controller, 3018 Display device, 3020 Input/output controller, 3022 Communication interface, 3024 Hard disk drive, 3026 DVD-ROM drive, 3030 ROM, 3040 Input/output chip, 3042 Keyboard

Claims (18)

An information processing device for predicting characteristics of a secondary battery based on characteristics of an electrode material of the secondary battery,
a material determination unit that determines a target material that is the electrode material that is the target of the prediction;
a characteristic determination unit that determines a target characteristic, which is the characteristic that is a target of the prediction;
an input value acquisition unit that acquires, in a case where the target characteristic is a reaction potential of a positive electrode when the target material is used as a positive electrode active material of the secondary battery, (i) an energy value of a LUMO level of a molecule of the target material, and (ii) an energy value of a level one level higher than the LUMO level of the molecule of the target material;
a prediction unit that inputs the values acquired by the input value acquisition unit into a prediction model and outputs a predicted value of the target property of the target material;
Equipped with
The predictive model is
The objective variable is the potential of the positive electrode when the material is used as a positive electrode active material of a secondary battery,
(a) the energy of a LUMO level of a molecule of the material, and (b) the energy of a level one level above the LUMO level of a molecule of the material,
The regression model,
Information processing device. An information processing device for predicting characteristics of a secondary battery based on characteristics of an electrode material of the secondary battery,
a material determination unit that determines a target material that is the electrode material that is the target of the prediction;
a characteristic determination unit that determines a target characteristic, which is the characteristic that is a target of the prediction;
when the target characteristic is a capacity when the target material is used as a positive electrode active material of the secondary battery, (iii) an input value acquisition unit that acquires a value per molecule of the number of reactions between the target material and a metal ion of the secondary battery derived from a molecular structure of the target material, and (iv) a value per molecule of the number of orbitals of the molecules of the target material that have an energy level equal to or lower than a work function of a metal ion of the secondary battery;
a prediction unit that inputs the values acquired by the input value acquisition unit into a prediction model and outputs a predicted value of the target property of the target material;
Equipped with
The predictive model is
The capacity of the material when it is used as a positive electrode active material of a secondary battery is set as the objective variable,
(c) the number of reactions between the material and the metal ions of the secondary battery derived from the molecular structure of the material, the number of reactions per molecule; and (d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ions of the secondary battery, the number of orbitals per molecule being included as a descriptor.
The regression model,
Information processing device. An information processing device for predicting characteristics of a secondary battery based on characteristics of an electrode material of the secondary battery,
a material determination unit that determines a target material that is the electrode material that is the target of the prediction;
a characteristic determination unit that determines a target characteristic, which is the characteristic that is a target of the prediction;
an input value acquisition unit that acquires, in the case where the target characteristic is an energy density when the target material is used as a positive electrode active material of the secondary battery, (i) an energy value of a LUMO level of a molecule of the target material, (ii) an energy value of the level one level higher than the LUMO level of the molecule of the target material, (iii) a value per molecule of the number of reactions between the target material and a metal ion of the secondary battery derived from a molecular structure of the target material, and (iv) a value per molecule of the number of orbitals of the molecule of the target material having an energy level equal to or lower than a work function of a metal ion of the secondary battery;
a prediction unit that inputs the values acquired by the input value acquisition unit into a prediction model and outputs a predicted value of the target property of the target material;
Equipped with
The predictive model is
The energy density when the material is used as a positive electrode active material of the secondary battery is set as an objective variable,
(a) energy of a LUMO level of a molecule of the material, (b) energy of a level one level higher than the LUMO level of a molecule of the material, (c) the number of reactions between the material and a metal ion of the secondary battery derived from a molecular structure of the material, the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material having an energy level equal to or lower than a work function of a metal ion of the secondary battery, the number of orbitals per molecule being included as a descriptor.
The regression model,
Information processing device. The regression model is represented by a linear regression equation or a nonlinear regression equation.
The information processing device according to any one of claims 1 to 3. A model construction unit that constructs the predictive model by machine learning;
Further equipped with
The model construction unit includes:
a first variable selection unit that selects M explanatory variables (M is a positive integer of 1 to N) from N explanatory variables (N is a positive integer of 2 or more) each of which represents a characteristic of the material, as the descriptor or candidates for the descriptor of the regression model;
having
The information processing device according to any one of claims 1 to 4. The first variable selection unit selects the M explanatory variables by a sparse modeling method.
The information processing device according to claim 5 . The first variable selection unit further includes:
determining a regression coefficient or partial regression coefficient corresponding to each of the M explanatory variables;
7. The information processing device according to claim 5 or 6. the material is an organic molecule or a salt thereof;
The N explanatory variables are:
(a) the energy of the LUMO level of the molecules of the material;
(b) the energy of a level one level above the LUMO level of the molecules of the material;
(c) the number of reactions of the material and metal ions of the secondary battery derived from the molecular structure of the material, the number of reactions per molecule; and
(d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ion of the secondary battery, the number of orbitals per molecule;
Including,
The N explanatory variables are:
(e) the variance term of the Hansen solubility parameters of the material and the electrolyte of the secondary battery;
(f) the polarization term of the Hansen solubility parameter;
(g) the hydrogen bond term of the Hansen solubility parameter;
(h) the dipole moment of the molecules of said material;
(i) a minimum partial charge density on the molecules of the material; and
(j) the number of elements other than carbon, hydrogen, and oxygen contained in the molecules of the material, which is the number of elements per molecule;
Further comprising at least one selected from the group consisting of:
The information processing device according to any one of claims 5 to 7. The N explanatory variables are:
(k) the energy of the level two levels above the LUMO level of the molecules of the material;
(l) the energy of the level three levels above the LUMO level of the molecules of the material;
(m) the energy of the level four levels above the LUMO level of the molecules of the material;
(n) the molecular weight of the molecule of the material;
(o) the number of carbon atoms conjugated by π bonds contained in the molecules of the material, which is the number of carbon atoms per molecule;
(p) the number of orbitals having an energy level equal to or less than 0 among the orbitals of the molecules of the material, the number of orbitals per molecule;
(q) the absolute value of the sum of the energy levels equal to or less than zero per molecule of the material; and
(r) the maximum partial charge density of the molecules of the material;
Further comprising:
The information processing device according to claim 8. the material is an organic molecule or a salt thereof;
The M explanatory variables are:
(a) the energy of the LUMO level of the molecules of the material;
(b) the energy of a level one level above the LUMO level of the molecules of the material;
(c) the number of reactions of the material and metal ions of the secondary battery derived from the molecular structure of the material, the number of reactions per molecule; and
(d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ion of the secondary battery, the number of orbitals per molecule;
At least two selected from
The information processing device according to any one of claims 5 to 9. The model construction unit includes:
a second variable selection unit that selects, based on a user's instruction, K explanatory variables (K is a positive integer not less than 1 and not more than M) from among the M explanatory variables as the descriptors;
Further comprising
The information processing device according to any one of claims 5 to 10. The model construction unit includes:
a third variable selection unit that selects N explanatory variables as candidates for the descriptor from among L explanatory variables (L is a positive integer equal to or greater than N) each of which represents a characteristic of the material;
and
The third variable selection unit is
determining, for each of the L explanatory variables, a degree of correlation with the other explanatory variables;
Selecting N explanatory variables from the L explanatory variables by leaving one explanatory variable out of a plurality of explanatory variables whose degree of correlation is stronger than a predetermined standard and excluding the other explanatory variables;
The information processing device according to any one of claims 5 to 11. An information processing device that constructs a prediction model for predicting characteristics of a secondary battery based on characteristics of an electrode material of the secondary battery,
the prediction model is a regression model in which the characteristic of the secondary battery is a response variable and one or more features related to the electrode material are each a descriptor;
The information processing device includes:
a first variable selection unit that selects M explanatory variables (M is a positive integer of 1 to N) from N explanatory variables (N is a positive integer of 2 or more) each of which represents a characteristic of a material, as the descriptor or candidates for the descriptor of the regression model;
Equipped with
The first variable selection unit
A group of teacher data is input in which, for each of a plurality of materials, a value of one characteristic related to the secondary battery when the material is used as the electrode material is associated with values of N explanatory variables related to features of the material;
selecting the M explanatory variables from the N explanatory variables by performing a regression analysis based on machine learning;
The N explanatory variables are:
(a) the energy of the LUMO level of the molecules of the material;
(b) the energy of a level one level above the LUMO level of the molecules of the material;
(c) the number of reactions of the material and metal ions of the secondary battery derived from the molecular structure of the material, the number of reactions per molecule; and
(d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ion of the secondary battery, the number of orbitals per molecule;
Including,
The N explanatory variables are:
(e) the variance term of the Hansen solubility parameters of the material and the electrolyte of the secondary battery;
(f) the polarization term of the Hansen solubility parameter;
(g) the hydrogen bond term of the Hansen solubility parameter;
(h) the molecular dipole moment of said material;
(i) a minimum partial charge density on the molecules of the material; and
(j) the number of elements other than carbon, hydrogen, and oxygen contained in the molecules of the material, which is the number of elements per molecule;
Further comprising at least one selected from the group consisting of:
Information processing device. A program for causing a computer to function as an information processing device according to any one of claims 1 to 13. 1. An information processing method for predicting characteristics of a secondary battery based on characteristics of an electrode material of the secondary battery, comprising:
A material determination step in which a computer determines a target material, which is the electrode material that is the target of the prediction;
a characteristic determination step in which the computer determines a target characteristic, the characteristic being the subject of the prediction;
When the target characteristic is a potential of a positive electrode when the target material is used as a positive electrode active material of the secondary battery, the computer acquires (i) an energy value of a LUMO level of a molecule of the target material, and (ii) an energy value of a level one level higher than the LUMO level of the molecule of the target material;
a prediction step in which the computer inputs the values acquired in the input value acquisition step into a prediction model and outputs a predicted value of the target property of the target material;
having
The predictive model is
The objective variable is the potential of the positive electrode when the material is used as a positive electrode active material of a secondary battery,
(a) the energy of a LUMO level of a molecule of the material, and (b) the energy of a level one level above the LUMO level of a molecule of the material,
The regression model,
Information processing methods. 1. An information processing method for predicting characteristics of a secondary battery based on characteristics of an electrode material of the secondary battery, comprising:
A material determination step in which a computer determines a target material, which is the electrode material that is the target of the prediction;
a characteristic determination step in which the computer determines a target characteristic, the characteristic being the subject of the prediction;
When the target characteristic is a capacity when the target material is used as a positive electrode active material of the secondary battery, the computer (iii) acquires a value per molecule of the number of reactions between the target material and a metal ion of the secondary battery derived from a molecular structure of the target material, and (iv) an input value acquisition step of acquiring a value per molecule of the number of orbitals of the molecules of the target material having an energy level equal to or lower than a work function of a metal ion of the secondary battery;
a prediction step in which the computer inputs the values acquired in the input value acquisition step into a prediction model and outputs a predicted value of the target property of the target material;
having
The predictive model is
The capacity of the material when it is used as a positive electrode active material of a secondary battery is set as the objective variable,
(c) the number of reactions between the material and the metal ions of the secondary battery derived from the molecular structure of the material, the number of reactions per molecule; and (d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ions of the secondary battery, the number of orbitals per molecule being included as a descriptor.
The regression model,
Information processing methods. 1. An information processing method for predicting characteristics of a secondary battery based on characteristics of an electrode material of the secondary battery, comprising:
A material determination step in which a computer determines a target material, which is the electrode material that is the target of the prediction;
a characteristic determination step in which the computer determines a target characteristic, the characteristic being the subject of the prediction;
an input value acquisition step in which, when the target characteristic is an energy density when the target material is used as a positive electrode active material of the secondary battery, the computer acquires (i) an energy value of a LUMO level of a molecule of the target material, (ii) an energy value of the level one level higher than the LUMO level of the molecule of the target material, (iii) a value per molecule of the number of reactions between the target material and a metal ion of the secondary battery derived from a molecular structure of the target material, and (iv) a value per molecule of the number of orbitals of the molecule of the target material having an energy level equal to or lower than a work function of a metal ion of the secondary battery;
a prediction step in which the computer inputs the values acquired in the input value acquisition step into a prediction model and outputs a predicted value of the target property of the target material;
having
The predictive model is
The energy density when the material is used as a positive electrode active material of a secondary battery is set as the objective variable,
(a) energy of a LUMO level of a molecule of the material, (b) energy of a level one level higher than the LUMO level of a molecule of the material, (c) the number of reactions between the material and a metal ion of the secondary battery derived from a molecular structure of the material, the number of reactions per molecule, and (d) the number of orbitals of the molecules of the material having an energy level equal to or lower than a work function of a metal ion of the secondary battery, the number of orbitals per molecule being included as a descriptor.
The regression model,
Information processing methods. 1. An information processing method for constructing a prediction model for predicting characteristics of a secondary battery based on characteristics of an electrode material of the secondary battery, comprising:
the prediction model is a regression model in which the characteristic of the secondary battery is a response variable and one or more features related to the electrode material are each a descriptor;
The information processing method includes:
a first variable selection step in which a computer selects M explanatory variables (M is a positive integer between 1 and N) from N explanatory variables (N is a positive integer between 2 and N) each of which represents a characteristic of a material, as the descriptor or candidates for the descriptor of the regression model;
having
The first variable selection step comprises:
a step of the computer acquiring a group of teacher data in which a value of one characteristic of the secondary battery when the material is used as the electrode material is associated with values of N explanatory variables related to features of the material, for each of a plurality of materials;
selecting the M explanatory variables from the N explanatory variables by performing a regression analysis based on machine learning by a computer;
Including,
The N explanatory variables are:
(a) the energy of the LUMO level of the molecules of the material;
(b) the energy of the level one level above the LUMO level of the molecules of the material;
(c) the number of reactions of the material and metal ions of the secondary battery derived from the molecular structure of the material, the number of reactions per molecule; and
(d) the number of orbitals of the molecules of the material having an energy level equal to or lower than the work function of the metal ion of the secondary battery, the number of orbitals per molecule;
Including,
The N explanatory variables are:
(e) the variance term of the Hansen solubility parameters of the material and the electrolyte of the secondary battery;
(f) the polarization term of the Hansen solubility parameter;
(g) the hydrogen bond term of the Hansen solubility parameter;
(h) the molecular dipole moment of said material;
(i) a minimum partial charge density on the molecules of the material; and
(j) the number of elements other than carbon, hydrogen, and oxygen contained in the molecules of the material, which is the number of elements per molecule;
Information processing methods.