JP4689755B1 - Battery internal state estimation device and battery internal state estimation method - Google Patents

Abstract

To efficiently learn parameters included in a battery simulation model.
In a battery internal state estimation device for estimating an internal state of a battery based on a simulation model of the battery, storage means (RAM10c) for storing a plurality of parameters included in the simulation model, and a discharge flowing from the battery to a load Detection means (I / F 10d) for detecting current, selection means (CPU 10a) for selecting a parameter to be subjected to adaptive learning in accordance with the value of the discharge current detected by the detection means, and parameters selected by the selection means On the other hand, it has an adaptive learning means (CPU10a) for performing adaptive learning.
[Selection] Figure 2

Description

  The present invention relates to a battery internal state estimation device and a battery internal state estimation method.

  As a method for estimating the internal state of the battery, for example, a technique disclosed in Patent Document 1 is known. That is, Patent Document 1 discloses a technique for estimating SOC (State of Charge) indicating a state of charge of a battery using a Kalman filter.

JP 2007-187534 A

  By the way, the technique disclosed in Patent Document 1 has a problem in that learning efficiency is low because parameters are adaptively learned in parallel.

  Therefore, an object of the present invention is to provide a battery internal state estimation device and a battery internal state estimation method capable of efficiently performing adaptive learning processing.

In order to solve the above-described problems, a battery internal state estimation device according to the present invention is a battery internal state estimation device that estimates the internal state of a battery based on a battery simulation model, and includes a plurality of parameters included in the simulation model. Storing means for storing; detecting means for detecting a discharge current flowing from the battery to a load; selecting means for selecting a parameter to be subjected to adaptive learning according to the value of the discharge current detected by the detecting means; Adaptive learning means for performing adaptive learning on the parameter selected by the selection means.
According to such a configuration, adaptive learning can be performed efficiently.

According to another invention, in addition to the above invention, the load has at least an electric motor for starting a prime mover, and the detection means detects a current when the prime mover is started by the electric motor, The selection unit selects a parameter according to a value of a current flowing through the electric motor, and the adaptive learning unit performs adaptive learning on the parameter selected by the selection unit.
According to such a configuration, it becomes possible to select a parameter based on the current flowing through the electric motor.

In addition to the above invention, another invention has at least a constant phase element (CPE) as an equivalent circuit of the anode and cathode of the battery as a component of the simulation model, and the constant phase element is in parallel The RC parallel unit composed of a plurality of connected resistors and capacitors is represented by an equivalent circuit connected in series, and the element values of the resistors and capacitors constituting each RC parallel unit are used as the parameters. A predetermined RC parallel unit predetermined according to the value of the discharge current is selected, and the adaptive learning means adaptively learns the resistance and capacitor element values selected by the selection means.
According to such a configuration, it is possible to efficiently learn the parameters of the constant phase element constituting the simulation model.

In addition to the above-mentioned invention, another invention has an internal resistance of the battery as a component of the simulation model, the resistance value of the internal resistance is the parameter, and the selection means The internal resistance is selected when a peak current flowing at the time of startup is detected.
According to such a configuration, it is possible to efficiently learn the internal resistance having a small element value.

In addition to the above-mentioned invention, another invention has a voltage source as a component of the simulation model, and has a concentration value of an electrolytic solution in the battery as a parameter relating to a voltage of the voltage source, and the selection means Is characterized in that when the discharge current is 0 or in the vicinity of 0, the concentration of the electrolytic solution is selected as an object of adaptive learning.
According to such a configuration, it is possible to efficiently learn the concentration of the electrolytic solution in a stable state.

According to another invention, in addition to the above-described invention, the parameter constitutes a state vector of an extended Kalman filter, and the adaptive learning means executes the adaptive learning on the state vector.
According to such a configuration, the parameters constituting the simulation model can be efficiently learned based on the extended Kalman filter.

The battery internal state estimation device of the present invention is a battery internal state estimation method for estimating the internal state of the battery based on a battery simulation model, and a storage step of storing a plurality of parameters included in the simulation model; A detection step for detecting a discharge current flowing from the battery to a load; a selection step for selecting a parameter to be subjected to adaptive learning according to the magnitude of the discharge current detected in the detection step; and a selection step selected in the selection step. An adaptive learning step for performing adaptive learning with respect to the parameters.
According to such a method, adaptive learning can be performed efficiently.

  According to the present invention, it is possible to provide a battery internal state estimation device and a battery internal state estimation method capable of efficiently performing adaptive learning processing.

It is a figure which shows the structural example of the battery internal state estimation apparatus which concerns on embodiment of this invention. It is a figure which shows the structural example of the control part shown in FIG. It is a figure for demonstrating the processing algorithm performed in this embodiment. It is an example of the simulation model of a lead acid battery. It is a figure which shows the impedance characteristic of the simulation model shown in FIG. It is a figure which shows the equivalent circuit of CPE shown in FIG. It is a figure which shows the relationship between CPE shown in FIG. 6, and an electrical double layer. It is a figure which shows the relationship between the discharge current and the parameter used as the object of adaptive learning in this embodiment. It is a figure which shows the voltage drop which arises in each component of a simulation model in this embodiment. It is a figure which shows the relationship between SOH estimated by this embodiment, and a measured value. It is a figure which shows the relationship between SOH estimated by this embodiment, and an error. 3 is a flowchart for explaining a flow of processing executed in the battery internal state estimation device shown in FIG. 1. It is a flowchart for demonstrating the detail of the process of step S22 of FIG.

  Next, an embodiment of the present invention will be described.

(A) Description of Configuration of Embodiment FIG. 1 is a diagram illustrating a configuration example of a battery internal state estimation device according to an embodiment of the present invention. As shown in FIG. 1, a battery internal state estimation device 1 of this embodiment includes a control unit 10, a voltage detection unit 11 (corresponding to a part of “detection means” in the claims), and a current detection unit 12 (invoice). Corresponding to a part of “detection means” in the section) and the discharge circuit 14 as main components, and the internal state of the lead storage battery 13 (corresponding to “battery” in the claims) is estimated. In this example, an alternator 15, a cell motor 16 (corresponding to “electric motor” in the claims), and a load 17 are connected to the lead storage battery 13 via the current detector 12. In the present embodiment, the case where the battery internal state estimation device 1 is mounted on a vehicle such as an automobile will be described as an example.

  As shown in FIG. 2, the control unit 10 includes a CPU (Central Processing Unit) 10a (corresponding to “selection means” and “adaptive learning means” in the claims), a ROM (Read Only Memory) 10b, and a RAM (Random Access). Memory) 10c (corresponding to “storage means” in claims) and I / F (Interface) 10d (corresponding to a part of “detecting means” in claims) are main components. The CPU 10a controls each part of the apparatus based on a program 10ba stored in the ROM 10b. The ROM 10b is constituted by a semiconductor memory, and stores the program 10ba and other information. The RAM 10c is constituted by a semiconductor memory, and stores the parameter 10ca and other information in a rewritable manner. The I / F 10d converts detection signals from the voltage detection unit 11 and the current detection unit 12 into digital signals and inputs them, and supplies a control signal to the discharge circuit 14 to control it.

  The voltage detection unit 11 detects the terminal voltage of the lead storage battery 13 and notifies the control unit 10 of it. The current detection unit 12 detects a current flowing through the lead storage battery 13 and notifies the control unit 10 of the current. The discharge circuit 14 has a semiconductor switch, for example, and discharges the electric power stored in the lead storage battery 13 based on the control of the control unit 10. The alternator 15 is driven by a prime mover (not shown) such as a reciprocating engine, for example, and generates DC power to charge the lead storage battery 13. The cell motor 16 is constituted by, for example, a DC motor, and is rotated by DC power supplied from the lead storage battery 13 to start the prime mover. The load 17 is composed of, for example, an automobile headlight, a wiper, a direction indication light, a navigation device, and other devices.

  FIG. 3 is a diagram for explaining an outline of a processing algorithm realized by executing the program 10ba. As shown in this figure, in this embodiment, a simulation model 30 of a lead storage battery 13 having a plurality of parameters is set. Then, an observation value is obtained by observing the target lead storage battery 13 and a calculated value corresponding to the observation value is obtained based on the simulation model 30. Deviations between these observed values and calculated values are calculated, and optimal parameters are estimated by adaptive learning using the extended Kalman filter 31. Then, the simulation model 30 can be optimized by updating the simulation model 30 with the estimated parameters. Using the simulation model 30 optimized in this way, SOC (State of Charge), SOF (State of Function), SOH (State of Health), and the like for estimating the internal state of the lead storage battery 13 are obtained by calculation. Can do. In the present embodiment, adaptive learning can be performed with high accuracy by selecting a parameter to be subjected to adaptive learning according to the value of the discharge current flowing out from the lead storage battery 13.

  In this specification, “adaptive learning” refers to a method in which a flexible and general model having parameters is prepared and the parameters are optimized statistically and adaptively by learning. In the following embodiment, an extended Kalman filter is used as an example of adaptive learning. However, the present invention is not limited to such a case. For example, adaptive learning using a neural network model or a genetic algorithm is used. It is also possible to use adaptive learning using a model. That is, any method can be used as long as it creates a learning target model and optimizes the parameters constituting the model based on the results obtained by observation.

  FIG. 4 is a diagram illustrating an example of the simulation model 30 (in this example, an electrical equivalent circuit) of the lead storage battery 13. In this example, the simulation model 30 includes a resistance R0, an inductance L, impedances Z1 and Z2, and a voltage source V0 as main components.

Here, the resistance R0 is an internal resistance whose main elements are the conductive resistance of the electrode of the lead storage battery 13 and the liquid resistance of the electrolytic solution. The inductance L is an inductive component caused by an electric field generated by a current flowing through the electrode of the lead storage battery 13 or the like. The inductance L is very small compared to the inductance value of the cable connected to the lead storage battery 13 and can be ignored as necessary. The impedance Z1 is an equivalent circuit corresponding to the anode of the lead storage battery 13 and the electrolytic solution in contact with the anode. The impedance Z1 basically has a characteristic based on the Butler-Volmer formula, and the constant phase element CPE1. It can be expressed as a parallel connection circuit with the resistor R1. Details of the impedance Z1 will be described later. The impedance Z2 is an equivalent circuit corresponding to the cathode of the lead storage battery 13 and the electrolytic solution in contact therewith, has a characteristic based on the aforementioned Butler-Volmer equation, and a parallel connection circuit of the constant phase element CPE2 and the resistor R2 Can be expressed as Details of the impedance Z2 will also be described later. Voltage source V0, the internal impedance is the ideal voltage source 0, a voltage source whose voltage is represented the electrolytic solution concentration C _N and the electrolyte concentration C _P in the vicinity of the anode in the vicinity of the cathode as a parameter.

  FIG. 5 is a diagram showing impedance characteristics of the equivalent circuit shown in FIG. In FIG. 5, the vertical axis represents the imaginary component of the impedance (Im (Z)), and the horizontal axis represents the real component of the impedance (Re (Z)). The thick line in the figure indicates the impedance characteristics of the equivalent circuit. In this example, as the frequency increases, the impedance of the equivalent circuit moves from the right to the left on the thick line, first moves in a locus so as to be asymptotic to the semicircle indicated by Z2, and then Z1 Draw a trajectory and move asymptotically to the semicircle indicated by. Then, at a high frequency, the real number component is asymptotic to a straight line of R0, and due to the inductance L, the impedance increases as the frequency increases.

  FIG. 6 shows an equivalent circuit of impedances Z1 and Z2 used in the present embodiment. In this example, the impedance Z1 is formed by connecting a plurality of RC parallel units in which resistors Ra1, Ra2 and Ra3 and capacitors Ca1, Ca2 and Ca3 are connected in parallel, in series. Specifically, the resistor Ra1 and the capacitor Ca1 are connected in parallel to form one RC parallel unit, and similarly, the resistor Ra2 and the capacitor Ca2 and the resistor Ra3 and capacitor Ca3 are connected in parallel to form an RC parallel unit. . As shown in FIG. 6, the voltage drops generated in the RC parallel units are denoted by ΔVa1, ΔVa2, and ΔVa3, respectively.

  The impedance Z2 is formed by connecting a plurality of RC parallel units in which resistors Rb1, Rb2, and Rb3 and capacitors Cb1, Cb2, and Cb3 are connected in parallel. In addition, as shown in FIG. 6, let voltage drop which arises in each RC parallel unit be (DELTA) Vb1, (DELTA) Vb2, (DELTA) Vb3, respectively.

  FIG. 7 is a diagram showing the relationship between the discharge current of the lead storage battery 13 and the electric double layer generated between the electrode and the electrolyte. FIG. 7 (A) shows the state of the electric double layer when the discharge current is small, and FIG. 7 (B) shows the state of the electric double layer when the discharge current is larger than FIG. 7 (A). 7 (C) shows the state of the electric double layer when the discharge current is larger than that in FIG. 7 (B). As shown in FIG. 7, as the current flows more, the electric double layer tends to spread. In the present embodiment, the state shown in FIG. 7A is represented by an RC parallel unit composed of a resistor Ra3 and a capacitor Ca3 and an RC parallel unit composed of a resistor Rb3 and a capacitor Cb3. The state shown in FIG. 7B is represented by an RC parallel unit composed of a resistor Ra2 and a capacitor Ca2, and an RC parallel unit composed of a resistor Rb2 and a capacitor Cb2. Further, the state shown in FIG. 7C is represented by an RC parallel unit composed of a resistor Ra1 and a capacitor Ca1 and an RC parallel unit composed of a resistor Rb1 and a capacitor Cb1.

(B) Description of schematic operation of embodiment Next, a schematic operation of the present embodiment will be described. In this embodiment, when starting the prime mover by the cell motor 16, the parameter of the simulation model 30 is selected according to the value of the discharge current flowing from the lead storage battery 13 to the cell motor 16, and adaptive learning is performed on the selected parameter. Adaptive learning is not performed for other parameters. FIG. 8 is a diagram showing the relationship between the discharge current flowing from the lead storage battery 13 and time when the prime mover is started by the cell motor 16. 9 is the same as FIG. 8, in the same case as FIG. 8, the voltage drop (VR0) generated in the resistor R0, the voltage drop (VL) generated in the inductance L, the voltage drop (VZ1) generated in the impedance Z1, and the voltage drop generated in the impedance Z2. (VZ2) is shown. In this example, the period (.tau.1) before the starter motor 16 is rotated, the discharge current is not flowing mostly parameters C _N and C _P indicates the electrolytic solution concentration during this period is subject to adaptive learning . In the period before the cell motor 16 is rotated, the concentration distribution of the electrolytic solution in the lead storage battery 13 is stable. By selecting these parameters and performing adaptive learning in such a period, an accurate concentration can be obtained. The value can be learned.

  When the rotation of the cell motor 16 is started, the discharge current starts to flow rapidly (inrush current flows). Since the cell motor 16 is a direct current brush motor, this inrush current becomes the largest when starting from a stopped state. The magnitude of the inrush current is considered to vary depending on, for example, the environmental temperature and the state of the prime mover. Specifically, when the prime mover is cooled because the engine is stopped for a long time or the ambient temperature is low, the viscosity of the lubricating oil is increased, so that the driving torque is large and the current is increased accordingly. In this example, the resistor R0 is selected and adaptive learning is performed in a period (period τ2) exceeding a predetermined threshold value (400 A in this example) defined in the vicinity of 500 A that is the peak current. Note that the value of the resistor R0 is adaptively learned during the period in which the current close to the maximum current flows, because the value of the resistor R0 is very small. Because it is easy to do.

  Next, in the first period (period τ3) in which the current flowing through the cell motor 16 is less than the peak current, Ra1, Ca1, Rb1, and Cb1 are selected as targets for adaptive learning. This period corresponds to the state of FIG. 7C, and this state is associated with the RC parallel unit composed of Ra1 and Ca1 and the RC parallel unit composed of Rb1 and Cb1. In a period in which the discharge current is 300A to 200A, Ra1, Ca1, Rb1, and Cb1 are selected and adaptive learning is executed.

  Next, Ra2, Ca2, Rb2, and Cb2 are selected as targets for adaptive learning in the period where the discharge current is 200A to 100A. This period corresponds to the state of FIG. 7B, and this state is associated with the RC parallel unit configured with Ra2 and Ca2 and the RC parallel unit configured with Rb2 and Cb2. In the period where the discharge current is 200A to 100A, Ra2, Ca2, Rb2, and Cb2 are selected and adaptive learning is executed.

  Next, Ra3, Ca3, Rb3, and Cb3 are selected as targets for adaptive learning in a period in which the discharge current is 100 A to 0 A. This period corresponds to the state of FIG. 7A, and this state is associated with the RC parallel unit configured with Ra3 and Ca3 and the RC parallel unit configured with Rb3 and Cb3. In a period in which the discharge current is 100 A to 0 A, Ra3, Ca3, Rb3, and Cb3 are selected, and adaptive learning is executed. When the prime mover is started by the cell motor 16, charging by the alternator 15 is started, so that the current flowing through the lead storage battery 13 becomes positive (see FIG. 8).

  As an adaptive learning method, for example, the detection values of the voltage detection unit 11 and the current detection unit 12 are acquired in units of a predetermined period (10 mS) to generate observation values based on the detection values. A predicted value based on 30 is generated. Then, it can be realized by comparing the observed value with the predicted value and optimizing the parameter selected based on the comparison result based on the extended Kalman filter. Note that, as described above, various methods exist as the optimal learning method, and therefore, an optimal method may be selected from a plurality of methods.

  By using the simulation model 30 in which adaptive learning has been performed as described above, for example, SOC, SOF, and SOH can be accurately calculated. By referring to these values, the internal state of the lead storage battery 13 can be accurately known.

  A specific example is shown. FIG. 10 is a diagram showing a correspondence relationship between SOH calculated (estimated) using the simulation model 30 of the present embodiment and the measured value of SOH. In this figure, the horizontal axis indicates the actual measurement value, the vertical axis indicates the estimated value, and each point indicates a sample. In FIG. 10, each point is almost gathered on a 45 degree line where the estimated value and the actually measured value coincide with each other, indicating that the validity of the estimation according to the present embodiment is high. FIG. 11 is a diagram showing the relationship between the SOH error and the number of samples. In this figure, the horizontal axis indicates the error (%) of SOH, and the vertical axis indicates the number of samples. As shown in this figure, the samples with errors within 10% account for the majority of the total. Also from this figure, it can be seen that the validity of this embodiment is high.

  As described above, in the present embodiment, the target parameter is selected according to the value of the discharge current, and adaptive learning is executed. Thereby, since the adaptive learning is performed at the timing at which each parameter appears most remarkably in the observed value, the internal state of the lead storage battery 13 can be accurately known by performing the adaptive learning efficiently. Moreover, the time required for calculation can be shortened by setting a limited parameter of the plurality of parameters as the target of adaptive learning. As a result, for example, it becomes possible to use a processor having a low computing capacity, so that the manufacturing cost of the apparatus can be reduced.

(C) Description of Detailed Operation of Embodiment Next, the detailed operation of this embodiment will be described with reference to the flowcharts shown in FIGS. FIG. 12 is a flowchart for explaining details of processing executed in the embodiment shown in FIG. 1. FIG. 13 is a flowchart for explaining details of step S22 in FIG. In the present embodiment, when the voltage drop due to the resistor R0 and the impedances Z1 and Z2 is ΔV in the equivalent circuit shown in FIG. 4, the predicted value of the voltage drop ΔV by the simulation model 30 is obtained as ΔVx. Then, the discharge capacity of the lead storage battery 13 is determined by comparing the predicted value ΔVx with a predetermined allowable value. Here, the inductance L is ignored because the impedance due to the wiring is larger. Further, it is assumed that the following relationship is established among the voltage drop ΔV, the terminal voltage V of the lead storage battery 13, and the voltage V0 of the voltage source corresponding to OCV (Open Circuit Voltage).

ΔV = V−V0 (1)

  In order to accurately determine the discharge capacity of the lead storage battery 13, it is necessary to accurately estimate the predicted value ΔVx of the voltage drop when the simulation model 30 (equivalent circuit) of the lead storage battery 13 is discharged with a predetermined pattern of current. is there. In the present embodiment, when the prime mover is started by the cell motor 16, the lead storage battery 13 is observed at discrete times of a predetermined time interval (ΔT), and the parameters of the simulation model 30 are adaptively learned based on the obtained observation data. Thus, the voltage drop ΔVx can be accurately estimated. When the flowchart shown in FIG. 12 is executed, the following processing is executed. This process is executed when, for example, the “ignition key” for starting the engine (motor) of the automobile is rotated.

In step S10, the CPU 10a adds a time interval ΔT (for example, 10 mS) to the time T _n at the time of the previous processing execution to obtain the current time T _{n + 1} . Specifically, since the process shown in FIG. 12 is executed at each time interval ΔT, the time is increased by ΔT each time the process is executed.

In step S <b> 11, the CPU 10 a acquires an observation value of the lead storage battery 13. Specifically, the CPU 10a turns on the discharge circuit 14 for a period shorter than ΔT (for example, a period of 10% of ΔT) and allows the discharge current to flow, and the terminal voltage of the lead storage battery 13 is supplied from the voltage detection unit 11. Obtained and stored in the variable V _{n + 1} , obtains the discharge current of the lead storage battery 13 from the current detection unit 12, and stores it in the variable In _{+ 1} . Further, SOC _{n + 1} is calculated based on a predetermined SOC calculation method. As a specific calculation method, for example, the measured value of the stable OCV at the start of the lead storage battery 13 (specifically, the terminal voltage (= V0) of the lead storage battery 13 measured in the period of τ1 in FIG. 8) and the current There are a method using a combination of integrated values and a method using current-voltage characteristics (I / V) in the operating environment, and either method may be used. Since the voltage detector 11 has a very high internal resistance and little current flows during the measurement, the voltage measured at the start of the lead storage battery 13 is hardly affected by the resistor R0 and the impedances Z1 and Z2. Suppose that the voltage V0 of the voltage source can be measured.

In step S12, the CPU 10a obtains a voltage drop ΔV _{n + 1} by substituting the observation value acquired in step S11 into the following equation.
ΔV _{n + 1} = V _{n + 1} −OCV _{n + 1} (2)
Here, OCV _{n + 1} can be calculated from the SOC _{n + 1} obtained in step S12 by the following equation.
OCV _{n + 1} = a · SOC _{n + 1} + b (3)

  In Equation (3), the values obtained in advance through experiments or the like can be used as the coefficients a and b. Such a value can be stored in the ROM 10b, for example. Further, since the coefficients a and b may have temperature dependency, the coefficients a and b corresponding to the temperature obtained from the temperature detection unit (not shown) are obtained from the table stored in the ROM 10b. Also good.

In step S13, the CPU 10a updates the Jacobian matrix F _n from the n-th observed value and the previous state vector estimated value. Here, in the present embodiment, the Jacobian matrix F _n is expressed as follows. Here, diag represents a diagonal matrix.
F _n = diag (1−ΔT / Ra1 _n · Ca1 _n , 1−ΔT / Ra2 _n · Ca2 _n , 1−ΔT / Ra3 _n · Ca3 _n , 1−ΔT / Rb1 _n · Cb1 _n , 1−ΔT / Rb2 _n · Cb2 _n , 1-ΔT / Rb3 _n · Cb3 _n , 1,1,1,1,1,1,1,1,1,1,1,1,1) (4)

In step S14, the CPU 10a sets ΔV _{n + 1} calculated from the observation value obtained by the observation in step S12 as the observation residual Y _{n + 1} .
Y _{n + 1} = ΔV _{n + 1} (5)

In step S15, the CPU 10a predicts one cycle ahead of the state vector. Specifically, for example, the voltage drop ΔVa1 _n of the leftmost RC parallel circuit in FIG. 6 is expressed by the following equation.
ΔVa1 _{n + 1} = ΔVa1 _n −ΔVa1 _n · ΔT / Ra1 _n · Ca1 _n (6)

Therefore, the state vector X _n ^T is represented by the following equation. T represents a transposed matrix.
^{_{X n T = (ΔVa1 n,}} ΔVa2 n, ΔVa3 n, ΔVb1 n, ΔVb2 n, ΔVb3 n, R0 n, Ra1 n, Ra2 n, Ra3 n, Ca1 n, Ca2 n, Ca3 n, Rb1 n, Rb2 n, _{Rb3 n, Cb1 n, Cb2 n} , Cb3 n) ··· (7)

Further, the input vector U _n ^T is expressed by the following equation.
^{_{U n T = (Δt · I}} n / Ca1 n, Δt · I n / Ca2 n, Δt · I n / Ca3 n, Δt · I n / Cb1 n, Δt · I n / Cb2 n, Δt · I n / Cb3 _n , 0,0,0,0,0,0,0,0,0,0,0,0,0) (8)

Therefore, the predicted value _{Xn + 1} | _n of one cycle ahead of _Xn is calculated by the following equation.
_{X n + 1 | n = F} n · X n + U n ··· (9)

Further, the Jacobian H _n ^T related to the observation model is expressed by the following equation.
H _n ^T = (1,1,1,1,1,1, I _n , 0,0,0,0,0,0,0,0,0,0,0,0) (10)
At this time, the system equation and the observation equation can be expressed by the following equations.
System equation: X _{n + 1} = F _n · X _n (11)
Observation equation: Y _{n + 1} = H _n ^T · X _n (12)

In step S16, the CPU 10a calculates an optimal Kalman gain using the predicted value _{Xn + 1} | _n one cycle ahead and the observed value Yn _{+ 1} , and calculates an updated value _{Xn + 1} | _{n + 1} of the state vector based on the optimal Kalman gain. calculate. In the calculation based on the normal extended Kalman filter, all parameters are updated based on the obtained state vector X _{n + 1} | _{n + 1. In} this embodiment, as will be described later, the parameter is determined according to the value of the discharge current. Update processing is executed only for predetermined parameters.

In step S17, the CPU 10a determines whether or not the current I _n observed in step S12 is I _n ≈0A, and if I _n ≈0 A (step S17: YES), the process proceeds to step S18. Otherwise (step S17: NO), the process proceeds to step S19. For example, in the period of τ1 shown in FIG. 8, since I _n ≈0A is established, in this case, the process proceeds to step S18. As a specific determination method, for example, a current value of the load 17 that is steadily flowing when the engine is stopped (for example, a current value of about several tens of mA that flows in an in-vehicle timepiece, a car security system, etc.) flows. If it is determined that I _n ≈0A, the process proceeds to step S18.

In step S18, CPU 10a updates the values of the parameters _{C N} and _{C P.} Here, the parameters C _N and C _P represent the electrolyte concentration of the cathode and the anode, respectively. The voltage of the voltage source V0 is expressed as a function V0 (C _N , C _P ) of these parameters C _N and C _P. In step S18, C _N and C _P are obtained from the observed voltage and current, and these parameters C _N and C _P are updated with the obtained values. It should be noted that, based on the parameters C _N and C _P determined in this way, the measured value of the stable OCV at the start of the lead storage battery 13 described above using the function V 0 (C _N , C _P ) (in detail, τ 1 in FIG. 8). The terminal voltage V0) of the lead storage battery 13 (see step S11) measured during the period can be obtained.

In step S19, the CPU 10a determines whether or not the current I _n observed in step S12 is I _n > 400A. If I _n > 400A (step S19: YES), the process proceeds to step S20. Otherwise (step S19: NO), the process proceeds to step S21. For example, in the period of τ2 shown in FIG. 8, since I _n > 400A is established, in that case, the process proceeds to step S20.

In step S20, CPU 10a, the state vector _X n + ₁ determined in step S16 _| by the value of the parameter R0 _n included in the _{n + 1,} and updates the value of the parameter R0 _n. Thereby, R0 _n becomes a new value.

In step S21, CPU 10a, when the current _{I n} which is observed in step S12 it is determined whether the 300A ≧ _I n> 0A, is 300A ≧ _I n> 0A satisfied (step S21: YES) step in The process proceeds to S22, and otherwise (step S21: NO), the process proceeds to step S23. For example, in FIG. 8, when the period of τ3 to τ5 is satisfied, 300A ≧ I _n > 0A is established, and the process proceeds to step S22.

  In step S22, the CPU 10a executes a process for updating the parameters Z1 and Z2. Details of this processing will be described later with reference to FIG.

  In step S23, the CPU 10a determines whether or not to repeatedly execute the process. If the process is repeatedly executed (step S23: YES), the process returns to step S10 and the same process as described above is repeated. In (Step S23: NO), the process proceeds to Step S24. For example, the process is repeated until a predetermined time (for example, several seconds) elapses after the prime mover is started, and when the predetermined time elapses, the process proceeds to step S24.

  In step S24, the CPU 10a estimates ΔVx. Specifically, the CPU 10a estimates the voltage drop ΔVx when the lead storage battery 13 is discharged with a predetermined current pattern based on the simulation model 30 using the parameters updated by the extended Kalman filter calculation. The predetermined current pattern described above can be determined, for example, by adding a current pattern of a load to be newly activated to the current discharge current.

As a specific calculation method, it can be expressed as follows using the relationship of Expression (5) and the value of a predetermined current pattern Ix _{n + 1} .
ΔVx _{n + 1} = ΔVa1 _{n + 1} + ΔVa2 _{n + 1} + ΔVa3 _{n + 1} + ΔVb1 _{n + 1} + ΔVb2 _{n + 1} + ΔVb3 _{n + 1} + R0 · Ix _{n + 1} (13)

  As a method for reducing the calculation load, although the accuracy is slightly lower than the method described above, for convenience, ΔVx = (Ra + Rb) × Ix or ΔVx = Ra × Ix, ΔVx = G (Ra , Rb, Ix) may be calculated from an experiment.

  In step S25, the CPU 10a compares the predicted voltage drop ΔVx with a predetermined allowable value ΔVlimit. If ΔVx is equal to or smaller than ΔVlimit (step S25: YES), the CPU 10a proceeds to step S26, and otherwise (step S25: If NO, the process proceeds to step S27.

  In step S26, the CPU 10a determines that the discharge capacity is sufficient. In step S27, the CPU 10a determines that the discharge capacity is insufficient. Then, the process ends.

  Next, with reference to FIG. 13, the details of the process of step S22 shown in FIG. 12 will be described. When the process shown in FIG. 13 is started, the following steps are executed.

In step S40, CPU 10a, when the current _{I n} which is observed in step S12 it is determined whether the _{300A ≧ I n> 200A, 300A} ≧ I n> 200A is satisfied (step S40: YES) step in The process proceeds to S41, and otherwise (step S40: NO), the process proceeds to step S42. For example, in FIG. 8, when it falls within the range of τ3, 300A ≧ I _n > 0A is established, and the process proceeds to step S41.

In step S41, the CPU 10a updates the values of the parameters Ra1 _n , Rb1 _n , Ca1 _n , and Cb1 _{n with} the parameter values included in the state vector X _{n + 1} | _{n + 1} obtained in step S16. Thereby, the parameters Ra1 _n , Rb1 _n , Ca1 _n , and Cb1 _n become new values.

In step S42, CPU 10a, when the current _{I n} which is observed in step S12 it is determined whether the _{200A ≧ I n> 100A, 200A} ≧ I n> 100A is satisfied (step S42: YES) step in The process proceeds to S43, and otherwise (step S42: NO), the process proceeds to step S44. For example, in FIG. 8, when it falls within the range of τ4, since 200A ≧ I _n > 100A is established, the process proceeds to step S43.

In step S43, the CPU 10a updates the values of the parameters Ra2 _n , Rb2 _n , Ca2 _n , and Cb2 _{n with} the parameter values included in the state vector X _{n + 1} | _{n + 1} obtained in step S16. As a result, the parameters Ra2 _n , Rb2 _n , Ca2 _n , and Cb2 _n become new values.

In step S44, CPU 10a, when the current _{I n} which is observed in step S12 it is determined whether the 100A ≧ _I n> 0A, is 100A ≧ _I n> 0A satisfied (step S44: YES) step in The process proceeds to S45, and in other cases (step S44: NO), the process returns (returns) to the original process. For example, in FIG. 8, when it falls within the range of τ5, since 100A ≧ I _n > 0A is established, the process proceeds to step S45.

In step S45, the CPU 10a updates the values of the parameters Ra3 _n , Rb3 _n , Ca3 _n , and Cb3 _{n with} the parameter values included in the state vector X _{n + 1} | _{n + 1} obtained in step S16. As a result, the parameters Ra3 _n , Rb3 _n , Ca3 _n , and Cb3 _n become new values. Then, the process returns to the original process.

  According to the above processing, the parameters of the simulation model 30 of the lead storage battery 13 can be adaptively learned using the extended Kalman filter. In addition, when adaptive learning is performed, the target parameter is selected according to the value of the discharge current, and adaptive learning is performed for the selected parameter. By performing adaptive learning, it is possible to perform learning accurately and efficiently.

(D) Modified Embodiments Each of the above-described embodiments is an example, and there are various modified embodiments other than this. For example, in the above embodiment, adaptive learning is performed using an extended Kalman filter, but other methods may be used. Specifically, adaptive learning may be performed using a neural network model, or adaptive learning may be performed using a genetic algorithm model.

  In the above embodiment, parameters are not updated for parameters other than the learning target. However, parameters other than the learning target may be excluded from the state vector calculation targets. This will be specifically described. For example, in the case of a general extended Kalman filter, the following prediction and update are repeatedly executed.

prediction

update

更新し、それ以外のパラメータについては更新しないようにしたが、例えば、式（１７）を更新の対象となるパラメータのみについて計算するようにしてもよい。具体的には、例えば、Ｒａ１，Ｒｂ１，Ｃａ１，Ｃｂ１が適応学習の対象となっている場合には、式（１７）を計算する際には、これらのパラメータのみについて計算を行い、それ以外のパラメータについては計算を行わないようにすることも可能である。このような方法によれば、対象外のパラメータに関する計算を省略することで、計算速度を向上させることができる。また、ＣＰＵ１０ａとして性能が低いものを使用することができるので、装置の製造コストを低減することが可能になる。

Although it is updated and other parameters are not updated, for example, equation (17) may be calculated only for the parameters to be updated. Specifically, for example, when Ra1, Rb1, Ca1, and Cb1 are targets of adaptive learning, when calculating equation (17), only these parameters are calculated, It is also possible not to calculate the parameters. According to such a method, it is possible to improve the calculation speed by omitting the calculation related to the non-target parameter. Further, since the CPU 10a having a low performance can be used, the manufacturing cost of the device can be reduced.

  In addition, the calculation is not performed only for the parameters to be updated, but the parameters to be updated can be weighted more heavily than other parameters. For example, when updating each parameter, it is also possible to set a weighting factor W, give a large weight to the parameter to be updated, and give a small weight to the other parameters. . As a specific example, W = 0.8 may be set for parameters to be updated, and W = 0.2 may be set for other parameters. In addition, the number mentioned above is an example to the last, and it cannot be overemphasized that other numbers may be sufficient.

  In the above embodiment, two impedances Z1 and Z2 are assumed. However, for example, it may be assumed that only one of them exists. Moreover, although impedance Z1, Z2 is set as the structure by three RC parallel units, respectively, the number of RC parallel units may differ in each of impedance Z1, Z2. Moreover, it may be 1 or 2 instead of 3, or 4 or more. In this case, the number of divisions τ3 to τ5 may be changed according to the number. Further, the division of τ3 to τ5 shown in FIGS. 8 and 9 is merely an example, and it goes without saying that other division methods may be used.

  In the above embodiment, the discharge current is caused to flow by the discharge circuit 14. However, since the current flowing to the cell motor 16 is temporally changed, the current flowing to the cell motor 16 without using the discharge circuit 14 is predetermined. You may make it sample and use with the period of. That is, the discharge circuit 16 may be excluded, the current and voltage during the rotation of the cell motor 16 may be sampled at a predetermined cycle (for example, a cycle of 10 mS), and the above-described processing may be executed.

In the above embodiment, the concentrations C _N and C _P of the cathode and anode of the electrolytic solution are selected as learning targets. However, V0 may be simply selected as the learning target. Alternatively, in addition to the cathode and anode concentrations C _N and C _P , the separator electrolyte concentration C _S may be added as a learning target, and V 0 may be obtained based on these C _N , C _P and C _S.

  In the above embodiment, the inductance L is ignored, but the inductance L may be added to the learning target. In that case, for example, application learning may be performed in a portion where the current change is large (for example, when the cell motor 16 is started).

  In the above embodiment, adaptive learning is executed based on the current flowing through the cell motor 16, but adaptive learning may be executed based on the current flowing through other loads. For example, when there is a motor for driving the vehicle (for example, a motor of a hybrid vehicle), adaptive learning may be performed based on the current flowing through the motor. In addition, for example, in the case of a photovoltaic power storage battery used in a general household, it is possible to use a current flowing in various loads used in the household (for example, an air conditioner in which a large current flows at startup). .

  Moreover, although the above embodiment demonstrated lead acid battery as an example, it is possible to apply this invention also about batteries other than this (for example, nickel cadmium battery etc.). In this case, the simulation model 30 may be changed according to the type of battery.

  In the above embodiment, the control device 10 is configured by a CPU, a ROM, a RAM, and the like, but may be configured by, for example, a DSP (Digital Signal Processor).

DESCRIPTION OF SYMBOLS 1 Battery internal state estimation apparatus 10 Control part 10a CPU (selection means, adaptive learning means)
10b ROM
10c RAM (storage means)
10d I / F (part of detection means)
11 Voltage detector (part of detection means)
12 Current detector (part of detection means)
13 Lead acid battery (battery)
14 Discharge circuit 15 Alternator 16 Cell motor (electric motor)
17 Load 30 Simulation model 31 Extended Kalman filter R Internal resistance Z1, Z2 Impedance

Claims (7)

In the battery internal state estimation device that estimates the internal state of the battery based on the battery simulation model,
Storage means for storing a plurality of parameters included in the simulation model;
Detecting means for detecting a discharge current flowing from the battery to the load;
Selection means for selecting a parameter to be subjected to adaptive learning according to the value of the discharge current detected by the detection means;
Adaptive learning means for performing adaptive learning on the parameter selected by the selection means;
A battery internal state estimation device comprising: The load has at least an electric motor for starting a prime mover,
The detection means detects a current when the prime mover is started by the electric motor,
The selection means selects a parameter according to a current value flowing through the electric motor,
The adaptive learning means performs adaptive learning on the parameter selected by the selection means;
The battery internal state estimation apparatus according to claim 1. The simulation model has at least a constant phase element (CPE) as an equivalent circuit of the anode and cathode of the battery, and the constant phase element includes a plurality of RC parallel units including resistors and capacitors connected in parallel. It is represented by an equivalent circuit connected in series, and the element values of the resistor and the capacitor constituting each RC parallel unit are the parameters.
The selection means selects a predetermined RC parallel unit predetermined according to the value of the discharge current,
The adaptive learning means adaptively learns the resistance and capacitor element values selected by the selection means;
The battery internal state estimation apparatus according to claim 1 or 2, wherein As a component of the simulation model, it has an internal resistance of the battery, the resistance value of the internal resistance is the parameter,
The selection means selects the internal resistance when detecting a peak current that flows when the electric motor starts.
The battery internal state estimation apparatus according to claim 2 or 3, wherein Having a voltage source as a component of the simulation model, and having a concentration value of the electrolyte in the battery as a parameter relating to the voltage of the voltage source;
The selection means, when the discharge current is 0 or in the vicinity of 0, selects the concentration of the electrolyte as an object of adaptive learning;
The battery internal state estimation device according to any one of claims 1 to 4, wherein The parameters constitute an extended Kalman filter state vector;
The adaptive learning means performs the adaptive learning on the state vector;
The battery internal state estimation apparatus according to any one of claims 1 to 5, wherein In the battery internal state estimation method for estimating the internal state of the battery based on the battery simulation model,
A storing step of storing a plurality of parameters included in the simulation model;
A detection step of detecting a discharge current flowing from the battery to the load;
A selection step of selecting a parameter to be subjected to adaptive learning according to the magnitude of the discharge current detected in the detection step;
An adaptive learning step for performing adaptive learning on the parameter selected in the selection step;
A battery internal state estimation method comprising:

Images