Evaluation device, evaluation method, and program

Abstract

This improves the efficiency of performance evaluation for energy storage devices with a three-dimensional electrode structure. [Solution] The evaluation device is an evaluation device for evaluating the performance of an energy storage device having a three-dimensional electrode structure, and comprises a control unit that performs an evaluation process to evaluate the performance of the energy storage device to be evaluated using a regression model constructed from the relationship between a structure matrix representing the electrode structure of an energy storage device with known performance and its performance, and using the structure matrix representing the electrode structure of the energy storage device to be evaluated as an explanatory variable.

Description

[Technical Field]

[0001] This disclosure relates to an evaluation device, an evaluation method, and a program. [Background technology]

[0002] Conventionally, we have been studying the optimization of the structure of energy storage devices with three-dimensional electrode structures. In order to evaluate the internal resistance during optimization, we have proposed a three-dimensional porous electrode model (3D-PEM), which is a transmission line model that takes into account the influence of the three-dimensional electrode structure (see, for example, Patent Document 1 and Non-Patent Document 1). Because the transmission line model replaces the complex reactions within the battery with a simple circuit diagram, it is possible to evaluate the internal resistance at a faster rate compared to so-called battery simulations that use a continuum model to solve the time evolution of reactions within the battery. We have also proposed introducing current dependence in the evaluation of internal resistance using such a transmission line model to improve the evaluation accuracy (see, for example, Non-Patent Document 2). Furthermore, we have proposed representing the three-dimensional electrode structure with a matrix called a geometry matrix and using it to evaluate the internal resistance (see, for example, Non-Patent Document 1), and evaluating the cell energy of the energy storage device using features extracted from this geometry matrix and internal resistance as feature quantities (explanatory variables) (see, for example, Non-Patent Document 3). [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] U.S. Patent No. 11568102 [Non-patent literature]

[0004] [Non-Patent Document 1] Miyamoto et al., iScience, 23, 101317 (2020) [Non-Patent Document 2] Miyamoto, ACS Phys. Chem Au 2024, 4 546-554 [Non-Patent Document 3] Miyamoto et al., Cell Reports Physical Science 2, 100504 (2021) [Overview of the project] [Problems that the invention aims to solve]

[0005] However, while the above method can evaluate the performance of energy storage devices, such as internal resistance and cell energy, with high accuracy and efficiency, there was a need to further improve the efficiency of performance evaluation.

[0006] This invention was made to solve these problems, and its main objective is to further improve the efficiency of performance evaluation of energy storage devices having a three-dimensional electrode structure. [Means for solving the problem]

[0007] To achieve the above-mentioned objectives, the inventors conducted diligent research. They acquired performance data for multiple three-dimensional electrode structures (more than 200 patterns, for example) using transmission line models, and conceived the idea of ​​creating a regression model by machine learning the relationship between the structural matrix representing the three-dimensional electrode structure and performance data such as internal resistance and energy. They found that by using this regression model, the performance of the energy storage device under evaluation can be evaluated (predicted) simply by using the structural matrix representing the electrode structure as the explanatory variable, thereby further improving the efficiency of performance evaluation, and thus completed this disclosure.

[0008] In other words, the evaluation apparatus of this disclosure is An evaluation apparatus for evaluating the performance of an energy storage device having a three-dimensional electrode structure, A control unit that executes an evaluation process for evaluating the performance of a power storage device to be evaluated, using, as an explanatory variable, a structure matrix representing the electrode structure of the power storage device to be evaluated, by using a regression model constructed from the relationship between the structure matrix representing the electrode structure of a power storage device with known performance and the performance. comprises the above.

[0009] The evaluation method of the present disclosure is an evaluation method for evaluating the performance of a power storage device having a three-dimensional electrode structure, comprising a step of executing an evaluation process for evaluating the performance of the power storage device to be evaluated, using, as an explanatory variable, a structure matrix representing the electrode structure of the power storage device to be evaluated, by using a regression model constructed from the relationship between the structure matrix representing the electrode structure of a power storage device with known performance and the performance. includes the above.

[0010] The program of the present disclosure causes one or more computers to implement the steps of the above-described evaluation method.

Advantages of the Invention

[0011] In the present disclosure, since the performance can be evaluated (predicted) simply by applying the structure matrix to a previously prepared regression model, the efficiency of performance evaluation of a power storage device having a three-dimensional electrode structure can be further improved. Further, by using, as an explanatory variable, a structure matrix that holds all the information of the three-dimensional electrode structure, the relationship between the structural characteristics and the performance can be accurately learned, and the performance can be evaluated with relatively high accuracy.

Brief Description of the Drawings

[0012] [Figure 1] Schematic explanatory diagram showing an example of an evaluation device 20. [Figure 2] Explanatory diagram showing an example of the structure of a power storage device 10. [Figure 3] Explanatory diagram of the dimensions and resolution of a three-dimensional electrode structure. [Figure 4] Example of generation of a structure matrix from a three-dimensional electrode structure. [Figure 5] Explanatory diagram showing an example of a three-dimensional electrode model. [Figure 6] A flowchart illustrating an example of the search process. [Figure 7] A diagram illustrating the procedure for creating a regression model and evaluating its accuracy. [Figure 8] Diagram illustrating the three-dimensional battery structure used as an example. [Modes for carrying out the invention]

[0013] (Evaluation device) Embodiments of the evaluation apparatus disclosed herein will be described below with reference to the drawings. Figure 1 is a schematic diagram showing an example of the evaluation apparatus 20. Figure 2 is a diagram showing an example of the structure of the energy storage device 10. Figure 3 is a diagram illustrating the dimensions and resolution of the three-dimensional electrode structure (referencing Figure S3 of Non-Patent Literature 2). Figure 4 is an example of generating a structure matrix from a three-dimensional electrode structure, where Figure 4A is an example of a three-dimensional electrode structure, Figure 4B is the structure of Figure 4A divided into 50 × 10 square elements (voxels), and Figure 4C is a structure matrix of the size (50 × 10) corresponding to the element division in Figure 4B. Figure 5 is a diagram showing an example of a three-dimensional electrode model, where Figure 5A is a conceptual diagram of a three-dimensional battery in which the positive and negative electrodes are not flat plates, and Figure 5B is a circuit diagram considering the structure of Figure 5A. The evaluation device 20 may be configured as a device that evaluates the performance of an energy storage device 10 under evaluation (hereinafter also referred to as energy storage device 10x) by using a regression model created by machine learning the relationship between the structural matrix representing the three-dimensional electrode structure of an energy storage device 10 with known performance (hereinafter also referred to as energy storage device 10a) and its internal resistance, and using the structural matrix representing the three-dimensional electrode structure of the energy storage device 10 as an explanatory variable.

[0014] The power storage device 10 includes, for example, a hybrid capacitor, a pseudo electric double layer capacitor, an alkali metal secondary battery such as lithium or sodium, an alkali metal ion battery, an air battery, a lead storage battery, and the like. Among these, as the power storage device 10, a secondary battery having an alkali metal ion as a carrier ion, particularly a lithium ion secondary battery, is preferable. Here, the power storage device 10 will be mainly described as being a lithium ion secondary battery. The power storage device 10 may include, for example, a positive electrode 12, a negative electrode 15, and an ion conductive medium 18 interposed between the positive electrode 12 and the negative electrode 15. The ion conductive medium 18 may be constituted by, for example, an electrolytic solution and a separator containing the electrolytic solution. Alternatively, the ion conductive medium 18 may be a solid electrolyte interposed between the positive electrode 12 and the negative electrode 15. The positive electrode 12 includes a current collector 14 and a positive electrode active material layer 13 formed on one surface of the current collector 14. The positive electrode active material layer 13 contains a positive electrode active material and may further contain a conductive material and a binder as required. Examples of the positive electrode active material include sulfides containing a transition metal element, composite oxides containing lithium and a transition metal element, and phosphate compounds containing iron. The positive electrode active material has, for example, a basic composition formula of Li (1-x) MnO2 (0 < x < 1, etc., the same below), Li (1-x) Mn2O4, etc., lithium manganese composite oxides, basic composition formulas of Li (1-x) CoO2, etc., lithium cobalt composite oxides, basic composition formulas of Li (1-x) NiO2, etc., lithium nickel composite oxides, basic composition formulas of Li (1-x) Ni a Co b Mn c O2 (a + b + c = 1), etc., lithium nickel cobalt manganese composite oxides, etc., can be used. Also, as the positive electrode active material, the basic composition formula is Li (1-x)Examples include lithium iron phosphate, such as FePO4. Note that the term "basic composition formula" means that other elements may also be included. The negative electrode 15 comprises a current collector 17 and a negative electrode active material layer 16 formed on one surface of the current collector 17. The negative electrode active material layer 16 contains a negative electrode active material and may further contain conductive materials and binders as needed. The negative electrode active material may include metallic lithium or its alloys, carbon materials, or lithium-containing composite oxides. Examples of negative electrode active materials include lithium, lithium alloys, inorganic compounds such as tin compounds, carbon materials capable of intercalating and deintercalating lithium ions, composite oxides containing multiple elements, and conductive polymers. Examples of carbon materials include coke, glassy carbons, graphites, non-graphitizable carbons, pyrolytic carbons, and carbon fibers. Of these, graphites such as artificial graphite and natural graphite are preferred. Examples of composite oxides include lithium titanium composite oxide and lithium vanadium composite oxide. The separator may contain, for example, an electrolyte containing a dissolved support salt. Examples of support salts include lithium salts such as LiPF6 and LiBF4. Examples of solvents for the electrolyte include carbonates, esters, ethers, nitriles, furans, sulforanes, and dioxolanes, which can be used individually or in combination. Specifically, examples of carbonates include cyclic carbonates such as ethylene carbonate, propylene carbonate, vinylene carbonate, butylene carbonate, and chloroethylene carbonate, as well as linear carbonates such as dimethyl carbonate, ethyl methyl carbonate, diethyl carbonate, ethyl-n-butyl carbonate, methyl-t-butyl carbonate, di-i-propyl carbonate, and t-butyl-i-propyl carbonate. Furthermore, the ion-conducting medium 18 can be a solid ion-conducting polymer, an inorganic solid electrolyte, a mixed material of an organic polymer electrolyte and an inorganic solid electrolyte, or an inorganic solid powder bound together by an organic binder.

[0015] The evaluation device 20 is a device that performs an evaluation process to evaluate the performance of an energy storage device 10 having a positive electrode 12 and a negative electrode 15. For example, the evaluation device 20 may take a structure matrix representing the electrode structure as an explanatory variable for an energy storage device 10x to be evaluated which has a predetermined electrode structure including the distribution of the positive electrode 12 and the negative electrode 15, and use a regression model 36 to output evaluation results of performance such as internal resistance and cell energy as the objective variable. The evaluation device 20 comprises a control unit 21, a storage unit 22, an input device 23, and a display device 24. The control unit 21 is configured as a microprocessor centered on a CPU and controls the entire device. The storage unit 22 is configured as a large-capacity storage device such as an HDD and stores a structure matrix 30 representing the electrode structure of an energy storage device 10a with known performance, known performance data 32, a machine learning program 34, a regression model 36, and an evaluation program 38. The memory unit 22 may store a structure generation program 40 for automatically generating the electrode structure of the energy storage device, and a structure matrix representation program 42 for representing the electrode structure as a structure matrix. The memory unit 22 may also store a data preparation program 44 for preparing known performance data 32 based on the electrode structure of the energy storage device, and a transmission line model 46 used for preparing the known performance data 32. The input device 23 includes a mouse, keyboard, etc., for various inputs. The display device 24 displays a screen, and is, for example, a liquid crystal display.

[0016] The control unit 21 uses a structure matrix representing the three-dimensional electrode structure of the energy storage device 10x under evaluation as an explanatory variable, and performs performance evaluation (evaluation process) of the energy storage device 10x under evaluation using a regression model 36. The "structure matrix (geometry matrix)" is a matrix that represents the three-dimensional electrode structure, as proposed in Non-Patent Literature 1. The structure matrix is ​​obtained by dividing the three-dimensional electrode structure (Figure 4A) into voxels (Figure 4B), as shown in Figure 4, and converting the divided electrode information into a matrix while maintaining its resolution (Figure 4C), thus retaining all electrode structure information. By using this structure matrix directly as a feature (explanatory variable), the relationship between structural features and performance can be accurately learned. Here, in Figure 4A, it is assumed that the structure in the depth direction is the same as the structure in the front. In Figure 4B, the electrode structure is divided into 50 × 10 voxels (the number of divisions is 50 × 10 = 500). Furthermore, in the structure matrix of Figure 4C, voxels corresponding to the positive electrode active material layer 13 (positive electrode block 25) are represented by 1, and voxels corresponding to the negative electrode active material layer 16 (negative electrode block 26) are represented by 3. From the viewpoint of improving evaluation accuracy, a larger number of divisions in the electrode structure is preferable, for example, 500 or more is preferable, 1000 or more is more preferable, and 2000 or more is even preferable. The structure matrix 30 can be any matrix that represents the three-dimensional electrode structure, and may be a two-dimensional matrix as in Figure 4C, or it may be a three-dimensional matrix that also includes information in the depth direction. The positive electrode block 25 and negative electrode block 26 do not have to be square (cubic) voxels, but may be rectangular (cuboid) blocks, for example. The "performance" to be evaluated may be, for example, 1 or more of the internal resistance and cell energy.

[0017] The structure matrix 30 is a structure matrix that represents the three-dimensional electrode structure of the energy storage device 10a with known performance, and is used as an explanatory variable when constructing the regression model 36. The structure matrix 30 may be created using a structure generation program 40 or a structure matrix conversion program 42. The structure generation program 40 may be, for example, a program that generates a three-dimensional electrode structure using the method disclosed in Non-Patent Literature 1, or it may be the Automatic Geometry Generator disclosed in Non-Patent Literature 1. The structure matrix conversion program 42 may be, for example, a program that represents a three-dimensional electrode structure as a structure matrix using the method shown in Figure 4, or it may be a program that converts a three-dimensional electrode structure into a structure matrix using the method disclosed in Non-Patent Literature 1.

[0018] The known performance data 32 is performance data of the energy storage device 10a with known performance, and is used as the target variable when constructing the regression model 36. The known performance data 32 may be, for example, one or more of the internal resistance and cell energy. The known performance data 32 may be derived using the data preparation program 44. The data preparation program 44 may be a program that derives performance data using the transmission line model 46. The data preparation program 44 may be, for example, a program that derives performance data using the transmission line model disclosed in any of Non-Patent Documents 1 to 3. The data preparation program 44 may also be a program that derives performance data using a model other than the transmission line model 46, for example, a continuum model.

[0019] The transmission line model 46 is constructed as a model that takes into account the effects of the three-dimensional battery structure, with the aim of evaluating the internal resistance of the energy storage device 10. This transmission line model 46 is a three-dimensional porous electrode model (3D-PEM) that replaces the complex reactions within the battery with a simple circuit diagram, and it is possible to perform evaluation of the energy storage device 10, such as its internal resistance, at high speed compared to so-called battery simulations that use a continuum model to solve the time evolution of reactions within the battery. By acquiring known performance data 32 using such a transmission line model, the known performance data 32 can be prepared more efficiently than by actual measurements or simulations using a continuum model, and the evaluation efficiency can be further improved.

[0020] As shown in Figure 5, the transmission line model 46 has a structure in which the positive electrode 12 and the negative electrode 15 are divided into blocks, and an electronic resistor 27, an interfacial resistor 29, and an ionic resistor 28 are connected to each block. This transmission line model 46 is defined as a two-layer circuit in which the electronic resistor 27 is connected between each block to the solid portion through which electrons conduct, and the ionic resistor 28 is connected between each block to the ion-conducting medium portion through which ions conduct. The electronic resistor 27 and the ionic resistor 28 are connected by the interfacial resistor 29 so that the movement of ions and electrons is switched by the electrode reaction. Furthermore, the transmission line model 46 has a structure in which the positive electrode block 25 and the negative electrode block 26, which are separated by the ion-conducting medium 18, are connected by the ionic resistor 28. This transmission line model 46 may also include a charge transfer resistor in which a reaction resistor and a film resistor are connected in series. In this regard, for example, immediately after discharge when the contribution of diffusion resistance is low, the interfacial resistance 29 largely depends on the charge transfer resistance, so the interfacial resistance 29 can be approximated as the charge transfer resistance and applied to the transmission line model 46 for treatment. For example, by reflecting the current-dependent reaction resistance and the film resistance in the charge transfer resistance, the internal resistance can be evaluated with higher evaluation accuracy. In the transmission line model 46, the black or white circles within the divided elements represent the centers of each element, and the centers of these elements are connected by resistors. Here, the black and white circles within the elements are called nodes. Because this model models a porous electrode, the inside of the electrode has a two-layer circuit structure consisting of an ion-conducting medium part through which ions flow, where the corresponding resistance is ion resistance, and a solid part through which electrons flow, where the corresponding resistance is electron resistance. Also, because the current carrier switches from ions to electrons due to the electrode reaction, the ion resistance 28 and the electron resistance 27 are connected by the interfacial resistance 29.

[0021] It should be noted that the interfacial resistance 29 is connected between the electronic resistance node and the ion resistance node, which are on the same coordinate system in Figure 5A. In lithium-ion batteries, for example, those using an electrolyte, the electrolyte is absorbed into the porous electrode. To accurately represent this, it is necessary to represent the structure in which the positive and negative electrodes in Figure 5A have a complex interplay of electrode portions and absorbed electrolyte portions at the micro level. Since it is difficult to handle this directly, this model eliminates the specific structural information, such as the interface between the electrode solid and the electrolyte, and approximates that the electrolyte and electrode material exist on average in the same space in a specified ratio. It should be noted that mainstream continuum models of lithium-ion batteries, including the reference documents such as Patent Document 1 and Non-Patent Document 1 mentioned above, also treat this similarly. Under this approximation, the coordinates of the centers (nodes) of each element in Figure 5A are the coordinates of a real battery (real-space coordinates), but the nodes for ion resistance and electronic resistance connected by the interfacial resistance in Figure 5B represent the same coordinates in real space. In other words, both the ion and electron nodes represent the same corresponding node in Figure 5A. Note that the interface between the electrode and electrolyte is defined by the area (and resistivity) of the interface between the electrolyte and the electrode, which is averaged within the element, and this information is not shown in Figure 5A because it is averaged.

[0022] The transmission line model described above has traditionally been used to evaluate internal resistance, but if the element is divided into many smaller parts to improve its accuracy, the number of divisions increases, which can lead to enormous computation time. Therefore, instead of using the transmission line model to evaluate the performance of all electrode structures, it is possible to evaluate the performance of a small number of electrode structures using the transmission line model, and then use a regression model created based on the evaluation results to evaluate the performance of the remaining electrode structures, thereby performing performance evaluation more efficiently.

[0023] The machine learning program 34 is executed by the control unit 21 and is a program that constructs a regression model 36 using machine learning from the structure matrix 30 and known performance data 32. The three-dimensional electrode structure used in machine learning may be, for example, 100 to 10,000 patterns, or 200 to 2,000 patterns. As for the machine learning method, regression methods such as Least absolute shrinkage and selection operator (LASSO), Support vector regression (SVR), and Random forest regression (RFR) can be suitably used. Of these, LASSO and RFR are preferred from the viewpoint of evaluation efficiency, with LASSO being more preferred. Also, from the viewpoint of evaluation accuracy, SVR and RFR are preferred, with SVR being more preferred. The hyperparameters of the regression model 36 can be determined by various optimization methods such as grid search, random search, and Bayesian optimization. The regression model 36 may be further optimized by cross-validation such as k-Fold Cross Validation. The coefficient of determination of the regression model 36 R 2 From the viewpoint of evaluation accuracy, a value of 0.6 or higher is preferable, 0.8 or higher is more preferable, and 0.9 or higher is even preferable.

[0024] The evaluation program 38 is executed by the control unit 21 and uses a regression model 36 to evaluate (predict) the performance of the energy storage device 10x under evaluation, using a structural matrix representing the electrode structure of the energy storage device under evaluation as an explanatory variable. Because this evaluation program 38 directly uses the structural matrix representing the electrode structure of the energy storage device 10x under evaluation as an explanatory variable to evaluate its performance, it eliminates the need for the cumbersome process of extracting explanatory variables that effectively represent the relationship between the objective variable and the three-dimensional electrode structure through trial and error. For example, Non-Patent Literature 3 predicts the energy of a three-dimensional battery using features and internal resistance extracted from the structural matrix as explanatory variables, requiring the extraction of explanatory variables through trial and error, but this disclosure eliminates such a process.

[0025] (Evaluation method) Next, the processing of the evaluation device 20 configured in this embodiment, in particular the evaluation method executed by the evaluation device 20, will be described. This evaluation method is a method for performing an evaluation process to evaluate the performance of an energy storage device 10x to be evaluated (performance unknown). For example, this evaluation method may evaluate performance such as internal resistance and cell energy using a regression model 36 with respect to an energy storage device 10x to be evaluated that has a predetermined electrode structure including the distribution of positive electrodes 12 and negative electrodes 15, using a structure matrix representing the electrode structure as an explanatory variable. This evaluation method includes the step of performing an evaluation process to evaluate the performance of an energy storage device 10x to be evaluated using a regression model 36 constructed from the relationship between a structure matrix 30 representing the electrode structure of an energy storage device 10x with known performance and performance (known performance data 32), and using a structure matrix representing the electrode structure of the energy storage device 10x to be evaluated as an explanatory variable.

[0026] The known performance data 32 used to construct the regression model 36 may be the internal resistance of a battery cell evaluated by the following procedure. For example, assuming that the internal resistance of the three-dimensional battery shown in Figure 5A is to be evaluated, first the electronic resistivity (ρ e [Ωcm] and ionic resistivity (ρ ion The [Ωcm] is determined from the electronic and ionic conductivity obtained from literature information or experiments. Also, the interfacial resistivity ρ ct [Ωcm 3 ] is determined from literature information or experiments. Next, the distance between the centers of each element in Figure 5B and the information of the cross-sectional area perpendicular to it, and ρ ion (ρ e Using ), a resistor R connects the centers of each element. ion [Ω](R e Calculate the [Ω]) interface resistance. Finally, the interface resistance R, represented by a square. ct to ρ ctThis is calculated using the volume of each element. This allows us to determine the resistance connecting all the element centers in Figure 5B. This resistance calculation method is described in detail in Non-Patent Literature 1 mentioned above. Once the resistance between the element centers in Figure 5B is determined, the internal resistance can be evaluated using standard methods such as the closed-circuit current method or the nodal potential method (Reference 1: Kenji Fukuto, Circuit Analysis by Graph Theory, Morikita Publishing Co., Ltd. (2014)). Since the internal resistance is the resistance between the positive and negative current collector foils in Figure 5A, for example, if the nodal potential method is used, the internal resistance can be evaluated by determining the potentials V1 and V2 (Figure 5B) of the nodes connected to the external power supply, and then using Kirchhoff's second law. Note that the element division in Figure 5A is not limited to the number of divisions shown in this example. For example, it is possible to select finer elements, such as further dividing each element in this figure into four. When determining the internal resistance, a larger number of element divisions in the electrode structure is preferable; for example, 500 or more is preferable, 1000 or more is more preferable, and 2000 or more is even preferable. As an example showing the dependence on the number of element divisions, the results of evaluating the dependence of the number of element divisions of the internal resistance for the three-dimensional battery structure described in Figure S2 of Non-Patent Literature 2 are shown in Table 8 below. It can be seen that increasing the number of element divisions significantly reduces the error in the internal resistance, bringing it closer to a precise solution.

[0027] This evaluation method may include, for example, a step of performing a structure matrix representation process to represent the three-dimensional electrode structure of the energy storage device 10x to be evaluated and the energy storage device 10a with known performance as a structure matrix, before performing the evaluation process, and a step of performing machine learning to construct a regression model 36 by learning the relationship between the structure matrix 30 representing the electrode structure of the energy storage device 10a with known performance and its performance. The structure matrix representation process may be a process that generates a structure matrix using the method shown in Figure 4, or it may be a process that converts the three-dimensional electrode structure into a structure matrix using the method disclosed in Non-Patent Literature 1.

[0028] In the evaluation apparatus 20 and evaluation method of this embodiment described above, performance can be evaluated (predicted) simply by fitting a structure matrix representing the electrode structure of the energy storage device 10x to be evaluated to a pre-fabricated regression model. Therefore, the efficiency of performance evaluation of energy storage devices 10 having a three-dimensional electrode structure can be further improved. Furthermore, by using a structure matrix that holds all the information of the three-dimensional electrode structure as an explanatory variable, the relationship between structural features and performance can be accurately learned, and performance can be evaluated with relatively high accuracy. The machine learning (regression) model disclosed here can predict the performance of an energy storage device 10x with unknown performance from a small number (e.g., 1000 patterns or less) of three-dimensional electrode structures and their performance data. The characteristic of this method is that the three-dimensional electrode structure is represented by a structure matrix, and the relationship between the structure matrix itself and performance is learned.

[0029] It goes without saying that the present invention is not limited in any way to the embodiments described above, and can be implemented in various forms as long as they fall within the technical scope of the present invention.

[0030] For example, in the embodiments described above, the present disclosure was described as an evaluation device and an evaluation method, but it is not limited thereto, and may also be a program that executes the evaluation method. This program causes one or more computers to implement each step of the evaluation method described above. This program may include the evaluation program 38, and may further include one or more of the machine learning program 34, the structure generation program 40, the structure matrix generation program 42, the data preparation program 44, and the search program described later. This program may be recorded on a computer-readable recording medium (e.g., hard disk, ROM, FD, CD, DVD, etc.), distributed from one computer to another via a transmission medium (communication network such as the Internet or LAN), or exchanged in any other form.

[0031] For example, in the embodiment described above, the evaluation device 20 may perform the evaluation process described above on a plurality of energy storage devices 10x with unknown performance, and execute a search process to search for a good electrode structure based on the evaluation results, or it may output a promising electrode structure derived from the search. The evaluation device 20 may have a search program (not shown), and the search program may be executed by the control unit 21 to perform the search process. The electrode structure of the energy storage device 10x with unknown performance may be generated by the structure generation program 40. Furthermore, the evaluation method may include the step of performing the evaluation process described above on a plurality of energy storage devices 10x with unknown performance, and executing a search process to search for a good electrode structure based on the evaluation results. Figure 6 is a flowchart of an example of the search process. In the search process, first, a new electrode structure is generated (step S100), and a structure matrix representing the electrode structure is obtained (step S110). Next, an evaluation process is performed to predict performance data from the structure matrix using a regression model (step S120), and the performance data obtained in the evaluation process and the electrode structure are associated and saved (step S130). Next, it is determined whether the evaluation process has been completed for all electrode structures within the search range (step S140). If it has not been completed, steps S100 to S140 are repeated. If it is determined in step S140 that the evaluation process has been completed for all electrode structures within the search range, an electrode structure corresponding to good performance data is searched for from the saved performance data and electrode structures (step S150), the promising electrode structure derived from the search is output (step S160), and the search process is terminated. This search process allows for the efficient search of high-performance electrode structures. Note that in the search process, the electrode structure corresponding to the best performance data may be searched for and output as a candidate, or electrode structures corresponding to several of the top-performing performance data may be searched for and output as candidates.

[0032] This disclosure may be any of the following [1] to

[10] . [1] An evaluation device for evaluating the performance of an energy storage device having a three-dimensional electrode structure, A control unit that performs an evaluation process to evaluate the performance of a storage device under evaluation, using a regression model constructed from the relationship between the structure matrix representing the electrode structure of a storage device with known performance and its performance, and using the structure matrix representing the electrode structure of the storage device under evaluation as an explanatory variable. An evaluation device equipped with the following features. [2] The performance used to construct the regression model is derived using a transmission line model, as described in [1]. [3] The evaluation apparatus described in [1] or [2], wherein the performance used to construct the regression model is internal resistance. [4] The regression model is constructed using one of the methods LASSO, SVR, and RFR, as described in any one of [1] to [3]. [5] The coefficient of determination of the regression model R 2 The evaluation device is one of the following [1] to [4], wherein the value is 0.6 or higher. [6] The structural matrix is ​​a matrix in which the electrode structure is divided into blocks and different numerical values ​​are assigned to the blocks of the positive electrode portion and the blocks of the negative electrode portion, as described in any one of [1] to [5]. [7] The evaluation apparatus according to any one of [1] to [6], wherein the control unit performs the evaluation process on a plurality of energy storage devices with different electrode structures and performs a search process to search for an electrode structure with good performance based on the evaluation results. [8] An evaluation apparatus according to any one of [1] to [7] for evaluating the performance of the energy storage device, which is a secondary battery using alkali metal ions as carrier ions. [9] An evaluation method for evaluating the performance of an energy storage device having a three-dimensional electrode structure, The step of performing an evaluation process to evaluate the performance of a storage device under evaluation, using a regression model constructed from the relationship between the structure matrix representing the electrode structure of a storage device with known performance and its performance, and using the structure matrix representing the electrode structure of the storage device under evaluation as an explanatory variable. An evaluation method that includes this. A program that causes one or more computers to carry out the steps of the evaluation method described in

[10] [9]. [Examples]

[0033] The following describes an example of an experimental case in which the evaluation method and evaluation apparatus of this disclosure were specifically examined.

[0034] To evaluate the effects of the present invention, a model was created to evaluate the internal resistance of a three-dimensional lithium-ion battery, as assessed by 3D-PEM, and its accuracy and computation time were evaluated. Figure 7 shows the procedure for creating a model (regression model) to predict internal resistance and evaluating its accuracy.

[0035] [Example of experiment] (Three-dimensional lithium-ion battery) For the study, two types of three-dimensional lithium-ion batteries were selected: an LFP / LTO system battery using lithium iron phosphate (LFP) as the positive electrode active material and lithium titanate (LTO) as the negative electrode active material, and an NMC532 / Graphite system battery using lithium nickel manganese cobalt composite oxide (NMC532) as the positive electrode active material and graphite as the negative electrode active material. Here, LFP is assumed to be LiFePO4 and LTO is assumed to be Li4Ti5O 12 Therefore, NMC532 is LiNi 0.5 Mn 0.3 Co 0.2 O2 was used as the ion conduction medium. The ion conduction medium was a gel electrolyte prepared by dissolving 1 M LiPF6 in a mixed solvent containing ethylene carbonate (EC) and dimethyl carbonate (DMC) in a 1:2 volume ratio, and adding a copolymer of vinylidene fluoride and hexafluoropropylene (p(VdFHFP)). The dimensions of the battery were 3000 μm × 600 μm × 3000 μm. The volume ratio of the positive electrode to the negative electrode was 1:1. The thickness of the separator was 20 μm.

[0036] (Generation of electrode structures and representation of structure matrices) In creating the regression model, 1000 electrode structures were randomly generated based on Non-Patent Document 1, and these electrode structures were represented by a structure matrix. The structure matrix is ​​obtained by dividing the three-dimensional electrode structure into voxels (elements) and converting the divided electrode information into a matrix while maintaining its resolution, thus retaining all electrode structure information.

[0037] (Construction and evaluation of regression models) First, the internal resistance of the 1000 generated electrode structures was evaluated using 3D-PEM (Figure 7(1)). For the internal resistance evaluation, the temperature T=298K and discharge current I=3.16mA / cm were used. 2 The resistance values ​​were calculated under the specified conditions. The number of element divisions for fabricating the equivalent circuit using 3D-PEM was set to 150 × 30. For the 3D-PEM resistance parameters, the values ​​in Tables 4-5 (referencing Tables S4-S5 of Non-Patent Literature 2) were used. These resistance parameters were calculated using the material parameters in Tables 1-3 (referencing Tables S1-S3 of Non-Patent Literature 2). For other specific calculation conditions, refer to Non-Patent Literature 2 (especially pages S2-S3).

[0038] Next, the dataset containing the structure matrix as the explanatory variable and internal resistance as the dependent variable was randomly divided into 800 model-building data and 200 test data (Figure 7(2)), and the hyperparameters of the regression model were determined using 5-fold cross validation with the 800 model-building data (Figure 7(3)). The regression models were constructed using three methods: Least absolute shrinkage and selection operator (LASSO), Support vector regression (SVR), and Random forest regression (RFR). These models have multiple hyperparameters, and their values ​​were determined by grid search in Figure 7(3). For LASSO, there is one hyperparameter called the regularization term, and its value is 10 in Figure 7(3). -6 ~10 6The validation error was minimized from 100 numerical values ​​equally spaced on a logarithmic scale within the specified range. Regarding SVR, there are three types of parameters: the kernel coefficient (γ) of the radial basis function (RBF), the regularization term (C), and the epsilon parameter (ε: error dead zone parameter). For γ and C, [10 -4 ,10 -3 ,10 -2 ,10 -1 ,1,10,10 2 ,10 3 From the eight types of [ ], and for ε, the value that minimized the validation error in Figure 7(3) was selected from two values ​​[0.001, 0.005], similar to LASSO. For RFR, the number of decision trees (n) and the maximum depth (max_depth) are hyperparameters, and these were determined in Figure 7(3) from [200, 250, 300, 350] and [30, 40, 50], respectively, similar to LASSO and SVR. Then, a regression model was created using the hyperparameters determined in Figure 7(3) and the 800 model creation data used (Figure 7(4)), and the model error was evaluated using 200 test data (Figure 7(5)). Then, the process in Figures 7(2) to (5) was repeated five times (Figure 7(6)), and the average of the errors of the five test data was calculated and taken as the model accuracy (Figure 7(7)).

[0039] (Results and Discussion) (1) Results for LFP / LTO systems Table 6 shows the accuracy of the regression model (mean RMSE and R 2 The average value of the coefficient of determination (R) and the time taken to predict the internal resistance of one structure are shown. In terms of accuracy, since the distribution of internal resistances of 1000 structures was within the range of 600Ω to 1100Ω, the root mean square error (RMSE) of all models was less than 9% of the internal resistance value. In this regard, when the RMSE of LASSO, SVR, and RFR are divided by 600Ω and expressed as percentages, they were 8.8%, 2.2%, and 3.7%, respectively. Of LASSO, SVR, and RFR, SVR and RFR showed particularly high accuracy. This is due to the coefficient of determination (R)2 It was also found that the value of

[0040] (2) Results of NMC532 / Graphite system Table 7 shows the accuracy of the regression model (average value of RMSE and average value of R 2 ), and the time taken for predicting the internal resistance of one structure. From the comparison with Table 6, this system also shows a similar trend to the LFP / LTO system. The prediction accuracy increases in the order of LASSO < RFR < SVR. Especially for RFR and SVR, the R 2 value exceeded 0.96. Regarding the prediction speed, SVR took the most time, and LASSO was the fastest. Compared with the internal resistance evaluation by 3D-PEM, LASSO, SVR, and RFR were respectively accelerated by 1000 times, 34 times, and 143 times.

[0041] From the above, it was found that in the above experimental examples, the internal resistance can be evaluated with high accuracy and high speed. Also, in the machine learning (regression) model of the above experimental examples, it was shown that the internal resistance of a power storage device with unknown performance can be evaluated from data of a small number of three-dimensional electrode structures and their internal resistances. It was inferred that performance other than internal resistance, such as cell energy, can be evaluated similarly.

[0042] [Reference Example] For reference, the dependence of internal resistance on the number of element divisions in 3D-PEM was investigated. Specifically, for the three-dimensional battery structure shown in Figure 8 (referencing Figure S2 of Non-Patent Literature 2), the number of element divisions was set to 50×10, 100×20, and 150×30, and the internal resistance was evaluated using 3D-PEM, and the error with the precise solution was calculated. The precise solution was the value obtained from simulation using a continuum model. The error was the error of the internal resistance evaluated by 3D-PEM relative to the precise solution. This investigation was performed on the NMC532 / Graphite system battery mentioned above. The calculation conditions for the continuum model and 3D-PEM were referred to on pages S2-S3 of Non-Patent Literature 2. The material parameters and resistance parameters used in the continuum model and 3D-PEM were those shown in Tables 2-3 and 5 (referencing Tables S2-S3 and S5 of Non-Patent Literature 2).

[0043] Table 8 summarizes the results of the example. As shown in Table 8, it was confirmed that increasing the number of element divisions significantly reduced the error in internal resistance, bringing the solution closer to a precise solution.

[0044] [Table 1]

[0045] [Table 2]

[0046] [Table 3]

[0047] [Table 4]

[0048] [Table 5]

[0049] [Table 6]

[0050] [Table 7]

[0051] [Table 8]

[0052] It goes without saying that the evaluation apparatus, evaluation method, and program disclosed herein are not limited in any way to the embodiments described above, and can be implemented in various forms as long as they fall within the technical scope of this disclosure. [Industrial applicability]

[0053] The evaluation apparatus, evaluation method, and program disclosed herein are applicable to the technical field of evaluating the characteristics of energy storage devices. [Explanation of symbols]

[0054] 10,10a,10x Energy storage device, 12 Positive electrode, 13 Positive electrode active material layer, 14 Current collector, 15 Negative electrode, 16 Negative electrode active material layer, 17 Current collector, 18 Ion conducting medium, 20 Evaluation device, 21 Control unit, 22 Memory unit, 23 Input device, 24 Display device, 25 Positive electrode block, 26 Negative electrode block, 27 Electronic resistance, 28 Ion resistance, 29 Interfacial resistance, 30 Structure matrix, 32 Known performance data, 34 Machine learning program, 36 Regression model, 38 Evaluation program, 40 Structure generation program, 42 Structure matrix generation program, 44 Data preparation program, 46 Transmission line model.

Claims

1. An evaluation apparatus for evaluating the performance of an energy storage device having a three-dimensional electrode structure, A control unit that performs an evaluation process to evaluate the performance of a storage device under evaluation, using a regression model constructed from the relationship between the structure matrix representing the electrode structure of a storage device with known performance and its performance, and using the structure matrix representing the electrode structure of the storage device under evaluation as an explanatory variable. An evaluation device equipped with the following features.

2. The evaluation apparatus according to claim 1, wherein the performance used in constructing the regression model is derived using a transmission line model.

3. The evaluation device according to claim 1 or 2, wherein the performance used in constructing the regression model is internal resistance.

4. The evaluation apparatus according to claim 1 or 2, wherein the regression model is constructed using one of the methods LASSO, SVR, and RFR.

5. The coefficient of determination R of the aforementioned regression model 2 The evaluation device according to claim 1 or 2, wherein the value is 0.6 or greater.

6. The evaluation apparatus according to claim 1 or 2, wherein the structural matrix is ​​a matrix in which the electrode structure is divided into blocks and different numerical values ​​are assigned to the blocks of the positive electrode portion and the blocks of the negative electrode portion.

7. The evaluation apparatus according to claim 1 or 2, wherein the control unit performs the evaluation process on a plurality of energy storage devices with different electrode structures, and performs a search process to search for an electrode structure with good performance based on the evaluation results.

8. An evaluation apparatus according to claim 1 or 2 for evaluating the performance of the energy storage device, which is a secondary battery using alkali metal ions as carrier ions.

9. An evaluation method for evaluating the performance of an energy storage device having a three-dimensional electrode structure, The step of performing an evaluation process to evaluate the performance of a storage device under evaluation, using a regression model constructed from the relationship between the structure matrix representing the electrode structure of a storage device with known performance and its performance, and using the structure matrix representing the electrode structure of the storage device under evaluation as an explanatory variable. An evaluation method that includes this.

10. A program that causes one or more computers to implement the steps of the evaluation method described in claim 9.

Images