Residual battery capacity prediction method

By constructing a regression formula based on discharge voltage and current, using explanatory variables such as overall slope, difference and reduction ratio, the problem of low prediction accuracy of battery capacity margin in the prior art is solved, and fast and accurate prediction of battery margin is achieved. Reduced the failure rate.

CN119013572APending Publication Date: 2024-11-22TOYOTA MOTOR KYUSHU
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202380027693.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2022-03-17
Filing Date
2023-02-16
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In the prior art, the battery capacity margin prediction method has a low accuracy and requires a long time to measure, so it is impossible to accurately predict the true margin of the battery in a short time.

Method used

By defining the target variables, extracting important and linear explanatory variables, derive the regression formula and predicting the margin of battery capacity using the time lapse of discharge voltage and current, using the overall slope, difference and reduction ratio of discharge voltage and current as explanatory variables, combined with The initial voltage is used as a constant term, and a regression formula is constructed for prediction.

Benefits of technology

The margin for high-precision prediction of battery capacity in a short period of time is achieved, which reduces the unqualification rate and improves the accuracy and efficiency of the judgment of battery quality and bad.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119013572A_ABST
    Figure CN119013572A_ABST
Patent Text Reader

Abstract

The invention provides a remaining battery capacity prediction method capable of predicting a true value in a short time by using discharge measurement of a battery. A remaining capacity prediction method of a battery capacity for predicting a remaining capacity of a battery that can be repeatedly used by charging, in which a target variable is defined to predict a remaining capacity of the battery; extracting a plurality of explanatory variables having high importance and linearity as elements for margin prediction; deriving a relational expression representing a relationship between the target variable and the explanatory variable as a regression formula; performing time lapse prediction on at least any one of the discharge voltage value and the discharge current of the battery by using a regression formula; a remaining amount of battery capacity is predicted.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a method for predicting the remaining capacity of a battery, which can predict the remaining capacity of batteries such as nickel-hydrogen batteries, nickel-cadmium batteries, nickel-zinc batteries, lithium-ion secondary batteries, rechargeable alkaline batteries, etc. that can be repeatedly used by charging. Background Art

[0002] In recent years, the movement to limit carbon dioxide emissions, which is a cause of global warming, has become active. As a result, for example, automobiles have been gradually shifting from driving methods such as engines using fossil fuels such as gasoline and light oil to hybrid methods, and further to driving methods such as electric motors using rechargeable secondary batteries.

[0003] Such recycling activities such as the reuse of secondary batteries are becoming more active. In view of this situation, a battery information processing system has been proposed as one of the methods for improving the estimation accuracy of the full charge capacity of a modular battery pack (see, for example, Patent Document 1). The system processes information for estimating the full charge capacity of a module and includes a storage device for storing a learned neural network model and an analysis device for estimating the full charge capacity of a secondary battery based on the AC impedance measurement result of the module using the learned neural network model.

[0004] In addition, in order to improve the estimation accuracy of the full charge capacity of the battery, a learned neural network model is also proposed (for example, refer to patent document 2). The neural network model is a model that has been learned based on the Nyquist diagrams of multiple modules with full charge capacity within the reference range. In the processing system, by using at least one feature quantity extracted from the Nyquist diagram of the module as a discriminant analysis of the explanatory variable, it is determined which group the module belongs to, the first group with full charge capacity within the reference range and the second group with full charge capacity outside the reference range. And, in the processing system, for example, in the case where it is determined that the module belongs to the first group, the learned neural network model is used to estimate the full charge capacity of the module.

[0005] In addition, a system for predicting with high accuracy whether the battery capacity can be restored by a capacity recovery process has been proposed (for example, see Patent Document 3). That is, in this system, when a capacity recovery process is performed on a module that is a nickel-hydrogen battery, the Mahalanobis-Taguchi method (MT method: Maharanobis-Taguchi System) using multiple explanatory variables is used to determine which group the module belongs to, a first group whose battery capacity is lower than a reference capacity and a second group whose battery capacity exceeds the reference capacity. In addition, the characteristic quantities of the multiple explanatory variables include at least two real number components of AC impedance drawn in the semicircular portion, at least two imaginary number components of AC impedance drawn in the semicircular portion, and at least one imaginary number component of AC impedance drawn in the straight line portion.

[0006] In addition, a control system for a secondary battery that predicts other parameters (internal resistance, SOC (state of charge), SOH (state of health), etc.) with high accuracy while performing abnormality detection has been proposed (for example, refer to patent document 4). The control system uses a Kalman filter to detect the difference between the observed value (voltage) at a certain moment and the voltage estimated using the prior state variable, pre-sets a threshold voltage and detects sudden abnormalities based on the detected differential voltage, specifically detecting micro short circuits, etc. Patent document 4 also records that it is preferred to use a neural network to learn the differential voltage data of the timing to determine and detect abnormalities or normality.

[0007] Prior art literature

[0008] Patent Literature

[0009] Patent Document 1: Japanese Patent Application Publication No. 2020-20604

[0010] Patent Document 2: Japanese Patent Application Publication No. 2020-21592

[0011] Patent Document 3: Japanese Patent Application Publication No. 2020-24153

[0012] Patent Document 4: WO2019 / 138286 Summary of the invention

[0013] Problems to be solved by the invention

[0014] In the models or systems described in these patent documents, even if a method of remaining capacity prediction is performed by measuring the internal resistance value or voltage value of a battery and using machine learning, etc., the remaining capacity prediction accuracy is poor and a long time is required for true value measurement.

[0015] In view of these circumstances, an object of the present invention is to provide a method for predicting the remaining capacity of a battery, which can predict a true value in a short time by using the discharge measurement of the battery.

[0016] Technical solutions to solve problems

[0017] The first battery capacity remaining amount prediction method of the present invention is a battery capacity remaining amount prediction method for predicting the remaining amount of a battery that can be repeatedly used by charging, wherein:

[0018] defining a target variable to predict the remaining capacity of the battery;

[0019] extracting a plurality of explanatory variables having high importance and linearity as elements for the margin prediction;

[0020] deriving a relational expression representing the relationship between the target variable and the explanatory variable as a regression formula;

[0021] Using the regression formula to predict at least one of the discharge voltage value and the discharge current of the battery over time;

[0022] The remaining capacity of the battery is predicted.

[0023] In the second remaining capacity prediction method of the present invention, among the explanatory variables, the reduction slope ([dV / dt] or [dA / dt]) of the discharge voltage or the discharge current over the entire measurement time, the reduction ratio per unit time of at least one of the discharge voltage value and the discharge current within a specified time range excluding the initial value, and the reduction amount per specified time interval, i.e., the difference, are used.

[0024] In addition, the initial voltage is used for the constant (intercept) in the regression formula.

[0025] The third battery capacity remaining prediction method of the present invention regularizes each of the explanatory variables to calculate the partial regression coefficient.

[0026] The partial regression coefficient is used for each coefficient of each target variable in the regression formula.

[0027] In the fourth remaining battery capacity prediction method of the present invention, in addition to the above-mentioned multiple explanatory variables, the regression formula also includes any one of the initial voltage value or the initial current value as a constant term. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flowchart showing a method for predicting the remaining amount of battery capacity using the remaining amount prediction method of battery capacity according to one embodiment of the present invention.

[0029] Figure 2A This is a flowchart showing a method of deriving a regression formula according to one embodiment of the present invention.

[0030] Figure 2B This is an explanatory diagram showing the relationship between explanatory variables and target variables constituting a regression formula according to one embodiment of the present invention.

[0031] Figure 3A The diagram shows the transient states of the good batteries and the bad batteries to be determined by the remaining battery prediction method according to one embodiment of the present invention.

[0032] Figure 3B This is a demand degree map used as a material for selecting explanatory variables used in the surplus quantity prediction method according to one embodiment of the present invention.

[0033] Figure 3C (A) is a graph showing the standard partial regression coefficients of the extracted explanatory variables for each element used in the remaining quantity prediction method as one embodiment of the present invention, Figure 3C (B) is a diagram showing analytical behavior of each element, which is a determination material for extracting each element used in the remaining quantity prediction method according to one embodiment of the present invention.

[0034] Figure 3D This is a table showing data such as partial regression coefficients used as materials for selecting explanatory variables used in the remaining quantity prediction method according to one embodiment of the present invention.

[0035] Figure 3E These are graphs plotting data used when calculating explanatory variables and their coefficients in a regression formula used in the remaining quantity prediction method according to one embodiment of the present invention.

[0036] Figure 3F This is an explanatory diagram showing the overall slope of the explanatory variables in the regression formula used in the remaining quantity prediction method according to one embodiment of the present invention.

[0037] Figure 3G This is an explanatory diagram showing differences in explanatory variables in a regression formula used in the remaining quantity prediction method according to one embodiment of the present invention.

[0038] Figure 3H It is an explanatory diagram showing the reduction ratio in the explanatory variables in the regression formula used in the remaining quantity prediction method according to one embodiment of the present invention.

[0039] Fig. 3I It is an explanatory diagram showing the change of the defective rate before and after using the remaining quantity prediction method according to one embodiment of the present invention.

[0040] Figure 4This is a diagram showing that the defect rate can be reduced to 10% or less by using the margin prediction method according to one embodiment of the present invention. DETAILED DESCRIPTION

[0041] Hereinafter, an embodiment of the present invention and an example thereof (specific example) will be described in detail with reference to the drawings.

[0042] <1. One Implementation Method>

[0043] Figure 1 A method for predicting the remaining battery capacity is represented, and the method includes the steps required to execute the remaining battery capacity prediction method of one embodiment of the present invention. In addition, in the present embodiment, among the batteries used as the object of remaining capacity prediction, HEV (Hybrid-Electric Vehicle) batteries (hereinafter referred to as "batteries") are used as the objects, but the batteries used as the objects in the remaining capacity prediction method of the present invention are of course not limited to this. That is, various batteries such as nickel-metal hydride (NiH) batteries, nickel-cadmium batteries, nickel-zinc batteries, lithium-ion secondary batteries, rechargeable alkaline batteries, etc., which can be used repeatedly by charging, can be used.

[0044] [Configuration steps of battery capacity remaining prediction method]

[0045] like Figure 1 As shown, the remaining capacity prediction method of the battery capacity of this embodiment is composed of the first step S1 to the fifth step S5, and the remaining capacity prediction is performed using the regression formula described later.

[0046] In the following embodiments, in the surplus prediction method of the present invention, "explanatory variable" and "objective variable (objective variable or response variable)" are used. In addition, these technical terms can be regarded as synonyms of "explanatory factor" and "target factor", respectively, but in mathematical concepts, "factor" is generally each component displayed in the form of product for one "term". On the other hand, in the present invention, if it is considered that the function form in the regression formula described later is in the form of a polynomial, each "explanatory variable" corresponds to and is equivalent to each "term", then the term "factor" is avoided and the former technical term "variable" is used.

[0047] The following uses Figure 1 The general flow of regression analysis according to one embodiment of the present invention will be described. Here, first, a method of determining a "regression formula" will be described.

[0048] In addition, in one embodiment of the present invention, in order to implement the method for predicting the remaining battery level, a "regression analysis" method is used. A general method for obtaining a "regression formula" used in the regression analysis is described in detail with reference to FIG. 2 as appropriate. Therefore, in the regression analysis, in particular, as an explanatory variable, Figure 2B The multiple factors shown are therefore configured to perform a "multiple regression analysis".

[0049] That is, in this multiple regression analysis, the dependent variable which is the target variable is predicted and explained using the independent variables which are the plurality of explanatory variables.

[0050] like Figure 2A As shown, the multivariate regression analysis is roughly composed of the following first step T1 to fourth step T4, thereby determining an appropriate regression formula.

[0051] The first step T1 determines the target variable (OV) and the explanatory variable (EV) associated with it. In this regression analysis, the degree of influence of the explanatory variable on the target variable is numerically expressed, and their relationship is expressed by a formula such as a straight line. In addition, Figure 2B It represents the relationship between these two intrinsic variables, namely the explanatory variables (EV1, EV2, ...) and the target variable (OV).

[0052] In addition, in this embodiment, in order to determine the explanatory variables as factors based on previously obtained data, etc., Elastic Net regression is used, which will be described later.

[0053] The second step T2 is to estimate the regression equation. The regression equation is the formula obtained as a result of regression analysis. Specifically, it is the functional form shown in the formula (1) described later. The regression equation is the value of the "coefficient" and "intercept" of the estimated formula, and the relationship between the target variable and the explanatory variable is summarized in one formula. The details will be described later. The coefficient here is sometimes called the "regression coefficient".

[0054] In addition, the coefficients in the multiple regression formula described later are sometimes called "partial regression coefficients." Regarding the partial regression coefficients, since they are coefficients that depend on the size and unit of the explanatory variable, it is meaningless to compare the coefficients of each explanatory variable. In order to compare these coefficients, they are standardized by using a method such as standard deviation regression to obtain a "standard partial regression coefficient." Based on the standard partial regression coefficients, even if the size or unit of the explanatory variable is different, the size of the relative coefficients can be compared, so the influence on the target variable can be compared (see Figure 3D ).

[0055] In addition, through this regression analysis, the coefficients of the explanatory variables serving as factors (i.e., regression coefficients) and the constant term (i.e., intercept) can be determined based on pre-obtained data, etc. However, for these coefficients and intercepts, for example, the least squares method (OLS: Ordinary Least Squares) can be used to calculate the "partial regression coefficient" and "regression constant (or simply referred to as constant") that minimize the sum of squares of the errors between the actual values ​​and the predicted values ​​of the dependent variable (target variable), respectively.

[0056] The third step T3 evaluates the (estimated) regression formula obtained in the second step T2. The benchmark indicators for this evaluation may be the accuracy of the regression formula (corrected value R2), coefficient (P value), statistical significance (i.e., significance of the regression formula: significance value F), or influence (t value).

[0057] In multiple regression analysis, each coefficient for each explanatory variable in the regression formula is sometimes referred to as a “partial regression coefficient.” Here, the respective definitions are omitted.

[0058] The “correction value R2” takes a value in the range of 0 to 1, and the closer it is to 1, the more accurate the regression formula is.

[0059] When the P value is less than 0.05, it can be judged that the relationship between the explanatory variable and the target variable is high.

[0060] If the “F value” is less than 0.05, it can be determined that the regression equation is significant.

[0061] The “t value” indicates that the larger the absolute value, the stronger the influence. As a guideline, when the absolute value of t is less than 2, it is determined that the explanatory variable has no influence on the target variable.

[0062] In addition, in the method of evaluating the regression formula, in addition to the method of calculating the coefficient of determination in the multivariate regression analysis, there is also a method of "residual analysis" that focuses on the deviation of the "residual". It is known that if the regression formula is appropriate with respect to the measured value data, the residual will change appropriately (dispersion uniformity). Therefore, in order to check this point, for example, a graph called "residual graph" in which the residuals of each data are arranged in order can be made and manually checked (in this graph, for example, the data number can be set on the horizontal axis and the residual value can be set on the vertical axis). In addition, in this residual graph, it is confirmed how the residual on the horizontal axis side changes relative to the "0" on the vertical axis. In addition, when the residuals are evenly dispersed relative to the vertical axis "0", there is no particularly big problem.

[0063] The fourth step T4 determines the evaluation with the highest accuracy or the like among the evaluations of the regression formula obtained in the third step T3, and uses it as the regression formula in the remaining capacity prediction method of the battery capacity of the present embodiment.

[0064] [Remainder prediction method]

[0065] Next, a method for predicting the remaining battery capacity is described. The method includes steps required for executing the remaining battery capacity prediction method according to one embodiment of the present invention. Figure 1 As shown, the remaining capacity prediction method is composed of a first step S1 to a fifth step S5.

[0066] In the first step S1, multiple "explanatory variables" and "constants" (these elements are described in detail in the second step S2) of the formula (1) described later are added together to form a single formula, which is defined as an "objective variable (OV)". In addition, the so-called "objective variable" here refers to a variable that receives the result produced by the cause of the object in the "explanatory variable".

[0067] In the second step S2, a plurality of "explanatory variables (EV)" are extracted. In addition, here, "explanatory variable EV" is also called "independent variable", which refers to a variable that becomes a cause. In the present invention, for the explanatory variable, Figure 3C As shown in (B), the explanatory variables with high cross-validation results and stable analytical behavior of Elastic Net regression are extracted. That is, in this embodiment, the definition and the like are described in detail in the examples described later, but as Figure 3C As shown in (A),

[0068] (30sec) Reduce the ratio

[0069] (120sec) Difference

[0070] Overall slope S (slope OCV)

[0071] These three factors were extracted as explanatory variables.

[0072] In addition, it is necessary and important to select highly linear elements based on the need to define these elements with the target variables described later, in other words, based on the need to be able to apply the principle of superposition. In addition, here, "Elastic Net" refers to a method of synthesizing "ridge regression" and "LASSO". In this Elastic Net, as with "LASSO (leastangle regression)", a reduced-dimensional model can be generated by generating coefficients with zero values. According to empirical studies, in the Elastic Net method, since the data is a predictor with high correlation, the performance is considered to be higher than that of LASSO. In addition, "cross-validation results" refers to a method of evaluating the generalization performance of each model.

[0073] In this embodiment, the "Elastic Net" method is used to solve the "regularization" problem with high accuracy compared to other methods. That is, the partial regression coefficient can be calculated with high accuracy. When describing the mathematical definition of Elastic Net, the following formula Z (problem) described using the norm α strictly defined between 0 and 1 and the non-negative normalization parameter λ can be a method to provide a solution strategy.

[0074] Z≡min[(1 / 2N)∑ N i =1(y i -β0-x i T β) 2 +λP α (β)]

[0075] Where, N: number of measurements

[0076] y i : Response data for measurement value i

[0077] x i : data (vector of P values ​​for measurement i, x i T is a tensor)

[0078] λ: A positive hyperparameter corresponding to a value of Lambda (e.g., a coefficient representing the influence of a regularization term, etc.), also called a normalization parameter.

[0079] β0: scalar

[0080] β: vector

[0081] α: A hyperparameter that adjusts the relative contribution of the L1 norm (Manhattan distance) and the L2 norm (Euclidean distance), where the domain is 0≤α≤1

[0082] Pα (β): Penalty term (normalization term)

[0083] Here, regarding the penalty term (normalization term) Pα(β), its specific detailed mathematical definition is omitted, but it can be obtained using a norm related to β, etc. Therefore, when α=1, Elastic Net is the same as LASSO. On the other hand, as α decreases toward 0, Elastic Net approaches Ridge Regression. In addition, this "Ridge Regression" is a method of estimating the coefficients of a multivariate regression model when the independent variables are strongly correlated. In addition, this ridge regression is used to solve the problem that the least squares estimate becomes incorrect when there is multicollinearity (there are strongly correlated independent variables) in the linear regression model, and the accuracy of the ridge regression estimate is higher than the least squares estimate.

[0084] Furthermore, since the graph drawn jumps when the coefficient becomes too large ("penalized"), a so-called normalization term is introduced as a parameter for adjusting the effect of the penalty.

[0085] (About explanatory variables)

[0086] In order to combine multiple explanatory variables to form the target variable, it is necessary to extract a linear function from each explanatory variable of the present embodiment. Therefore, among the expected main elements selected as the causes of the composition of the explanatory variables described later, the following three elements are extracted as important and linear elements, especially elements that have a significant impact on the target variable.

[0087] Right now:

[0088] 1) "Overall slope S (= dv / dt)" indicates "the time slope of the initial voltage or open circuit voltage (OCV (Open Circuit Voltage: resting potential) as a whole". Hereinafter, this "overall slope S" is sometimes referred to as "slope OCV". That is, it is defined by the voltage change rate over the entire time from the initial moment (t=0) when the (initial voltage or) open circuit voltage (OCV) is provided to a specific measurement time range, that is, the voltage change within the entire time interval to the specific measurement time range, that is, the voltage slope relative to the entire time. In addition, it is defined by "the first explanatory variable: x1".

[0089] 2) “Differential”

[0090] It is defined by "Second explanatory variable: x2".

[0091] 3) “Reduction ratio (=dv' / dt')”

[0092] It is defined as "third explanatory variable: x3".

[0093] Although the details will be described later, the target variable y obtained using these explanatory variables x, that is, the regression formula, can be expressed as the following formula (1).

[0094] y=[(k1)·(x1)]+[(k2)·(x2)]+[(k3)·(x3)]+

[0095] c···(1)

[0096] Among them, x1, x2, x3: explanatory variables

[0097] k1, k2, k3: coefficients

[0098] c: constant (intercept)

[0099] In addition, generally speaking, it is also necessary to remove the explanatory variables with (multi)collinearity between the extracted explanatory variables in advance. Here, the so-called (multi)collinearity refers to the existence of a linear correlation between two explanatory variables. In other words, in the multiple regression model, there is a combination of high correlation coefficients between the explanatory variables.

[0100] In addition, regarding the method of extracting the explanatory variable as the element, for example, the elements of the explanatory variable are determined by multiple regression analysis, and the surface evaluation of the regression formula (regression model) obtained is performed. In addition, as a development of multiple regression analysis, regularization (Lasso regression) can also improve the accuracy of the regression formula, and is therefore an important method. That is to say, this "regularized regression" is a method that can perform characteristic representation as partial regression coefficients by multiplying factors with strong multicollinearity by coefficients as a whole, as described in the previous paragraph. Therefore, the coefficients of the regression formula are reduced by pulling together the overall coefficient multipliers (there is also a reason why regularized regression is also called reduction estimation). Therefore, detailed description is omitted here, but based on the insights gained from the inventors of the present application through many years of knowledge and experience, in the embodiment, a technique is implemented to perform multiple regression analysis again after deriving the explanatory variables once, and return to the regression formula without reduction.

[0101] For example, as a method for obtaining the regression formula, in this embodiment, if using Figure 2A As described above, a method consisting of the first step T1 to the fourth step T4 is used.

[0102] In addition, the purpose of performing this multiple regression analysis is to use the dependent variable ("x" in the regression formula described later) to predict and explain the target variable ("y" in the regression formula described later), and the predictive ability and explanatory ability of the explanatory variable can be regarded as the influence of the explanatory variable on the target variable. Therefore, using the above-mentioned least squares method (OLS) and the like, the above-mentioned partial regression coefficient and regression constant can be obtained, which minimizes the sum of squares of the error between the measured value and the predicted value of the target variable.

[0103] (i) “Overall slope S”: the first explanatory variable

[0104] That is, Figure 3F As shown, the overall slope S (slope OCV) (=dv / dt) is a mathematically linear function defined by the overall slope (over the entire time axis range) in a graph related to the measurement time [sec] relative to the discharge voltage [V].

[0105] In addition, the overall Figure 3F As shown in FIG. 1 , the discharge voltage V drops from the discharge voltage value V0 at the initial time t0 to the discharge voltage V at a specific value V S (hereinafter referred to as the "minimum effective discharge voltage value V S ”) at a specific time t S The change of discharge voltage [V] in a specific time range T up to 100 ms.

[0106] In each battery used in the present embodiment, if the specific time Δt is set, the temporal transition (change) of the decrease phenomenon of the discharge voltage is uniquely determined within the specific time. That is, the decrease curve of the discharge voltage within the specific time (hereinafter sometimes referred to as the "decline function") can be uniquely drawn (determined). In this way, if the decrease curve of the discharge voltage can be uniquely drawn (determined), the above-mentioned overall slope S (= dv / dt) is determined by a linear linear function regardless of the specific time range T set to what extent, and regardless of the functional form (decline curve) of the decrease function. In other words, based on the discharge voltage value V0 at the initial moment, that is, the discharge start moment (t=0), the specific time range T, and the discharge voltage value V after the end of the specific time range T. S , uniquely determines the overall slope S (= dv / dt). That is,

[0107] S=dv / dt

[0108] =(V0-V S ) / T··· (2)

[0109] (ii) “Difference D”: Second explanatory variable

[0110] The difference D is found by the inventors of the present invention. In the graph measuring the time course of the discharge voltage, the element representing the amount of reduction per specified time (for example, the number of seconds required each time) is a significant contributing factor in predicting the remaining battery capacity. It is an important element in particular in nickel-metal hydride (NiH) batteries, etc., and therefore it is added as an explanatory variable.

[0111] In addition, in this embodiment, for example, data of a voltage value (Vmes) at a measurement time (referred to as a "measurement point") and a voltage value (Vmes-30) at a measurement time 30 seconds before the measurement point are used.

[0112] D=(Vmes-30)-(Vmes)···(3)

[0113] to define.

[0114] (iii) “Reduction ratio”: the third explanatory variable

[0115] This "reduction ratio (=dv' / dt')" is, for example, Figure 3H As shown, for the entire time region T' (= T-t0) effective for prediction sampling after removing only the inherent time (hereinafter referred to as "initial time t0") from the discharge start time, that is, the initial time (t=0), the time transition (change) of the discharge voltage drop phenomenon is uniquely determined for each battery.

[0116] That is, even for the reduction ratio (=dv' / dt') of the third element, the decrease curve (decrease function) of the discharge voltage within the time zone T' is uniquely drawn (determined), as in the case of the above-mentioned overall slope S. In this way, if the decrease curve (decrease function) of the discharge voltage can be uniquely drawn (determined), then, as in the case of the above-mentioned overall slope S (=dv / dt) as the first element, the reduction ratio (=dv' / dt') of the second element, regardless of the extent of the time zone T', and regardless of the function form (decrease curve) of the decrease function, the overall slope within the time zone T' is determined by a linear first-order function. In other words, the reduction ratio of the third element is uniquely determined based on the charge and discharge voltage V within a specific time range T' of any area where the inherent initial time t0 is subtracted from the initial moment and the specific time zone T' (=T-t0). That is,

[0117] Reduction ratio = dv′ / dt′

[0118] =(V0-V0'-Vs) / (T-T') ··· (4)

[0119] Here, the voltage drop in the initial time t0 during the 10 seconds after the start of discharge is more significant than that in the subsequent time and is not preferably used for the prediction of the remaining capacity. Therefore, it is considered necessary to remove it from the reduction ratio of the second factor for accurate remaining capacity prediction.

[0120] The above are the three elements used as explanatory variables, and in addition, a constant c is used. In addition, the explanatory variables constituting these three elements use the above-mentioned overall slope, difference, and reduction ratio in this embodiment, but depending on the type of battery used as the object of the remaining power prediction, it is of course not necessarily limited to the above-mentioned explanatory variables, and as long as a better remaining power prediction can be performed, of course, the explanatory variable can also be used.

[0121] In the third step S3, a regression formula is derived. In this regression formula, as described above, the variables with the greatest influence on the target variable are used as explanatory variables, and linear elements are used in each of the explanatory variables, and these are added together to define the target variable.

[0122] That is, if the linearity of each explanatory variable (x1, x2, x3, ..., xn) is used and the target variable that adds (combines) them is defined by y, the following relationship (regression formula) holds.

[0123] y≡[(k1)·(x1)]+[(k2)·(x2)]+[(k3)·(x3)]···+[(kn)·(xn)]+c

[0124] Among them, x1, x2, x3, ···, xn are explanatory variables,

[0125] k1, k2, k3, ..., kn: coefficients

[0126] c: constant (intercept)

[0127] In addition, in this embodiment, since there are three explanatory variables, the relationship (regression formula) in this embodiment is as described above,

[0128] y=[(k1)·(x1)]+[(k2)·(x2)]+[(k3)·(x3)]+c···(1)

[0129] Among them, x1, x2, x3: explanatory variables,

[0130] k1, k2, k3: coefficients

[0131] c: constant (intercept)

[0132] In addition, in the regression formula (1), as described above, it is also required to derive a high-precision regression formula, and the estimated regression formula is evaluated in advance to select a high-precision regression formula. In addition, the method for deriving the regression formula has been described with reference to FIG. 2, so its description is omitted here, but its specific example will be described in the embodiment described later.

[0133] In the fourth step S4, the remaining battery capacity is predicted over time using a regression formula. For example, in each nickel-metal hydride (NiH) battery with a rated voltage of 8V, the current capacity based on discharge is predicted for each nickel-metal hydride (NiH) battery using the regression formula (1) obtained in step S3.

[0134] That is, in this embodiment, by performing DC discharge for about 3 minutes, for example, a constant current is applied to each nickel-metal hydride (NiH) battery for about 3 minutes, thereby monitoring the voltage value of the battery capacity. Then, by using the voltage value obtained at this time, for each nickel-metal hydride (NiH) battery, the transient phenomenon of the remaining capacity for about 3 minutes is detected, so that various data can be collected, and by using the above-mentioned regression formula (1), the time lapse prediction of the remaining capacity thereafter can be performed.

[0135] In step 5 S5, the time transition of the battery capacity of each NiH battery is predicted using the regression formula in step 4 S4, and the usable time when the effective voltage of each NiH battery drops to within the range (sometimes referred to as the "termination voltage") is predicted, and the quality of the NiH battery is judged.

[0136] [Effect]

[0137] Therefore, in the past, when judging the quality of batteries, AC-IR sorting was performed to measure the internal resistance, but in this method, for example, in order to reduce the failure rate in the quality sorting process to a certain level, a long measurement time is required. In other words, if the measurement is performed in a short time, the prediction accuracy of the quality is low and a large number of defective products are mixed.

[0138] On the other hand, according to this embodiment of the present invention, by selecting regression formula (1) and using the regression formula (1) to perform current discharge for about 2 minutes (120 seconds) to 3 minutes (180 seconds), it is possible to obtain insights into the preferred waveform (function) for predicting the remaining battery amount. Therefore, by using the regression formula (1), only performing current discharge for about 2 or 3 minutes can achieve good battery quality judgment. As a result, this is related to the rapid sorting and effective use of reusable batteries such as secondary batteries, and can further promote carbon neutrality.

[0139] <One Implementation Method (Specific Example)>

[0140] Next, a method of predicting the remaining battery capacity by executing the remaining battery capacity prediction method described in the above embodiment of the present invention will be described in detail with reference to an example (specific example) that has been actually implemented.

[0141] According to this embodiment, which will be described in detail later, by using the remaining capacity prediction method of one embodiment described above, a regression formula suitable for predicting the time transition of the battery capacity of a nickel-metal hydride (NiH) battery can be obtained. This will be described in detail below.

[0142] In this embodiment, when estimating explanatory variables, Figure 3A Symbols A to D are used to indicate that the above-mentioned qualified products are judged to meet the above requirements in a transient state where the voltage (V) value of the nickel-metal hydride (NiH) battery decreases from the rated voltage of 8V to the current capacity value required for actual use (for example, 6.3Ah or more) by DC discharge. On the other hand, symbols E to H are used to indicate that the products are judged to be unqualified.

[0143] (Derivation of regression formula)

[0144] Next, the explanatory variable is obtained using a waveform diagram of symbols A to D representing qualified products. In this case, as described above, the method for deriving the explanatory variable may be a method that extracts a high cross-validation result of R / F (random forest). In addition, methods such as K-split cross-validation may be used. Furthermore, when deriving the explanatory variable, it is necessary and important to select a high linearity based on the need and the definition of the target variable. In addition, in the present invention, in particular, when estimating and deriving the explanatory variable, various methods may be used, not limited to the above-mentioned methods or techniques. For example, as a correct estimation method, it is specifically described below, but Elastic Net, etc. may be used.

[0145] That is, in this embodiment, a plurality of explanatory variables are estimated and derived by the Elastic Net method, and the regression formula (1)′ of the above-mentioned one embodiment is obtained using these derived explanatory variables and constants.

[0146] For example, as described in one embodiment described above, that is, as described in the third step S3, when the linearity of each explanatory variable (x1, x2, x3, ..., xn) is utilized and the target variable obtained by adding (combining) them is defined by y, the following relationship holds.

[0147] y≡[(k1)·(x1)]+[(k2)·(x2)]+[(k3)·(x3)]···+[(kn)·(xn)]+c

[0148] Among them, k1, k2, k3, ..., kn: coefficients

[0149] c: constant (intercept)

[0150] In addition, here, in this embodiment, the number of explanatory variables is 3, and therefore the relational expression in this embodiment is as follows.

[0151] y=[(k1)·(x1)]+[(k2)·(x2)]+[(k3)·(x3)]+c···(1)

[0152] Among them, x1, x2, x3: explanatory variables

[0153] k1, k2, k3: coefficients

[0154] c: constant (intercept)

[0155] Furthermore, assuming that a high cross-validation result of R / F (random forest) is extracted, for example, in this embodiment, from Figure 3B As shown in the demand graph, it can be seen that the demand is contributed in the order of difference 180, 120s▲%, overall slope OCV, difference 120, 180s▲%, △30, difference 60, 30s▲%, △120, etc. However, in this prediction method, among the factors with high contribution to the demand, the overall slope OCV and difference 120 are preferably used as explanatory variables, but other factors with high contribution to the demand, i.e., difference 180, 120s▲%, 180s▲%, △30, difference 60, △120, etc. are not preferred as explanatory variables, and the accuracy deteriorates. The reason will be referred to Figure 3C (B) and Figure 3D In addition, the definitions of the symbols that may cause ambiguity in the above description are explained below.

[0156] 1) 120s▲%: The reduction ratio of the discharge voltage [V] within 120 seconds from any time (but excluding 10 seconds from the start),

[0157] 2) Slope OCV: Overall slope (= dv / dt),

[0158] 3) 180s▲%: The reduction ratio of the discharge voltage [V] within 180 seconds from any time (but excluding 10 seconds from the start),

[0159] 4) 30s▲%: The reduction ratio of the discharge voltage [V] for 30 seconds from any time (but excluding 10 seconds from the start),

[0160] 5) △30: The overall ratio of the measurement from the initial moment to 30 seconds

[0161] 6) △120: The overall ratio of the measurement from the initial moment to 120 seconds

[0162] On the other hand, in this embodiment, based on the knowledge accumulated by the inventor of this application and the insights gained from many years of experience, for example, through the method of Elastic Net, the following Figure 3C The analysis shown in (B) and Figure 3D , we know that as an explanatory variable, Figure 3C As shown in (A), using the slope OVC for x1, using the difference 120 (120 sec difference) for x2, and using 30s▲% (30 sec reduction ratio) for x3 brings good accuracy to the remaining battery capacity prediction.

[0163] That is, to explain this specifically, for example, in order to introduce a square error function (cost function) and minimize the square error function, Figure 3C In (B), for example, the factors with normalization parameter λ around 0.01000 are analyzed. In order to regularize the explanatory factors, extract the partial regression coefficients, and then select the factors with high influence, for example, to avoid multicollinearity and extract the relative influence, according to the relationship formula, that is, according to

[0164] Standardized partial regression coefficient (β)

[0165] = Partial regression coefficient (B) × standard deviation of independent variable ÷ standard deviation of dependent variable

[0166] Calculate the standardized partial regression coefficients.

[0167] Therefore, by selecting coefficients with large contributions from the standard partial regression coefficients, the following elements can be extracted.

[0168] No.11: 30s▲%(30sec reduction ratio)

[0169] No.20: Difference 120 (120sec difference)

[0170] No.25: Slope OCV

[0171] In addition, if Figure 3D As shown in , for tolerance (tolerance, likelihood), the data obtained are 0.902, 0.776, and 0.823 respectively. In addition, as Figure 3E As shown, for these explanatory variables, their linearity is confirmed respectively.

[0172] Then, for example, calculate Figure 3EThe coefficients (partial regression coefficients) k1, k2, k3 of the explanatory variables are shown in the table. To this end, as described above, the least squares method (OLS) can be used to find the partial regression coefficient that minimizes the sum of squares of the error between the measured value and the predicted value of the target variable, and then the regression constant can be found.

[0173] (Specific method for determining partial regression coefficients and constants)

[0174] Next, a method for determining the coefficients k1, k2, and k3 of each explanatory variable will be described.

[0175] For example, in this embodiment, referring to Figure 3E The partial regression coefficients (k1 to k3) and the regression constant (C) are explained. In addition, the coefficient values ​​here are always just an example of the situation used in the remaining capacity prediction of nickel-metal hydride (NiH) batteries. Depending on the type of battery used in the remaining capacity prediction or the measuring instrument, these explanatory variables and coefficients may sometimes vary greatly (be different).

[0176] (i) The first partial regression coefficient k1 (with Figure 3F The coefficients related to the slope "Slope OVC (overall slope S)" shown are:

[0177] The coefficient k1 is the partial regression coefficient of the first explanatory variable x1. In this embodiment, Figure 3E In (C), the numerical values ​​on the graph are read approximately as approximate numbers and calculated roughly.

[0178] k1=0.30-(-1.0) / 0.04-(-0.11)

[0179] =8.66

[0180] ≈8.7

[0181] In the present embodiment, the actual correct coefficient k1 is calculated based on strict data, and a value of k1=8.731 is obtained.

[0182] (ii) The second partial regression coefficient k2( Figure 3G (This figure shows the coefficient of the difference D at 120 seconds from 30 seconds ago):

[0183] The coefficient k2 is the partial regression coefficient of the second explanatory variable x2. In this embodiment, Figure 3E In (B), the numerical values ​​on the graph are read approximately as approximate numbers and calculated roughly.

[0184] k2=-[0.80-(-0.47)] / [0.01-(-0.02)]

[0185] =-1.27 / 0.03

[0186] ≈-42.3

[0187] In the present embodiment, the actual correct coefficient k2 is calculated based on strict data, and a value of k2=-42.449 is obtained.

[0188] (iii) The third partial regression coefficient k3( Figure 3H The reduction ratio shown is 30s▲% coefficient):

[0189] The coefficient k3 is the partial regression coefficient of the third explanatory variable x3. In this embodiment, Figure 3E In (A), the numerical values ​​on the graph are read approximately as approximate numbers and calculated roughly.

[0190] k3=1.7-(-1.5) / [0.002-(-0.002)]

[0191] =3.2 / 0.004

[0192] ≈800

[0193] In the present embodiment, the actual correct coefficient k3 is calculated based on strict data, and a value of k3=802.946 is obtained.

[0194] (iv) Regarding the constant (intercept) c:

[0195] Furthermore, regarding the constant (intercept) c, for example, by reading Figure 3E When the data recorded in the figure is calculated approximately,

[0196] y≈c+(8.7×25)+(-42.5×20)+(803×1)

[0197] =120

[0198] However, in order to set the initial time to 3 minutes and discard the inherent time (e.g., 120 seconds) from the start time that is not suitable for margin prediction, y=120 is set, which becomes

[0199] c≈-50.5

[0200] In the case of this embodiment, the actual correct constant (intercept) was calculated based on strict data and a value of c=-57.617 was obtained.

[0201] Therefore, as shown in the following formula, the regression formula used in this embodiment can be uniquely determined.

[0202] y=[(8.7)·(x1)]+[(-42.5)·(x2)]+[(802.9)·(x3)]-57.6···(1)′

[0203] Where, y: target variable

[0204] x1~x3:Explanatory variables

[0205] In addition, the regression formula (1)' is only applicable to the present embodiment. The remaining battery capacity prediction method of the present invention is certainly not limited to this mathematical formula and coefficients.

[0206] (effect)

[0207] Therefore, when the correlation between the predicted current capacity and the measured current capacity [Ah] was investigated using the regression formula (1)′ of this embodiment, it was found that Fig. 3I As shown in the experimental data, the failure rate of the predicted value of the residual capacity in the sample N (=130) batteries in the simple measurement of the previous method AC-IR was 34%, but when the regression formula (1)' was used for prediction, the failure rate was reduced to about 7%.

[0208] Moreover, according to the present embodiment, it can be seen that the measurement time of the residual capacity when the defect rate is reduced to 7% by using the regression formula (1)' here is about 2 minutes, which can achieve a significant time reduction compared with the past. In addition, in the present embodiment, the remaining capacity is predicted based on the discharge current of the nickel-hydrogen (NiH) battery, but in the present invention, the remaining capacity of any battery as the object can be predicted based on the time lapse prediction of at least one of the discharge voltage (V) and the discharge current (I).

[0209] (Effect)

[0210] Thus, in this embodiment, if Figure 4 As shown, in the past, in order to achieve a failure rate of 34%, the discharge voltage measurement required about 20 minutes, but in the discharge voltage measurement of about 2 minutes, it is possible to achieve a failure rate of 7% which is significantly lower than the target of 10%. That is, in the existing prediction method, for example, the failure rate is 34%, but according to the prediction method of the present invention, it can be reduced to 7%, and the pass rate can be reduced by 27%. Moreover, since the time for the battery pass / fail judgment operation for the reuse of the used battery can be greatly saved, that is, the time is actually shortened from 20 minutes to about 2 minutes, the time waste of the pass / fail judgment operation can be saved, and the judgment operation efficiency is greatly improved.

[0211] Above, an embodiment and an example of the present invention are described. Finally, the description of the above-mentioned embodiment and example is an example of the present invention, and the present invention is not limited to the above-mentioned embodiment and example. Therefore, even outside the above-mentioned embodiment and example, various changes can be made according to the design, etc. as long as they do not deviate from the technical idea of ​​the present invention. In addition, the effects recorded in this specification are only illustrative and not limiting. In addition, there may be other effects.

[0212] In addition, the drawings or data in the above-mentioned embodiments and examples are illustrative and conceptual. The various components, etc., may be inconsistent with the present embodiments or examples depending on the different batteries used, and may not be consistent with the batteries actually used. In addition, the numerical values ​​of the data, etc. in the drawings also include situations that are greatly different depending on the batteries used. In addition, as long as the same effects as those described in the above-mentioned embodiments and examples can be obtained, various effects can be expected.

[0213] Description of Figure Numbers

[0214] c···Constant (intercept)

[0215] EV: explanatory variable

[0216] k1···Overall slope S (slope OCV) (=dv / dt) (coefficient of the first explanatory variable)

[0217] k2···Difference D (coefficient of the second explanatory variable)

[0218] k3···Reduction ratio (=dv' / dt') (Coefficient of the third explanatory variable)OV···Target variable (objective variable or response variable)OLS···Least squares method

[0219] OCV···Initial voltage or open circuit voltage

[0220] P value···Coefficient

[0221] R2···Accuracy (correction value),

[0222] T'... specific time zone (=T-t0)

[0223] V···Charge and discharge voltage

[0224] F value...Significance value: Statistical significance (i.e., the significance of the regression formula)

[0225] t value···Influence

[0226] t0···initial time

[0227] ts···specific time

[0228] t=0···Initial time (discharge start time)

[0229] T···Specific time range

[0230] T'···Time Zone

[0231] V0···Discharge voltage value at the initial moment

[0232] V···Discharge voltage

[0233] Vs···specific value (minimum effective discharge voltage value).

Claims

1. A method for predicting the remaining capacity of a battery, for predicting the remaining capacity of a battery that can be repeatedly used by charging, characterized in that: defining a target variable to predict the remaining capacity of the battery; extracting a plurality of explanatory variables having high importance and linearity as elements for the margin prediction; deriving a relational expression representing the relationship between the target variable and the explanatory variable as a regression formula; Using the regression formula to predict at least one of the discharge voltage value and the discharge current of the battery over time; as well as The remaining capacity of the battery is predicted.

2. The method for predicting remaining battery capacity according to claim 1, characterized in that: Among the explanatory variables, the reduction slope ([dV / dt] or [dA / dt]) of the discharge voltage or the discharge current over the entire measurement time, the reduction ratio per unit time of at least one of the discharge voltage value and the discharge current within a specified time range excluding the initial value, and the difference in the amount of reduction per specified time interval are used. In addition, the initial voltage is used as the constant, ie, the intercept, in the regression formula.

3. The method for predicting remaining battery capacity according to claim 1 or 2, characterized in that: Regularize each of the explanatory variables to calculate the partial regression coefficients, The partial regression coefficient is used for each coefficient of each target variable in the regression formula.

4. The method for predicting remaining battery capacity according to any one of claims 1 to 3, characterized in that: In addition to the above-mentioned multiple explanatory variables, the regression formula also includes any one of the initial voltage value or the initial current value as a constant term.

Citation Information

Patent Citations

  • Battery data processing system, estimation method for secondary battery capacity, battery pack and method for manufacturing the battery pack

    JP2020020604A

  • Battery information processing system, battery information processing method, battery pack and manufacturing method thereof

    JP2020024153A

  • Abormality detection device for secondary battery, abnormality detection method, and program

    WO2019138286A1