Estimation device and estimation method

Abstract

An estimation device (1) comprises an estimation unit (3) that estimates at least a parameter using a storage battery state space model (21) that includes at least a resistance as a parameter, such estimation performed on the basis of the current of a storage battery (50) detected by a current detection device (51) and the voltage of the storage battery (50) detected by a voltage detection device (52). The estimation unit (3) comprises: an inverse sigmoid transformation parameter estimation unit (31) that estimates an inverse sigmoid transformation parameter obtained by performing inverse sigmoid transformation of the parameter; and a sigmoid transformation unit (32) that obtains as an estimated value of the parameter a sigmoid transformation parameter obtained by performing sigmoid transformation of the estimated inverse sigmoid transformation parameter.

Description

Estimation device and estimation method

[0001] The present disclosure relates to an estimation apparatus and an estimation method for estimating the state, parameters, etc. of an apparatus or device such as a motor or a storage battery that constitutes a system.

[0002] When a system includes a storage battery, estimation of state information such as SOC (State Of Charge) and parameters such as resistance and capacity is important for deterioration diagnosis to diagnose the deterioration state of the storage battery. Patent Document 1 listed below discloses a technology for estimating the state and parameters of the storage battery using time-series data of current measurement values ​​and voltage measurement values ​​during operation of the storage battery.

[0003] Specifically, in Patent Document 1, the parameters of an equivalent circuit model of a storage battery are logarithmically transformed to obtain logarithmic transformed parameter values, which are used as state variables, and the state variables are used to iteratively estimate logarithmic transformed parameter values ​​from a state equation and an output equation using a Kalman filter based on the charge / discharge current and the terminal voltage.Then, the iteratively estimated logarithmic transformed parameter values ​​are subjected to an inverse logarithmic transformation to obtain estimated parameter values, which are antilogarithmic numbers corresponding to the logarithmic transformed parameter values.

[0004] JP 2014-74682 A

[0005] However, the estimation device of Patent Document 1 does not impose upper constraints on the parameters, and only considers non-negative lower constraints, so that depending on the input conditions to the estimation device, the estimated values ​​of the parameters may fluctuate significantly, which may result in a decrease in estimation accuracy. Furthermore, if the estimated values ​​of the parameters become unrealistic values, the estimation process calculations will diverge, which forces the estimation process to be interrupted, resulting in a problem of being unable to perform efficient estimation.

[0006] The present disclosure has been made in view of the above, and aims to provide an estimation device that can accurately and efficiently estimate states and parameters.

[0007] In order to solve the above-mentioned problems and achieve the object, an estimation device according to the present disclosure includes an estimation unit that estimates at least parameters using a state space model of the storage battery that includes at least resistance as a parameter, based on the current of the storage battery detected by a current detection device and the voltage of the storage battery detected by a voltage detection device. The estimation unit includes an inverse sigmoid transformation parameter estimation unit and a sigmoid transformation unit. The inverse sigmoid transformation parameter estimation unit estimates inverse sigmoid transformation parameters obtained by inverse sigmoid transformation of the parameters. The sigmoid transformation unit sigmoidally transforms the estimated inverse sigmoid transformation parameters to obtain the sigmoid transformation parameters as parameter estimates.

[0008] The estimation device according to the present disclosure has the advantage of being able to estimate states and parameters accurately and efficiently.

[0009] 1 is a block diagram showing an example of the configuration of an estimation system including an estimation device according to an embodiment; 2 is a block diagram showing an example of a hardware configuration for realizing the functions of an estimation device according to an embodiment; 3 is a diagram showing an example of an equivalent circuit of a storage battery used in an estimation device according to an embodiment; 4 is a diagram illustrating an inverse sigmoid transformation according to an embodiment; 5 is a diagram illustrating a first specific example for explaining the robustness of estimation calculation according to an embodiment; 6 is a diagram illustrating a second specific example for explaining the robustness of estimation calculation according to an embodiment;

[0010] Hereinafter, an estimation apparatus and an estimation method according to embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Note that, although the following embodiments will be described by taking as an example a case where a storage battery is used as a target for estimating states and parameters, the target may be an apparatus or device other than a storage battery.

[0011] FIG. 1 is a block diagram showing an example configuration of an estimation system 100 including an estimation device 1 according to an embodiment. The estimation system 100 according to the embodiment includes the estimation device 1, a storage battery 50, a current detection device 51, and a voltage detection device 52. The estimation device 1 according to the embodiment also includes a model unit 2 and an estimation unit 3. The estimation unit 3 includes an inverse sigmoid transformation parameter estimation unit 31 and a sigmoid transformation unit 32. The model unit 2 includes a storage battery state space model 21, and the inverse sigmoid transformation parameter estimation unit 31 includes a Kalman filter 31a. The Kalman filter 31a is a filter suitable for a state estimation method using sequential estimation. Specifically, a type of nonlinear Kalman filter, such as an extended Kalman filter (EKF) or an unscented Kalman filter (UKF), can be used. Note that the Kalman filter 31a is merely an example and is not limiting. Another filter, such as a particle filter, may be used instead of the Kalman filter 31a.

[0012] The storage battery 50 can be mounted on vehicles such as railway cars, electric vehicles, and hybrid electric vehicles. The storage battery 50 mounted on the vehicle is a power source that provides driving force to the vehicle and powers the vehicle's lighting, air conditioners, and other devices. The storage battery 50 also receives regenerative power during braking and is charged with power from wayside facilities. These operating modes cause the storage battery 50 to repeatedly charge and discharge, causing the internal state of the storage battery 50 to change over time. The estimation device 1 according to the embodiment estimates parameters representing the internal state of the storage battery 50 to grasp the state of the storage battery 50 and evaluate the health of the storage battery 50. A typical example of the storage battery 50 is a lithium-ion battery, but is not limited to this example. The storage battery 50 may also be another type of storage battery, such as a nickel-metal hydride battery.

[0013] The current detection device 51 detects the current of the storage battery 50, i.e., the discharge current and charge current flowing in and out of the storage battery 50. The detected current I is input to the estimation unit 3. The voltage detection device 52 detects the voltage of the storage battery 50, i.e., the voltage between the terminals of the storage battery 50. The detected voltage V is input to the estimation unit 3.

[0014] 2 is a block diagram showing an example of a hardware configuration for realizing the functions of the estimation device 1 according to the embodiment. When realizing the functions of the estimation device 1 according to the embodiment, as shown in FIG. 2, the configuration can include a processor 91 that performs calculations, a memory 92 that stores programs read by the processor 91, and an interface 93 that inputs and outputs signals.

[0015] The processor 91 is an example of a computing means. The processor 91 may be a computing means called a microprocessor, a microcomputer, a CPU (Central Processing Unit), or a DSP (Digital Signal Processor). Examples of the memory 92 include non-volatile or volatile semiconductor memory such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable ROM), and EEPROM (Electrically EPROM), as well as a magnetic disk, a flexible disk, an optical disk, a compact disk, a minidisk, and a DVD (Digital Versatile Disc).

[0016] The memory 92 stores a program that executes the functions of the estimation device 1 according to the embodiment. The processor 91 exchanges necessary information via the interface 93 and executes the program stored in the memory 92, thereby performing the processing described below. The calculation results by the processor 91 can be stored in the memory 92.

[0017] 3 is a diagram showing an example of an equivalent circuit of the storage battery 50 used in the estimation device 1 according to the embodiment. When the storage battery 50 used in the estimation device 1 according to the embodiment is a lithium ion battery, the equivalent circuit has an open circuit voltage OCV and a direct current resistance R ０ and the diffusion resistance R ｄ and the diffusion capacitance C ｄ The diffusion resistance R ｄ and diffusion capacitance Cｄ is an approximate expression of the Warburg impedance caused by the diffusion process of lithium ions inside the storage battery 50. ０ is the diffusion resistance R ｄ It should be noted that when the storage battery state space model 21 is constructed in a simplified manner, the diffusion capacitance may be omitted.

[0018] Other known methods for approximating the Warburg impedance include, for example, two types of models expressed as n-th order (n is a natural number) linear equivalent circuits. The first model is a Foster-type circuit that performs approximation using the sum of a finite number of infinite series expansions. The second model is a Cauer-type circuit that performs approximation using continued fraction expansions. The configurations of these Foster-type and Cauer-type circuits are well known, and therefore will not be described here. Note that the configuration of the equivalent circuit model of the storage battery 50 is not limited to the above. Furthermore, the storage battery state space model 21 is not limited to being configured based on an equivalent circuit model, and may be configured based on other models such as an electrochemical model.

[0019] Returning to the explanation of FIG. 1, the storage battery state space model 21 is ０ , diffusion resistance R ｄ and diffusion capacitance C ｄ 1 is a state space model of the storage battery 50 expressed by including the above as parameters. The estimation unit 3 estimates the state of the storage battery 50 and the parameters of the storage battery state space model 21 using the storage battery state space model 21 based on the current I of the storage battery 50 detected by the current detection device 51 and the voltage V of the storage battery 50 detected by the voltage detection device 52. In FIG. 1 , what is written as "state estimated value" is an estimated value of the state of the storage battery 50, and what is written as "sigmoid transformation parameter estimated value" is an estimated value of the parameters of the storage battery state space model 21. Details of these estimated values ​​will be described later.

[0020] The inverse sigmoid transformation parameter estimation unit 31 estimates inverse sigmoid transformation parameters obtained by inverse sigmoid transformation of the parameters of the storage battery 50. Furthermore, the sigmoid transformation unit 32 obtains sigmoid transformation parameters obtained by sigmoid transformation of the inverse sigmoid transformation parameters estimated by the inverse sigmoid transformation parameter estimation unit 31 as estimated values ​​of the parameters of the storage battery 50.

[0021] As described above, the inverse sigmoid transformation parameter estimation unit 31 includes a Kalman filter 31a. The inverse sigmoid transformation parameter estimation unit 31 sequentially estimates inverse sigmoid transformation parameters using the Kalman filter 31a based on the current I and the voltage V. The parameters estimated in this estimation process are inverse sigmoid transformation parameters. These inverse sigmoid transformation parameters are calculated using a state equation and an output equation, which will be described later. Furthermore, in this estimation process, a state estimate value is calculated using the state equation. An example of the state estimate value is the SOC. The inverse sigmoid transformation parameters estimated by the inverse sigmoid transformation parameter estimation unit 31 are passed to the sigmoid transformation unit 32. The sigmoid transformation unit 32 outputs values ​​obtained by sigmoid-transforming the inverse sigmoid transformation parameters to the outside of the estimation device 1 as sigmoid transformation parameter estimates.

[0022] As described above, the estimation unit 3 calculates an estimated value using the inverse sigmoid transformed parameters, and then performs sigmoid transformation on the estimated value to estimate the parameters of the storage battery 50. That is, the estimation unit 3 does not directly estimate the parameter θ, but estimates a certain X obtained by transforming the parameter θ. In this paper, this X, i.e., the transformed parameter, is appropriately referred to as the "inverse sigmoid transformed parameter."

[0023] Next, the calculation process performed by the inverse sigmoid transformation parameter estimation unit 31 will be described with reference to some equations and some drawings.

[0024] First, the state space model of the storage battery 50 is formulated between the detected current I and voltage V by the following equations (1) and (2).

[0025]

[0026] The above equation (1) is called a state equation, and the above equation (2) is called an output equation. ｄ is the overpotential due to the diffusion component, FCC is the full charge capacity, and I ｏｆｆ is the offset current, and R ｄ , C ｄ are the diffusion resistance and diffusion capacitance, respectively. ｏｃｖ (SOC) is the open circuit voltage of the storage battery 50 corresponding to the state of charge SOC, and R ０ is the component of the series resistance described above. Note that the state space model of the storage battery expressed by equations (1) and (2) is an example, and various formulations are possible without being limited to this.

[0027] In the estimation device 1 according to the embodiment, the storage battery state space model 21 is formulated by the following equations (3) and (4).

[0028]

[0029] In the expanded system represented by the above equation (3), the state of charge SOC and overvoltage V, which were state variables in equation (1), ｄ In addition, the full charge capacity FCC and offset current I ｏｆｆ , DC resistance R ０ , diffusion resistance R ｄ and diffusion capacitance C ｄ are also incorporated into the state variables. Using such a state equation has the advantage of being able to estimate the state and parameters simultaneously.

[0030] As described above, the estimation device 1 according to the embodiment calculates an estimated value using the inverse sigmoid transformation parameter X that has been inversely transformed. Fig. 4 is a diagram illustrating the inverse sigmoid transformation according to the embodiment. Fig. 4 shows the curve of an asymmetric logistic function, which is an example of a desymmetrized sigmoid function.

[0031] There are a wide variety of examples of sigmoid functions, including the cumulative distribution functions of various continuous probability distributions such as the normal distribution, the logistic distribution, the Cauchy distribution, the Student's t distribution, and the Student's z distribution. The estimation device 1 according to the embodiment uses a sigmoid function based on the logistic function, among various sigmoid functions. The logistic function referred to here is a concept that includes not only the symmetric logistic function but also the asymmetric logistic function.

[0032] 4, the horizontal axis represents the inverse sigmoid transformation parameter X, and the vertical axis represents the parameter θ before transformation. The numerical range of the vertical axis, 0 to 600, is the value obtained when the parameter θ is ｄ The diffusion time constant τ is the range of possible values ​​when ｄ is the diffusion resistance R ｄ and diffusion capacitance C ｄ The product of (= R ｄ ×C ｄ The asymmetric logistic function is represented by f(X), and the upper and lower limits of the asymmetric logistic function f(X) are respectively θ ｓｕｐ , θ ｉｎｆ Then, the parameter θ corresponding to the asymmetric logistic function f(X) can be expressed by the following equation (5).

[0033]

[0034] Furthermore, the vertical axis parameter θ is converted into an inverse sigmoid transformation parameter X by the following equation (6). The inverse sigmoid transformation parameter X is a value converted by the inverse function of the asymmetric logistic function f(X).

[0035]

[0036] The state equation and output equation when the inverse sigmoid transformation is used can be expressed as the following equations (7) and (8).

[0037]

[0038] In the above formulas (7) and (8), f ｓ(θ) represents a sigmoid function with θ as an argument. ａ = f ａ (θ) and f ａ (θ) represents the inverse sigmoid function with θ as an argument. The inverse sigmoid function is the inverse function of the sigmoid function. For example, R ｄ ａ is the inverse sigmoid transformed diffusion resistance R ｄ represents f ｓ (R ｄ ａ ) is the diffusion resistance R ｄ It should be noted that the formulation in equation (7) is only an example, and some parameters may be used as state variables without being subjected to inverse sigmoid transformation.

[0039] In the estimation device 1 according to the embodiment, the use of the inverse sigmoid transformation provides the following advantages: (a) The order of the inverse sigmoid transformation parameter X, which is an estimation parameter, is consistent, which facilitates tuning of the setting value for estimation and reduces the risk of significant digit cancellation. Note that the setting value for estimation here refers to a hyperparameter, and in the case of a Kalman filter, for example, refers to the value of the covariance matrix of normal white noise assumed to be added to each of the state equation and output equation, the initial value of the estimation parameter, etc. (b) The estimated value estimated by the inverse sigmoid transformation parameter estimator 31 is always within the range of the upper and lower limit constraints defined by the inverse sigmoid function, which allows knowledge about the estimation parameter to be reflected.

[0040] In the example of FIG. 4, the inverse sigmoid transformation parameter X changes in the range of −∞<X<∞, while the parameter θ is in the range of θ ｉｎｆ <θ<θ ｓｕｐ That is, by using the inverse sigmoid transformation, the parameter θ can be set within the range of the upper and lower limit constraints defined by the inverse sigmoid function. ０ , R ｄ is 10 －４ is of the order of τ ｄ is 10 ２whereas the inverse sigmoid transformed R ０ ａ , R ｄ ａ , τ ｄ ａ is the original parameter R ０ , R ｄ , τ ｄ For a wide range of values, 10 ０ That is, by using the inverse sigmoid transformation, the order of the inverse sigmoid transformation parameter X is automatically normalized. Due to these advantages, the estimation device 1 according to the embodiment can realize robust estimation of the parameter θ.

[0041] 5 and 6 are diagrams showing first and second specific examples, respectively, for explaining the robustness of the estimation calculation in the embodiment. The horizontal axis of FIG. 5 represents the diffusion time constant τ ｄ Inverse sigmoid transformation parameter X τｄ 6, the horizontal axis represents the diffusion resistance R ｄ Inverse sigmoid transformation parameter X Ｒｄ The vertical axis of FIG. 5 represents the inverse sigmoid transformation parameter X τｄ The diffusion time constant τ obtained by sigmoid transformation of ｄ The vertical axis of FIG. 6 represents the value of the inverse sigmoid transformation parameter X Ｒｄ The diffusion resistance R obtained by sigmoid transformation of ｄ The inverse sigmoid transformation parameter X τｄ , X Ｒｄ Even if the value of is of the same order, the diffusion time constant τ ｄ and diffusion resistance R ｄ The values ​​of the diffusion time constant τ ｄ and diffusion resistance R ｄ It can be seen that the estimated value of the diffusion time constant τ ｄ For ｄ , ｓｕｐ and the lower limit τ ｄ , ｉｎｆ The value of the diffusion resistor R shown in FIG.ｄ For ｄ , ｓｕｐ and the lower limit R ｄ , ｉｎｆ The value of will be reflected.

[0042] 7 is a diagram illustrating the relationship between the sigmoid function used in the variable transformation process of the embodiment and other functions. In FIG. 7, an asymmetric logistic function, which is an example of the sigmoid function, is shown by a solid line. An example of the asymmetric logistic function can be expressed by the following equation (9):

[0043]

[0044] In the above formula (9), s is the gentleness of the curve, and p is a setting parameter that sets the strength of asymmetry. When p=1, it becomes symmetrical, and as p becomes smaller in the range of 0<p<1, the asymmetry becomes stronger. For simplicity, in formula (9), the upper limit is set to θ ｓｕｐ = 1, the lower limit is θ ｉｎｆ = 0.

[0045] Specifically, Fig. 7 shows the curve of the asymmetric logistic function in equation (9) with p = 0.1 and s = 0.1, and the base a of the function is Napier's constant e, as a solid line. Note that the base a of the asymmetric logistic function does not have to be Napier's constant e, and the value of the base a is arbitrary. The fact that the value of the base a is arbitrary and the preferred ranges of p and s will be described later.

[0046] 7 also shows a symmetric logistic function, an example of a sigmoid function, with a dashed line. Specifically, FIG. 7 shows a curve with a dashed line when p = 1, s = 1, and the base of the function is Napier's constant e in equation (9). When p = 1, as shown in FIG. 7, when X = 0, f(X) = 0.5, and the function is rotationally symmetric about the point (0, 0.5). Note that, like the asymmetric logistic function, the base of the symmetric logistic function does not need to be Napier's constant e, and the value of the base can be arbitrary. These asymmetric logistic functions and symmetric logistic functions can be suitably used as the sigmoid function in the estimation device 1 of this embodiment.

[0047] For comparison, an exponential function whose base is Napier's constant e is shown by a dashed line in Fig. 7. As shown in Fig. 7, when the inverse sigmoid transformation parameter X is in the range of -∞<X<0, the asymmetric logistic function f(X) and the exponential function e Ｘ Between them, f(X) ≒ e Ｘ Therefore, it can be seen that the same effect can be obtained as when using logarithmic transformation.

[0048] When logarithmic transformation, which is a conventional method, is used, as described in the section [Problems to be Solved by the Invention], there are no upper limit constraints on the parameters, and only non-negative lower limit constraints are considered, so that the estimated values ​​of the parameters may fluctuate greatly depending on the input conditions to the estimation device 1, which may result in a decrease in the estimation accuracy, including that of other state variables such as SOC. Furthermore, when logarithmic transformation is used, if the estimated values ​​of the parameters become unrealistic values, the accuracy of the parameter values ​​obtained during the estimation process cannot be guaranteed, so the estimation process must be interrupted, which may result in an inefficient estimation process.

[0049] In contrast, when the inverse sigmoid transformation is used as in the method of the embodiment, both the upper and lower limit constraints are taken into account for the parameter θ, so the estimated value of the parameter does not fluctuate significantly, making it possible to suppress a decrease in estimation accuracy. Also, in the method of the embodiment, when the asymmetric logistic function shown by the solid line in Figure 7 is used, for example, the value of the parameter θ corresponding to the value of the inverse sigmoid transformation parameter X obtained in the region of X>0 is the upper limit value θ ｓｕｐ Since this is close to the value of , it may be an unrealistic value. However, even if the estimation process enters such a region, it may return to a region where estimation process is possible in subsequent estimation processes. In the case of the conventional method, if the region where X>0 is entered, the estimation process calculation may diverge, which may lead to a situation where the estimation process must be interrupted. On the other hand, in the method of the embodiment, the estimation process calculation does not diverge, so it is possible to continue the estimation process.

[0050] Finally, the fact that the base values ​​of the asymmetric logistic function and the symmetric logistic function are arbitrary, as well as the preferred ranges of p and s in the asymmetric logistic function, will be explained using some mathematical formulas.

[0051] First, the exponential function a with base a －Ｘ and the exponential function b with base b －（Ｘ／ｓ） Consider these a －Ｘ and b －（Ｘ／ｓ） The relationship of the following equation (10) holds true.

[0052]

[0053] Since a and b are the bases of the logarithmic function, a, b > 0. From the above equation (10), changing the bases a and b is equivalent to changing the value of s. Therefore, it can be seen that the value of the base a in the above equation (9) is arbitrary.

[0054] In the above equation (9), the function f when s=1 and p=1 is １ (X)=1 / (1+a －Ｘ ) and the function f when p = 1 and a is Napier's number e. ２ (X)=1 / (1+e －（Ｘ／ｓ） ) Consider the function f １ (X) is a monotonically decreasing function when 0<a<1, and is a monotonically increasing function when a>1, as shown in FIG. ２ (X) is a monotonically decreasing function when s<0, and is a monotonically increasing function as shown in Fig. 7 when s>0. Although Fig. 5 and Fig. 6 show an example of a sigmoid function that is a monotonically increasing function, it goes without saying that a sigmoid function that is a monotonically decreasing function may also be selected.

[0055] Furthermore, when a sigmoid function that is a monotonically increasing function is selected, the function f(X) shown in equation (9) can be transformed into the following equation (11).

[0056]

[0057] Using the last transformation formula in the above formula (11), the curve of the function f(X) is considered. First, the function f(X) is a （Ｘ／ｓ） <<1 with a ＸApproaching a （Ｘ／ｓ） >>1 approaches 1. For example, when a>1, the function f(X) approaches 1 as X=0 approaches 0. Ｘ As X gets larger, it gets closer to 1. Conversely, when 0<a<1, as X gets smaller around X=0, it gets closer to 1, and as X gets larger, a Ｘ Furthermore, as p and s approach 0, the function f(X) approaches min{a Ｘ , 1}, i.e., a Ｘ and 1, whichever is smaller. Conversely, the closer p and s are to 1, the closer the function f(X) is to the standard logistic function. Therefore, by making p and s closer to 0, it is possible to obtain an effect equivalent to logarithmic conversion in the region of X<0 without exceeding the upper limit constraint in the region of X>1.

[0058] However, if p and s are set too close to 0, the value of the second derivative of the function f(X) will become very large near X = 0, which may adversely affect estimation calculations using the sigmoid transformation and inverse sigmoid transformation, and may also reduce the calculation accuracy of the function f(X). Therefore, it is preferable to typically set the values ​​of p and s to about p = s = 0.1. Setting these values ​​makes it possible to obtain a function that provides the above-mentioned benefits while eliminating these concerns.

[0059] The ability to set upper and lower limit constraints on estimated parameters is an advantage of using a sigmoid transformation based on a sigmoid function and an inverse sigmoid transformation based on an inverse sigmoid function. The advantage of being able to normalize estimated parameters over a wide range is an advantage of using an asymmetric sigmoid function as the sigmoid function.

[0060] As described above, the estimation device according to the embodiment includes an estimation unit that estimates at least parameters using a state space model of the storage battery, which is expressed based on the current and voltage of the storage battery and includes at least resistance as a parameter. The estimation unit includes an inverse sigmoid transformation parameter estimation unit and a sigmoid transformation unit. The inverse sigmoid transformation parameter estimation unit estimates inverse sigmoid transformation parameters obtained by inverse sigmoid transformation of the parameters. The sigmoid transformation unit obtains sigmoid transformation parameters as parameter estimates by sigmoid transformation of the estimated inverse sigmoid transformation parameters. With this estimation device configured in this manner, the parameter estimates obtained by the estimation processing calculation fall within the set upper and lower limit values, preventing divergence of the estimation processing calculation itself. This allows the estimation processing to continue and enables efficient execution of the estimation processing calculation. Furthermore, with the estimation device according to the embodiment, since both upper and lower limit constraints are taken into account for the parameters, the parameter estimates do not fluctuate significantly, thereby suppressing a decrease in estimation accuracy.

[0061] In the estimation device according to the embodiment, the state space model may include an inverse sigmoid transformation parameter in the state variables, and the inverse sigmoid transformation parameter estimation unit may estimate the state variables by recursive estimation. A Kalman filter may be used for the recursive estimation.

[0062] Furthermore, the estimation device according to the embodiment may estimate parameters by batch estimation so as to minimize the error between the voltage and the voltage calculated by the state space model. A specific method for parameter estimation by batch estimation may be an approach using an output error method based on the nonlinear least squares method. More specifically, by using a known nonlinear least squares method with the sum of squares of the error between the time series values ​​of the measured voltage and the time series values ​​of the estimated voltage output by the state space model as an evaluation function, parameters can be estimated so as to minimize the evaluation function under appropriate initial value settings. Specific examples of the nonlinear least squares method include the Levenberg-Marquardt method and the Gauss-Newton method, which are based on the gradient or gradient approximation of the evaluation function, and the Nelder-Mead method, which does not use a gradient. Alternatively, metaheuristics such as particle swarm optimization, simulated annealing, and genetic algorithms may be used.

[0063] Furthermore, an estimation method according to an embodiment estimates at least parameters using a state space model of a storage battery, the state space model including at least resistance as a parameter, based on the current of the storage battery detected by a current detection device and the voltage of the storage battery detected by a voltage detection device. The estimation method may include the following steps: an estimation step and a conversion step. In the estimation step, the parameters are subjected to an inverse sigmoid transformation to estimate inverse sigmoid-transformed parameters; and in the conversion step, the inverse sigmoid-transformed parameters estimated in the estimation step are subjected to a sigmoid transformation to obtain sigmoid-transformed parameters as estimated values ​​of the parameters. According to this estimation method, the estimated values ​​of the parameters obtained by the estimation process fall within the set upper and lower limit values. This prevents the estimation process from diverging, allowing the estimation process to continue and the estimation process to be performed efficiently. Furthermore, according to the estimation method according to an embodiment, both upper and lower limit constraints are taken into account for the parameters, preventing large fluctuations in the estimated values ​​of the parameters and suppressing a decrease in estimation accuracy.

[0064] The configurations shown in the above embodiments are merely examples, and may be combined with other known technologies, and parts of the configurations may be omitted or modified without departing from the spirit of the invention.

[0065] For example, when the estimation device 1 is distributed and installed on a plurality of railway vehicles, the functions of the inverse sigmoid transformation parameter estimation unit and the sigmoid transformation unit in the estimation unit described above may be distributed among the plurality of estimation devices. Furthermore, the state space model does not necessarily have to be included in the estimation device. When the estimation device is configured to be able to communicate with a ground processing device or a cloud server on the Internet, the state space model of the storage battery, for example, may be included in the processing device or the cloud server.

[0066] REFERENCE SIGNS LIST 1 Estimation device, 2 Model unit, 3 Estimation unit, 21 Storage battery state space model, 31 Inverse sigmoid transformation parameter estimation unit, 31a Kalman filter, 32 Sigmoid transformation unit, 50 Storage battery, 51 Current detection device, 52 Voltage detection device, 91 Processor, 92 Memory, 93 Interface, 100 Estimation system.

Claims

1. The system includes an estimation unit that estimates at least the parameters using a state-space model of the battery, which is expressed with at least resistance as a parameter, based on the current of the battery detected by a current detection device and the voltage of the battery detected by a voltage detection device. The estimation unit, An inverse sigmoid transform parameter estimation unit estimates inverse sigmoid transform parameters obtained by performing an inverse sigmoid transform on the aforementioned parameters, A sigmoid transform unit obtains a sigmoid transform parameter obtained by performing a sigmoid transform on the estimated inverse sigmoid transform parameter, and uses this parameter as an estimated value. An estimation device characterized by comprising:

2. The sigmoid transform and inverse sigmoid transform processes use a sigmoid function based on the logistic function. The estimation device according to feature 1.

3. The sigmoid transform and inverse sigmoid transform processes use an asymmetrical sigmoid function. The estimation device according to feature 1.

4. The state-space model includes the inverse sigmoid transformation parameter in its state variables, The inverse sigmoid transform parameter estimation unit estimates the state variables by sequential estimation. The estimation device according to any one of claims 1 to 3.

5. The aforementioned sequential estimation uses a Kalman filter. The estimation device according to feature 4.

6. The parameters are estimated by batch estimation so that the error between the voltage and the voltage calculated by the state-space model is minimized. The estimation device according to any one of claims 1 to 3.

7. The state-space model of the battery further includes diffusion capacity as a parameter. The estimation device according to any one of claims 1 to 3.

8. The state-space model includes at least one of the following as state variables: charge level, full charge capacity, and offset current. The estimation device according to any one of claims 1 to 3.

9. An estimation method for estimating at least the parameters of a battery, which are expressed using a state-space model of the battery that includes at least resistance as a parameter, based on the current of the battery detected by a current detection device and the voltage of the battery detected by a voltage detection device, The estimation method described above is An estimation step of estimating the inverse sigmoid transformed parameters obtained by inverse sigmoid transforming the aforementioned parameters, A transformation step to obtain a sigmoid transformation parameter obtained by performing a sigmoid transformation on the estimated inverse sigmoid transformation parameter is used as the estimated value of the parameter. An estimation method characterized by including the following.

10. The sigmoid transform and inverse sigmoid transform processes use a sigmoid function based on the logistic function. The estimation method according to feature 9.

11. The sigmoid transform and inverse sigmoid transform processes use an asymmetrical sigmoid function. The estimation method according to claim 9 or 10.