State estimation model learning apparatus, method, and program

Abstract

To learn a state estimation model for accurately estimating an internal state of a measuring object by reducing labor for impedance measurement.SOLUTION: A state estimation model learning device includes: an input part which accepts a plurality of pieces of learning data, which is a combination of impedance at each frequency measured by an impedance method for each of a plurality of measuring objects and an internal state of the measuring objects determined in advance; a learning data generation part which generates learning data for a virtual measuring object by internally dividing at least two pieces of learning data of the received learning data; and a learning part which learns a state estimation model for outputting the internal state of the measuring object using the impedance at each frequency as input, based on the received plurality of pieces of learning data and the generated learning data.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The disclosed technology relates to a state estimation model learning apparatus, method, and program. [Background technology]

[0002] Conventionally, a method using machine learning has been known to determine the degree of degradation of batteries such as lithium-ion batteries, by examining the relationship between impedance obtained by measurement using the AC impedance method and the degradation characteristics of the battery (Patent Document 1).

[0003] Furthermore, there is a known technique for predicting the battery's cycle life using machine learning based on the voltage of the battery's early charge-discharge cycles (Patent Document 2). [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2016-090346 [Patent Document 2] Japanese Patent Publication No. 2019-113524 [Overview of the project] [Problems that the invention aims to solve]

[0005] However, neither of the above-mentioned Patent Documents 1 and 2 takes into account individual differences in batteries, and if there is variation in the characteristics of the batteries used for machine learning, accurate state estimation is not possible.

[0006] The disclosed technology was developed in view of the above points, and aims to provide a state estimation model learning device, method, and program that can reduce the effort required for impedance measurement and learn a state estimation model that accurately estimates the internal state of an object being measured. [Means for solving the problem]

[0007] A first aspect of this disclosure is a state estimation model learning device, comprising: an input unit that receives a plurality of learning data sets, each of which is a combination of impedances for each frequency measured by the impedance method and the internal state of the object to be measured, which has been determined in advance; a learning data generation unit that generates learning data for a virtual object to be measured by dividing at least two of the plurality of learning data sets received; and a learning unit that learns a state estimation model that takes the impedances for each frequency as input and outputs the internal state of the object to be measured, based on the plurality of learning data sets received and the generated learning data.

[0008] A second aspect of this disclosure is a state estimation model learning method, wherein an input unit receives a plurality of learning data sets, each of which is a combination of impedances for each of a plurality of objects to be measured measured by an impedance method and the internal state of the object to be measured determined in advance; a learning data generation unit generates learning data for a virtual object to be measured by dividing at least two of the received plurality of learning data sets into internal learning data sets; and a learning unit learns a state estimation model that takes the impedances for each frequency as input and outputs the internal state of the object to be measured, based on the plurality of received learning data sets and the generated learning data.

[0009] A third aspect of this disclosure is a state estimation model learning program, which receives a plurality of learning data sets, each of which is a combination of impedance for each of a plurality of objects to be measured measured by the impedance method and the internal state of the object to be measured determined in advance; generates learning data for a virtual object to be measured by dividing at least two of the received plurality of learning data sets into internal units; and causes a computer to learn a state estimation model that takes the impedance for each frequency as input and outputs the internal state of the object to be measured, based on the plurality of received learning data sets and the generated learning data. [Effects of the Invention]

[0010] According to the disclosed technology, it is possible to learn a state estimation model that reduces the labor of impedance measurement and accurately estimates the internal state of the measurement object.

Brief Description of the Drawings

[0011] [Figure 1] It is a schematic block diagram of an example of a computer that functions as the state estimation model learning device and the state estimation device of the present embodiment. [Figure 2] It is a diagram for explaining a measurement method by the impedance method. [Figure 3] It is a block diagram showing the configuration of the state estimation model learning device of the present embodiment. [Figure 4] It is a diagram for explaining a method of generating learning data for a virtual battery. [Figure 5] It is a diagram showing an example of variations in the internal ratio. [Figure 6] It is a diagram showing an example of a state estimation model. [Figure 7] It is a diagram showing an example of an activation function. [Figure 8] It is a block diagram showing the configuration of the state estimation device of the present embodiment. [Figure 9] It is a flowchart showing the state estimation model learning processing routine of the state estimation model learning device of the present embodiment. [Figure 10] It is a flowchart showing the state estimation processing routine of the state estimation device of the present embodiment. [Figure 11] It is a graph showing an example of internally dividing the impedance characteristics of three lithium-ion batteries. [Figure 12] It is a graph showing experimental results.

Modes for Carrying Out the Invention

[0012] Hereinafter, an example of an embodiment of the disclosed technology will be described while referring to the drawings. In each of the drawings, the same or equivalent components and parts are given the same reference numerals. Also, the dimensional ratios in the drawings are exaggerated for the convenience of explanation and may be different from the actual ratios.

[0013] <Overview of this embodiment> It is conceivable to use the AC impedance method of lithium-ion batteries and machine learning to estimate the internal states such as the temperature, charge amount, and degree of deterioration of batteries such as lithium-ion batteries. In this case, learning is performed in advance using the impedance characteristics of a specific battery and the internal state to be estimated. Then, the internal state is estimated from the impedance of a battery with an unknown internal state. However, since there are individual differences in batteries, when using a battery different from the one used for learning, the estimation of the internal state may not be performed accurately. If learning is performed using a large number of batteries, due to the property called generalization of machine learning, the individual differences can be absorbed and the state can be estimated with high accuracy, but the measurement is time-consuming.

[0014] Therefore, in this embodiment, impedance characteristics obtained by internally dividing the impedance characteristics, which are the impedances for each frequency of two or more batteries, are used to generate the impedance characteristics of a plurality of virtual batteries by changing the internal division ratio. Machine learning is performed using this impedance characteristic as learning data, and the internal state is estimated from the impedance characteristic of a new battery with an unknown internal state.

[0015] Specifically, a large number of impedance characteristics obtained by internally dividing the impedance characteristics of two or more batteries are generated by changing the internal division ratio to obtain the impedance characteristics of virtual batteries. By using the impedance characteristics of virtual batteries, the impedance characteristics of a large number of batteries are no longer required, and the measurement effort is significantly reduced.

[0016] Furthermore, when estimating the state of a battery using machine learning based on impedance characteristics obtained by the AC impedance method, even if there are differences between the battery used to acquire the training data necessary for machine learning and the battery being estimated due to individual battery variations, the battery state can still be accurately estimated.

[0017] <Configuration of the state estimation model learning device according to this embodiment> Figure 1 is a block diagram showing the hardware configuration of the state estimation model learning device 10 of this embodiment.

[0018] As shown in Figure 1, the state estimation model learning device 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, storage 14, an input unit 15, a display unit 16, and a communication interface (I / F) 17. Each component is connected to the others via a bus 19 so as to be able to communicate with each other.

[0019] The CPU 11 is a central processing unit that executes various programs and controls various parts. Specifically, the CPU 11 reads a program from the ROM 12 or storage 14 and executes the program using the RAM 13 as a working area. The CPU 11 controls each of the above components and performs various calculations according to the program stored in the ROM 12 or storage 14. In this embodiment, the ROM 12 or storage 14 stores a state estimation model learning program for learning a state estimation model. The state estimation model learning program may be a single program or a group of programs consisting of multiple programs or modules.

[0020] ROM12 stores various programs and data. RAM13 temporarily stores programs or data as a working area. Storage14 consists of an HDD (Hard Disk Drive) or SSD (Solid State Drive) and stores various programs, including the operating system, and various data.

[0021] The input unit 15 includes a pointing device such as a mouse and a keyboard, and is used for various types of input.

[0022] The input unit 15 receives learning data as input, which is a combination of the impedance for each of the multiple batteries measured by the impedance method at each frequency and the internal state of the battery (temperature and charge state) determined in advance. Specifically, as shown in Figure 2, a battery 34 whose charge state has been determined in advance is placed in a constant temperature bath 32, and the impedance for each frequency is measured by the impedance method using an impedance meter 30, a switch 36, and a charge / discharge circuit 38. At this time, the temperature of the battery 34 is controlled by the constant temperature bath 32, and the impedance of the battery 34 at each frequency is measured using the impedance meter 30. The combination of the measured impedance for each frequency, the temperature of the constant temperature bath 32, and the charge state of the battery 34 determined in advance is used as learning data. This learning data is prepared for each of the multiple batteries, and the input unit 15 receives the learning data for each of the multiple batteries. These multiple batteries are batteries of the same type.

[0023] The display unit 16 is, for example, a liquid crystal display and displays various information. The display unit 16 may also function as an input unit 15 by employing a touch panel system.

[0024] The communication interface 17 is an interface for communicating with other devices, and standards such as Ethernet (registered trademark), FDDI (Fiber Distributed Data Interface), and Wi-Fi (registered trademark) are used.

[0025] Next, the functional configuration of the state estimation model learning device 10 will be described. Figure 3 is a block diagram showing an example of the functional configuration of the state estimation model learning device 10.

[0026] Functionally, the state estimation model learning device 10 comprises a learning database (DB) 20, a learning data generation unit 22, a learning unit 24, and a model storage unit 26, as shown in Figure 3.

[0027] The learning database 20 stores learning data for each of the multiple batteries that have been input.

[0028] The learning data generation unit 22 generates learning data for the virtual battery by dividing at least two of the received learning data into internal units. Specifically, the learning data generation unit 22 generates learning data for the virtual battery by dividing at least two learning data whose internal states correspond to each other.

[0029] For example, as shown in Figure 4, the impedance characteristic Z represents the impedance at each frequency of three training data sets whose internal states (temperature and charge state) are approximately identical. A , Z B , Z C The point is divided internally according to the following equation to obtain the impedance characteristic Z of the virtual battery X. X This generates the impedance characteristics obtained by plotting the real part Z' and imaginary part Z'' of the impedance measured for each frequency. As an example, impedance characteristics are generated by plotting the real part Z' and imaginary part Z'' of the impedance measured for each of 26 frequencies.

[0030] JPEG0007900816000001.jpg2189

[0031] However, α, β, and γ are the impedance characteristics Z of the three training data. A , Z B , Z C This is the internal division ratio when dividing the value internally. As shown in Figure 5, various internal division ratios are set to generate multiple training data sets for the virtual battery.

[0032] In FIG. 5, an example using 81 types of internal ratios of α:β:γ = 1:1:9, ···, 9:9:1 is shown. In the example of this embodiment, the impedance characteristics Z A , Z B , Z C of three learning data with substantially the same internal state (temperature and charge state) are prepared in 8 sets, and by using 81 types of internal ratios, learning data for 81×8 = 648 virtual batteries are generated.

[0033] Based on the learning data for a plurality of received batteries and the learning data for a plurality of generated virtual batteries, the learning unit 24 learns a state estimation model that takes the impedance for each frequency as an input and outputs the internal state of the battery.

[0034] Specifically, the state estimation model, which is a neural network, is learned so as to take the impedance for each frequency of the learning data as an input and output the internal state (temperature and charge state) of the battery of the learning data.

[0035] More specifically, since there are a real part and an imaginary part for the impedance for each of the 26 measured frequencies, as shown in FIG. 6, in the state estimation model, the input data is 52 - dimensional. The state estimation model has two hidden layers, and the number of nodes in the first and second layers are 32 and 16 respectively. As the activation function, the ReLU function shown in FIG. 7 is used. From the output layer, two - dimensional values representing the temperature SoT (°C) and the charge state SoC (%) are output.

[0036] The learned state estimation model is stored in the model storage unit 26.

[0037] <Configuration of the state estimation device according to this embodiment> FIG. 1 above is a block diagram showing the hardware configuration of the state estimation device 50 of this embodiment.

[0038] As shown in Figure 1 above, the state estimation device 50 has the same configuration as the state estimation model learning device 10, and the ROM 12 or storage 14 stores a state estimation program for estimating the internal state of the battery.

[0039] The input unit 15 accepts frequency-specific impedances measured by the impedance method for the battery to be estimated as input. For example, the input unit 15 accepts a 52-dimensional value consisting of the real and imaginary parts of the impedances for 26 frequencies measured by the impedance method for the battery to be estimated as input.

[0040] Next, the functional configuration of the state estimation device 50 will be described. Figure 8 is a block diagram showing an example of the functional configuration of the state estimation device 50.

[0041] Functionally, the state estimation device 50 comprises a model storage unit 60 and a state estimation unit 62, as shown in Figure 8.

[0042] The model storage unit 60 stores the state estimation model learned by the state estimation model learning device 10.

[0043] The state estimation unit 62 inputs the impedance for each frequency received into the state estimation model and estimates the internal state of the battery to be estimated.

[0044] Specifically, a 52-dimensional value consisting of the real and imaginary parts of the impedance measured for each of the 26 frequencies using the impedance method for the battery under estimation is input into the state estimation model, and two-dimensional values ​​representing the internal state of the battery, namely temperature (SoT) (°C) and charge state (SoC) (%), are estimated.

[0045] <Operation of the state estimation model learning device according to this embodiment> Next, the operation of the state estimation model learning device 10 according to this embodiment will be described.

[0046] Figure 9 is a flowchart showing the flow of the state estimation model learning process by the state estimation model learning device 10. The state estimation model learning process is performed when the CPU 11 reads the state estimation model learning program from the ROM 12 or storage 14, loads it into the RAM 13, and executes it. The state estimation model learning device 10 is also input with learning data, which consists of a combination of the impedance for each of the multiple batteries measured by the impedance method for each frequency and the internal state of the battery (temperature, charge state) that has been determined in advance, and is stored in the learning database 20.

[0047] In step S100, the CPU 11, acting as a learning data generation unit 22, generates learning data for the virtual battery by dividing at least two of the received learning data into smaller parts. Specifically, the learning data generation unit 22 generates learning data for the virtual battery by dividing at least two learning data whose internal states correspond to each other. At this time, multiple sets of learning data for the virtual battery are generated by changing the division ratio.

[0048] In step S102, the CPU 11, acting as a learning unit 24, learns a state estimation model that takes impedance for each frequency as input and outputs the internal state of a battery, based on the learning data for the multiple batteries received and the learning data for the multiple virtual batteries that have been generated. The CPU 11 stores the learned state estimation model in the model storage unit 26 and terminates the state estimation model learning process.

[0049] <Operation of the state estimation device according to this embodiment> Next, the operation of the state estimation device 50 according to this embodiment will be described.

[0050] Figure 10 is a flowchart showing the flow of the state estimation process by the state estimation device 50. The state estimation process is performed when the CPU 11 reads the state estimation program from the ROM 12 or storage 14, loads it into the RAM 13, and executes it. The state estimation device 50 is also input with the impedance for each frequency measured by the impedance method for the battery to be estimated.

[0051] In step S110, the CPU 11 obtains the frequency-specific impedance measured by the impedance method for the input battery to be estimated.

[0052] In step S112, the CPU 11, acting as a state estimation unit 62, inputs the received impedance for each frequency into the state estimation model and estimates the internal state of the battery to be estimated.

[0053] In step S114, the CPU 11 outputs the estimation result from step S112 via the display unit 16 and terminates the state estimation process.

[0054] <Examples> An example of estimating the battery temperature using the state estimation model learning device 10 and state estimation device 50 described above will be explained.

[0055] Here, four types of lithium-ion batteries, A, B, C, and D, were prepared. Using three types of lithium-ion batteries, A, B, and C, the impedance characteristics of a virtual battery were generated, and a state estimation model was trained.

[0056] Figure 11(A) shows the impedance characteristics of three types of lithium-ion batteries A, B, and C at different frequencies. Figure 11(B) shows the impedance characteristics ZX of a virtual battery X generated by internally dividing the impedance characteristics of the three types of lithium-ion batteries A, B, and C in a 1:1:9 ratio.

[0057] Figure 12(A) shows the estimated temperature of lithium-ion battery D using its impedance characteristics (vertical axis) and the actual temperature (horizontal axis). It can be seen that the estimated value (vertical axis) is close to the actual temperature (horizontal axis).

[0058] Furthermore, Figure 12(B) shows the result when the impedance characteristics of a virtual battery were generated using three types of lithium-ion batteries B, C, and D, and the state estimation model was trained to estimate the temperature of lithium-ion battery A. Figure 12(B) is a graph showing the result of estimating the temperature of lithium-ion battery A using the impedance characteristics of lithium-ion battery A (vertical axis) and the actual temperature (horizontal axis). It can be seen that the estimated value (vertical axis) is close to the actual temperature (horizontal axis). Thus, it was found that the temperature estimation was successful even when the impedance characteristics of a battery not used in training were used to estimate the temperature.

[0059] As described above, the state estimation model learning device according to this embodiment receives multiple learning data sets, which are combinations of impedances for each frequency measured by the impedance method and the internal state of a battery determined in advance. By dividing at least two of these learning data sets internally, it generates learning data for a virtual battery and learns a state estimation model. This reduces the effort required for impedance measurement and enables the learning of a state estimation model that accurately estimates the internal state of a battery.

[0060] Furthermore, the state estimation device according to this embodiment estimates the internal state of the battery using a state estimation model learned by a state estimation model learning device. This allows for accurate estimation of the internal state of the battery.

[0061] <Variation> It should be noted that the present invention is not limited to the embodiments described above, and various modifications and applications are possible without departing from the spirit of the invention.

[0062] For example, the explanation used the case where the object being measured is a battery, but it is not limited to this. For example, the internal state of electronic devices other than batteries, or living organisms such as vegetables and fruits, may also be estimated as the object being measured.

[0063] Furthermore, while we have described the estimation of battery temperature and charge state, the method is not limited to these. The battery degradation state may also be estimated. Alternatively, the battery temperature, charge state, and degradation state may be estimated, or any combination of the battery temperature, charge state, and degradation state may be estimated.

[0064] Furthermore, although the explanation described an example where the state estimation model learning device and the state estimation device are configured as separate devices, the explanation is not limited to this, and the state estimation model learning device and the state estimation device may be configured as a single device.

[0065] Furthermore, the various processes that the CPU reads and executes in each of the above embodiments may be executed by various processors other than the CPU. Examples of such processors include GPUs (Graphics Processing Units), FPGAs (Field-Programmable Gate Arrays), and other PLDs (Programmable Logic Devices) whose circuit configurations can be changed after manufacturing, as well as dedicated electrical circuits that are processors with circuit configurations specifically designed to execute particular processes, such as ASICs (Application Specific Integrated Circuits). In addition, the state estimation model learning process and the state estimation process may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (for example, multiple FPGAs, and a combination of a CPU and an FPGA). More specifically, the hardware structure of these various processors is an electrical circuit that combines circuit elements such as semiconductor elements.

[0066] Furthermore, although the above embodiments describe a configuration in which the state estimation model learning program and the state estimation program are pre-stored (installed) in the storage 14, the invention is not limited thereto. The program may be provided in a form stored on a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), DVD-ROM (Digital Versatile Disk Read Only Memory), and USB (Universal Serial Bus) memory. Alternatively, the program may be provided in a form that can be downloaded from an external device via a network. [Explanation of symbols]

[0067] 10 State Estimation Model Learning Device 11 CPU 14 Storage 15 Input section 16 Display section 20 Learning Databases 22. Training Data Generation Unit 24 Learning Department 26 Model Memory Unit 30 Impedance Meter 32 Constant temperature bath 34 Batteries 50 State Estimation Device 60 Model Memory Unit 62 State Estimation Unit

Claims

1. An input unit that receives multiple learning data sets, each consisting of a combination of the impedance measured for each of several objects by the impedance method at each frequency and the internal state of the object determined in advance. A learning data generation unit generates learning data for a virtual measurement target by dividing at least three of the multiple learning data received, A learning unit learns a state estimation model that takes the impedance for each frequency as input and outputs the internal state of the object to be measured, based on the multiple learning data received and the generated learning data. Includes, The learning data generation unit generates, for each of a predetermined plurality of internal division ratios for internally dividing the impedance of at least three learning data, by internally dividing the impedance at each frequency using the internal division ratio for at least three learning data corresponding to the internal state, thereby generating a combination of the obtained impedance at each frequency and the internal state as learning data for the virtual object to be measured, for each internal division ratio. The object to be measured is a battery, The aforementioned internal state is temperature and degradation state, and the state estimation model learning device.

2. The state estimation model learning device according to claim 1, wherein the plurality of objects to be measured are of the same type.

3. The input unit receives multiple learning data sets, each consisting of a combination of the impedance measured for each of the multiple objects to be measured using the impedance method at each frequency, and the internal state of the object to be measured, which has been determined in advance. The learning data generation unit generates learning data for a virtual measurement target by dividing at least two of the received learning data into individual data. The learning unit learns a state estimation model that takes the impedance for each frequency as input and outputs the internal state of the object to be measured, based on the multiple learning data received and the generated learning data. This includes, The learning data generation unit generates, for each of a predetermined plurality of internal division ratios for internally dividing the impedance of at least three learning data, by internally dividing the impedance at each frequency using the internal division ratio for at least three learning data corresponding to the internal state, thereby generating a combination of the obtained impedance at each frequency and the internal state as learning data for the virtual object to be measured, for each internal division ratio. The object to be measured is a battery, The aforementioned internal state is temperature and degradation state, and this is a state estimation model learning method.

4. The system accepts multiple learning data sets, each consisting of a combination of the impedance measured for each of several objects using the impedance method at each frequency and the internal state of the object determined in advance. By dividing at least two of the received training data into individual data, training data for the virtual object to be measured is generated. Based on the multiple training data received and the generated training data, a state estimation model is trained that takes the impedance for each frequency as input and outputs the internal state of the object to be measured. This includes, In generating the aforementioned training data, for each of a predetermined plurality of internal division ratios for internally dividing the impedance of at least three training data, the impedance for each frequency is internally divided using the internal division ratio for at least three training data corresponding to the internal state, thereby generating training data for the virtual object to be measured, where the combination of the obtained impedance for each frequency and the internal state is used for each internal division ratio. The object to be measured is a battery, The aforementioned internal state is temperature and degradation state. A state estimation model learning program that instructs a computer to perform this task.

Images