JP2021176131A - Learning model generation method, program, storage medium and learned model - Google Patents

Abstract

To provide a novel learning model generation method etc. for an electrochemical device having electrolyte.SOLUTION: A learning model generation method, which generates a learning model in which evaluation of an electrochemical device having electrolyte is determined using a computer, at least includes: an acquisition step (S12) in which information including electrolyte component information, which is component information included in the electrolyte, and the evaluation of the electrochemical device is acquired as teacher data by the computer; a learning step (S15) which causes the computer to learn on the basis of the plurality of pieces of teacher data acquired in the acquisition step (S12); and a generation step (S16) which causes the computer to generate a learning model on the basis of the result learned in the learning step (S15). The learning model outputs evaluation using input information, which is unknown information different from the teacher data, as input. The input information is information at least including the electrolyte component information.SELECTED DRAWING: Figure 1

Description

The present disclosure relates to a learning model generation method, a program, a storage medium for storing the program, and a trained model.

Patent Document 1 discloses an electrolytic solution capable of providing an electrochemical device having low resistance and excellent high temperature storage characteristics and cycle characteristics.

Patent Document 2 discloses an optimization analysis device and a storage medium in which an optimization analysis program is stored.

International Publication No. 2019/031315 International Publication No. 99/07543

It is an object of the present disclosure to provide a novel learning model generation method, a program, a storage medium for storing a program, and a trained model for an electrochemical device including an electrolytic solution.

The present disclosure is a learning model generation method for generating a learning model in which the evaluation of an electrochemical device including an electrolytic solution is determined by using a computer.
At least, the acquisition step (S12) in which the computer acquires the information including the electrolyte component information which is the information of the components contained in the electrolytic solution and the evaluation of the electrochemical device as teacher data.
Based on the plurality of teacher data acquired in the acquisition step (S12), the learning step (S15) learned by the computer and the learning step (S15).
Based on the result of learning in the learning step (S15), the generation step (S16) in which the computer generates the learning model, and
With
The learning model outputs the above evaluation by inputting input information which is unknown information different from the above teacher data.
The input information is information including at least the electrolyte component information.
Regarding the learning model generation method.

The present disclosure is a learning model generation method for generating a learning model in which the optimum electrolyte component information for obtaining an evaluation of a target electrochemical device is determined by using a computer.
At least, the acquisition step (S12) in which the computer acquires the information including the electrolyte component information which is the information of the components contained in the electrolyte and the evaluation of the electrochemical device including the electrolyte as teacher data.
Based on the plurality of teacher data acquired in the acquisition step (S12), the learning step (S15) learned by the computer and the learning step (S15).
A generation step (S16) in which the computer generates a learning model based on the result of learning in the learning step (S15), and
With
The learning model takes input information, which is unknown information different from the teacher data, as input, and outputs optimum electrolyte component information for obtaining a target evaluation.
The input information is information including at least the evaluation information.
It is also related to the learning model generation method.

The teacher data further includes at least one type of information selected from the group consisting of device information and charge / discharge information.
The input information further includes at least one type of information selected from the group consisting of the device information and charge / discharge information.
The device information is information on the configuration of the electrochemical device, and is
The charge / discharge information is preferably information on conditions for charging / discharging the electrochemical device.

The learning step (S15) is preferably learned by regression analysis and / or ensemble learning in which a plurality of regression analyzes are combined.

The present disclosure is a program in which a computer uses a learning model to determine the evaluation of an electrochemical device containing an electrolyte.
In the input step (S22) in which the computer inputs the input information,
In the determination step (S23) in which the computer determines the evaluation,
An output step (S24) in which the computer outputs the evaluation determined in the determination step (S23), and
With
The learning model learns at least the information including the electrolytic solution component information, which is the information of the components contained in the electrolytic solution, and the evaluation, as teacher data.
The input information is information including at least the electrolyte component information, and is unknown information different from the teacher data.
Also related to the program.

The present disclosure is a program in which a computer uses a learning model to determine optimal electrolyte component information for obtaining an evaluation of a target electrochemical device.
In the input step (S22) in which the computer inputs the input information,
The determination step (S23) in which the computer determines the optimum electrolyte component information, and
An output step (S24) in which the computer outputs the optimum electrolytic solution component information determined in the determination step (S23), and
With
The learning model learns at least the information including the electrolyte component information which is the information of the components contained in the electrolytic solution and the evaluation of the electrochemical device including the electrolytic solution as teacher data.
The input information is information including at least the evaluation information, and is unknown information different from the teacher data.
Also related to the program.

The teacher data further includes at least one type of information selected from the group consisting of device information and charge / discharge information.
The input information further includes at least one type of information selected from the group consisting of the device information and charge / discharge information.
The device information is information on the configuration of the electrochemical device, and is
The charge / discharge information is preferably information on conditions for charging / discharging the electrochemical device.

The evaluation preferably includes information on at least one selected from the group consisting of resistance, capacity, gas generation, charge / discharge efficiency and rate characteristics of the electrochemical device.

The electrolytic solution component information may include at least one kind of information selected from the group consisting of LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of the components contained in the electrolytic solution. preferable.

The present disclosure also relates to a storage medium that stores the above program.

In the present disclosure, the electrolyte component information input to the input layer of the neural network is calculated based on the weighting coefficient of the neural network, and the evaluation of the electrochemical device is output from the output layer of the neural network. , A trained model for making computers work,
The weighting coefficient is obtained by learning using at least the electrolyte component information and the evaluation as teacher data.
The electrolytic solution component information is information on the components contained in the electrolytic solution.
The above evaluation is an evaluation of an electrochemical device provided with the above electrolytic solution.
Also related to trained models.

In the present disclosure, the evaluation information input to the input layer of the neural network is calculated based on the weighting coefficient of the neural network, and the optimum electrolytic solution for obtaining the target evaluation from the output layer of the neural network. A trained model for making a computer work so that it outputs component information.
The weighting coefficient is obtained by learning using at least the electrolyte component information and the evaluation as teacher data.
The electrolytic solution component information is information on the components contained in the electrolytic solution.
The above evaluation is an evaluation of an electrochemical device provided with the above electrolytic solution.
Also related to trained models.

The teacher data further includes at least one type of information selected from the group consisting of device information and charge / discharge information.
At least one type of information selected from the group consisting of the device information and the charge / discharge information is further input to the input layer.
The device information is information on the configuration of the electrochemical device, and is
The charge / discharge information is preferably information on conditions for charging / discharging the electrochemical device.

According to the present disclosure, it is possible to provide a novel learning model generation method, a program, a storage medium for storing a program, and a trained model for an electrochemical device including an electrolytic solution.

It is a figure which shows the structure of the learning model generator. It is a figure which shows the structure of a user apparatus. This is an example of a decision tree. This is an example of a feature space divided by a decision tree. This is an example of SVM. This is an example of a feature space. This is an example of a neuron model of a neural network. This is an example of a neural network. This is an example of teacher data. It is a flowchart which shows the operation of the learning model generator. It is a flowchart which shows the operation of a user apparatus.

Hereinafter, a learning model according to an embodiment of the present disclosure will be described. The following embodiments are specific examples, do not limit the technical scope, and can be appropriately changed within a range that does not deviate from the purpose.

(1) Outline FIG. 1 is a diagram showing a configuration of a learning model generator. FIG. 2 is a diagram showing a configuration of a user device.
The learning model is generated by the learning model generation device 10, which is one or more computers, acquiring and learning the teacher data. The generated learning model is a so-called learned model, which is implemented in a general computer or terminal, downloaded as a program or the like, or distributed in a state of being stored in a storage medium, and is one or more computers. It is used in the user device 20.

The learning model can output the correct answer for unknown information different from the teacher data. Furthermore, the learning model can be updated so that the correct answer is output for various input data.

(2) Configuration of Learning Model Generation Device 10 The learning model generation device 10 generates a learning model used in the user device 20 described later.
The learning model generation device 10 is a device having a so-called computer function. The learning model generation device 10 may include a communication interface such as a NIC and a DMA controller, and may be able to communicate with the user device 20 and the like via a network. Although the learning model generation device 10 shown in FIG. 1 is shown as one device, it is preferable that the learning model generation device 10 is compatible with cloud computing. Therefore, the hardware configuration of the learning model generation device 10 does not need to be housed in one housing or provided as a group of devices. For example, the resources of the hardware learning model generator 10 are dynamically connected / disconnected according to the load.
The learning model generation device 10 has a control unit 11 and a storage unit 14.

(2-1) Control unit 11
The control unit 11 is, for example, a CPU and controls the entire learning model generation device 10. The control unit 11 appropriately functions each of the functional units described later, and executes the learning model generation program 15 stored in advance in the storage unit 14. The control unit 11 has functional units such as an acquisition unit 12 and a learning unit 13.

Of the control units 11, the acquisition unit 12 acquires the teacher data input to the learning model generation device 10, and stores the acquired teacher data in the database 16 constructed in the storage unit 14. The teacher data may be directly input to the learning model generation device 10 by a person who uses the learning model generation device 10, or may be acquired from another device or the like via a network.

The method of acquiring teacher data by the acquisition unit 12 is not particularly limited. Teacher data is information for generating a learning model that achieves a learning purpose. Here, the learning purpose is either to output the evaluation of the electrochemical device including the electrolytic solution, or to output the optimum electrolytic solution component information for obtaining the evaluation of the target. Details will be described later.

The learning unit 13 extracts a learning data set from the teacher data stored in the storage unit 14 and automatically performs machine learning. A training dataset is a set of data for which the correct answer to an input is known. The learning data set extracted from the teacher data depends on the learning purpose. A learning model is generated by the learning unit 13 performing learning.

(2-2) Machine learning The machine learning method performed by the learning unit 13 is not particularly limited as long as it is supervised learning using a learning data set. Models or algorithms used in supervised learning include regression analysis, decision trees, support vector machines, neural networks, ensemble learning, random forests, and the like. In addition, supervised learning may be performed for each class after class classification is performed in advance. At that time, the classification may be supervised or unsupervised.

Regression analysis is, for example, linear regression analysis, multiple regression analysis, logistic regression analysis. Regression analysis is a method of fitting a model between input data (explanatory variable) and training data (objective variable) using the least squares method or the like. The dimension of the explanatory variable is 1 in linear regression analysis and 2 or more in multiple regression analysis. In logistic regression analysis, a logistic function (sigmoid function) is used as a model. Further, when the explanatory variables have many dimensions, it is preferable to perform the regression analysis by performing the dimension compression such as the principal component regression analysis and the partial least squares regression analysis.

A decision tree is a model for combining multiple classifiers to generate complex discriminant boundaries. Details of the decision tree will be described later.

A support vector machine is an algorithm that generates two classes of linear discriminant functions. Details of the support vector machine will be described later.

A neural network is a model of a network formed by connecting neurons of the human cranial nerve system at synapses. A neural network, in a narrow sense, means a multi-layer perceptron using the backpropagation method. Typical neural networks include convolutional neural networks (CNN) and recurrent neural networks (RNN). CNN is a type of feedforward neural network that is not fully coupled (loosely coupled). The details of the neural network will be described later.

Ensemble learning is a method of improving discrimination performance by combining a plurality of models. Techniques used by ensemble learning are, for example, bagging, boosting, and random forest. Bagging is a method in which a plurality of models are trained using a bootstrap sample of training data, and the evaluation of new input data is determined by a majority vote by the plurality of models. Boosting is a method of weighting training data according to the learning result of bagging so that erroneously identified learning data is trained more intensively than correctly identified learning data. Random forest is a method of generating a decision tree group (random forest) consisting of a plurality of decision trees having low correlation when a decision tree is used as a model. The details of Random Forest will be described later.

(2-2-1) Decision tree A decision tree is a model for combining a plurality of classifiers to obtain a complex discriminant boundary (non-linear discriminant function, etc.). The discriminator is, for example, a rule regarding the magnitude relationship between the value of a certain feature axis and the threshold value. As a method of constructing a decision tree from learning data, for example, there is a divide-and-conquer method in which a rule (discriminator) for dividing a feature space into two is repeatedly obtained. FIG. 3 is an example of a decision tree constructed by the divide-and-conquer law. FIG. 4 represents a feature space divided by the decision tree of FIG. In FIG. 4, the training data is indicated by a white circle or a black circle, and each training data is classified into a white circle class or a black circle class according to the decision tree shown in FIG. FIG. 3 shows nodes numbered 1 to 11 and links linking the nodes and labeled Yes or No. In FIG. 3, terminal nodes (leaf nodes) are indicated by squares, and non-terminal nodes (root nodes and internal nodes) are indicated by circles. Terminal nodes are nodes numbered 6 to 11, and nonterminal nodes are nodes numbered 1 to 5. Each terminal node is indicated by a white circle or a black circle representing the training data. Each nonterminal node has a classifier. Classifier is a rule for determining the size relationship between the value of the feature axis x _1, x ₂ and the threshold a to e. The label attached to the link indicates the judgment result of the classifier. In FIG. 4, the classifiers are shown by dotted lines, and the areas divided by the classifiers are numbered with corresponding nodes.

In the process of constructing an appropriate decision tree by the divide-and-conquer law, it is necessary to consider the following three points (a) to (c).
(A) Selection of feature axes and thresholds for constructing the classifier.
(B) Determining the terminal node. For example, the number of classes to which the training data contained in one terminal node belongs. Or, select how far to prun the decision tree (getting the same subtree with the same root node).
(C) Class assignment by majority vote to the terminal node.

For example, CART, ID3 and C4.5 are used as the learning method of the decision tree. As shown in FIGS. 3 and 4, CART is a method of generating a binary tree as a decision tree by dividing the feature space into two for each feature axis in each node other than the terminal node.

（ｂ）交差エントロピー（逸脱度）

（ｃ）ジニ係数

上式において、確率Ｐ（Ｃ_ｉ｜ｔ）は、ノードｔにおけるクラスＣ_ｉの事後確率であり、言い換えると、ノードｔにおいてクラスＣ_ｉのデータが選ばれる確率である。式（１−３）の第２式において、確率Ｐ（Ｃ_ｊ｜ｔ）は、クラスＣ_ｉのデータがｊ（≠ｉ）番目のクラスに間違われる確率であるので、第２式は、ノードｔにおける誤り率を表す。式（１−３）の第３式は、全てのクラスに関する確率Ｐ（Ｃ_ｉ｜ｔ）の分散の和を表す。
不純度を評価関数としてノードを分割する場合、例えば、当該ノードにおける誤り率、および、決定木の複雑さで決まる許容範囲まで、決定木を剪定する手法が用いられる。 In learning using a decision tree, it is important to divide the feature space at the optimum division candidate points in the non-terminal node in order to improve the discrimination performance of the training data. An evaluation function called impureness may be used as a parameter for evaluating the division candidate points of the feature space. As the function I (t) representing the purity of the node t, for example, the parameters represented by the following equations (1-1) to (1-3) are used. K is the number of classes.
(A) Error rate at node t

(B) Cross entropy (deviance)

(C) Gini coefficient

In the above equation, the probability P (C _i | t) is _{the posterior probability of the class C i at} the node t, in other words, the probability that _{the data of the class C i} is selected at the node t. In the second equation of the equation (1-3), the probability P (C _j | t) is the _{probability that the data of the class C i} is mistaken for the j (≠ i) th class. Represents the error rate at t. The third equation of equation (1-3) represents the sum of the variances of the _{probabilities P (C i | t) for all classes.}
When dividing a node using impureness as an evaluation function, for example, a method of pruning the decision tree to an allowable range determined by the error rate at the node and the complexity of the decision tree is used.

(2-2-2) Support Vector Machine The Support Vector Machine (SVM) is an algorithm for finding a two-class linear discrimination function that realizes the maximum margin. FIG. 5 is a diagram for explaining SVM. The two-class linear discrimination function represents the discrimination hyperplanes P1 and P2, which are hyperplanes for linearly separating the training data of the two classes C1 and C2 in the feature space shown in FIG. In FIG. 5, the training data of class C1 is shown by a circle, and the learning data of class C2 is shown by a square. The margin of the identification hyperplane is the distance between the training data closest to the identification hyperplane and the identification hyperplane. FIG. 5 shows a margin d1 of the identification hyperplane P1 and a margin d2 of the identification hyperplane P2. In SVM, the optimum identification hyperplane P1 which is the identification hyperplane that maximizes the margin is required. The minimum value d1 of the distance between the training data of one class C1 and the optimal identification hyperplane P1 is equal to the minimum value d1 of the distance between the training data of the other class C2 and the optimal identification hyperplane P2.

学習データセットＤ_Ｌは、学習データ（特徴ベクトル）ｘ_ｉと、教師データｔ_ｉ＝｛−１，＋１｝との対の集合である。学習データセットＤ_Ｌの要素数は、Ｎである。教師データｔ_ｉは、学習データｘ_ｉがクラスＣ１，Ｃ２のどちらに属するのかを表す。クラスＣ１はｔ_ｉ＝−１のクラスであり、クラスＣ２はｔ_ｉ＝＋１のクラスである。 5, represented by the training dataset D _L used in the two-class problem supervised learning following equation (2-1).

Training dataset _{D L} is the learning data (feature vector) _{x i,} = teacher data _{t i {- 1, + 1} } is a set of pairs of. The number of elements of the training dataset D _L is N. The teacher data t _i represents whether the learning data x _i belongs to the classes C1 and C2. Class C1 is a class of _t i = -1, class C2 is the class of _t i = + 1.

これらの２つの式は、以下の１つの式（２−４）で表される。

5, holds for all learning data x _i, the linear discriminant function is normalized is represented by the following two formulas (2-2) and (2-3). w is a coefficient vector and b is a bias.

These two equations are represented by the following one equation (2-4).

式（２−６）において、ρ（ｗ）は、クラスＣ１，Ｃ２のそれぞれの学習データｘ_ｉを識別超平面Ｐ１，Ｐ２の法線ベクトルｗ上に射影した長さの差の最小値を表す。式（２−６）の「ｍｉｎ」および「ｍａｘ」の項は、それぞれ、図５において符号「ｍｉｎ」および符号「ｍａｘ」で示された点である。図５において、最適識別超平面は、マージンｄが最大となる識別超平面Ｐ１である。 When the identification hyperplanes P1 and P2 are represented by the following equation (2-5), the margin d is represented by the equation (2-6).

In the formula (2-6), ρ (w) denotes the class C1, the respective minimum length difference which is projected onto the normal vector w of learning data _{x i} identification hyperplane P1, P2 of C2 .. The terms "min" and "max" in formula (2-6) are the points represented by the symbols "min" and "max" in FIG. 5, respectively. In FIG. 5, the optimum identification hyperplane is the identification hyperplane P1 having the maximum margin d.

スラック変数ξ_ｉは、学習時のみに使用され、０以上の値をとる。図６には、識別超平面Ｐ３と、マージン境界Ｂ１，Ｂ２と、マージンｄ３とが示されている。識別超平面Ｐ３の式は式（２−５）と同じである。マージン境界Ｂ１，Ｂ２は、識別超平面Ｐ３からの距離がマージンｄ３である超平面である。 FIG. 5 represents a feature space in which two classes of training data are linearly separable. FIG. 6 is a feature space similar to that of FIG. 5, and represents a feature space in which the two classes of training data cannot be linearly separated. When the two classes of training data are not linearly separable, the following equation (2-7) extended by introducing the _{slack variable ξ i into equation (2-4) can be used.}

The slack variable ξ _i is used only during learning and takes a value of 0 or more. FIG. 6 shows the identification hyperplane P3, the margin boundaries B1 and B2, and the margin d3. The equation of the identification hyperplane P3 is the same as that of equation (2-5). The margin boundaries B1 and B2 are hyperplanes in which the distance from the identification hyperplane P3 is the margin d3.

When the slack variable ξ _i is 0, equation (2-7) is equivalent to equation (2-4). At this time, as shown by the circles or open squares in Figure 6, the learning data x _i that satisfies the equation (2-7) is correctly identified in the margin d3. The distance between the learning data x _i and the identification hyperplane P3 is margin d3 more.

When the slack variable ξ _i is greater than 0 and less than or equal to 1, the training data x _i satisfying equation (2-7) exceeds the margin boundaries B1 and B2, as shown by the hatched circle or square in FIG. However, it does not exceed the identification hyperplane P3 and is correctly identified. The distance between the learning data x _i and the identification hyperplane P3 is less than the margin d3.

When the slack variable ξ _{i is larger than 1, the training data x i} satisfying the equation (2-7) exceeds the identification hyperplane P3 and is erroneously recognized, as shown by the black circle or square in FIG. Will be done.

In this way, _{by using the equation (2-7) in which the slack variable ξ i is introduced, the training data x i} can be identified even when the training data of the two classes cannot be linearly separated.

評価関数Ｌ_ｐの出力値を最小化する解（ｗ、ξ）を求める。式（２−８）において、第２項のパラメータＣは、誤認識に対するペナルティの強さを表す。パラメータＣが大きいほど、ｗのノルム（第１項）よりも誤認識数（第２項）を小さくする方を優先する解が求められる。 From the above description, the sum of the slack variables ξ _{i of all the training data x i} represents the upper limit of the number of misrecognized learning data x _i. Here, the evaluation function L _p is defined by the following equation (2-8).

Find the solution (w, ξ) that minimizes the output value of the evaluation function L _p. In the formula (2-8), the parameter C of the second term represents the strength of the penalty for misrecognition. As the parameter C is larger, a solution that gives priority to reducing the number of false recognitions (second term) rather than the norm of w (first term) is required.

式（３−１）において、入力ｘ、出力ｙおよび重みｗは、すべてベクトルであり、θは、バイアスであり、φは、活性化関数である。活性化関数は、非線形関数であり、例えば、ステップ関数（形式ニューロン）、単純パーセプトロン、シグモイド関数又はＲｅＬＵ（ランプ関数）である。 (2-2-3) Neural network FIG. 7 is a schematic diagram of a neuron model of a neural network. FIG. 8 is a schematic diagram of a three-layer neural network configured by combining the neurons shown in FIG. 7. As shown in FIG. 7, the neuron outputs an output y for a plurality of inputs x (inputs x1, x2, x3 in FIG. 7). Each input x (inputs x1, x2, x3 in FIG. 7) is multiplied by a corresponding weight w (weights w1, w2, w3 in FIG. 7). The neuron outputs the output y using the following equation (3-1).

In equation (3-1), the input x, the output y and the weight w are all vectors, θ is the bias, and φ is the activation function. The activation function is a non-linear function, such as a step function (formal neuron), a simple perceptron, a sigmoid function or a ReLU (ramp function).

In the three-layer neural network shown in FIG. 8, a plurality of input vectors x (input vectors x1, x2, x3 in FIG. 8) are input from the input side (left side in FIG. 8), and the output side (right side in FIG. 8). A plurality of output vectors y (output vectors y1, y2, y3 in FIG. 8) are output from. This neural network is composed of three layers L1, L2, and L3.

In the first layer L1, the input vectors x1, x2, x3 are input by applying corresponding weights to each of the three neurons N11, N12, and N13. In FIG. 8, these weights are collectively referred to as W1. The neurons N11, N12, and N13 output the feature vectors z11, z12, and z13, respectively.

In the second layer L2, the feature vectors z11, z12, and z13 are input by applying corresponding weights to each of the two neurons N21 and N22. In FIG. 8, these weights are collectively referred to as W2. The neurons N21 and N22 output the feature vectors z21 and z22, respectively.

In the third layer L3, the feature vectors z21 and z22 are input by applying corresponding weights to each of the three neurons N31, N32 and N33. In FIG. 8, these weights are collectively referred to as W3. The neurons N31, N32, and N33 output output vectors y1, y2, and y3, respectively.

The operation of the neural network includes a learning mode and a prediction mode. In the learning mode, the weights W1, W2, and W3 are learned using the learning data set. In the prediction mode, prediction such as identification is performed using the learned parameters of the weights W1, W2, and W3.

The weights W1, W2, and W3 can be learned by, for example, an error backpropagation method. In this case, the information about the error is transmitted from the output side to the input side, in other words, from the right side to the left side in FIG. The backpropagation method learns by adjusting the weights W1, W2, and W3 in each neuron so as to reduce the difference between the output y when the input x is input and the true output y (teacher data). It is a method to do. Examples of the method for optimizing the weight include general methods such as stochastic gradient descent, RMSrop, and Adamax.

The neural network can be configured to have more than three layers. A machine learning method using a neural network having four or more layers is known as deep learning.

(2-2-4) Random forest Random forest is a kind of ensemble learning, and is a method of enhancing discrimination performance by combining a plurality of decision trees. In learning using a random forest, a group (random forest) consisting of multiple decision trees with low correlation is generated. The following algorithms are used to generate and identify random forests.
(A) The following is repeated from m = 1 to M.
(A) From N d-dimensional learning data, m bootstrap samples Z _m are generated.
(B) Using Z _m as training data, each node t is divided according to the following procedure to generate m decision trees.
(I) From d features, d'features are randomly selected. (D'<d)
(Ii) From the selected d'features, the features and division points (threshold values) that give the optimum division of the training data are obtained.
(Iii) The node t is divided into two at the obtained division point.
(B) Output a random forest consisting of m decision trees.
(C) For the input data, the identification result of each decision tree of the random forest is obtained. The identification result of the random forest is determined by the majority of the identification results of each decision tree.

In learning using a random forest, the correlation between decision trees can be lowered by randomly selecting a predetermined number of features to be used for identification at each non-terminal node of the decision tree.

(2-3) Storage unit 14
The storage unit 14 shown in FIG. 1 is an example of a recording medium, and is composed of, for example, a flash memory, a RAM, an HDD, or the like. The storage unit 14 stores in advance the learning model generation program 15 executed by the control unit 11. A database 16 is constructed in the storage unit 14, and a plurality of teacher data acquired by the acquisition unit 12 are stored and appropriately managed. The database 16 stores a plurality of teacher data, for example, as shown in FIG. Note that FIG. 9 shows a part of the teacher data stored in the database 16. In addition to the teacher data, the storage unit 14 may store information for generating a learning model such as a learning data set and test data.

(3) Teacher data It was found that there is a correlation between the electrolyte component information and the evaluation of the electrochemical device.
Therefore, the teacher data acquired to generate the learning model includes at least the electrolyte component information and the evaluation information as shown below. Further, from the viewpoint of improving the accuracy of the output value, it is preferable to include at least one kind of information selected from the group consisting of device information and charge / discharge information. Of course, the teacher data may include information other than the information shown below. It is assumed that the database 16 of the storage unit 14 in the present disclosure stores a plurality of teacher data including the following information.

(3-1) Electrolytic solution component information The electrolytic solution component information is information on components contained in the electrolytic solution.
The electrolyte component information includes, for example, the LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density, molecular weight, solubility parameter, molecular surface area, dipole moment, volume, and number of unsaturated bonds of the component. Information such as density, electrolyte viscosity, water content, purity, type, and usage amount can be mentioned.

The LUMO level and the HOMO level may be values obtained by calculation (simulation) or may be values obtained by measuring by a known method. The above calculation can be performed based on the density functional theory (DFT method), the molecular orbital method, or the like, and may be performed using commercially available calculation software (for example, Gaussian manufactured by Gaussian, Winmosar manufactured by Crossability). ..

The molecular structure may be one in which the molecular structure of the electrolytic solution component is illustrated by a structural formula or the like, or one in which the molecular structure is expressed by a known description method such as SMILES notation or fingerprint.

The melting point and boiling point may be values obtained by calculation (simulation) or may be values obtained by measuring by a known method. The above calculation may be performed using commercially available calculation software (for example, Material studio manufactured by BIOVIA, COSMOtherm manufactured by MOLSIS).
Since the melting point and boiling point are physical properties as an aggregate of molecules rather than a single molecule, more accurate output can be obtained by using this information.

The electron density is preferably the electron density of the most negatively charged portion. The electron density may be a value obtained by calculation (simulation), or may be a value obtained by measuring by a known method. The above calculation can be performed based on the density functional theory (DFT method), the molecular orbital method, or the like, and may be performed using commercially available calculation software (for example, Gaussian manufactured by Gaussian).

The molecular weight may be a value obtained from the structure (chemical formula) of the molecule, or may be a value obtained by measuring by a known method.

The solubility parameter is a physical property value defined by the square root of the aggregation energy density, and is a numerical value indicating the dissolution behavior of the solvent. For example, the solubility parameter of Hildebrand and the solubility parameter of Hansen can be mentioned. The solubility parameter may be a value obtained by calculation (simulation) or a value obtained by measuring by a known method. The above calculation may be performed using commercially available calculation software (for example, HSPiP).

The molecular surface area is the surface area of one molecule of the above component, and may be a value obtained by calculation (simulation) or a value obtained by measuring by a known method. The above calculation may be performed using commercially available calculation software (for example, Winmosar manufactured by Crossability Co., Ltd.).

The dipole moment may be a value obtained by calculation (simulation), or may be a value obtained by measuring by a known method. The above calculation may be performed using commercially available calculation software (for example, Winmosar manufactured by Crossability Co., Ltd.).

The volume is the volume of one molecule of the component, and may be a value obtained by calculation (simulation) or a value obtained by measuring by a known method. The above calculation may be performed using commercially available calculation software (for example, Winmosar manufactured by Crossability Co., Ltd.).

The number of unsaturated bonds is the number of double bonds or triple bonds in one molecule of the component, and can be obtained from, for example, the structure (chemical formula) of the molecule.

The density may be a value measured by a density meter or the like. It can also be obtained by calculation (simulation).

The electrolytic solution viscosity is the viscosity when a predetermined amount of the above components is added to a predetermined electrolytic solution. The electrolyte viscosity may be a value measured by a viscometer or the like. It can also be obtained by calculation (simulation). The above calculation may be performed using commercially available calculation software (for example, J-OCTA manufactured by JSOL).

The water content may be a value measured by a known method such as the Karl Fischer method.

The purity may be a value measured by a known method such as chromatography or NMR.

The electrolyte component information includes, among other things, the LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density, molecular weight, solubility parameter, molecular surface area, dipole moment, volume, and number of unsaturated bonds. It preferably contains at least one information selected from the group consisting of density, electrolyte viscosity, water content, purity, type and amount used.
It is more preferable to include at least one information selected from the group consisting of LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of the above components.
It is more preferable to include at least one information selected from the group consisting of LUMO level, HOMO level, molecular structure, melting point, boiling point and electron density of the above components.
It is even more preferable to include at least one information selected from the group consisting of LUMO level, HOMO level, molecular structure, melting point and boiling point of the above components.
It is even more preferable to include at least one information selected from the group consisting of LUMO level, HOMO level and molecular structure of the above components.
It is particularly preferable to include at least one type of information selected from the group consisting of the LUMO level and the HOMO level of the above components.
Most preferably, it contains information on the LUMO level of the above components.
Since this information has a particularly strong correlation with the evaluation of the electrochemical device, it is possible to obtain a more accurate output by using this information.

The electrolyte component information includes LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density, molecular weight, solubility parameter, molecular surface area, dipole moment, volume, number of unsaturated bonds, density, and electrolysis. It is also preferable to include at least two types of information selected from the group consisting of liquid viscosity, water content, purity, type and amount used.
It is more preferable to include at least two kinds of information selected from the group consisting of LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of the above components.
It is more preferable to include at least two kinds of information selected from the group consisting of LUMO level, HOMO level, molecular structure, melting point, boiling point and electron density of the above components.
It is even more preferable to include at least two types of information selected from the group consisting of LUMO level, HOMO level, molecular structure, melting point and boiling point of the above components.
It is even more preferable to include at least two types of information selected from the group consisting of LUMO level, HOMO level and molecular structure of the above components.
It is particularly preferable to include information on the LUMO level and the HOMO level of the above components.
Since this information has a particularly strong correlation with the evaluation of the electrochemical device, it is possible to obtain a more accurate output by using two or more kinds of this information.

Examples of the electrolytic solution components are shown below, but the present invention is not limited thereto. Further, the classification of "solvent", "electrolyte salt", "additive" and the like is not limited to the following examples, and can be appropriately classified according to the type of component, the amount used and the like.
As the electrolytic solution component, one type may be used alone, or two or more types may be used in combination in any combination and ratio.

Examples of the electrolytic solution component include a solvent. The solvent preferably contains at least one selected from the group consisting of carbonates and carboxylic acid esters.

The carbonate may be a cyclic carbonate or a chain carbonate.

The cyclic carbonate may be a non-fluorinated cyclic carbonate or a fluorinated cyclic carbonate.

As the non-fluorinated cyclic carbonate, a non-fluorinated saturated cyclic carbonate is preferable.

The fluorinated cyclic carbonate may be a fluorinated saturated cyclic carbonate or a fluorinated unsaturated cyclic carbonate.

The chain carbonate may be a non-fluorinated chain carbonate or a fluorinated chain carbonate.

The carboxylic acid ester may be a cyclic carboxylic acid ester or a chain carboxylic acid ester.

The cyclic carboxylic acid ester may be a non-fluorinated cyclic carboxylic acid ester or a fluorinated cyclic carboxylic acid ester.

The chain carboxylic acid ester may be a non-fluorinated chain carboxylic acid ester or a fluorinated chain carboxylic acid ester.

An ionic liquid can also be used as the solvent. An "ionic liquid" is a liquid composed of ions in which an organic cation and an anion are combined.

The solvent is preferably a non-aqueous solvent, and the electrolytic solution is preferably a non-aqueous electrolytic solution.

Examples of the electrolyte component include an electrolyte salt. As the electrolyte salt, in addition to lithium salt, ammonium salt and metal salt, liquid salt (ionic liquid), inorganic polymer type salt, organic polymer type salt and the like can be used in the electrolyte solution. Any one can be used.

As the electrolyte salt of the electrolytic solution for a lithium ion secondary battery, a lithium salt is preferable.
Any of the above lithium salts can be used, for example, inorganic lithium salts, lithium tungstates, lithium carboxylates, lithium salts having an S = O group, lithiumimide salts, lithiummethide salts, and fluorine-containing organics. Examples include lithium salts.

As the electrolyte salt of the electrolytic solution for the electric double layer capacitor, an ammonium salt is preferable. Examples of the ammonium salt include tetraalkyl quaternary ammonium salt, spirocyclic bipyrrolidinium salt, imidazolium salt, N-alkylpyridinium salt, N, N-dialkylpyrrolidinium salt and the like.
Further, as the electrolyte salt for the electric double layer capacitor, a lithium salt or a magnesium salt can also be used.

Further, various additives are also mentioned as the electrolytic solution component.

Examples of the additive include a dicarboxylic acid complex salt in which the complex center element is boron, a dicarboxylic acid complex salt in which the complex center element is phosphorus, and a dicarboxylic acid complex salt in which the complex center element is aluminum. Can be mentioned.

As the additive, an acid anhydride having a 5-membered ring structure, a 6-membered ring structure or another cyclic structure is also preferable. The acid anhydride may be substituted with a halogen atom such as a fluorine atom.

Examples of the additive include a nitrile compound, a compound having an isocyanato group (isocyanate), a cyclic sulfonic acid ester, a polyethylene oxide, a fluorinated saturated cyclic carbonate, an unsaturated cyclic carbonate, a compound having a triple bond, and a carboxylic acid anhydride. Examples include overcharge inhibitors and other auxiliaries.

Examples of the overcharge inhibitor include biphenyl, biphenyl derivatives, terphenyls, terphenyl derivatives, partial hydrides of terphenyl derivatives, various aromatic compounds and their partial fluorinated compounds, fluorine-containing anisole compounds, and aromatic acetates. Aromatic carbonates, toluene derivatives and the like can be mentioned.

Examples of the other auxiliary agents include hydrocarbon compounds, fluorine-containing aromatic compounds, carbonate compounds, ether compounds, ketone compounds, acid anhydrides, ester compounds, amide compounds, sulfur-containing compounds, nitrogen-containing compounds, and phosphorus-containing compounds. Examples thereof include compounds, boron-containing compounds and silane compounds.

Examples of the above-mentioned additives include cyclic and chain carboxylic acid esters, ether compounds, nitrogen-containing compounds, boron-containing compounds, organic silicon-containing compounds, nonflammable (flame-retardant) agents, surfactants, and highly dielectricized additives. Examples thereof include cycle property and rate property improving agents, sulfone compounds, ionic conductive compounds, fluorophosphate lithium salts, lithium salts having an S = O group, metal oxides, glass and the like.

The electrolytic solution may be a gel-like (plasticized) gel electrolytic solution combined with a polymer material.

Of course, the electrolyte component information may include information other than the above. The teacher data in FIG. 9 includes each of the above items, which is the electrolyte component information, but some of the items are not shown.

(3-2) Device information Device information is information on the configuration of an electrochemical device.
The device information can include, for example, at least one type of information selected from the group consisting of electrode information, separator information, and electrolyte amount information.

(3-2-1) Electrode information Electrode information is information on electrodes constituting an electrochemical device.
The electrode information can include positive electrode information and negative electrode information.

The electrode information (positive electrode information or negative electrode information) includes the type and composition ratio of the active material, the electrode density, the type and amount of the binder, the type and amount of the conductive auxiliary agent (conductive material), and the particle size of the active material (the particle size of the active material). Information such as (primary particle size or secondary particle size), BET specific surface area, type and amount of thickener, type and thickness of current collector, and the like can be mentioned.

As each of the above information, a value obtained by measuring by a known method may be used. Further, if necessary, a value obtained by calculation (simulation) can be used.

Information on the particle size (primary particle size or secondary particle size) and BET specific surface area of the active material is particularly preferably included in the positive electrode information.

The above electrode information includes, among others, the type and composition ratio of the active material, the electrode density, the type and amount of the binder, the type and amount of the conductive auxiliary agent (conductive material), and the particle size of the active material (primary particle size or secondary). It preferably contains at least one information selected from the group consisting of (next particle size) and BET specific surface area, type and amount of thickener, and type and thickness of current collector.
It is more preferable to include at least one information selected from the group consisting of the type and composition ratio of the active material, the electrode density, the type and amount of the binder, and the type and amount of the conductive auxiliary agent.
It is more preferable to include information on at least one selected from the group consisting of the type and composition ratio of the active material, the electrode density, and the type and amount of the binder.
It is particularly preferred to include information on the type and composition ratio of the active material and at least one selected from the group consisting of the type and amount of binder.
Since this information has a particularly strong correlation with the evaluation of the electrochemical device, it is possible to obtain a more accurate output by using this information.

The above electrode information includes the type and composition ratio of the active material, the electrode density, the type and amount of the binder, the type and amount of the conductive auxiliary agent (conductive material), and the particle size of the active material (primary particle size or secondary particle size). ) And the BET specific surface area, the type and amount of the thickener, and at least two types of information selected from the group consisting of the type and thickness of the current collector.
It is more preferable to include at least two types of information selected from the group consisting of the type and composition ratio of the active material, the electrode density, the type and amount of the binder, and the type and amount of the conductive auxiliary agent.
It is more preferable to include at least two types of information selected from the group consisting of the type and composition ratio of the active material, the electrode density, and the type and amount of the binder.
It is particularly preferred to include information on the type and composition ratio of the active material and at least two types selected from the group consisting of the type and amount of binder.
Since this information has a particularly strong correlation with the evaluation of the electrochemical device, it is possible to obtain a more accurate output by using two or more kinds of this information.

Examples of the positive electrode and the negative electrode are shown below, but the present invention is not limited thereto.

The positive electrode for a lithium ion secondary battery is composed of a positive electrode active material layer containing a positive electrode active material and a current collector.

The positive electrode active material is not particularly limited as long as it can electrochemically occlude and release lithium ions. For example, a lithium-containing transition metal composite oxide, a lithium-containing transition metal phosphoric acid compound, a sulfur-based material, and the like. Examples include conductive polymers.

Examples of the shape of the particles of the positive electrode active material include a lump, a polyhedron, a sphere, an elliptical sphere, a plate, a needle, and a columnar shape. Further, the primary particles may be aggregated to form secondary particles.

The positive electrode active material layer can be formed from a positive electrode mixture. The positive electrode mixture preferably contains a binder, a thickener, and a conductive auxiliary agent (conductive material) in addition to the positive electrode active material.
Examples of the binder include resin-based polymers, rubber-like polymers, thermoplastic elastomer-like polymers, soft resin-like polymers, fluorine-based polymers, and polymer compositions having ionic conductivity. ..
Examples of the thickener include cellulose derivatives.
Examples of the conductive auxiliary agent (conductive material) include metal materials and carbon materials.

Examples of the material of the current collector for the positive electrode include a metal material and a carbon material.

The negative electrode for a lithium ion secondary battery is composed of a negative electrode active material layer containing a negative electrode active material and a current collector.

Examples of the negative electrode active material include a carbonaceous material capable of occluding / releasing lithium, a metal oxide material capable of occluding / releasing lithium, a lithium metal, various lithium alloys, a thium-containing metal composite oxide material, and the like. can.

The negative electrode active material layer can be formed from a negative electrode mixture. The negative electrode mixture preferably contains a binder, a thickener, and a conductive auxiliary agent (conductive material) in addition to the negative electrode active material.

Examples of the binder, thickener, and conductive auxiliary agent include the same binders, thickeners, and conductive auxiliary agents that can be used for the positive electrode.

Examples of the material of the current collector for the negative electrode include copper, nickel, stainless steel and the like.

In an electric double layer capacitor, at least one of a positive electrode and a negative electrode is a polar electrode. Examples of the polar electrode include a polar electrode mainly composed of activated carbon.

The non-polarizing electrode is mainly composed of a carbon material that can occlude and occlude lithium ions reversibly, and it is preferable to use a carbon material in which lithium ions are occluded. In this case, a lithium salt is used as the electrolyte.

Stainless steel, aluminum, titanium or tantalum can be preferably used as the current collector of the polar electrode mainly composed of activated carbon. Stainless steel, copper or nickel is preferably used as the current collector of the electrode mainly composed of a carbon material capable of reversibly occluding and releasing lithium ions.

Of course, the electrode information may include information other than the above. The teacher data in FIG. 9 includes each of the above items, which are electrode information, but some of them are not shown.

(3-2-2) Separator information Separator information is information on separators constituting an electrochemical device.

Examples of the separator information include information such as the type, thickness, porosity, and average pore diameter of the separator.
As each of the above information, a value obtained by measuring by a known method may be used. Further, if necessary, a value obtained by calculation (simulation) can be used.

The separator information preferably includes at least one type of information selected from the group consisting of the type, thickness, porosity and average pore size of the separator.
It is more preferable to include at least one type of information selected from the group consisting of the type, thickness and porosity of the separator.
It is more preferable to include at least one type of information selected from the group consisting of the thickness and porosity of the separator.
Since this information has a particularly strong correlation with the evaluation of the electrochemical device, it is possible to obtain a more accurate output by using this information.

The separator information preferably includes at least two types of information selected from the group consisting of the type, thickness, porosity and average pore diameter of the separator.
It is more preferable to include at least two types of information selected from the group consisting of the type, thickness and porosity of the separator.
It is more preferable to include information on the thickness and porosity of the separator.
Since this information has a particularly strong correlation with the evaluation of the electrochemical device, it is possible to obtain a more accurate output by using two or more kinds of this information.

Examples of the separator include those made of a resin, glass fiber, an inorganic substance, etc., which are made of a material stable to the electrolytic solution, and are in the form of a porous sheet, woven cloth, non-woven fabric, etc., which have excellent liquid retention properties. Etc. are preferable, but the present invention is not limited to these.

Of course, the separator information may include information other than the above. The teacher data in FIG. 9 includes each of the above items, which is separator information, but some of the above items are omitted.

(3-2-3) Electrolyte amount information The electrolyte amount information is information on the amount of the electrolyte in the electrochemical device (the amount of the electrolyte injected into the electrochemical device).

(3-3) Charge / discharge information The charge / discharge information is information on conditions for charging / discharging an electrochemical device.
Examples of the charge / discharge information include information such as voltage, current, and temperature during charge / discharge.

The charge / discharge information preferably includes at least one type of information selected from the group consisting of charge / discharge voltage, current, and temperature, and is selected from the group consisting of charge / discharge voltage and current. It is more preferable to include at least one kind of information.
Since this information has a particularly strong correlation with the evaluation of the electrochemical device, it is possible to obtain a more accurate output by using this information.

The charge / discharge information preferably includes at least two types of information selected from the group consisting of charge / discharge voltage, current, and temperature, and more preferably includes information on charge / discharge voltage and current.
Since this information has a particularly strong correlation with the evaluation of the electrochemical device, it is possible to obtain a more accurate output by using two or more kinds of this information.

Of course, the charge / discharge information may include information other than the above. The teacher data in FIG. 9 includes each of the above items, which are charge / discharge information, but some of them are not shown.

(3-4) Evaluation Evaluation is information on evaluation of an electrochemical device including an electrolytic solution.
The evaluation can include information on the characteristics of the electrochemical device, for example, information on the resistance, capacity, gas generation amount, charge / discharge efficiency, rate characteristics, etc. of the electrochemical device.
The characteristics of the electrochemical device can be measured by an experiment using the electrochemical device.

Examples of the resistance include IV resistance and impedance. The resistance may be an initial resistance (immediately after production), or may be a resistance after a charge / discharge cycle or storage.

The IV resistor is obtained, for example, by charging an electrochemical device at a predetermined temperature / constant current to a predetermined capacity, discharging the electrochemical device at a predetermined temperature / current, measuring a voltage at a predetermined time point, and discharging the device. It can be obtained by calculating the resistance from the voltage drop of.

The impedance can be obtained, for example, by charging an electrochemical device at a predetermined temperature and a constant current so as to have a predetermined capacity, and then applying a predetermined AC voltage amplitude at a predetermined temperature.

Examples of the capacity include an initial capacity (immediately after production), a capacity after a charge / discharge cycle, a capacity after storage, and the like. The capacity after the charge / discharge cycle or after storage can also be expressed as a ratio (capacity retention rate) to the initial capacity.

The capacity may be a discharge capacity. For example, by charging a predetermined voltage with a predetermined current at a predetermined temperature to a predetermined voltage and then discharging the battery to a predetermined voltage at a predetermined temperature and current. You can ask.

Examples of the gas generation amount include a gas generation amount during the charge / discharge cycle, a gas generation amount during storage, and the like.
The amount of gas generated can be obtained by measuring the volume of the electrochemical device before and after the charge / discharge cycle or storage by the Archimedes method or the like, and from the volume change before and after.

The charge / discharge efficiency can be expressed by the ratio of the discharge amount to the charge amount in a predetermined charge / discharge cycle.

The above rate characteristics are the magnitude of the discharge capacity when charging / discharging at a predetermined current value and the discharge capacity at different current values when the current value for discharging the reference capacity of the electrochemical device in 1 hour is 1C. Represents the ratio of.
For the above rate characteristics, for example, using a secondary battery, the battery is charged to 4.3 V by the constant current method of 0.2 C, then discharged to 3.0 V at 0.2 C to obtain the 0.2 C discharge capacity, and then the 0.2 C discharge capacity is obtained. , Charge to 4.3V at 0.2C, then discharge to 3.0V at 2C to determine the 2C discharge capacity. The value of the discharge capacity at the time of 2C constant current discharge with respect to the discharge capacity at the time of 0.2C constant current discharge can be evaluated as the rate characteristic value.
Rate characteristic value (2C) = (2C discharge capacity / 0.2C discharge capacity) x 100 (%)

In each of the above-mentioned measurements, the charge / discharge cycle is, for example, charging to a predetermined voltage with a predetermined current at a predetermined temperature and then discharging to a predetermined voltage at a predetermined temperature / current. It can be carried out by setting it as a cycle and repeating this cycle a predetermined number of times.

In each of the measurements described above, storage can be performed, for example, by keeping the electrochemical device at a predetermined temperature for a predetermined period of time.

The measurement of each of the above-mentioned characteristics may be carried out at room temperature, at a high temperature (for example, 40 ° C. or higher), or at a low temperature (for example, 5 ° C. or lower).
Further, the voltage-related measurement may be performed at a high voltage (for example, 4.3 V or more) or a low voltage (for example, less than 4.3 V).
The charge / discharge cycle and storage described above may be performed at room temperature, at a high temperature (for example, 40 ° C. or higher), or at a low temperature (for example, 0 ° C. or lower).

The evaluation preferably includes information on at least one selected from the group consisting of resistance, capacity, gas generation, charge / discharge efficiency and rate characteristics of the electrochemical device.
It is more preferable to include information on at least one selected from the group consisting of resistance, capacity, gas generation amount and charge / discharge efficiency of the electrochemical device.
It is more preferred to include information about at least one selected from the group consisting of resistance, capacity and gas generation of the electrochemical device.
It is more preferred to include information about at least one selected from the group consisting of resistance and capacitance of the electrochemical device.

The electrochemical device in the present disclosure includes a lithium ion secondary battery, a lithium ion capacitor, a capacitor (hybrid capacitor, an electric double layer capacitor), a radical battery, a solar cell (particularly a dye-sensitized solar cell), a lithium ion primary battery, and the like. Examples thereof include fuel cells, various electrochemical sensors, electrochromic elements, electrochemical switching elements, aluminum electrolytic capacitors, tantalum electrolytic capacitors, etc., and lithium ion secondary batteries, lithium ion capacitors, electric double layer capacitors are preferable, and lithium ion secondary capacitors are used. Batteries are more preferred.

The above evaluation is preferably an evaluation of a lithium ion secondary battery including an electrolytic solution.

The teacher data in FIG. 9 includes each of the above items, which are evaluations, but some of them are not shown.

(4) Operation of Learning Model Generation Device 10 The outline of the operation of the learning model generation device 10 will be described below with reference to FIG. 10.
First, in step S11, the learning model generation device 10 activates the learning model generation program 15 stored in the storage unit 14. As a result, the learning model generation device 10 operates based on the learning model generation program 15, and starts generating the learning model.

In step S12, the acquisition unit 12 acquires a plurality of teacher data based on the learning model generation program 15.

In step S13, the acquisition unit 12 stores a plurality of teacher data in the database 16 constructed in the storage unit 14. The storage unit 14 stores a plurality of teacher data and manages them appropriately.

In step S14, the learning unit 13 extracts the learning data set from the teacher data stored in the storage unit 14. The A data set to be extracted is determined according to the learning purpose of the learning model generated by the learning model generation device 10. The dataset is based on teacher data.

In step S15, the learning unit 13 performs learning based on the extracted plurality of data sets.

In step S16, a learning model according to the learning purpose is generated based on the result learned by the learning unit 13 in step S15.

This completes the operation of the learning model generator 10. The order of operations of the learning model generation device 10 can be changed as appropriate. The generated learning model is used by being implemented in a general computer or terminal, being downloaded as software or an application, or being distributed in a state of being stored in a storage medium.

(5) Configuration of User Device 20 FIG. 2 shows the configuration of the user device 20 used by the user in the present embodiment. Here, the user is a person who inputs some information to the user device 20 or outputs some information. The user device 20 uses the learning model generated by the learning model generation device 10.

The user device 20 is a device having a computer function. The user device 20 may include a communication interface such as a NIC and a DMA controller, and may be able to communicate with the learning model generation device 10 and the like via a network. Although the user device 20 shown in FIG. 2 is shown as one device, it is preferable that the user device 20 supports cloud computing. Therefore, the hardware configuration of the user device 20 does not need to be housed in one housing or provided as a group of devices. For example, the resources of the hardware user device 20 are dynamically connected / disconnected according to the load.

The user device 20 includes, for example, an input unit 24, an output unit 25, a control unit 21, and a storage unit 26.

(5-1) Input unit 24
The input unit 24 is, for example, a keyboard, a touch panel, a mouse, or the like. The user can input information to the user device 20 via the input unit 24.

(5-2) Output unit 25
The output unit 25 is, for example, a display, a printer, or the like. The output unit 25 can output the result of analysis by the user device 20 using the learning model as well.

(5-3) Control unit 21
The control unit 21 is, for example, a CPU and executes control of the entire user device 20. The control unit 21 has functional units such as an analysis unit 22 and an update unit 23.

The analysis unit 22 of the control unit 21 analyzes the input information input via the input unit 24 using a learning model as a program stored in advance in the storage unit 26. The analysis performed by the analysis unit 22 is preferably performed by using the machine learning method described above, but is not limited thereto. The analysis unit 22 can output a correct answer even for unknown input information by using the learning model that has been learned in the learning model generation device 10.

The update unit 23 updates the learning model stored in the storage unit 26 to the optimum state in order to obtain a high-quality learning model. The update unit 23 optimizes the weighting between neurons in each layer, for example, in a neural network.

(5-4) Storage unit 26
The storage unit 26 is an example of a recording medium, and is composed of, for example, a flash memory, a RAM, an HDD, and the like. The storage unit 26 stores in advance the learning model executed by the control unit 21. A plurality of teacher data are stored in the database 27 in the storage unit 26, and each of them is appropriately managed. In addition, information such as a learning data set may be stored in the storage unit 26. The teacher data stored in the storage unit 26 is the above-mentioned information such as electrolyte component information and evaluation.

(6) Operation of the User Device 20 The outline of the operation of the user device 20 will be described below with reference to FIG. Here, the user device 20 is in a state in which the learning model generated by the learning model generation device 10 is stored in the storage unit 26.

First, in step S21, the user device 20 activates the learning model stored in the storage unit 26. The user device 20 operates based on the learning model.

In step S22, the user who uses the user device 20 inputs the input information via the input unit 24. The input information input via the input unit 24 is sent to the control unit 21.

In step S23, the analysis unit 22 of the control unit 21 receives the input information from the input unit 24, analyzes the information, and determines the information output by the output unit. The information determined by the analysis unit 22 is sent to the output unit 25.

In step S24, the output unit 25 outputs the result information received from the analysis unit 22.

In step S25, the update unit 23 updates the learning model to the optimum state based on the input information, the result information, and the like.

This completes the operation of the user device 20. The order of operations of the user device 20 can be changed as appropriate.

(7) Specific Examples In the following, specific examples using the above-mentioned learning model generation device 10 and user device 20 will be described.

(7-1) Resistance Learning Model Here, a resistance learning model that outputs the resistance of a lithium ion secondary battery as an evaluation will be described.

(7-1-1) Resistance learning model generator 10
In order to generate the resistance learning model, the resistance learning model generator 10 is at least
Electrolytic solution component information including information on LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of components contained in the electrolytic solution, and
Electrode information (device information) including information on the types and composition ratios of positive and negative electrode active materials, electrode densities, types and amounts of binders, and types and amounts of conductive aids,
Separator information (device information) including information on separator type, thickness and porosity,
Electrolyte volume information (device information) including information on electrolyte injection volume,
Charging / discharging information including information on voltage, current and temperature during charging / discharging,
Resistance information and
Multiple teacher data must be obtained, including. The resistance learning model generation device 10 may acquire other information.

The resistance learning model generator 10 performs learning based on the acquired teacher data.
Electrolytic solution component information including information on LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of components contained in the electrolytic solution, and
Electrode information (device information) including information on the types and composition ratios of positive and negative electrode active materials, electrode densities, types and amounts of binders, and types and amounts of conductive aids,
Separator information (device information) including information on separator type, thickness and porosity,
Electrolyte volume information (device information) including information on electrolyte injection volume,
Charging / discharging information including information on voltage, current and temperature during charging / discharging,
It is possible to generate a resistance learning model that outputs resistance information as input.

(7-1-2) User device 20 using a resistance learning model
The user device 20 is a device capable of using the resistance learning model. The user who uses the user device 20
Electrolytic solution component information including information on LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of components contained in the electrolytic solution, and
Electrode information (device information) including information on the types and composition ratios of positive and negative electrode active materials, electrode densities, types and amounts of binders, and types and amounts of conductive aids,
Separator information (device information) including information on separator type, thickness and porosity,
Electrolyte volume information (device information) including information on electrolyte injection volume,
Charging / discharging information including information on voltage, current and temperature during charging / discharging,
Is input to the user device 20.
The user device 20 determines the resistance information using the resistance learning model. The output unit 25 outputs the determined resistance information.

(7-2) Capacity Learning Model Here, a capacity learning model that outputs the capacity of a lithium ion secondary battery as an evaluation will be described.

(7-2-1) Capacitive learning model generator 10
In order to generate the capacity learning model, the capacity learning model generation device 10 is at least
Electrolytic solution component information including information on LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of components contained in the electrolytic solution, and
Electrode information (device information) including information on the types and composition ratios of positive and negative electrode active materials, electrode densities, types and amounts of binders, and types and amounts of conductive aids,
Separator information (device information) including information on separator type, thickness and porosity,
Electrolyte volume information (device information) including information on electrolyte injection volume,
Charging / discharging information including information on voltage, current and temperature during charging / discharging,
Capacity information and
Multiple teacher data must be obtained, including. The capacity learning model generation device 10 may acquire other information.

The capacity learning model generation device 10 performs learning based on the acquired teacher data, thereby performing learning.
Electrolytic solution component information including information on LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of components contained in the electrolytic solution, and
Electrode information (device information) including information on the types and composition ratios of positive and negative electrode active materials, electrode densities, types and amounts of binders, and types and amounts of conductive aids,
Separator information (device information) including information on separator type, thickness and porosity,
Electrolyte volume information (device information) including information on electrolyte injection volume,
Charging / discharging information including information on voltage, current and temperature during charging / discharging,
It is possible to generate a capacity learning model with input as input and capacity information as output.

(7-2-2) User device 20 using a capacitance learning model
The user device 20 is a device capable of using the capacity learning model. The user who uses the user device 20
Electrolytic solution component information including information on LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of components contained in the electrolytic solution, and
Electrode information (device information) including information on the types and composition ratios of positive and negative electrode active materials, electrode densities, types and amounts of binders, and types and amounts of conductive aids,
Separator information (device information) including information on separator type, thickness and porosity,
Electrolyte volume information (device information) including information on electrolyte injection volume,
Charging / discharging information including information on voltage, current and temperature during charging / discharging,
Is input to the user device 20.
The user device 20 determines the capacity information using the capacity learning model. The output unit 25 outputs the determined capacity information.

(7-3) Electrolyte component learning model Here, an electrolytic solution component learning model that outputs an optimum electrolytic solution component will be described.

(7-3-1) Electrolyte component learning model generator 10
In order to generate the electrolytic solution component learning model, the electrolytic solution component learning model generator 10 is at least
Electrolytic solution component information including information on LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of components contained in the electrolytic solution, and
Electrode information (device information) including information on the types and composition ratios of positive and negative electrode active materials, electrode densities, types and amounts of binders, and types and amounts of conductive aids,
Separator information (device information) including information on separator type, thickness and porosity,
Electrolyte volume information (device information) including information on electrolyte injection volume,
Charging / discharging information including information on voltage, current and temperature during charging / discharging,
Evaluation information including information on resistance, capacity, gas generation, charge / discharge efficiency and rate characteristics,
Multiple teacher data must be obtained, including. The electrolytic solution component learning model generation device 10 may acquire other information.

The electrolyte component learning model generator 10 performs learning based on the acquired teacher data, thereby performing learning.
Electrode information (device information) including information on the types and composition ratios of positive and negative electrode active materials, electrode densities, types and amounts of binders, and types and amounts of conductive aids,
Separator information (device information) including information on separator type, thickness and porosity,
Electrolyte volume information (device information) including information on electrolyte injection volume,
Charging / discharging information including information on voltage, current and temperature during charging / discharging,
Evaluation information including information on resistance, capacity, gas generation, charge / discharge efficiency and rate characteristics,
It is possible to generate an electrolytic solution component learning model that outputs the optimum electrolytic solution component information for obtaining the target evaluation.

(7-3-2) User device 20 using the electrolyte component learning model
The user device 20 is a device capable of using the electrolytic solution component learning model. The user who uses the user device 20
Electrode information (device information) including information on the types and composition ratios of positive and negative electrode active materials, electrode densities, types and amounts of binders, and types and amounts of conductive aids,
Separator information (device information) including information on separator type, thickness and porosity,
Electrolyte volume information (device information) including information on electrolyte injection volume,
Charging / discharging information including information on voltage, current and temperature during charging / discharging,
Evaluation information including information on resistance, capacity, gas generation, charge / discharge efficiency and rate characteristics,
Is input to the user device 20.
The user device 20 uses the electrolytic solution component learning model to determine the optimum electrolytic solution component information for obtaining the evaluation of the target. The output unit 25 outputs the determined electrolytic solution component information.

(8) Features (8-1)
The learning model generation method of the present embodiment is a learning model generation method for generating a learning model in which the evaluation of the electrochemical device including the electrolytic solution is determined by using a computer. The learning model generation method includes acquisition step S12, learning step S15, and generation step S16. In the acquisition step S12, the computer acquires the teacher data. The teacher data includes electrolyte component information and evaluation. The electrolytic solution component information is information on the components contained in the electrolytic solution. The evaluation is an evaluation of an electrochemical device. In the learning step S15, the computer learns based on the plurality of teacher data acquired in the acquisition step S12. In the generation step S16, a learning model is generated based on the result learned by the computer in the learning step S15. The learning model takes input information as input and outputs evaluation. The input information is unknown information different from the teacher data. The input information is at least information including electrolyte component information.

Further, as described above, the evaluation is determined by using the learning model in which the electrolyte component information and the evaluation are trained as teacher data in the computer as a program. The learning model includes an input step S22, a determination step S23, and an output step S24. In the input step S22, the input information which is the information including the electrolytic solution component information and is unknown information different from the teacher data is input. In the determination step S23, the evaluation is determined using the learning model. The output step S24 outputs the evaluation determined in the determination step S23.

Conventionally, in the evaluation of an electrochemical device equipped with an electrolytic solution, a performance test of an electrochemical device has been actually performed using various electrolytic solutions. In such a conventional evaluation method, a lot of time and steps are required to perform the evaluation, and improvement of the evaluation method has been required.
Further, as shown in Patent Document 2 (International Publication No. 99/07543), a program using a neural network or the like is designed in order to output optimum information in a different field, but electrochemical including an electrolytic solution is provided. In the special field of devices, programs using neural networks have not been designed.
The learning model generated by the learning model generation method of the present embodiment can be evaluated using a computer. It is possible to reduce a lot of time and process required in the past. Further, by reducing the number of processes, it is possible to reduce the number of personnel required for the evaluation, and the cost for the evaluation can also be reduced.

(8-2)
The learning model generation method of the present embodiment is a learning model generation method in which the optimum electrolyte component for obtaining the evaluation of the target electrochemical device is determined by using a computer. The acquisition step S12, the learning step S15, and the generation step S16 are included. In the acquisition step S12, the computer acquires the teacher data. The teacher data includes electrolyte component information and evaluation. The electrolytic solution component information is information on the components contained in the electrolytic solution. The evaluation is an evaluation of an electrochemical device including an electrolytic solution. In the learning step S15, the computer learns based on the plurality of teacher data acquired in the acquisition step S12. The generation step S16 generates a learning model based on the result learned by the computer in the learning step S15. The learning model takes input information as input and outputs electrolyte component information. The input information is unknown information different from the teacher data. The input information is at least information including evaluation information.

Further, as described above, the electrolytic solution component information is determined by using the learning model in which the electrolytic solution component information and the evaluation are learned as teacher data in the computer as a program. The program includes an input step S22, a determination step S23, and an output step S24. In the input step S22, input information which is information including evaluation information and is unknown information different from the teacher data is input. In the determination step S23, the learning model is used to determine the optimum electrolyte component information for obtaining the evaluation of the target. The output step S24 outputs the electrolytic solution component information determined in the determination step S23.

With conventional evaluation methods, if the electrochemical device is poorly evaluated, further research and improvement may be required to find the optimum electrolyte component to obtain the target evaluation, which takes a lot of time and time. A process was needed.
In the learning model generated by the learning model generation method of the present embodiment, it is possible to determine the optimum electrolytic solution component for obtaining the evaluation of the target by using a computer. This makes it possible to reduce the time, process, personnel, cost, etc. for selecting the optimum electrolytic solution component.

(8-3)
In the learning model generation method and program of (8-1) to (8-2) described above, the teacher data may further include at least one kind of information selected from the group consisting of device information and charge / discharge information. preferable. In this aspect, the input information preferably further includes at least one type of information selected from the group consisting of the device information and charge / discharge information.
The device information is information on the configuration of the electrochemical device, and the charge / discharge information is information on conditions for charging / discharging the electrochemical device.
The teacher data preferably contains information on many items, and the larger the number of teacher data, the more preferable. This makes it possible to obtain a more accurate output.

(8-4)
In the learning step S15 of the learning model generation method of the present embodiment, it is preferable to perform learning by regression analysis and / or ensemble learning in which a plurality of regression analyzes are combined.

The evaluation of the learning model as a program of the present embodiment preferably includes information on at least one selected from the group consisting of resistance, capacity, gas generation amount, charge / discharge efficiency, and rate characteristics of the electrochemical device.
The electrolytic solution component information may include at least one kind of information selected from the group consisting of LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of the components contained in the electrolytic solution. preferable.
Since these electrolyte component information have a strong correlation with the above-mentioned evaluation items for the electrochemical device, more accurate output can be obtained by using this information.

(8-5)
The learning model as a program of the present embodiment may be distributed via a storage medium in which the program is stored.

(8-6)
The trained model of the present embodiment is a trained model learned in the learning model generation method.
The trained model of the present embodiment performs an operation based on the weighting coefficient of the neural network on the electrolytic solution component information which is the information of the components contained in the electrolytic solution input to the input layer of the neural network, and performs the calculation based on the weighting coefficient of the neural network. It is a trained model for making a computer function so as to output an evaluation of an electrochemical device including the above-mentioned electrolyte from the output layer of the above. The weighting coefficient is obtained by learning using at least electrolyte component information and evaluation as teacher data.

(8-7)
The trained model of the present embodiment performs an operation based on the weighting coefficient of the neural network on the evaluation information of the electrochemical device including the electrolytic solution input to the input layer of the neural network, and outputs the neural network. It is a trained model for making the computer function so as to output the optimum electrolyte component information for obtaining the target evaluation. The weighting coefficient is obtained by learning using at least electrolyte component information and evaluation as teacher data.

(8-8)
In the trained models (8-6) to (8-7) described above, the teacher data preferably further includes at least one type of information selected from the group consisting of device information and charge / discharge information. In this embodiment, it is preferable that at least one type of information selected from the group consisting of the device information and charge / discharge information is further input to the input layer.
The device information is information on the configuration of the electrochemical device, and the charge / discharge information is information on conditions for charging / discharging the electrochemical device.

(8-9)
The learning model generation method, program, and trained model of the present disclosure use a relatively small amount (for example, 30% by mass or less, preferably 20% by mass or less, more preferably 10% by mass or less with respect to the electrolytic solution). It can be particularly preferably used when the information of the electrolytic solution component (electrolyte solution additive) is used as the input information and the evaluation of the electrochemical device (preferably a lithium ion secondary battery) is used as the output information.

(9)
Although the embodiments of the present disclosure have been described above, it will be understood that various modifications of the forms and details are possible without departing from the purpose and scope of the present disclosure described in the claims. ..

Next, the present disclosure will be described with reference to examples, but the present disclosure is not limited to such examples.

Example 1
Using the information of the electrolyte component (electrolyte additive) as the input information, the negative electrode resistance after the storage test was predicted and output.
The input information is the HOMO level, molecular structure, boiling point, electron density, molecular weight, molecular surface area, number of unsaturated bonds, density, electrolyte viscosity, and water content of the electrolytic solution additive. Both input / output information was standardized at the time of learning. Partial least squares regression was used for learning.
When learning was performed using the above input information, it was found that the mean square error was 0.016, and it was possible to predict with a certain degree of accuracy.

Example 2
When learning was performed in the same manner as in Example 1 except that the LUMO level was added to the input information, the mean square error was 0.012, and the prediction accuracy was improved as compared with Example 1.

S12 Acquisition step S15 Learning step S16 Generation step S22 Input step S23 Determination step S24 Output step

Claims (13)

A learning model generation method that generates a learning model that uses a computer to determine the evaluation of an electrochemical device containing an electrolyte.
At least, an acquisition step (S12) in which the computer acquires information including the electrolyte component information which is the information of the components contained in the electrolytic solution and the evaluation of the electrochemical device as teacher data.
A learning step (S15) in which the computer learns based on the plurality of teacher data acquired in the acquisition step (S12), and a learning step (S15).
A generation step (S16) in which the computer generates the learning model based on the result of learning in the learning step (S15).
With
The learning model outputs the evaluation by inputting input information which is unknown information different from the teacher data.
The input information is information including at least the electrolyte component information.
Learning model generation method. It is a learning model generation method that generates a learning model that uses a computer to determine the optimum electrolyte component information for obtaining an evaluation of a target electrochemical device.
At least, an acquisition step (S12) in which a computer acquires information including electrolyte component information, which is information on components contained in the electrolytic solution, and evaluation of an electrochemical device including the electrolytic solution, as training data.
A learning step (S15) in which the computer learns based on the plurality of teacher data acquired in the acquisition step (S12), and a learning step (S15).
A generation step (S16) in which the computer generates a learning model based on the result of learning in the learning step (S15).
With
The learning model takes input information, which is unknown information different from the teacher data, as input, and outputs optimum electrolyte component information for obtaining a target evaluation.
The input information is information including at least the evaluation information.
Learning model generation method. The teacher data further includes at least one type of information selected from the group consisting of device information and charge / discharge information.
The input information further includes at least one type of information selected from the group consisting of the device information and charge / discharge information.
The device information is information on the configuration of the electrochemical device, and is
The charge / discharge information is information on conditions for charging / discharging the electrochemical device.
The learning model generation method according to claim 1 or 2. In the learning step (S15), learning is performed by regression analysis and / or ensemble learning in which a plurality of regression analyzes are combined.
The learning model generation method according to any one of claims 1 to 3. A computer is a program that uses a learning model to determine the evaluation of an electrochemical device with an electrolyte.
In the input step (S22) in which the computer inputs the input information,
In the determination step (S23) in which the computer determines the evaluation,
An output step (S24) in which the computer outputs the evaluation determined in the determination step (S23), and
With
The learning model learns at least the information including the electrolytic solution component information, which is the information of the components contained in the electrolytic solution, and the evaluation, as teacher data.
The input information is information including at least the electrolyte component information, and is unknown information different from the teacher data.
program. A program in which a computer uses a learning model to determine optimal electrolyte component information to obtain an evaluation of a target electrochemical device.
In the input step (S22) in which the computer inputs the input information,
In the determination step (S23), in which the computer determines the optimum electrolyte component information,
An output step (S24) in which the computer outputs the optimum electrolytic solution component information determined in the determination step (S23), and
With
The learning model learns at least the information including the electrolyte component information which is the information of the components contained in the electrolytic solution and the evaluation of the electrochemical device including the electrolytic solution as teacher data.
The input information is information including at least the evaluation information, and is unknown information different from the teacher data.
program. The teacher data further includes at least one type of information selected from the group consisting of device information and charge / discharge information.
The input information further includes at least one type of information selected from the group consisting of the device information and charge / discharge information.
The device information is information on the configuration of the electrochemical device, and is
The charge / discharge information is information on conditions for charging / discharging the electrochemical device.
The program according to claim 5 or 6. The evaluation includes information on at least one selected from the group consisting of resistance, capacity, gas generation, charge / discharge efficiency and rate characteristics of the electrochemical device.
The program according to any one of claims 5 to 7. The electrolytic solution component information includes at least one information selected from the group consisting of LUMO level, HOMO level, molecular structure, melting point, boiling point, electron density and molecular weight of the components contained in the electrolytic solution.
The program according to any one of claims 5 to 8. A storage medium in which the program according to any one of claims 5 to 9 is stored. The computer functions so as to perform an operation based on the weighting coefficient of the neural network on the electrolyte component information input to the input layer of the neural network and output the evaluation of the electrochemical device from the output layer of the neural network. It ’s a trained model to make you
The weighting coefficient is obtained by learning using at least the electrolyte component information and the evaluation as teacher data.
The electrolytic solution component information is information on the components contained in the electrolytic solution.
The evaluation is an evaluation of an electrochemical device including the electrolytic solution.
Trained model. The evaluation information input to the input layer of the neural network is calculated based on the weighting coefficient of the neural network, and the optimum electrolyte component information for obtaining the target evaluation from the output layer of the neural network is output. A trained model for making a computer work, like
The weighting coefficient is obtained by learning using at least the electrolyte component information and the evaluation as teacher data.
The electrolytic solution component information is information on the components contained in the electrolytic solution.
The evaluation is an evaluation of an electrochemical device including the electrolytic solution.
Trained model. The teacher data further includes at least one type of information selected from the group consisting of device information and charge / discharge information.
At least one type of information selected from the group consisting of the device information and charge / discharge information is further input to the input layer.
The device information is information on the configuration of the electrochemical device, and is
The charge / discharge information is information on conditions for charging / discharging the electrochemical device.
The trained model according to claim 11 or 12.

Images