Secondary battery deterioration location estimation system, secondary battery control system, and secondary battery deterioration location estimation method

Abstract

To provide a secondary battery degradation point estimation system capable of efficiently and accurately estimating the state of degradation at each degradation point of a secondary battery, and to provide a secondary battery control system and a secondary battery degradation point estimation method.SOLUTION: A secondary battery degradation point estimation system is configured to estimate equivalent circuit model parameters of a secondary battery and use a learning model machine-trained using equivalent circuit model parameters and degradation point information as teacher data to output degradation point information estimated from the estimated equivalent circuit model parameters.SELECTED DRAWING: Figure 17

Description

Technical Field

[0001] The present invention relates to a secondary battery deterioration location estimation system, a secondary battery control system, and a secondary battery deterioration location estimation method.

Background Art

[0002] Techniques for detecting the status of each of the positive and negative electrodes inside a secondary battery are known (see, for example, Patent Document 1). In this technique, the unique charge-discharge curves of each of the positive electrode material and the negative electrode material as basic data and the charge-discharge curve of the battery under test as a measured value are acquired, and the charge-discharge curves of the positive and negative electrodes inside the battery under test are obtained using these and predetermined correction parameters.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the above technique, in order to acquire the charge-discharge curve of the battery under test as a measured value, it is necessary to perform a set of charge-discharge operations from full discharge to full charge. That charge-discharge operation needs to be performed separately from normal operation, and further, it needs to be performed with a constant and low-rate charge-discharge current. Therefore, energy and time losses occur in order to estimate the status of each of the positive and negative electrodes of the secondary battery (specifically, the deterioration state for each deterioration location).

[0005] In view of the above circumstances, an object of the present invention is to provide a secondary battery deterioration location estimation system, a secondary battery control system, and a secondary battery deterioration location estimation method that can efficiently and highly accurately estimate the deterioration state for each deterioration location of a secondary battery.

Means for Solving the Problems

[0006] The deterioration location estimation system for a secondary battery according to the present invention includes A measurement unit that measures the AC impedance, current, and voltage of a secondary battery, and based on the AC impedance, current, and voltage of the secondary battery measured by the measurement unit, the a parameter estimation unit that estimates the equivalent circuit model parameters of the secondary battery, and deterioration location information true value and a deterioration location information output unit that outputs the deterioration location information estimated from the equivalent circuit model parameters estimated by the parameter estimation unit using a learning model that has been machine-learned with the equivalent circuit model parameters and the deterioration location information as teacher data. The equivalent circuit model parameters are Foster-type equivalent circuit model parameters or Cowell-type equivalent circuit model parameters, and the learning model is characterized by being machine-learned by random forest It is provided with.

[0007] The secondary battery control system according to the present invention includes the deterioration location estimation system for the secondary battery, and based on the deterioration location information output from the deterioration location information output unit, setting the upper and lower voltage limits of the secondary battery, setting the current limit value of the secondary battery, and And a control unit that executes at least one of the abnormality detection of the secondary battery.

[0008] The method for estimating the deterioration location of a secondary battery according to the present invention includes A measurement step of measuring the AC impedance, current, and voltage of a secondary battery, and based on the AC impedance, current, and voltage of the secondary battery measured in the measurement step, the a parameter estimation step of estimating the equivalent circuit model parameters of the secondary battery, and deterioration location information true value and a deterioration location information output step of outputting the deterioration location information estimated from the equivalent circuit model parameters estimated in the parameter estimation step using a learning model that has been machine-learned with the equivalent circuit model parameters and the deterioration location information as teacher data, and is executed using a computer. The equivalent circuit model parameters are Foster-type equivalent circuit model parameters or Cowell-type equivalent circuit model parameters, and the learning model is characterized by being machine-learned by random forest To do.

Effect of the Invention

[0009] According to the present invention, it is possible to efficiently and highly accurately estimate the deterioration state for each deterioration location of the secondary battery.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, the present invention will be described along with preferred embodiments. Note that the present invention is not limited to the embodiments shown below, and can be appropriately changed without departing from the gist of the present invention. Also, in the embodiments shown below, there are some places where the illustration and description of some configurations are omitted. For the details of the omitted technology, well-known or widely-known technologies are appropriately applied within the range where there is no contradiction with the content described below.

[0012] FIG. 1 is a diagram showing a schematic configuration of a secondary battery deterioration location estimation system 1 and a secondary battery control system 100 according to an embodiment of the present invention. The deterioration location estimation system 1 shown in this figure is a system that estimates the deterioration state for each deterioration location of an in-vehicle or stationary secondary battery (a lithium-ion battery in this embodiment). In the following description, the "deterioration state for each deterioration location" is referred to as "deterioration location information".

[0013] Examples of the "deterioration location information" include the active Li (lithium) capacity, the positive electrode charge reserve capacity, and the negative electrode capacity. The active Li capacity [mAh] is the amount of Li that can be involved in charge and discharge inside the secondary battery, and decreases in response to deterioration. The positive electrode charge reserve capacity [mAh] is the capacity within the positive electrode that is not used within the range of the depth of discharge, and increases in response to deterioration. Furthermore, the negative electrode capacity [mAh] is the charge and discharge-capable capacity of the negative electrode, and decreases in response to deterioration.

[0014] The secondary battery control system 100 includes the deterioration location estimation system 1, and based on the deterioration location information output from the deterioration location estimation system 1, executes a process of setting the upper and lower voltage limits and the limiting values of the charge and discharge current of the secondary battery in use, and a process of detecting an abnormality in the secondary battery in use.

[0015] As shown in FIG. 1, the deterioration location estimation system 1 includes a CPU (Central Processing Unit) 10 and a memory 11 such as a RAM (Random Access Memory) and a ROM (Read Only Memory). The memory 11 stores a program for the CPU 10 to execute the estimation process of the deterioration location information and the learning data described later. The CPU 10 includes a parameter estimation unit 101 and a deterioration location information output unit 102.

[0016] The secondary battery control system 100 includes a CPU 110 and a memory 111 such as a RAM and a ROM. In addition to the program for the CPU 110 to execute the process of setting the upper and lower voltage limits and the charge and discharge current limit values of the secondary battery in use and the process of detecting abnormalities in the secondary battery in use, the memory 111 stores deterioration location information, the upper and lower voltage limits of the secondary battery, and setting information of the charge and discharge current limit values.

[0017] Here, the learning model stored in the memory 11 is a learning model obtained by machine learning using the parameters of the equivalent circuit model of the secondary battery (hereinafter referred to as "equivalent circuit model parameters") and the true values of the deterioration location information as teacher data. This learning model is created in advance by performing the tests described later. In the deterioration location estimation system 1 of the present embodiment, the parameter estimation unit 101 of the CPU 10 estimates the equivalent circuit model parameters by a predetermined method described later, and the deterioration location information output unit 102 of the CPU 10 uses the learning model stored in the memory 11 to estimate the deterioration location information from the estimated equivalent circuit model parameters and outputs it to the secondary battery control system 100.

[0018] The creation of a learning model will be described below. Figure 2 is a diagram showing the equivalent circuit model of a secondary battery. In the present embodiment, the Foster-type equivalent circuit model shown in Figure 2 is used as the equivalent circuit model of the secondary battery. In this Foster-type equivalent circuit model, L0 represents an inductance component, Rn represents a resistance component, and Cn represents a capacitance component. Also, n is an integer indicating the number of stages of the RC parallel circuit, and the resistance component connected in series before the first-stage RC parallel circuit is R0. In the Foster-type equivalent circuit model of the present embodiment, the inductance component L0, the resistance components R0 to R6, and the capacitance components C1 to C6 correspond to the equivalent circuit model parameters.

[0019] The Foster-type equivalent circuit model parameters can be obtained by known methods such as the electrochemical impedance method, the estimation method using a Kalman filter, and the method using Bayesian estimation. The electrochemical impedance method is a method of measuring the impedance spectrum for each frequency using a frequency response analyzer (FRA) and a potentiostat / galvanostat. By analyzing the Nyquist plot showing the impedance spectrum obtained by this measurement, the electrode characteristics of the secondary battery can be obtained without disassembling the secondary battery (i.e., non-destructively). Specifically, information regarding frequency characteristics such as complex impedance is acquired, a Nyquist plot showing the impedance spectrum is created, an equivalent circuit model corresponding to the Nyquist plot is set, and the set equivalent circuit model is fitted to the impedance spectrum. By this fitting analysis, the parameters (equivalent circuit model parameters) that make the frequency response of the equivalent circuit model equivalent to the impedance spectrum are set for each component L0, R0 to R6, C1 to C6 of the equivalent circuit model.

[0020] As a simpler method compared to the electrochemical impedance method, estimation methods using the above-mentioned Kalman filter (see, for example, Japanese Patent Application Laid-Open No. 2012-149947) and methods using Bayesian estimation (Japanese Patent Application Laid-Open No. 2018-72090) are known. These methods are suitable for practical applications such as electric vehicle batteries and stationary batteries.

[0021] Figure 3 is a flowchart showing the procedure for creating a learning model. In creating the learning model shown in this flowchart, 11 sample batteries were prepared and experiments were conducted to obtain true values used in machine learning. As the sample battery, NCR18650GA (manufactured by Panasonic) was selected.

[0022] First, an accelerated degradation test was performed on 10 sample batteries, and the AC impedance of the sample batteries was measured at predetermined timings (step S1). In the accelerated degradation test, charge and discharge with a discharge depth of 0% to 100% was performed 100 cycles in an environment at 10°C. The measurement of the AC impedance was performed at timings of 0 cycle, 20 cycles, 40 cycles, 70 cycles, and 100 cycles.

[0023] In addition, in order to obtain information on the degradation location in the initial state, charge and discharge tests and the like were performed using one sample battery different from the above-mentioned 10 sample batteries (step S1). Here, the one sample battery was completely discharged once from the initial state and then disassembled, and analysis and charge and discharge tests were performed to obtain information on the degradation location in the initial state. Specifically, single-pole cells of the positive electrode and the negative electrode were created from the disassembled sample battery, and the amount of active Li remaining in the negative electrode was quantified by ICP (Inductively Coupled Plasma) emission analysis. Also, the positive electrode active material capacity was quantified by charging and discharging the single-pole cells of the positive electrode and the negative electrode. Furthermore, the positive electrode discharge reserve capacity was quantified from the difference between the positive electrode active material capacity and the battery capacity.

[0024] In the accelerated degradation test of 10 sample batteries, the discharge curves synthesized from the charge-discharge curves of the initial single cells were fitted to the charge-discharge curves of the degraded batteries, and the estimated values of the active Li capacity, the positive electrode discharge reserve capacity, and the negative electrode capacity after degradation were obtained (Reference: K. Honkura, H. Honbo, Y. Koshikawa, and T. Horiba: State Analysis of Lithium-Ion batteries Using Discharge Curves, ECS Transactions, 13(19)61 / 73(2008)).

[0025] Confirm that the information necessary for performing machine learning has been obtained (step S2), and proceed to the next step. Specifically, confirm that the information on the degradation location of the initial state of one sample battery and the information on the degradation location of 10 sample batteries at 0 to 100 cycles, which are used as the true values of the degradation location information in machine learning, have been obtained. In addition, confirm that information such as the measured values of the alternating current impedance necessary for setting the equivalent circuit model parameters has been obtained. If it cannot be confirmed that the information necessary for machine learning has been obtained (step S2: NO), return to step S1.

[0026] After the degradation location information used as the true value in machine learning and the information necessary for setting the equivalent circuit model parameters are complete (step S2: YES), first, set the equivalent circuit model parameters (step S3). In this step, set the equivalent circuit model as described above, and fit the equivalent circuit model to the impedance spectrum to set the equivalent circuit model parameters. Specifically, measure the alternating current impedance at 0 cycles, 20 cycles, 40 cycles, 70 cycles, and 100 cycles during the accelerated degradation test, and based on the measurement results, fit the equivalent circuit model to the impedance spectrum using the electrochemical software ZVIEW manufactured by SCRIBNER to obtain the equivalent circuit model parameters for each cycle.

[0027] Figure 4 is a table showing the equivalent circuit model parameters of the specimen battery and the information on the degradation location. As shown in this table, the equivalent circuit model parameters at 0 cycles, 20 cycles, 40 cycles, 70 cycles, and 100 cycles of 10 specimen batteries are set. Note that although the cycle numbers are shown in the table of Figure 4, in actual machine learning, the cycle numbers are not used as information. Also, in the table of Figure 4, Li_cap [mAh] is the active Li capacity, p_reserve [mAh] is the positive electrode charge reserve capacity, and n_cap [mAh] is the negative electrode capacity.

[0028] Next, as shown in Figure 3, by performing machine learning using the equivalent circuit model parameters set in step S3 and the true values of the degradation location information as teacher data, the correlation between the equivalent circuit model parameters and the degradation location information is obtained (step S4). As the machine learning algorithm, various known ones can be adopted, but in this embodiment, random forest is adopted. The features of this random forest include being fast due to its ability to perform parallel learning and being able to calculate the importance of features.

[0029] I(j), which is the importance of a certain feature j, is expressed by the following formula (1). By calculating the importance, it is possible to know which feature contributes to the branching of the sub-regression tree. [Equation] n: The number of nodes having the feature j as a branching condition N p (i): The number of samples of a certain node i E p (i): The MSE (Mean Square Error) of a certain node i N l (i): The number of samples of the left node among the child nodes of i E l (i): The MSE of the left node among the child nodes of i N r (i): The number of samples of the right node among the child nodes of i E r (i): MSE of the right node among the child nodes of i

[0030] Here, the relationship between the equivalent circuit model parameters and the degradation location information of the secondary battery is complex, and it is difficult for humans to predict which degradation location information is affected by the change in the equivalent circuit model parameters. In contrast, random forest is suitable for analysis in a state where it is not known which equivalent circuit model parameter and which degradation location information affect each other because it creates regression trees by randomly adopting features.

[0031] As described above, various known machine learning algorithms can be adopted. For example, it is also generally common to adopt neural networks, support vector machines, etc. (References: Japanese Patent Application Laid-Open No. 2021-086666, Japanese Patent Publication No. 2016-536605, etc.).

[0032] Using the equivalent circuit model parameters and degradation location information shown in the table of FIG. 4 as input data, machine learning was performed using the regression model of random forest. At that time, from the viewpoint of making the value of each degradation location information estimated from the equivalent circuit model parameters approach the value of the degradation location information given in advance as the true value, the learning model was set as the mean squared error E deg to be a model that minimizes. [Number] m: Total number of samples DI t (i): True value of degradation location information in the i-th sample DI p (i): Estimated value of degradation location information in the i-th sample

[0033] FIG. 5 is a table showing the hardware and software environments. The hardware and software environments used in creating the learning model in this experiment are as shown in the table of FIG. 5.

[0034] Hyperparameter tuning was performed using k-fold cross-validation and grid search. K-fold cross-validation is a method of dividing a given dataset into k parts, using one data as evaluation data, and evaluating the remaining data as learning data. In this method, the evaluation is repeated while changing the combination of evaluation data and learning data, and finally, a highly accurate estimated value is obtained by calculating the average of the evaluation values. In this experiment, the number of cross-validation splits was set to 2 in accordance with the fact that the number of datasets was 50.

[0035] Grid search is a method of comprehensively evaluating the search range (combinations) of predetermined setting values in order to adjust hyperparameters and selecting the best setting value. Figure 6 is a table showing the search range of grid search. The search range of grid search in this experiment was set as shown in the table of Figure 6.

[0036] Figure 7 is a table showing the relationship between the estimated value and the true value of each degradation location information. Figure 8 is a graph showing the relationship between the estimated value and the true value of the active Li capacity, Figure 9 is a graph showing the relationship between the estimated value and the true value of the positive electrode charge reserve capacity, and Figure 10 is a graph showing the relationship between the estimated value and the true value of the negative electrode capacity.

[0037] The evaluation of the estimation results of each degradation location information is performed based on how close the estimated value of each degradation location information is to the true value. Furthermore, in order to perform a quantitative evaluation, the coefficient of determination R of each degradation location information is calculated according to the following formula (3). 2 is calculated according to the following formula (3).

Equation

[0038] In addition, the importance of the feature amount calculated according to the above formula (1) and the visualization of the sub-decision tree of the learning model by a graph creation tool (for example, Graphviz (Version 2.47.0) etc.) were used for analysis.

[0039] The vertical axis of the graphs in FIGS. 8 to 10 is the true value of each deterioration location information, and the horizontal axis of the graphs in FIGS. 8 to 10 is the estimated value of each deterioration location information. In these graphs, if the true value and the estimated value are plotted on a straight line at 45°, it means that the estimation can be performed with high accuracy. From the graphs in FIGS. 8 to 10, it can be confirmed that the active Li capacity, the positive electrode charge reserve capacity, and the negative electrode capacity can be estimated with high accuracy.

[0040] Next, the creation of the learning model according to another embodiment will be described. FIG. 11 is a diagram showing an equivalent circuit model of a secondary battery. In the present embodiment, as the equivalent circuit model of the secondary battery, the Cowell type equivalent circuit model shown in FIG. 11 is used. In this Cowell type equivalent circuit model, L0 represents an inductance component, Rn represents a resistance component, and Cn represents a capacitance component. The Cowell type equivalent circuit model is an equivalent circuit model in which each of n resistance components R0 to Rn is connected in parallel between n capacitance components C1 to Cn connected in series. In the Cowell type equivalent circuit model of the present embodiment, the inductance component L0, the resistance components R0 to R6, and the capacitance components C1 to C6 correspond to the equivalent circuit model parameters. The Cowell type equivalent circuit model parameters can be obtained by the same method as the above-described Foster type equivalent circuit model parameters.

[0041] The creation of the learning model of the present embodiment is also performed according to the procedure shown in the flowchart of FIG. 3 described above. The difference from the above-described embodiment is that in step S3, the Cowell type equivalent circuit model parameters are set, and in step S4, the correlation between the Cowell type equivalent circuit model parameters and the deterioration location information is obtained by performing machine learning using the Cowell type equivalent circuit model parameters and the true value of the deterioration location information as teacher data.

[0042] FIG. 12 is a table showing the equivalent circuit model parameters of the specimen battery and the information on the degradation location. As shown in this table, the equivalent circuit model parameters of 10 specimen batteries at 0 cycles, 20 cycles, 40 cycles, 70 cycles, and 100 cycles are set. Although the number of cycles is shown in the table of FIG. 12, in actual machine learning, the number of cycles is not used as information.

[0043] Using the equivalent circuit model parameters and the information on the degradation location shown in the table of FIG. 12 as input data, machine learning was performed using the regression model of the random forest. At that time, from the viewpoint of making the value of each degradation location information estimated from the equivalent circuit model parameters approach the value of the degradation location information given in advance as the true value, the learning model was set as the mean squared error E deg minimizing model represented by the above formula (2).

[0044] The hardware and software environments specified in creating the learning model in this experiment are as shown in the table of FIG. 5. Also, the tuning of hyperparameters and the search range of grid search are the same as those in the above-described embodiment (see FIG. 6).

[0045] FIG. 13 is a table showing the relationship between the estimated value and the true value of each degradation location information. FIG. 14 is a graph showing the relationship between the estimated value and the true value of the active Li capacity, FIG. 15 is a graph showing the relationship between the estimated value and the true value of the positive electrode charge reserve capacity, and FIG. 16 is a graph showing the relationship between the estimated value and the true value of the negative electrode capacity.

[0046] The evaluation of the estimation result of each degradation location information is performed based on how close the estimated value of each degradation location information is to the true value. Further, in order to perform a quantitative evaluation, the coefficient of determination R 2 of each degradation location information is calculated according to the above formula (3). Also, the importance of the feature calculated according to the above formula (1) and the visualization of the sub-decision tree of the learning model by a graph creation tool (for example, Graphviz (Version 2.47.0), etc.) were used for the analysis.

[0047] The vertical axis of the graphs in FIGS. 14 to 16 represents the true value of each deterioration location information, and the horizontal axis of the graphs in FIGS. 14 to 16 represents the estimated value of each deterioration location information. In these graphs, if the true value and the estimated value are plotted on a straight line at a 45° angle, it means that the estimation can be made with high accuracy. From the graphs in FIGS. 14 to 16, it can be confirmed that the active Li capacity, the positive electrode charge reserve capacity, and the negative electrode capacity can be estimated with high accuracy.

[0048] The learning model created by the method described above is stored in the memory 11 (see FIG. 1) of the deterioration location estimation system 1 of the present embodiment. During the operation of an electric vehicle or a stationary battery equipped with the deterioration location estimation system 1 and the secondary battery control system 100, the parameter estimation unit 101 (see FIG. 1) of the deterioration location estimation system 1 estimates the equivalent circuit model parameters from the measured values of the AC impedance and the like, and the deterioration location information output unit 102 uses the learning model stored in the memory 11 to output the deterioration location information estimated from the equivalent circuit model parameters to the secondary battery control system 100. Hereinafter, the estimation of the deterioration location information during the operation of an electric vehicle or a stationary battery will be described.

[0049] FIG. 17 is a flowchart showing the processing of the deterioration location estimation system 1 and the secondary battery control system 100 of the present embodiment. As shown in this flowchart, the CPU 10 of the deterioration location estimation system 1 executes measurements of the AC impedance, current, voltage, etc. of the secondary battery (step S11). Next, the parameter estimation unit 101 of the CPU 10 estimates the equivalent circuit model parameters of the secondary battery from the measured AC impedance, etc. of the secondary battery (step S12). The method for estimating the equivalent circuit model parameters in this step is the same as that at the time of creating the learning model.

[0050] Next, the deterioration location information output unit 102 of the CPU 10 uses the learning model stored in the memory 11 to output the deterioration location information estimated from the equivalent circuit model parameters estimated in step S12 to the secondary battery control system 100 (step S13). Examples of the deterioration location information include the active Li amount, the positive electrode discharge reserve capacity, the negative electrode capacity, and the like.

[0051] Next, the CPU 110 of the secondary battery control system 100 executes a process of setting the upper and lower voltage limits of the secondary battery, a process of setting the current limit value of the secondary battery, and a process of detecting an abnormality of the secondary battery based on the inputted degradation location information (step S14).

[0052] As described above, the degradation location estimation system 1 according to the present embodiment estimates the equivalent circuit model parameters of the secondary battery based on measurement values of the AC impedance and the like, and uses a learning model in which the equivalent circuit model parameters and the degradation location information are machine-learned as teacher data to output degradation location information (active Li capacity, positive electrode reserve capacity, negative electrode capacity, etc.) estimated from the estimated equivalent circuit model parameters. As a result, it is possible to estimate the degradation location information of the secondary battery without having to perform charging and discharging from full discharge to full charge separately from normal operation. Therefore, it is possible to estimate the degradation location information of the secondary battery efficiently without causing energy and time losses. Also, as described above, by using a learning model that has learned the correlation between the equivalent circuit model parameters and each degradation location information, the accuracy of estimating the degradation location information of the secondary battery can be ensured.

[0053] Further, the secondary battery control system 100 according to the present embodiment executes a process of setting the upper and lower voltage limits and the current limit value for the purpose of suppressing degradation of the secondary battery, and a process of detecting an abnormality such as a short circuit of the secondary battery. Since the secondary battery control system 100 according to the present embodiment executes these processes based on information on detailed degradation locations inside the secondary battery such as the active Li capacity, the positive electrode charge reserve capacity, and the negative electrode capacity, it is possible to suppress insufficient control amount, excessive control, and false detection.

[0054] The present invention has been described based on the embodiments. However, the present invention is not limited to the above embodiments, and modifications may be made without departing from the spirit of the present invention, or known and well-known techniques may be appropriately combined. For example, although the Foster-type equivalent circuit model and the Cowell-type equivalent circuit model are cited as examples of the equivalent circuit model of the secondary battery, other equivalent circuit models of the secondary battery may be used. Further, the above-described deterioration location information is an example, and other information such as the negative electrode reserve capacity in which a correlation is recognized with respect to the equivalent circuit model parameters is also included in the deterioration location information.

Explanation of Signs

[0055] 1: Deterioration Location Estimation System 100: Secondary Battery Control System 101: Parameter Estimation Unit 102: Deterioration Location Information Output Unit 110: CPU (Control Unit)

Claims

1. A measurement unit that measures the AC impedance, current, and voltage of a secondary battery, a parameter estimation unit that estimates the equivalent circuit model parameters of the secondary battery based on the AC impedance, current, and voltage of the secondary battery measured by the measurement unit, a degradation location information output unit that outputs the degradation location information estimated from the equivalent circuit model parameters estimated by the parameter estimation unit using a learning model that has been machine-learned with the equivalent circuit model parameters and the true value of the degradation location information as teacher data comprising: the equivalent circuit model parameters are Foster-type equivalent circuit model parameters or Kauer-type equivalent circuit model parameters, the learning model is a secondary battery degradation location estimation system characterized by being machine-learned by random forest.

2. The secondary battery degradation location estimation system according to Claim 1, wherein the degradation location information is at least one of the active lithium capacity, the positive electrode charge reserve capacity, and the negative electrode capacity.

3. The secondary battery degradation location estimation system according to Claim 1 or 2, and a control unit that executes at least one of the processes of setting the upper and lower limit voltages of the secondary battery, setting the current limit value of the secondary battery, and detecting an abnormality of the secondary battery based on the degradation location information output from the degradation location information output unit A secondary battery control system comprising:

4. A measurement step of measuring the AC impedance, current, and voltage of a secondary battery, a parameter estimation step of estimating the equivalent circuit model parameters of the secondary battery based on the AC impedance, current, and voltage of the secondary battery measured in the measurement step, a degradation location information output step of outputting the degradation location information estimated from the equivalent circuit model parameters estimated in the parameter estimation step using a learning model that has been machine-learned with the equivalent circuit model parameters and the true value of the degradation location information as teacher data executed using a computer, the equivalent circuit model parameters are Foster-type equivalent circuit model parameters or Kauer-type equivalent circuit model parameters, the learning model is a secondary battery degradation location estimation method characterized by being machine-learned by random forest.

Images