Model parameter estimation device and model parameter estimation method

JPWO2024079867A5Active Publication Date: 2024-11-08MITSUBISHI ELECTRIC CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
JP2024551013
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2022-10-14
Filing Date
2022-10-14
Publication Date
2024-11-08
Estimated Expiration
2042-10-14
Patent Text Reader

Abstract

This model parameter estimation device comprises: a state amount calculation unit (332) for calculating the state amount which indicates the state of a storage battery (1) with respect to a measurement value of current, on the basis of a state equation obtained by inputting a value to a non-linear parameter (φ) of a model which represents the storage battery (1) by using the non-linear parameter (φ) and a linear parameter (ψ), and of time-series data of current and terminal voltage of the storage battery (1); a linear parameter estimation unit (333) for estimating a linear parameter (ψ) which minimizes an error between a measurement value of the terminal voltage and an estimation value of the terminal voltage calculated on the basis of the model, the state amount, and the measurement value of the current; and a non-linear parameter update unit (331) for repeating updates of the value of the non-linear parameter (φ) until the value satisfies the convergence condition such that the minimized error becomes small.
Need to check novelty before this filing date? Find Prior Art

Description

Model parameter estimation device and model parameter estimation method

[0001] The present application relates to a model parameter estimation device and a model parameter estimation method.

[0002] Stationary energy storage systems that utilize renewable energy are becoming more commonplace to reduce environmental impact. Electric vehicles such as EVs (Electric Vehicles), HEVs (Hybrid Electric Vehicles), and PHVs (Plug-in Hybrid Vehicles) are also becoming commercially available, and the development of electric aircraft is also progressing.

[0003] However, these devices use storage batteries such as lithium-ion batteries, but it is known that the performance of storage batteries deteriorates with use. Therefore, there is a demand for a model parameter estimation technique for modeling storage batteries with high accuracy, in order to estimate the state of charge (SOC), diagnose deterioration, and predict the lifespan of storage batteries with high accuracy.

[0004] Therefore, a model parameter estimation technique using a continuous-time system identification method based on the output error method has been proposed (see, for example, Patent Document 1). Specifically, by applying a nonlinear optimization method to time-series data of current and voltage during charging and discharging of a storage battery, model parameters are estimated so that the output voltage of a storage battery model constructed as a continuous-time system matches the measured voltage. With this method, by appropriately selecting initial values ​​for estimation, it is possible to estimate model parameters with high accuracy while taking into account the nonlinearity of OCV (Open Circuit Voltage) characteristics.

[0005] JP 2014-86313 A (paragraphs 0017-0018, FIG. 1, paragraphs 0044-0045, FIG. 5)

[0006] However, the method proposed in Patent Document 1 requires setting initial values ​​for all parameters prior to estimation, but depending on how the initial values ​​are set, the estimated values ​​may diverge or converge to a local solution that differs from the true value. Therefore, it is necessary to set initial values ​​based on prior information about the target system, making it difficult to apply this method to systems with little prior information.

[0007] The present application discloses a technique for solving the above-mentioned problems, and aims to provide a model parameter estimation device and a model parameter estimation method that can estimate parameters with little information.

[0008] The model parameter estimation device disclosed in the present application is characterized by comprising: a state quantity calculation unit that calculates a state quantity indicating the state of the storage battery with respect to a measured current based on a state equation obtained by substituting values ​​for nonlinear parameters in a state space model that expresses a storage battery using nonlinear parameters and linear parameters, and time-series data of measured current and terminal voltage of the storage battery; a linear parameter estimation unit that estimates the linear parameters that minimize an error between the estimated value of the terminal voltage calculated based on the state space model, the state quantity, and the measured current, and the measured terminal voltage; and a nonlinear parameter update unit that repeatedly updates the values ​​of the nonlinear parameters until a predetermined convergence condition is satisfied so that the minimized error becomes small.

[0009] The model parameter estimation method disclosed in the present application is characterized by including: a step of setting initial values ​​of nonlinear parameters of a state space model that represents a storage battery using nonlinear parameters and linear parameters; a state quantity calculation step of calculating state quantities that indicate the state of the storage battery with respect to the measured current based on a state equation obtained by substituting values ​​for the nonlinear parameters in the state space model and time series data of measured current and terminal voltage of the storage battery; a linear parameter estimation step of estimating the linear parameters that minimize an error between the estimated value of the terminal voltage calculated based on the state space model, the state quantities, and the measured current and the measured terminal voltage; and a nonlinear parameter update step of repeatedly updating the values ​​of the nonlinear parameters until a predetermined convergence condition is satisfied so that the minimized error becomes small.

[0010] According to the model parameter estimation device or model parameter estimation method disclosed in the present application, model parameters can be estimated by setting the initial values ​​of only the nonlinear parameters, making it possible to estimate parameters with little information.

[0011] FIG. 1 is a block diagram for explaining the configuration of a model parameter estimation device according to a first embodiment. FIG. 2 is a block diagram illustrating an example of a hardware configuration of a portion that executes calculation processing of the model parameter estimation device according to the first embodiment. FIG. 3 is a diagram illustrating an equivalent circuit of a storage battery that is a target for model parameter estimation. FIG. 4 is a diagram illustrating a Foster-type equivalent circuit as another model of a storage battery that is a target for model parameter estimation. FIG. 5 is a flowchart illustrating the operation of the model parameter estimation device or a model parameter estimation method according to the first embodiment. FIG. 6 is a graph illustrating time-dependent changes in measured voltage values, measured current values, and estimated current values ​​by a model when the model parameter estimation device or the model parameter estimation method according to the first embodiment is applied to a storage battery system. FIG. 7 is a graph illustrating the relationship between the quantity of electricity and the open circuit voltage as an OCV function estimated when the model parameter estimation device or the model parameter estimation method according to the first embodiment is applied to a storage battery system.

[0012] 1 to 7 are diagrams illustrating the configuration and operation of a model parameter estimation device and a model parameter estimation method according to a first embodiment. Fig. 1 is a block diagram of a storage battery system including a storage battery that is an estimation target and the model parameter estimation device, illustrating the configuration of the model parameter estimation device. Fig. 2 is a block diagram illustrating an example of the hardware configuration of a portion that executes the calculation processing of the model parameter estimation device.

[0013] Fig. 3 is a diagram showing an equivalent circuit that expresses a storage battery, the object of which model parameters are to be estimated, using a DC component and a relaxation component of an overvoltage, Fig. 4 is a diagram showing a Foster-type equivalent circuit, as another model of a storage battery, that expresses diffusion impedance by connecting multiple CR parallel elements in series, and Fig. 5 is a flowchart showing the model parameter estimation operation in the model parameter estimation device, i.e., the model parameter estimation method.

[0014] 6 is a diagram in which three graphs are arranged vertically, each showing a time-dependent change in the measured voltage of a battery, a time-dependent change in the measured current of the battery, and a time-dependent change in the estimated current value by the model when the model parameter estimation device or the model parameter estimation method is applied to a certain battery system, with the horizontal axis representing the common (synchronized) time and the vertical axis representing the measured voltage, measured current, and estimated current, respectively. Also, FIG. 7 is a graph showing the estimated OCV function, with the horizontal axis representing the quantity of electricity normalized by a certain nominal capacity and the vertical axis representing the open-circuit voltage.

[0015] Hereinafter, a model parameter estimation device and a model parameter estimation method according to embodiments of the present application will be described in detail with appropriate reference to the drawings. Note that the same reference numerals in the various drawings indicate the same or corresponding parts.

[0016] 1 , a model parameter estimation device 3 according to the first embodiment is provided in a storage battery system 100 having a storage battery 1 in order to diagnose the storage battery 1. The model parameter estimation device 3 is connected to a detection unit 2 having a current detection device 21 that detects a current, which is an output characteristic of the storage battery 1, and a voltage detection device 22 that detects a voltage, and is configured to estimate model parameters of the storage battery 1.

[0017] The storage battery 1 is typically assumed to be a lithium-ion battery, but it may also be other types of storage batteries such as lead-acid batteries, nickel-metal hydride batteries, nickel-cadmium batteries, and all-solid-state batteries. The storage battery 1 to be diagnosed may be a single-cell storage battery or a storage battery module in which multiple cells are connected in any combination of series and parallel connections. It may also be a pack in which modules are further connected in any combination of series and parallel connections, and the unit that constitutes a single storage battery 1 is arbitrary.

[0018] The model parameter estimation device 3 includes a time-series data acquisition unit 31 that acquires time-series data of measurement values ​​from the current value I and voltage value V detected by the detection unit 2, and a model acquisition unit 32 that acquires a storage battery model and sets an initial value of a nonlinear parameter φ, which will be described later. The model parameter estimation device 3 further includes a parameter estimation unit 33 that estimates model parameters from the time-series data acquired by the time-series data acquisition unit 31, the storage battery model set by the model acquisition unit 32, and the initial value of the nonlinear parameter φ. Note that, although an example in which the model acquisition unit 32 sets the initial values ​​has been shown here, an initial value setting unit may be provided separately or within the parameter estimation unit 33 to set (or read) the initial values.

[0019] The parameter estimation unit 33 includes a nonlinear parameter update unit 331 that updates the nonlinear parameter φ, and a state quantity calculation unit 332 that calculates a state quantity of the storage battery model based on a state equation of the storage battery model when the nonlinear parameter φ is fixed. The parameter estimation unit 33 further includes a linear parameter estimation unit 333 that estimates a linear parameter ψ based on the storage battery model and the state quantity, and a determination unit 334 that determines whether the parameter estimation has converged based on a convergence determination condition.

[0020] As shown in FIG. 2, the model parameter estimation device 3 may be configured as a single piece of hardware 30 including a processor 30a and a storage device 30b. Although not shown, the storage device 30b includes a volatile storage device such as a random access memory and a non-volatile auxiliary storage device such as a flash memory. Alternatively, a hard disk auxiliary storage device may be included instead of the flash memory. The processor 30a executes a program input from the storage device 30b. In this case, the program is input from the auxiliary storage device to the processor 30a via the volatile storage device. The processor 30a may output data such as calculation results to the volatile storage device of the storage device 30b, or may store the data in the auxiliary storage device via the volatile storage device.

[0021] That is, the functions of each unit constituting the model parameter estimation device 3, for example, the functions of the model acquisition unit 32, the state quantity calculation unit 332, the linear parameter estimation unit 333, and the nonlinear parameter update unit 331, are realized by software, firmware, or a combination thereof. The software and firmware are written as programs and stored in the storage device 30b. The processor 30a reads out the programs stored in the storage device 30b and executes them to realize the functions of each unit of the model parameter estimation device 3.

[0022] <General Description of the Technology of the Present Application> The technical content of the model parameter estimation device 3 and the model parameter estimation method of the present application will first be described using a general problem setting that is not limited to storage batteries. First, it is assumed that the state space model of the target system is expressed as shown in Equation (1).

[0023] where x is a vector representing a state quantity, u is an input, y is an output, and f and g are nonlinear vector-valued functions. x and φ y are the nonlinear parameter vectors in the state equation and output equation, respectively, and ψ is the linear parameter vector.

[0024] What we want to solve here is the time {t k |k=0,...,N}, the time series input / output data {u(t k ) |k=0,...,N},{y(t k ) |k=0,...,N} is given, the parameter θ is found to minimize the evaluation function J(θ) based on the sum of squares of the error ε between the output y (estimated value) of the model shown in equation (2) and the measured value y.

[0025] where θ is a vector that lists the linear parameter ψ and the nonlinear parameter φ, and ψ:=[ψ x Τ , ψ y Τ ] Τ Whereas, θ:=[φ Τ , ψ Τ ] Τ It is as defined as follows.

[0026] Since this is a nonlinear optimization problem, it is possible to obtain a local optimum using a known nonlinear optimization method. For example, the gradient of the evaluation function with respect to the parameters, or the estimated output y(t k The Gauss-Newton method or the Levenberg-Marquardt method, which are based on the gradient of the parameters (θ, θ), can be used. In addition, a gradient-free method, i.e., a method not based on gradient information, such as the Nelder-Mead method, can also be used, and there is no limitation on the specific method of nonlinear optimization.

[0027] To obtain the gradient related to the parameters, a method based on the calculation of a sensitivity equation such as that described in Patent Document 1 can be used. Alternatively, a gradient approximation method using one-sided or two-sided differences may be used, or an adjoint method using an adjoint variable may be used.

[0028] Although equation (1) describes a continuous-time state space model, the target system may be considered in discrete time. In this case, the state space model is written as in equation (3).

[0029] However, x k is a vector representing the state quantity at time k, u k and y k are the input and output at time k, respectively, and f d and g d are each a nonlinear vector-valued function. To convert a continuous-time system to a discrete-time system, known methods can be used, such as zero-order hold or first-order hold.

[0030] On the other hand, when dealing with a continuous-time system as is, it is possible to calculate the state quantity at any time from the input time-series data by applying a known numerical solution method for differential equations to the original state equation.Typical numerical solutions for differential equations include the Euler method or the Runge-Kutta method.

[0031] However, a common issue with output error methods that deal with nonlinear optimization problems is that parameter estimates are dependent on initial values. Therefore, as mentioned in the Background Art section, unless the initial values ​​of the estimated parameters are set appropriately, a globally optimal solution cannot be obtained. Therefore, we focus on the structure of the state space model in equation (1) and consider reducing the dependency on initial values. First, by arranging the model outputs vertically in chronological order, we obtain equation (4).

[0032] In other words, the linear parameter ψ and the nonlinear parameter φ can be expressed separately. In this case, the output vector Y:=[y(t 0 ), ..., y(t N )] Τ and the error vector E:=[ε(t 0 ), ..., ε(t N )] Τ The relational expression (5) holds true for Y = G(φ)ψ + E (5)

[0033] Therefore, when the nonlinear parameter φ in equation (5) is fixed, the linear least squares solution of the linear parameter ψ is obtained by the linear least squares method as shown in equation (6). ψ(φ) = (G Τ (φ)G(φ)) -1 G Τ (φ)Y (6) When the linear least squares solution of equation (6) is used, the problem to be solved is transformed from problem P1 of equation (7) to problem P2 of equation (8).

[0034]

[0035] In other words, problem P1, which originally required estimation of both the linear parameter ψ and the nonlinear parameter φ, has been transformed into a problem (problem P2) that requires estimation of only the nonlinear parameter φ. If a solution for the nonlinear parameter φ in problem P2 can be obtained (fixed), then the solution for the linear parameter ψ can be immediately obtained by substituting this into equation (6).

[0036] Since problem P2 is also a nonlinear optimization problem, it is possible to use various known nonlinear optimization methods as described above, as in the case of solving problem P1. When problem P1 is converted into problem P2, the only estimated parameter is the nonlinear parameter φ, which has the following advantages.

[0037] First, it is possible to reduce the dependency on initial values. When solving the original problem P1, it was necessary to set initial values ​​for all of the linear parameters ψ and nonlinear parameters φ, whereas in problem P2, the linear parameters ψ can be found as a linear least-squares solution without any initial values. Therefore, it is possible to solve the problem by setting only the initial value of the nonlinear parameter φ. Therefore, the problem becomes simpler in terms of estimating the true values ​​of the parameters.

[0038] Second, the computational complexity is reduced when applying many nonlinear optimization methods. For example, when using a method that uses gradient information related to the model output parameters, the size of the gradient vector or Jacobian matrix is ​​reduced because the gradient information of the linear parameter ψ is no longer necessary. From another perspective, the reduction in the number of estimated parameters is expected to reduce the number of iterations required until the parameters converge.

[0039] Third, numerical stability is improved. Generally, when the order of magnitude of the values ​​of the estimated parameters differs significantly, problems such as cancellation of significant digits tend to occur. Therefore, it is expected that such problems will be less likely to occur by reducing the number of estimated parameters.

[0040] Next, specific methods for obtaining a solution will be explained for the cases where the linear parameter ψ includes a linear equality constraint and where it includes a linear equality constraint and a linear inequality constraint.

[0041] <1: Case where Linear Parameters Include Linear Equality Constraints> When the linear parameters ψ include linear equality constraints, the least squares problem with equality constraints for the linear parameters ψ is formulated as problem P3 in equation (9).

[0042]

[0043] Here, C eq and d eq are a matrix and a vector that respectively describe the linear equality constraints for the linear parameter ψ. In order to solve the problem P3 using the same idea as before, when the nonlinear parameter φ is fixed at a certain value, G = G(φ), and the sub-problem SP3 of the problem P3 is considered, it becomes as shown in equation (10).

[0044] Since the sub-problem SP3 is a linear least squares problem with equality constraints, it can be solved by the Lagrange multiplier method. Specifically, the sub-problem SP3 can be solved by finding ψ that minimizes the evaluation function L shown in equation (11) using the Lagrange multiplier vector λ.

[0045] Therefore, when the point at which the gradients of the evaluation function L due to ψ and λ become 0 is considered, the equation (12) is obtained.

[0046] This can be summarized as equation (13), and the solution can be found immediately.

[0047] Therefore, to solve problem P3, it is sufficient to alternately repeat updating of nonlinear parameters by a nonlinear optimization method and calculation of linear parameters by the Lagrange multiplier method.

[0048] <2: Case where linear parameters include linear equality constraints and linear inequality constraints> When linear parameters include linear equality constraints and linear inequality constraints, the least squares problem with inequality constraints of the linear parameters is formulated as problem P4 of equation (14).

[0049] Here, C ineq and d ineq are a matrix and a vector describing the linear inequality constraints on the linear parameter ψ, respectively. Considering the subproblem SP4 of problem P4 in the same way as problem P3, we obtain equation (15).

[0050]

[0051] As a method for solving sub-problem SP4, an interior point method, a sequential quadratic programming method, or the like can be used. In this case, iterative calculations are required, but since sub-problem SP4 is a convex optimization problem, a globally optimal solution can be obtained. Of course, since sub-problem SP3 is included in sub-problem SP4, such a solution method using iterative calculations can also be applied to sub-problem SP3. Therefore, to solve problem P4, it is sufficient to repeatedly update the nonlinear parameters using a nonlinear optimization method and calculate the linear parameters using a solution method for a convex optimization problem.

[0052] From here, specific methods for applying the nonlinear optimization method to problems P1, P3, and P4 will be explained, focusing on problem P4. Note that problems P1 and P3 are included in problem P4, so it is sufficient to explain the method focusing on problem P4. In fact, problem P1 is a special case of problem P4 in which there are no inequality constraints and no equality constraints. Problem P3 is also a special case of problem P4 in which there are no inequality constraints.

[0053] The optimal solution of the linear parameter ψ for a certain nonlinear parameter φ in the subproblem SP4 is expressed as ψ = ψ * (φ), the evaluation function to be minimized in problem P4 is given by equation (16).

[0054]

[0055] ψ * Since (φ) can be obtained by solving sub-problem SP4 using the method described above, problem P4 can be solved by finding the nonlinear parameter φ that minimizes L(φ). Below, we will explain three nonlinear optimization methods in order: the steepest descent method, which is a type of gradient method; the Gauss-Newton method; and the Nelder-Mead method, which is a gradient-free method.

[0056] <Steepest Descent Method> When the steepest descent method is used, the update equation for the nonlinear parameter φ is Equation (17).

[0057] Here, α is a predetermined positive value, and may be determined by a line search or the like each time an update is performed.

[0058] If it is difficult to directly calculate the gradient, a one-sided difference or a two-sided difference can be used as the gradient approximation. For example, in the case of a one-sided difference, φ j A sufficiently small Δ j By perturbing the equation by only , the partial differential can be approximately calculated as in equation (18).

[0059] However, e j is a unit vector whose length is the same as the nonlinear parameter φ and whose jth element is 1. In other words, φ+Δ j e jis a perturbation value obtained by perturbing only the j-th element of the nonlinear parameter φ. In this case, the update equation for the nonlinear parameter φ is given by equation (19).

[0060] In calculating the evaluation function L, it is necessary to calculate the state quantity for each value of the nonlinear parameter φ, estimate the linear parameter ψ, and calculate the output estimated value.

[0061] <Gauss-Newton Method> The update formula for the Gauss-Newton method is formula (20). However, Λ(ψ) is expressed by formula (21), and ∂ hat y(t k , φ, ψ * (φ)) / ∂φ j is the sensitivity function, i.e., a nonlinear parameter φ of the output estimate. j is the partial differential with respect to

[0062]

[0063] When it is difficult to directly calculate the sensitivity function, an approximation of the sensitivity function is used instead of the sensitivity function. In other words, a gradient approximation is used instead of the gradient itself. In this case, too, the calculation formula when using a one-sided difference is Equation (22).

[0064]

[0065] <Nelder-Mead Method> For example, when the Nelder-Mead method is used, the number of iterations is l (el) and the number of nonlinear parameters is n (n) within one parameter update. φ For n φ +1 variable φ l,1 , φ l,2 , ..., φ l,nφ+1 Using the evaluation function value L(φ l,1 ), L(φ l,2 ), ..., L(φ l,nφ+1 ) is calculated. Furthermore, evaluation function values ​​are calculated for parameter values ​​that have undergone four types of operations: mirroring, enlarging, contracting, and reducing. Then, using the evaluation function values ​​calculated in this way, the parameters are adjusted to φ l From φ l+1 Update to.

[0066] Thus, depending on the nonlinear optimization method, it may be necessary to find performance functions or output estimates for a plurality of different nonlinear parameter estimates in order to update the nonlinear parameter estimates.

[0067] Although the explanation has been omitted because it is outside the scope of the present application, even when the original problem P1 includes at least one of an equality constraint and an inequality constraint on the nonlinear parameter φ, it is possible to apply a known nonlinear optimization method with constraint conditions to update the nonlinear parameter φ. For example, a local solution can be obtained by applying the above-mentioned interior point method, sequential quadratic programming, or the like.

[0068] The above is the content of the parameter estimation technology of the present application, which has been described for a general problem. Based on this content, a case where this technology is applied to the storage battery 1 will be described.

[0069] <Application to a storage battery> As explained in FIG. 1, the current detection device 21 detects the current of the storage battery 1 and outputs the detected current value I. The voltage detection device 22 detects the terminal voltage of the storage battery 1 and outputs the detected voltage value V. However, if the storage battery 1 is configured with a series-parallel connection of multiple cells, the detected current and voltage may be those of a single cell or multiple cells that are part of the storage battery 1. In the following, the sampling period of the time series data is defined as t s The following explanation will be given assuming that the sampling period is t s does not have to be fixed, but may be variable.

[0070] The time series data acquisition unit 31 acquires a time series {t k |k=0, 1, . . . , N}, the time series data of the detected current {I(t k ) |k=0, 1, . . . , N} and the time series data of the detected voltage {V(t k ) |k = 0, 1, ..., N}. For simplicity, the sampling period t s Considering the case where is fixed, the initial time t 0 For t k = t 0 +kt sHowever, the time-series data may be held internally by the time-series data acquisition unit 31, or may be acquired via an external PC, server, cloud, or the like.

[0071] The model acquisition unit 32 holds a storage battery model for the storage battery 1 to be estimated, and this model is used in the parameter estimation unit 33. The parameters of the storage battery model, including initial values, are updated as appropriate based on the parameters estimated by the parameter estimation unit 33. The storage battery model held by the model acquisition unit 32 may be originally held internally by the model parameter estimation device 3, or may be acquired from an external PC, server, cloud, or the like.

[0072] As a storage battery model, for example, an equivalent circuit model as shown in FIG. 0 represents the DC component of the overvoltage resulting from the electrolyte resistance, collector metal resistance, contact resistance, charge transfer resistance, etc. of the storage battery 1. Also, the resistive element R d and capacitor element C d represents the relaxation component of the overvoltage resulting from the diffusion component of the storage battery 1.

[0073] At this time, the state space model of the storage battery 1 is described, for example, as in equations (23) to (25).

[0074]

[0075] Here, q is the quantity of electricity, I is the detected current value (detected current), and I off is the current offset error included in the detected current, q d is the capacitor element C d The charge stored in OCV is a function representing the OCV characteristic depending on the quantity of electricity q of the storage battery 1 (OCV function f OCV ) where the quantity of electricity q may be a normalized value, in which case the OCV function f OCV is also a function of the normalized electric quantity q.

[0076] When a discrete-time state space model is used, for example, equations (23) to (25) are transformed by zero-order hold into equations (26) to (28) and used.

[0077]

[0078] As another storage battery model, a Foster-type equivalent circuit model in which multiple CR parallel elements are connected in series may be used, as shown in Fig. 4. The equivalent circuit model in Fig. 4 is a more detailed storage battery model than the equivalent circuit model in Fig. 3 because it has three CR parallel elements connected in series. Of course, the number of CR parallel elements connected in series may be any number.

[0079] In addition, when the diffusion impedance is approximated by a Foster type circuit, each CR parallel element has a diffusion resistance R d and diffusion capacitance C d (See, for example, Kuhn, Estelle, et al., "Modeling Ni-mH battery using Cauer and Foster structures," Journal of power sources 158.2 (2006): 1490-1497.)

[0080] Here, n d is the approximation order. For example, the approximation order n d When = 3, the spreading impedance is approximated by three stages of CR parallel elements as in the equivalent circuit model of FIG.

[0081] In this case, the equations (27) and (28) become the equations (30) and (31).

[0082] However, k i is as shown in equation (32). i =4 / ((2i-1) 2 π 2 ) ... (32)

[0083] Note that other models may be used as the storage battery model; for example, a Cauer-type circuit may be used. Alternatively, a so-called electrochemical model that attempts to directly describe the internal phenomena of the storage battery 1 without using an equivalent circuit may be used. Various known models of the storage battery 1 may be used, and are not limited as long as they can be expressed within the framework of Equation (1) or Equation (3). However, for simplicity, the following description will be made assuming the equivalent circuit model described in FIG. 3.

[0084] OCV function f representing the OCV characteristics OCV Typically, a linear or piecewise linear function is used for the electric quantity q. For example, when a polynomial is used, it is expressed as in equation (33), and a parameter linear expression is possible.

[0085] However, m p represents the degree of the polynomial, and c i Haq i is the coefficient of

[0086] Alternatively, a piecewise linear function (linear spline curve) shown in equation (34) is used.

[0087] However, m s (≧2) is the number of nodes, and in equation (34), the function q = p 1 , p 2 , ..., p ms m s By point m s - is divided into one section, m s When ≧3, the equality constraint satisfies equation (35). i p i+1 +b i = a i+1 p i+1 +b i+1 for i=1,…,m s -2... (35)

[0088] Alternatively, a cubic spline curve of equation (36) is used.

[0089] However, m s When ≧3, the equality constraint satisfies equation (37).

[0090]

[0091] Alternatively, spline curves of any other degree may be used.

[0092] A piecewise polynomial such as a spline curve can also be expressed as a parameter linear function. As an example, a piecewise linear function can be expressed as in equation (38). However, δ i,k is as shown in equation (39).

[0093]

[0094] Similarly, other cases such as cubic spline curves can also be expressed as parameter linear expressions. OCV It is also possible to consider a case where both a linear parameter ψ and a nonlinear parameter φ are included, in which case the function form becomes as shown in equation (40).

[0095] However, g OCV is a vector-valued function, and φ OCV and ψ OCV are the OCV functions f OCV is a vector of nonlinear parameters φ and linear parameters ψ contained in

[0096] Other OCV functions f OCV It is also possible to consider a case where only the nonlinear parameter φ is included as the OCV function f OCV The function form of is not necessarily limited to a form including the linear parameter ψ. However, when a large number of linear parameters ψ are included, the advantage of the approach of converting problem P1 into problem P2 in this application is particularly strong. However, in the following, we will use equation (34) as the OCV function f OCV The case where it is used as

[0097] The nonlinear parameter updater 331 updates the time series data of the detected current {I(t k ) |k=0, 1, . . . , N} and the time series data of the detected voltage {V(t k) |k=0, 1, ..., N} and the battery model, the nonlinear parameter φ of the battery model is updated, where the nonlinear parameter φ is expressed as in Equation (41).

[0098] That is, it refers to the parameters included in the state equation and the nonlinear parameters (parameters that cannot be linearly expressed) included in the output equation in the state space model of the storage battery 1. However, it is not necessary to include all of the nonlinear parameters φ in the estimation parameters, for example, when the values ​​of certain nonlinear parameters are known.

[0099] The elements of the nonlinear parameter φ may vary depending on how the state space model of the storage battery 1 is created. For example, the OCV function f OCV the nonlinear parameter φ OCV If φ does not contain φ, then the nonlinear parameter φ also necessarily contains φ. OCV In addition, the parameters included in the state equations of Equation (24) and Equation (30) are the time constant τ d and the current offset error I off However, depending on the modeling method, other parameters may also be included in the state equation.

[0100] For example, when a multi-stage CR parallel element is expressed by approximating the diffusion impedance as in equation (29), the time constant is τ d In other cases, there will be only one parameter, i.e., the number of time constant parameters will be equal to the number of CR parallel elements. Also, for example, if the state equation for CR overvoltage in equation (42) is used instead of the state equation for CR overvoltage in equation (24), and the output equation is changed to equation (43) instead of equation (25), the nonlinear parameter φ will be used instead of the linear parameter ψ. d Include.

[0101] V=R 0 (I-I off ) + V d +f OCV (q) (43)

[0102] However, in this case, τ d and C dConsidering that both of these are included in the nonlinear parameter φ, it is better to use equation (24) in order to make the most of the advantages of the technical idea of ​​the present application.

[0103] As a specific method for updating the nonlinear parameter φ, it is possible to use a known nonlinear optimization method based on iterative calculations, as described above.

[0104] A predetermined value is used as the initial value of the nonlinear parameter φ. If possible, the predetermined value is determined by prior knowledge of the true value of the nonlinear parameter φ. For example, the diffusion time constant τ d It is known that the time is usually several tens to several hundreds of seconds in a lithium ion battery, so the initial value is set to, for example, 100 seconds. off Since is a very small value, it is set to, for example, 0 A as an initial value.

[0105] These settings are highly appropriate for general lithium-ion batteries and current sensors, and are independent of the series-parallel connection configuration of the battery 1, which is advantageous for product installation in that initial value tuning is not required in many cases. Also, if the results of characteristic evaluation of the battery 1 when it was new or data on a specification sheet are available, these may be used. Also, if parameter estimation has been performed in the past, the past estimation results may be used.

[0106] The initial values ​​may be set in the model acquisition unit 32, the nonlinear parameter update unit 331, or a separately provided initial value setting unit (not shown). In this case, the initial values ​​may be set automatically based on the above-mentioned prior knowledge, or the initial values ​​may be set based on the input after prior information, initial value candidates, etc. are presented via an input I / F (not shown).

[0107] Note that the update of the nonlinear parameter φ may include the operations of a state quantity calculation unit 332 and a linear parameter estimation unit 333, which will be described later. For example, when gradient information regarding the nonlinear parameter φ of the evaluation function is used to update the nonlinear parameter φ, calculations similar to those of the state quantity calculation unit 332 and the linear parameter estimation unit 333 are required.

[0108] The state quantity calculation unit 332 calculates the time series data of the detected current {I(t k ) |k=0, 1, ..., N} and the storage battery model, the time series values ​​of the state quantities in the storage battery model (i.e., the state space model of the storage battery 1) are calculated. For example, the state quantities in the state space model of Equations (23) to (25) are the quantity of electricity q and the charge q d Refers to...

[0109] As described above, either a continuous-time or discrete-time state space model may be used to calculate the time series values ​​of state quantities based on a state space model. In the former case, a known numerical solution method such as the fourth-order Runge-Kutta method may be used, while in the latter case, calculations may be performed sequentially according to the model. Alternatively, calculations may be performed based on the solution of a state equation.

[0110] The linear parameter estimation unit 333 estimates the time series data of the detected current {I(t k ) |k=0, 1, . . . , N} and the time series data of the detected voltage {V(t k ) |k=0, 1, ..., N}, the battery model, and the state quantities, a linear parameter ψ is estimated, and a model output of voltage is calculated. Specifically, since the linear parameter ψ can be expressed separately in the output equation as in equation (44), it can be expressed as in equation (45).

[0111]

[0112] Therefore, the linear least squares solution of the linear parameters ψ for the estimated values ​​of the nonlinear parameters φ can be estimated as Equation (46). ψ(φ) = (G Τ (φ) G(φ) -1 G Τ (φ)ν (46) Therefore, problem P1, which is a voltage error minimization problem based on the output error method in the storage battery 1, can be converted to problem P2 using equation (46). In addition, by using the estimated value of this linear parameter ψ, the model output of the voltage can also be calculated.

[0113] If an equality constraint is included, problem P3 can be considered taking the equality constraint into account. For example, if the OCV function f OCVIf is a linear spline curve (piecewise linear function), it must satisfy equation (35), so ψ OCV,i = [a i b i ] Τ Then, for equation (47), the equality constraint equation becomes equation (48). i = [p i+1 1] Τ ...(47)

[0114]

[0115] In addition, the OCV function f OCV If is a cubic spline curve, it must satisfy equation (37), so ψ OCV,i = [a i b i c i d i ] Τ Then, for equation (49), the equality constraint equation becomes equation (50).

[0116]

[0117] By formulating the problem in this way, it becomes possible to treat the problem to be solved as problem P3 and sub-problem SP3. The solution method in this case is as already explained. Note that the constraint equations are not limited to those explained in the above example, and may be any single or multiple equations expressed as linear constraint equations.

[0118] It is also possible to consider the case where inequality constraints on the linear parameters ψ are included. For example, 0 and C d Regarding R 0,min ≦R 0 ≦R 0,max , C d,min ≦C d ≦C d,max If it is desired to reflect prior knowledge regarding the upper and lower limit constraints, the inequality constraints can be expressed as in equation (51).

[0119]

[0120] This makes it possible to treat the problem to be solved as problem P4 and sub-problem SP4. The solution method in this case is as already explained. Note that the constraint inequalities are not limited to those explained in the above example, and may be any single or multiple inequalities expressed as linear constraint inequalities.

[0121] Here, the OCV function f OCV We have explained the polynomial approximation method and spline approximation method of R. 0 , R d , C d , τ d The electrical quantity dependence of the circuit parameters, such as , may be modeled. The temperature dependence of the circuit parameters may also be modeled using a similar method. However, when modeling the temperature dependence, it is necessary to separately measure the temperature of the storage battery 1, and in that case, the model parameter estimation device 3 needs to acquire the output of a temperature detector. Note that the well-known Arrhenius equation, etc., can be used as the temperature model.

[0122] The determination unit 334 uses the time series data of the detected voltage {V(t k ) |k=0, 1, ..., N}, the voltage model output, the linear parameter estimates, and the nonlinear parameter estimates, and the convergence of the parameter estimates and voltage estimates is determined. If a predetermined criterion is met, the estimation is terminated and the model parameters are output externally. If the criterion is not met, the process returns to the nonlinear parameter update unit 331 and the update of the nonlinear parameter φ is repeated.

[0123] Specifically, the j-th element of the l-th estimated value of the parameter θ is expressed as θ j l When this is the case, it is determined whether or not the formula (52) is satisfied, as in Patent Document 1.

[0124] However, n θ is the number of elements of θ, ε θ is a predetermined sufficiently small value.

[0125] Alternatively, it is determined whether or not the equation (53) is satisfied for only the nonlinear parameter φ.

[0126] However, n φ is the number of elements of φ, ε φ is a predetermined sufficiently small value.

[0127] Alternatively, it is determined whether or not the formula (54) is satisfied based on the root-mean-square error (RMSE) regarding the voltage.

[0128] where ε V is a predetermined sufficiently small value.

[0129] Alternatively, the judgment may be based on the mean absolute error (MAE) related to voltage. Alternatively, the convergence condition may be judged based on whether the number of iterations of estimating the nonlinear parameter φ has reached a predetermined upper limit. Alternatively, a combined judgment using a plurality of the above methods may be performed. Alternatively, any other convergence judgment method may be used. Since various publicly known methods are known for judging convergence of estimated parameters in nonlinear optimization, the convergence judgment method is not limited.

[0130] The final parameter estimates of the parameter estimation unit 33 may be stored in the model acquisition unit 32 and used as the initial values ​​for the next model parameter estimation, or may be stored on an external PC, server, cloud, etc.

[0131] <Processing Procedure of Model Parameter Estimation Device> Next, the processing procedure of the model parameter estimation device 3 according to the first embodiment, that is, the model parameter estimation method, will be described with reference to the flowchart in Fig. 5. However, here, the most typical case, that is, the case where a nonlinear optimization method using a gradient approximation is used, will be described.

[0132] First, the time-series data acquisition unit 31 acquires the current value I of the storage battery 1 output by the current detection device 21 and the voltage value V of the storage battery 1 output by the voltage detection device 22 as time-series data of the measured values ​​(step S1310). Next, the model acquisition unit 32 acquires a storage battery model from an external source or stored internally, and sets or reads the initial value of the nonlinear parameter φ (step S1320). As the storage battery model, for example, a state space model shown in equations (23) to (25) or equations (26) to (28) corresponding to equation (3) is acquired.

[0133] The acquired time-series data, storage battery model, and initial value are output to the parameter estimation unit 33, and the following repeated calculations (steps S2331 to S2400) are performed in each section within the parameter estimation unit 33.

[0134] In step S2331, the nonlinear parameter update unit 331 updates the nonlinear parameter φ according to a nonlinear optimization method, which is a type of gradient method based on iterative calculation, so that the error minimized for the linear parameter ψ becomes smaller. When using a gradient approximation value based on one-sided difference, for example, in the gradient calculation, the gradient approximation value is calculated based on an output estimate value for a perturbation value for each element of the previous nonlinear parameter update value (previous value), which will be described later. However, when calculating a gradient approximation value based on one-sided difference or the like, an output estimate value for the previous value is also used. Note that, at the first time, a predetermined value is set as the initial value of the nonlinear parameter (for example, τ d = 100, I off =0, etc.).

[0135] In step S2332, the state quantity calculation unit 332 calculates the state quantities of the storage battery model based on the state equation of the storage battery model using the updated nonlinear parameter φ. In step S2333, the linear parameter estimation unit 333 estimates the linear parameter ψ and the output value of the storage battery model based on the state quantities output from the state quantity calculation unit 332.

[0136] Then, in step S2334, the determination unit 334 determines the convergence of the parameter estimation in the parameter estimation unit 33 based on a predetermined convergence determination condition. If it is determined that the convergence determination condition is satisfied ("Yes" in step S2400), the parameter estimation is terminated and the parameters of the estimation result are output to the outside. At this time, the parameters of the estimation result may be held in the model acquisition unit 32.

[0137] On the other hand, if it is determined that the convergence determination condition is not satisfied ("No" in step S2400), the process proceeds to step S3331, and the execution of each step from step S3331 onwards continues.

[0138] In step S3331, the nonlinear parameter update unit 331 sets a perturbation value for each nonlinear parameter element. In step S3332, the state quantity calculation unit 332 calculates the state quantities of the storage battery model for each perturbation value for each nonlinear parameter element based on the state equation of the storage battery model using the perturbation values ​​for each nonlinear parameter element set by the nonlinear parameter update unit 331.

[0139] In step S3333, the linear parameter estimation unit 333 estimates the linear parameters ψ and the output values ​​of the storage battery model for each perturbation value of each nonlinear parameter element, based on the state quantities for each perturbation value of each nonlinear parameter element output from the state quantity calculation unit 332. Then, the execution of each step from step S2331 onwards described above is continued.

[0140] The above is an example of the processing procedure of the model parameter estimation device 3 according to the first embodiment, that is, an example of the operation of the model parameter estimation method. Note that the content and order of the processing described in Fig. 5 are merely an example, and the content and order are not limited to these.

[0141] Next, a result of applying the model parameter estimation device 3 or the model parameter estimation method according to the first embodiment to time-series data of current and voltage during operation of a certain battery system 100 will be described with reference to Fig. 6. In this application, the current and voltage in the battery system 100 with a series-parallel connection configuration are converted into cells using the battery 1.

[0142] Here, the storage battery 1 is modeled as a Foster type circuit shown in FIG. 1 , R 2 , R 3 and C 1 , C 2 , C 3 is assumed to follow equation (29), which is a finite approximation of the diffusion impedance. OCV As for m s = 10 and a cubic spline curve with equal intervals was used. For simplicity, the nonlinear parameter φ is d and the linear parameter ψ is R 0 , C d and all parameters of the cubic spline curve.

[0143] In the time-series data of current I in the upper part of Figure 6, the first half shows free charging and discharging, and the second half shows constant-power charging. In both cases, it can be seen that the actual data in the middle (measured values: solid line) and the model output in the lower part (estimated values: dashed line) match with high accuracy.

[0144] In addition, the OCV function f OCV is shown in Figure 7. The horizontal axis indicates a standard capacitance q typ The quantity of electricity q / q normalized by typ The estimation results are obtained by smoothly connecting each point and calculating the non-linear electric quantity q / q typ This allows us to model the nonlinear fluctuations in OCV with respect to the voltage. As a result, we can accurately estimate the voltage in Figure 6 (middle: actual data, bottom: model output), including subtle irregularities during charging.

[0145] Embodiment 2 In the first embodiment, the processing when the OCV function is unknown has been described. In the second embodiment, the processing when the OCV function is known will be described. Note that the configuration of the model parameter estimation device and the basic operation of the model parameter estimation method are the same as those in the first embodiment, so the description of the similar parts will be omitted and the figures used in the first embodiment will be used.

[0146] OCV function f OCVis known, in the state space model of the storage battery 1 expressed by equations (23) to (25), equation (25) is max can be expressed as in equation (55) using

[0147] At this time, the OCV function f OCV Since is known, the nonlinear parameter φ is expressed as in equation (56): φ = [τ d I off q max ] Τ ...(56)

[0148] Furthermore, if the output equation is transformed, it becomes as in equation (57), and since it can be expressed in a parameter linear manner as in the first embodiment, the linear parameter ψ becomes as in equation (58).

[0149] In this case, unlike the first embodiment, the full charge capacity q max This has the advantage that it is possible to estimate the

[0150] In addition, the OCV function f OCV For more detailed information, the positive electrode potential function f p and the negative electrode potential function f n is known, and the OCV function f OCV can be expressed as equation (59) using electrode parameters.

[0151] However, among the electrode parameters, q p,0 and q n,0 are the initial charge amounts of the positive and negative electrodes, respectively, and q p,max and q n,max are the full charge capacities of the positive and negative electrodes, respectively. When expressed in this way, the nonlinear parameter φ is expressed as in equation (60). φ = [τ d I off q p,0 q p,max q n,o q n,max ] Τ (60) The linear parameter ψ is the same as in equation (58).

[0152] In this case, the OCV function f OCVBy modeling these in more detail, it is possible to diagnose the deterioration of the storage battery 1 in more detail. Specifically, it is possible to distinguish and grasp the decrease in positive electrode capacity, decrease in negative electrode capacity, and difference in the amount of electricity between the positive electrode and the negative electrode that accompany the deterioration of the storage battery 1. The difference in the amount of electricity between the positive electrode and the negative electrode is said to be a phenomenon that occurs mainly due to the growth of a coating (SEI: Solid Electrolyte Interface) formed on the surface of the negative electrode and the deposition of lithium, and causes a decrease in the full charge capacity of the storage battery 1.

[0153] As described above, in the model parameter estimation device 3 or model parameter estimation method of the present invention, the updating of the nonlinear parameter φ and the estimation of the linear parameter ψ are separated. Alternatively, the nonlinear parameter φ is updated by considering a problem in which a linear least-squares solution of linear parameters is embedded, such as problem P2. This provides three major advantages during optimization calculations, as described above.

[0154] First, the initial value dependency is reduced because only the nonlinear parameter φ needs to be set as the initial value. Second, the amount of calculation is reduced. Third, numerical stability is improved.

[0155] In estimating the linear parameter ψ, it is possible to obtain an optimal solution for unconstrained linear least squares problems, linear least squares problems with linear equality constraints, and linear inequality constraint problems. In particular, for unconstrained linear least squares problems, a linear least squares solution can be obtained without iterative calculations. Furthermore, for linear least squares problems with linear equality constraints, a global optimal solution can be obtained without iterative calculations using the Lagrange multiplier method.

[0156] These advantages are more effective when the storage battery model is more complex and the number of parameters to be estimated increases. Therefore, this method is also useful when more detailed modeling and deterioration diagnosis are performed by estimating some or all of the electrode parameters after separately storing the positive electrode potential characteristics and the negative electrode potential characteristics as the OCV characteristics stored in advance.

[0157] Here, when the continuous-time system identification method based on the output error method described in Patent Document 1 is reexamined, although OCV characteristic data is not required, a method for estimating model parameters in the case where OCV characteristic data is not available is not clearly described. In practical use, model parameter estimation techniques that do not assume OCV characteristic data are also important. For example, since it is known that OCV characteristics fluctuate due to degradation, etc., estimation of model parameters including OCV characteristics is important for improving the accuracy of estimation and application to degradation diagnosis.

[0158] Furthermore, when reusing storage batteries, modeling may be required under circumstances where there is little or no prior information on the true values ​​of model parameters, including the OCV characteristics. When the OCV characteristics are parameterized by a certain function and included in the model parameters for collective estimation, the method described in Patent Document 1 has the problems of initial value setting and convergence described above, making it more difficult to converge to an optimal value.

[0159] In contrast to this, in the present application, the OCV characteristics, which are the relationship between the quantity of electricity (which may be the SOC or a normalized quantity of electricity) and the OCV, are expressed by a function including parameters, and then the parameters are estimated, thereby making it possible to estimate the OCV characteristics even if they are unknown.

[0160] In particular, if the OCV function f is expressed as a function including a large number of linear parameters ψ, such as a polynomial or piecewise polynomial, the feature of the present invention, which separates the process of updating the nonlinear parameters φ from the process of estimating the linear parameters ψ, can be utilized. This makes it possible to easily estimate all parameters without having to carefully set the initial values ​​of the large number of linear parameters ψ. Furthermore, as described above, a global solution can be obtained in estimating the linear parameters ψ even if a linear equality constraint is included. Therefore, the OCV function f can be expressed as a piecewise polynomial including a linear equality constraint. OCV It is possible to easily estimate parameters even when dealing with the following. Note that by using a piecewise polynomial instead of a polynomial, it becomes easier to accurately represent the curve, including local fluctuations.

[0161] Although various exemplary embodiments and examples are described in this application, the various features, aspects, and functions described in one or more embodiments are not limited to the application of a particular embodiment, but may be applied to the embodiments alone or in various combinations. Therefore, countless modifications not illustrated are contemplated within the scope of the technology disclosed in this specification. For example, this includes cases where at least one component is modified, added, or omitted, or where at least one component is extracted and combined with components of another embodiment.

[0162] As described above, according to the model parameter estimation device 3 of the present application, the state of the storage battery 1 (for example, the quantity of electricity q, the charge q) with respect to the measured value of the current value I is estimated based on a state equation obtained by substituting a value (for example, an initial value) for the nonlinear parameter φ in a state space model that expresses the storage battery 1 using the nonlinear parameter φ and the linear parameter ψ, and time-series data of the measured values ​​of the current (current value I) and the terminal voltage (voltage value V) of the storage battery 1. d The control system is configured to include a state quantity calculation unit 332 that calculates a state quantity indicating the state of the nonlinear parameter φ, a linear parameter estimation unit 333 that estimates a linear parameter ψ that minimizes the error between the measured value of the voltage value V and an estimated value of the voltage value V, the estimated value being calculated based on the state space model, the state quantity, and the measured value of the current value I, a determination unit 334 that determines whether the estimated linear parameter ψ has converged, and a nonlinear parameter update unit 331 that repeatedly updates the value of the nonlinear parameter φ until a predetermined convergence condition is satisfied so that the minimized error becomes smaller until the determination unit 334 determines that convergence has occurred. Therefore, by limiting the initial value setting to only the nonlinear parameter φ, it is possible to estimate parameters with less information.

[0163] Furthermore, the dependence on the initial value can be reduced, the amount of calculation can be reduced, and numerical stability can be improved. In this case, since the initial value can be a generally known and highly valid value for the storage battery, estimation can be performed without additional information when updating the system.

[0164] In particular, the nonlinear parameter update unit 331 can easily converge by updating the value of the nonlinear parameter φ using the gradient or gradient approximation of the estimated value of the voltage value V with respect to the nonlinear parameter φ, or by using a gradient-free nonlinear optimization method.

[0165] Alternatively, when updating the value of the nonlinear parameter φ, the nonlinear parameter updating unit 331 generates a perturbation value for each nonlinear parameter φ by perturbing the previous value, which is the value before the update; the state quantity calculating unit 332 calculates a state quantity for each perturbation value; the linear parameter estimating unit 333 estimates a linear parameter ψ for the state quantity for each perturbation value; and the nonlinear parameter updating unit 331 calculates a gradient approximation for the nonlinear parameter φ of the estimated value of the terminal voltage (voltage value V) based on at least the estimated value of the linear parameter ψ for the state quantity for each perturbation value, and updates the previous value based on the calculated gradient approximation, thereby achieving more reliable convergence.

[0166] In this case, the nonlinear parameter φ is the charge rate of the storage battery 1 and the full charge capacity q max and the diffusion time constant τ d and the offset current (current offset error I off ) and the linear parameter ψ is the DC resistance of the storage battery 1 and the diffusion resistance R d If at least one of the capacitance and the capacitance of the capacitor is included, the state of the storage battery 1 can be reliably evaluated.

[0167] In this case, if the linear parameter estimation unit 333 estimates the linear parameters ψ using a linear least squares solution to the linear least squares problem, the amount of calculation can be reduced.

[0168] In this case, the state space model includes an OCV function f that represents the relationship between the quantity of electricity q and the open circuit voltage and includes a linear parameter ψ. OCV By including the above, the storage battery 1 can be accurately modeled.

[0169] Furthermore, the OCV function f OCVis a polynomial of the electrical quantity q, and if the linear parameter ψ includes a coefficient of the polynomial, it is possible to model the nonlinear variation of the OCV relative to the electrical quantity q.

[0170] If the state space model includes a linear equality constraint on the linear parameter ψ and the linear parameter estimation unit 333 is configured to estimate the linear parameter ψ by solving a linear least-squares problem including the linear equality constraint by the Lagrange multiplier method, convergence can be reliably achieved by alternately repeating the update of the nonlinear parameter φ and the calculation of the linear parameter ψ by the Lagrange multiplier method.

[0171] The state space model represents the relationship between the quantity of electricity q and the open circuit voltage, and includes an OCV function f of a piecewise polynomial including a linear parameter ψ. OCV where the linear parameter ψ includes a linear parameter of the piecewise polynomial, and the linear equality constraint includes a linear equality constraint at a node of the piecewise polynomial, the feature of the present application in which the process of updating the nonlinear parameter φ and the process of estimating the linear parameter ψ are separated can be fully utilized.

[0172] Alternatively, the state space model may include a linear inequality constraint on the linear parameter ψ, and the linear parameter estimation unit 333 may be configured to estimate the linear parameter ψ by solving a linear least-squares problem including the linear inequality constraint, thereby making it possible to obtain an optimal solution.

[0173] The state space model includes OCV characteristic information that represents the relationship between the charge rate and the open circuit voltage of the storage battery 1, and the nonlinear parameter φ is a function of the full charge capacity q max If the battery 1 is configured to include the above, the deterioration of the battery 1 can be diagnosed in detail.

[0174] In this case, the state space model includes positive electrode potential characteristic information representing the relationship between the positive electrode electric quantity and the positive electrode potential of the storage battery 1, and negative electrode potential characteristic information representing the relationship between the negative electrode electric quantity and the negative electrode potential of the storage battery 1, and the nonlinear parameter φ is a function of the full charge capacity q max Instead, if the configuration includes at least one of the positive electrode capacity, the negative electrode capacity, the positive electrode electrical quantity, and the negative electrode electrical quantity, the deterioration of the storage battery 1 can be diagnosed in more detail.

[0175] Furthermore, according to the model parameter estimation method of the present application, a step (S1320) of setting an initial value of the nonlinear parameter φ of a state space model that expresses the storage battery 1 using the nonlinear parameter φ and the linear parameter ψ, a state equation obtained by substituting a value for the nonlinear parameter φ for the state space model, and a state of the storage battery (for example, the quantity of electricity q, the charge q) for the measured value of the current value I based on the time-series data of the measured values ​​of the current (current value I) and the terminal voltage (voltage value V) of the storage battery 1 are estimated. d The method includes a state quantity calculation step (S2332) of calculating a state quantity indicating (V) (V), a linear parameter estimation step (S2333) of estimating a linear parameter ψ that minimizes the error between the measured voltage value V and an estimated value of the voltage value V, the estimated value being calculated based on the state space model, the state quantity, and the measured value of the current value I, a determination step (S2334) of determining whether the estimated linear parameter ψ has converged, and a nonlinear parameter update step (S2331) of repeatedly updating the value of the nonlinear parameter ψ until it is determined that the linear parameter ψ has converged and a predetermined convergence condition is satisfied so that the minimized error becomes smaller. Therefore, by limiting the initial value setting to only the nonlinear parameter ψ, it is possible to estimate parameters with less information.

[0176] Furthermore, the dependence on the initial value can be reduced, the amount of calculation can be reduced, and numerical stability can be improved. In this case, since the initial value can be a generally known and highly valid value for the storage battery, estimation can be performed without additional information when updating the system.

[0177] In the nonlinear parameter updating step (S2331), the value of the nonlinear parameter φ can be easily converged by using the gradient or gradient approximation of the estimated value of the voltage value V with respect to the nonlinear parameter φ, or by using a gradient-free nonlinear optimization method.

[0178] Alternatively, when updating the value of the nonlinear parameter φ, the method may further include a perturbation value setting step (S3331) for generating perturbation values ​​for each (element of) the nonlinear parameter φ by perturbing the previous value, which is the value before the update, a perturbation state quantity calculation step (S3332) for calculating a state quantity for each perturbation value, and a perturbation linear parameter estimation step (S3333) for estimating a linear parameter ψ for the state quantity for each perturbation value, and the nonlinear parameter updating step (S2331) may be configured to calculate a gradient approximation for the nonlinear parameter φ of the estimated value of the terminal voltage (voltage value V) based on at least the estimated linear parameters for the state quantity for each perturbation value, and update the previous value based on the calculated gradient approximation, thereby ensuring more reliable convergence.

[0179] In this case, the nonlinear parameter φ is the charge rate of the storage battery 1 and the full charge capacity q max and the diffusion time constant τ d and the offset current (current offset error I off ) and the linear parameter ψ is the DC resistance of the storage battery 1 and the diffusion resistance R d If at least one of the capacitance and the capacitance of the capacitor is included, the state of the storage battery 1 can be reliably evaluated.

[0180] In the linear parameter estimation step (S2333), if the linear parameter ψ is estimated using a linear least squares solution to the linear least squares problem, the amount of calculation can be reduced.

[0181] In this case, the state space model includes an OCV function f that represents the relationship between the quantity of electricity q and the open circuit voltage and includes a linear parameter ψ. OCV By including the above, the storage battery 1 can be accurately modeled.

[0182] Furthermore, the OCV function f OCV is a polynomial of the electrical quantity q, and if the linear parameter ψ includes a coefficient of the polynomial, it is possible to model the nonlinear variation of the OCV relative to the electrical quantity q.

[0183] The state space model includes a linear equality constraint on the linear parameter ψ, and in the linear parameter estimation step (S2333), if the linear parameter ψ is estimated by solving a linear least squares problem including the linear equality constraint using the Lagrange multiplier method, convergence can be ensured by alternately repeating the update of the nonlinear parameter φ and the calculation of the linear parameter ψ using the Lagrange multiplier method.

[0184] The state space model represents the relationship between the quantity of electricity q and the open circuit voltage, and includes an OCV function f of a piecewise polynomial including a linear parameter ψ. OCV where the linear parameter ψ includes a linear parameter of the piecewise polynomial, and the linear equality constraint includes a linear equality constraint at a node of the piecewise polynomial, the feature of the present application in which the process of updating the nonlinear parameter φ and the process of estimating the linear parameter ψ are separated can be fully utilized.

[0185] Alternatively, the state space model may include a linear inequality constraint on the linear parameter ψ, and the linear parameter estimation step (S2333) may be configured to estimate the linear parameter ψ by solving a linear least-squares problem including the linear inequality constraint, thereby making it possible to obtain an optimal solution.

[0186] The state space model includes OCV characteristic information that represents the relationship between the charge rate and the open circuit voltage of the storage battery 1, and the nonlinear parameter φ is a function of the full charge capacity q max If the battery 1 is configured to include the above, the deterioration of the battery 1 can be diagnosed in detail.

[0187] In this case, the state space model includes positive electrode potential characteristic information representing the relationship between the positive electrode electric quantity and the positive electrode potential of the storage battery 1, and negative electrode potential characteristic information representing the relationship between the negative electrode electric quantity and the negative electrode potential of the storage battery 1, and the nonlinear parameter φ is a function of the full charge capacity q max Instead, if the configuration includes at least one of the positive electrode capacity, the negative electrode capacity, the positive electrode electrical quantity, and the negative electrode electrical quantity, the deterioration of the storage battery 1 can be diagnosed in more detail.

[0188] 1: storage battery, 21: current detection device, 22: voltage detection device, 3: model parameter estimation device, 31: time series data acquisition unit, 32: model acquisition unit, 33: parameter estimation unit, 331: non-linear parameter update unit, 332: state quantity calculation unit, 333: linear parameter estimation unit, 334: determination unit, I: current value, I off : current offset error, q: quantity of electricity, q max : Full charge capacity, R d : diffusion resistance, V: voltage value, τ d : Diffusion time constant.

Claims

1. a state quantity calculation unit that calculates a state quantity indicating a state of the target system with respect to a measured value of the input, based on a state equation obtained by substituting values ​​for the nonlinear parameters in a state space model that expresses the target system using nonlinear parameters and linear parameters, and based on time-series data of the measured values ​​of each of the inputs and outputs of the target system; a linear parameter estimating unit that estimates the linear parameters that minimize an error between an estimated value of an output calculated based on the state space model, the state quantity, and the measured value of the input, and the measured value of the output; and a nonlinear parameter update unit that repeatedly updates the values ​​of the nonlinear parameters until a predetermined convergence condition is satisfied so that the minimized error becomes smaller; A model parameter estimation device comprising:

2. The model parameter estimation device described in Claim 1, characterized in that the state quantity calculation unit calculates the state quantity assuming that the target system is a storage battery, the input is the current of the storage battery, and the output is the terminal voltage of the storage battery.

3. 3. The model parameter estimation device according to claim 1, wherein the nonlinear parameter update unit updates the values ​​of the nonlinear parameters by utilizing a gradient or a gradient approximation of the estimated value of the output with respect to the nonlinear parameters, or by using a gradient-free nonlinear optimization method.

4. the nonlinear parameter update unit generates perturbation values ​​of the nonlinear parameters by perturbing previous values, which are values ​​before the update, when updating the values ​​of the nonlinear parameters; the state quantity calculation unit calculates a state quantity for each of the perturbation values; the linear parameter estimation unit estimates linear parameters for the state quantity for each of the perturbation values; The nonlinear parameter update unit calculates a gradient approximation value for the nonlinear parameter of the estimated value of the output based on an estimate value of the linear parameter for the state quantity for at least each of the perturbation values, and updates the previous value based on the calculated gradient approximation value.

3. The model parameter estimation device according to claim 1 or 2.

5. the nonlinear parameters include at least one of a charging rate, a full charge capacity, a diffusion time constant, and an offset current of a sensor used to measure the current of the storage battery; 3. The model parameter estimation device according to claim 2, wherein the linear parameters include at least one of a DC resistance, a diffusion resistance, and a capacitance of the storage battery.

6. 3. The model parameter estimation device according to claim 2, wherein the linear parameter estimation unit estimates the linear parameters using a linear least-squares solution of a linear least-squares problem.

7. 7. The model parameter estimation device according to claim 6, wherein the state space model includes an OCV function that represents a relationship between an electrical quantity and an open circuit voltage and that includes the linear parameters.

8. the OCV function is a polynomial of the electrical quantity, 8. The apparatus of claim 7, wherein the linear parameters include coefficients of the polynomial.

9. the state space model includes linear equality constraints on the linear parameters; 6. The model parameter estimation device according to claim 2, wherein the linear parameter estimation unit estimates the linear parameters by solving a linear least squares problem including the linear equality constraints by a Lagrange multiplier method.

10. The state space model includes an OCV function that represents a relationship between an electrical quantity and an open circuit voltage and is a piecewise polynomial function including the linear parameters; the linear parameters include linear parameters of the piecewise polynomial; 10. The model parameter estimator of claim 9, wherein the linear equality constraints include linear equality constraints at the knots of the piecewise polynomial.

11. the state space model includes linear inequality constraints on the linear parameters; 3. The model parameter estimation device according to claim 1, wherein the linear parameter estimation unit estimates the linear parameters by solving a linear least squares problem including the linear inequality constraint.

12. the state space model includes OCV characteristic information representing a relationship between a charging rate and an open circuit voltage of the storage battery; 7. The model parameter estimation device according to claim 2, wherein the nonlinear parameters include a full charge capacity.

13. the state space model includes positive electrode potential characteristic information representing a relationship between a positive electrode electric quantity and a positive electrode potential of the storage battery, and negative electrode potential characteristic information representing a relationship between a negative electrode electric quantity and a negative electrode potential of the storage battery, 13. The model parameter estimation device according to claim 12, wherein the nonlinear parameters include at least one of a positive electrode capacity, a negative electrode capacity, a positive electrode electric quantity, and a negative electrode electric quantity, instead of the full charge capacity.

14. A step of setting initial values ​​of nonlinear parameters of a state space model that represents a target system using nonlinear parameters and linear parameters; a state quantity calculation step of calculating a state quantity indicating a state of the target system with respect to the measured value of the input, based on a state equation obtained by substituting values ​​for the nonlinear parameters for the state space model and time-series data of the measured values ​​of the input and output of the target system; a linear parameter estimation step of estimating the linear parameters that minimize an error between the estimated value of the output calculated based on the state space model, the state quantity, and the measured value of the input, and the measured value of the output; and a nonlinear parameter updating step of repeatedly updating the values ​​of the nonlinear parameters until a predetermined convergence condition is satisfied so that the minimized error becomes smaller; A method for estimating model parameters, comprising:

15. A model parameter estimation device as described in claim 14, characterized in that, in the step of calculating the initial value, the initial value is set assuming that the target system is a storage battery, and, in the step of calculating the state quantity, the state quantity is calculated assuming that the input is the current of the storage battery and the output is the terminal voltage of the storage battery.

16. 16. The method for estimating model parameters according to claim 14 or 15, wherein in the nonlinear parameter updating step, values ​​of the nonlinear parameters are updated using a gradient or a gradient approximation of the estimated value of the output for the nonlinear parameters, or using a gradient-free nonlinear optimization technique.

17. the nonlinear parameters include at least one of a charging rate, a full charge capacity, a diffusion time constant, and an offset current of a sensor used to measure the current of the storage battery; 16. The method of claim 15, wherein the linear parameters include at least one of a DC resistance, a diffusion resistance, and a capacitance of the storage battery.

18. the state space model includes linear inequality constraints on the linear parameters; 16. The model parameter estimation method according to claim 14, wherein in the linear parameter estimation step, the linear parameters are estimated by solving a linear least squares problem including the linear inequality constraint.

19. the state space model includes OCV characteristic information representing a relationship between a charging rate and an open circuit voltage of the storage battery; 16. The method of claim 15, wherein the non-linear parameters include a full charge capacity.

20. the state space model includes positive electrode potential characteristic information representing a relationship between a positive electrode electric quantity and a positive electrode potential of the storage battery, and negative electrode potential characteristic information representing a relationship between a negative electrode electric quantity and a negative electrode potential of the storage battery, 20. The method for estimating model parameters according to claim 19, wherein the nonlinear parameters include at least one of a positive electrode capacity, a negative electrode capacity, a positive electrode electric quantity, and a negative electrode electric quantity, instead of the full charge capacity.