Manufacturing method, generating apparatus, estimation apparatus

Abstract

A generation device (1) comprises: an acquisition unit (121) that acquires identification information, one or more pieces of which are imparted to a component of a certain level from among a plurality of components, and which are imparted such that the type and quantity of a lower-level component can be identified; a generation unit (122) that, by learning operation data for each of at least one piece of identification information, generates a trained model corresponding to each piece of identification information for estimating the state of a battery; and an output unit (123) that outputs the generated trained model.

Description

Technical Field

[0006] , ,

[0001] The present disclosure relates to a technique for generating a learned model of a battery including a plurality of components configured in a hierarchical structure.

Background Art

[0002] Patent Document 1 discloses a technique for obtaining measurement values of a battery, determining an operation regime of the battery from the obtained measurement values, selecting a non-linear regression model corresponding to the determined operation regime, inputting the measurement values into the selected non-linear regression model, and estimating a state of charge of the battery.

[0003] However, since the non-linear regression model of Patent Document 1 is not generated in consideration of differences in the types and numbers of components constituting the battery, further improvement is required to improve the estimation accuracy of the state of charge.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

[0005] The present disclosure has been made in view of such problems, and an object thereof is to provide a technique for generating a learned model capable of accurately estimating the state of a battery.

[0006] A manufacturing method in one aspect of the present disclosure is a method for manufacturing a trained model of a battery comprising a plurality of hierarchically configured components in a generating device, wherein the processor of the generating device acquires one or more identification pieces of information assigned to a component of a certain hierarchy among the plurality of components, acquires battery operation data corresponding to each identification piece of information, and learns the acquired operation data for each of the one or more identification pieces of information to generate a trained model corresponding to each identification piece of information that estimates the state of the battery, and outputs the generated trained model.

[0007] According to this disclosure, it is possible to generate a trained model that can estimate the state of a battery with high accuracy. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows an example of the overall configuration of the information processing system in Embodiment 1 of this disclosure. [Figure 2] This block diagram shows an example of the configuration of the generating apparatus in Embodiment 1 of the present disclosure. [Figure 3] This is a block diagram showing an example of the configuration of the application device in Embodiment 1 of the present disclosure. [Figure 4] This block diagram shows an example of the battery configuration in Embodiment 1 of the present disclosure. [Figure 5] This figure shows an example of the components of a battery. [Figure 6] This flowchart shows an example of the process by which the assigning device generates identification information in Embodiment 1 of this disclosure. [Figure 7] This is a flowchart that continues from Figure 6. [Figure 8] This flowchart shows an example of the process by which the generation device generates a trained model in Embodiment 1 of this disclosure. [Figure 9] This figure shows an example of the process when the battery uploads operational data in Embodiment 1 of this disclosure. [Figure 10]This flowchart shows an example of the process when the granting device downloads a trained model to the battery in Embodiment 1 of this disclosure. [Figure 11] This flowchart shows an example of the processing in the utilization phase of the trained model in Embodiment 1 of this disclosure. [Figure 12] This block diagram shows an example of the configuration of the generating apparatus in Embodiment 2 of the present disclosure. [Figure 13] This flowchart shows a first example of the processing of the generating apparatus in Embodiment 2 of this disclosure. [Figure 14] This flowchart shows a second example of the processing of the generating apparatus in Embodiment 2 of the present disclosure. [Modes for carrying out the invention]

[0009] (Knowledge forming the basis of this disclosure) In recent years, research has been conducted on generating trained models that can accurately estimate the state of a battery, such as the State of Charge (SOC), from operational data collected from various types of batteries manufactured by various manufacturers, and then training the model on this collected data. Batteries are hierarchically composed of multiple components, such as cells, blocks containing cells, modules containing blocks, and battery packs containing modules. If the number and types of components constituting the battery differ, the characteristics of the battery will also change accordingly. Therefore, if a trained model is generated using operational data as training data without considering these differences in characteristics, it may not be possible to obtain a trained model that can accurately estimate the state of the battery.

[0010] In the aforementioned Patent Document 1, a nonlinear regression model is selected according to the operating regime determined from battery measurements. However, the operating regime is defined by at least one parameter selected from among the battery's charge state, the fact that the battery is charging or discharging, the charging or discharging rate, the ambient or internal temperature, the average value of the battery's voltage measurements, and the battery's impedance or measured values ​​of the battery's healthy state. Therefore, the nonlinear regression model in Patent Document 1 does not take into account the differences in the number and types of components that make up the battery, and cannot estimate the state of the battery with high accuracy.

[0011] This disclosure is made to solve these problems and aims to provide a technology for generating a trained model that can estimate the state of a battery with high accuracy.

[0012] A manufacturing method in one aspect of the present disclosure is a method for manufacturing a trained model of a battery comprising a plurality of hierarchically configured components in a generating device, wherein the processor of the generating device acquires one or more identification pieces of information assigned to a component of a certain hierarchy among the plurality of components, acquires battery operation data corresponding to each identification piece of information, and learns the acquired operation data for each of the one or more identification pieces of information to generate a trained model corresponding to each identification piece of information that estimates the state of the battery, and outputs the generated trained model.

[0013] In this configuration, one or more identification pieces of information are acquired for a component at a certain hierarchical level among multiple components. Battery operation data corresponding to the acquired identification pieces is then acquired, and a trained model corresponding to each identification piece is generated by training the acquired operation data for each identification piece. This makes it possible to generate a trained model for each identification piece that takes into account the differences in the number and types of components that make up the battery. As a result, it is possible to generate a trained model that can estimate the state of the battery with high accuracy.

[0014] In the above manufacturing method, each identification information may be provided with the types and numbers of components in the lower hierarchy being identifiable.

[0015] According to this configuration, each identification information is provided with the types and numbers of components in the lower hierarchy being identifiable. Therefore, it is possible to obtain operation data with the same types and numbers as the components in the lower hierarchy. As a result, it is possible to generate a learned model for each identification information using operation data of batteries with the same types and numbers of components in the lower hierarchy.

[0016] In the above manufacturing method, based on the operation data corresponding to one or more first identification information in the first hierarchy, a first learned model corresponding to each first identification information is generated, the learning cost or learning error of the first learned model is calculated, and when the calculated learning cost or the learning error is greater than a threshold value, based on the operation data corresponding to one or more second identification information in a second hierarchy different from the first hierarchy, a second learned model corresponding to each second identification information may be generated.

[0017] According to this configuration, when the learning cost or learning error of the first learned model generated based on the operation data corresponding to each first identification information in the first hierarchy is greater than the threshold value, a learned model corresponding to each second identification information in the second hierarchy is generated. As a result, it is possible to search for a learned model in another hierarchy that can make the learning cost or learning error lower.

[0018] In the above manufacturing method, the plurality of components may include a first component and a second component having a different hierarchy from the first component, the one or more first identification information is information for identifying the first component, and the one or more second identification information may be information for identifying the second component.

[0019] According to this configuration, it is possible to search for a learned model in another hierarchy that can make the learning cost or learning error lower.

[0020] In the above manufacturing method, if the training cost or training error of the first trained model is less than or equal to the threshold, the first trained model may be determined as the trained model to be trained.

[0021] With this configuration, if the training cost or training error of the first trained model is below a threshold, the first trained model is selected as the trained model to be generated. Therefore, thereafter, the first trained model can be generated and updated using operational data corresponding to the first identification information, and trained models with low training costs or high accuracy can be generated.

[0022] In the above manufacturing method, if the accuracy of the second trained model is lower than the reference accuracy, a third trained model corresponding to each third identification information may be generated based on the operational data corresponding to one or more third identification information of a third layer different from the first and second layers.

[0023] In this configuration, if the accuracy of the second pre-trained model generated based on the operational data corresponding to each second identification piece is lower than the reference accuracy, a third pre-trained model corresponding to each third identification piece is generated. This allows for the exploration of pre-trained models at different hierarchical levels that can lower the training cost or training error.

[0024] In the above manufacturing method, if the accuracy of the second trained model is higher than the reference accuracy, the second trained model may be selected as the trained model to be trained.

[0025] With this configuration, if the accuracy of the second trained model is higher than the reference accuracy, the second trained model is selected as the trained model to be generated. Therefore, the second trained model can be updated using the operational data corresponding to each second identification piece, and a more accurate trained model can be generated.

[0026] In the above manufacturing method, the learning cost may be calculated based on at least one of the following: the number of trained models generated, the amount of operational data used to generate the trained models, and the processing load of the processor when generating the trained models.

[0027] With this configuration, the training cost can be accurately estimated because it is calculated based on at least one of the following: the number of trained models generated, the amount of operational data used to generate the trained models, and the processing load during the generation of the trained models.

[0028] In the above manufacturing method, the one or more identification pieces of information include one or more type identification pieces of information that identify each component by type.

[0029] This configuration makes it easy to acquire operational data for each type of component.

[0030] In the above manufacturing method, each identification piece of information includes one or more individual identification pieces of information that identify each component piece individually.

[0031] This configuration makes it easy to obtain individual operational data for each component.

[0032] In the above manufacturing method, the plurality of components may include a cell, a block containing the cell, a module containing the block, and a battery pack containing the module.

[0033] This configuration allows for the generation of a trained model using operational data corresponding to one of the following hierarchical levels: cell, block, module, or battery pack.

[0034] A generating apparatus in another aspect of the present disclosure is a generating apparatus for generating a trained model of a battery including a plurality of hierarchically configured components, the generating apparatus comprising a processor, the processor which acquires one or more identification pieces of information assigned to a component of a certain hierarchical level among the plurality of components, acquires operational data corresponding to each identification piece of information, and generates the trained model corresponding to each identification piece of information for estimating the state of the battery by learning the acquired operational data for each of the one or more identification pieces of information, and outputs the generated trained model.

[0035] This configuration makes it possible to provide a production apparatus that can obtain the effects and advantages of the above-described manufacturing method.

[0036] An estimation device in yet another aspect of the present disclosure is an estimation device for estimating the state of a battery comprising a plurality of hierarchically configured components, wherein the estimation device comprises a processor, the processor performs the following processes: acquiring battery operation data, inputting the operation data into a trained model to estimate the state of the battery, and outputting state information indicating the estimated state, the trained model is a model generated by learning the operation data corresponding to one or more identification pieces assigned to a component of a certain hierarchical level among the plurality of components, for each of the one or more identification pieces.

[0037] This configuration allows for the estimation of the battery state using a trained model generated with operational data corresponding to one or more identifiers assigned to a certain hierarchical level. This enables high-accuracy estimation of the battery state using a trained model that takes into account differences in the number and types of components that make up the battery.

[0038] A method for assigning identification information in an assigning device for assigning identification information to a battery comprising a first component and a second component including the first component, the method being used to assign identification information, the device acquires first configuration information indicating a configuration according to the type of the first component, generates first identification information that identifies the first component by type using the first configuration information, outputs the first identification information, acquires second configuration information indicating a configuration according to the type of the second component, the second configuration information includes the first identification information and the number of the first component, generates second identification information that identifies the second component by type using the second configuration information, and outputs the second identification information.

[0039] According to this configuration, first identification information is generated from the first configuration information to identify the first component by type. Second identification information is generated from the second configuration information to identify the second component by type. Here, the second configuration information includes the first identification information and the number of the first component.

[0040] Therefore, it is possible to easily obtain operating data for batteries with the same type and number of lower-level components (first-order components). This makes it possible to generate a trained model that takes into account the differences in the types and numbers of components that make up the battery.

[0041] In the above identification information assignment method, the battery further comprises a third component including the second component, and further acquires third configuration information indicating a configuration according to the type of the third component, the third configuration information including the second identification information, the number of the second component, and connection information indicating the connection configuration of the second component, and further generates third identification information that identifies the third component by type using the third configuration information, and outputs the third identification information.

[0042] According to this configuration, third identification information is generated from the third configuration information to identify the third components by type. Here, the third configuration information includes the second identification information, the number of second components, and connection information for the second components.

[0043] Therefore, it is possible to easily obtain operating data for batteries that have the same type and number of lower-level components, as well as the same connection configuration. This makes it possible to generate a trained model that takes into account the differences in the type, number, and connection configuration of the components that make up the battery.

[0044] In the above method for assigning identification information, a fourth configuration information indicating the configuration of each individual third component may be acquired, the fourth configuration information may include the third identification information and the serial number of the third component, and further, a fourth identification information that identifies each individual third component may be generated using the fourth configuration information, and the fourth identification information may be output.

[0045] With this configuration, since the fourth identification information is generated using the third identification information and the fourth configuration information including the serial number of the third component, individual operational data for the third component can be easily obtained.

[0046] In the above method for assigning identification information, a fifth configuration information indicating the configuration of each individual second component may be acquired, the fifth configuration information including the fourth identification information and the manufacturing number of the second component, and a fifth identification information that identifies each individual second component may be generated using the fifth configuration information, and the fifth identification information may be output.

[0047] With this configuration, since the fifth identification information is generated using the fourth identification information and the fourth configuration information including the serial number of the second component, individual operational data of the second component can be easily obtained.

[0048] An assigning device in yet another aspect of the present disclosure is an assigning device for assigning identification information to a battery comprising a first component and a second component including the first component, wherein the assigning device comprises a processor, the processor acquires first configuration information indicating a configuration according to the type of the first component, generates first identification information for identifying the first component by type using the first configuration information, outputs the first identification information, acquires second configuration information indicating a configuration according to the type of the second component, the second configuration information includes the first identification information and the number of the first component, generates second identification information for identifying the second component by type using the second configuration information, and outputs the second identification information.

[0049] This configuration provides an information-granting device that can achieve the same effects as the above-described method of granting identification information.

[0050] This disclosure can also be implemented as a program that causes a computer to execute the characteristic configurations included in such a manufacturing method, estimation apparatus, identification information assignment method, and assignment apparatus, or as a system operated by this program. It goes without saying that such a computer program can be distributed via a computer-readable non-temporary recording medium such as a CD-ROM or via a communication network such as the Internet.

[0051] The embodiments described below are all specific examples of this disclosure. The numerical values, shapes, components, steps, and order of steps shown in the following embodiments are examples only and are not intended to limit this disclosure. Furthermore, among the components in the following embodiments, those not described in the independent claim representing the highest-level concept will be described as optional components. In addition, the contents of each embodiment can be combined.

[0052] (Embodiment 1) Figure 1 shows an example of the overall configuration of an information processing system in Embodiment 1 of this disclosure. The information processing system includes a generation device 1, an input device 2, one or more batteries 3, and an input terminal 4. The generation device 1 to the input terminal 4 are each connected to each other via a network NT. The network NT consists of a wide-area communication network, such as an internet communication network or a mobile phone communication network.

[0053] Generation device 1 is, for example, a cloud server composed of one or more computers. Generation device 1 generates a trained model for estimating the state of batteries using battery operating data and sends the generated trained model to assignment device 2. Assignment device 2 is, for example, a cloud server composed of one or more computers. Assignment device 2 assigns identification information to each battery 3. Assignment device 2 acquires operating data from battery 3 and stores it in memory. Assignment device 2 sends the stored operating data to generation device 1 as needed. Assignment device 2 sends the trained model received from generation device 1 to battery 3.

[0054] Battery 3 is installed in a vehicle, for example. Battery 3 includes multiple components arranged hierarchically. The vehicle is, for example, an electric car, an electric bicycle, and an electric kick scooter. Input terminal 4 receives input of configuration information for battery 3 and transmits the received configuration information to assignment device 2.

[0055] Figure 2 is a block diagram showing an example of the configuration of the generation device 1 in Embodiment 1 of the present disclosure. The generation device 1 includes a communication circuit 11, a processor 12, and a memory 13. The communication circuit 11 connects the generation device 1 to the network NT. The communication circuit 11 transmits the trained model to the assignment device 2. The communication circuit 11 receives operational data from the assignment device 2.

[0056] The processor 12 is composed of, for example, a central processing unit. The processor 12 includes an acquisition unit 121, a generation unit 122, and an output unit 123. The acquisition unit 121 acquires one or more identification pieces of information assigned to components at a certain level among the multiple components that make up the battery 3. Each identification piece of information is assigned in a way that allows identification of the type and number of components at a lower level. The acquisition unit 121 also acquires operational data corresponding to each identification piece of information from the assignment device 2.

[0057] Figure 5 shows an example of the components of battery 3. Battery 3 is composed of multiple components, including a battery pack 51, a module 52, a block 53, and a cell 54. Block 53 contains multiple cells 54 connected in parallel. Module 52 contains multiple blocks 53 connected in series. Battery pack 51 is composed of multiple modules 52 connected in series and in parallel. In the example in Figure 5, battery pack 51 contains four modules 52. Battery pack 51 is composed of two groups of modules connected in parallel, each containing two modules 52 connected in series. Note that block 53 may consist of one cell 54. Also, module 52 may consist of one block 53. Also, battery pack 51 may consist of one module 52. Thus, battery 3 is hierarchically composed of the battery pack 51, module 52, block 53, and cell 54 in this order.

[0058] The identification information of battery 3 includes type identification information (hereinafter referred to as "type ID") that identifies each component by type, and individual identification information (hereinafter referred to as "individual ID") that identifies each component individually. The type ID includes cell type ID, block type ID, module type ID, and pack type ID. The cell type ID is identification information that identifies cell 54 by type. The block type ID is identification information that identifies block 53 by type. The module type ID is identification information that identifies module 52 by type. The pack type ID is identification information that identifies battery pack 51 by type.

[0059] The individual ID includes the cell individual ID, block individual ID, module individual ID, and pack individual ID. The cell individual ID is identification information that identifies each cell 54 individually. The block individual ID is identification information that identifies each block 53 individually. The module individual ID is identification information that identifies each module 52 individually. The pack individual ID is identification information that identifies each battery pack 51 individually.

[0060] Refer back to Figure 2. The generation unit 122 generates a trained model corresponding to each piece of identification information for estimating the state of the battery 3 by learning the operational data acquired by the acquisition unit 121 for each piece of identification information. The battery state to be estimated is, for example, SOC (state of charge), SOH (state of health), or a sign of failure. A sign of failure is, for example, the remaining lifespan.

[0061] The operational data includes, for example, at least one of the current, voltage, and temperature of battery 3. Furthermore, the operational data includes the state of battery 3. This state includes, for example, at least one of the SOC, SOH, and whether or not there are signs of failure. Furthermore, the operational data includes a timestamp indicating the date and time the operational data was generated and identification information for battery 3. This identification information is, for example, the pack individual ID described later.

[0062] The output unit 123 outputs the trained model generated by the generation unit 122. For example, the output unit 123 transmits the generated trained model to the assignment device 2 using the communication circuit 11.

[0063] Memory 13 consists of non-volatile rewritable storage devices such as solid-state drives and hard disk drives. Memory 13 stores operational data acquired from the data acquisition device 2.

[0064] Figure 3 is a block diagram showing an example of the configuration of the granting device 2 in Embodiment 1 of the present disclosure. The granting device 2 includes a communication circuit 21, a processor 22, and a memory 23. The communication circuit 21 connects the granting device 2 to the network NT. The communication circuit 21 transmits operational data to the generation device 1 and a trained model to the battery 3. The communication circuit 21 receives a trained model from the generation device 1 and operational data from the battery 3.

[0065] The processor 22 is composed of, for example, a central processing unit. The processor 22 includes an acquisition unit 221, a generation unit 222, and an output unit 223. Here, one of the components of the battery 3 is called the first component. The components that include the first component are called the second component. The components that include the second component are called the third component.

[0066] The acquisition unit 221 acquires first configuration information, which indicates the configuration according to the type of first component, from the input terminal 4 using the communication circuit 21. The first configuration information includes, for example, characteristic data showing the relationship between the remaining capacity and voltage of the first component. Specifically, the characteristic data is data showing the relationship between the SOC and OCV.

[0067] The generation unit 222 generates a first type ID that identifies the first component by type based on the first configuration information acquired by the acquisition unit 221. The generation unit 222 also assigns the first type ID to the first component by associating the generated first type ID with the first configuration information and storing it in the identification information database 231.

[0068] The acquisition unit 221 acquires second configuration information from the input terminal 4 via the communication circuit 21, which indicates the configuration according to the type of second component. The second configuration information includes the first type ID and the number of first components that make up the second component.

[0069] The generation unit 222 generates a second type ID that identifies the second component by type based on the second configuration information acquired by the acquisition unit 221. The generation unit 222 also assigns the second type ID to the second component by associating the generated second type ID with the second configuration information and storing it in the identification information database 231.

[0070] The acquisition unit 221 acquires third configuration information, which indicates the configuration according to the type of third component, from the input terminal 4 via the communication circuit 21. The third configuration information includes the second type ID, the number of second components, and connection information indicating the connection configuration of the second components.

[0071] The generation unit 222 generates a third type ID that identifies the third component by type based on the third component information acquired by the acquisition unit 221. The generation unit 222 also assigns the third type ID to the third component by associating the generated third type ID with the third component information and storing it in the identification information database 231.

[0072] The acquisition unit 221 acquires fourth configuration information, which indicates the configuration specific to each individual third component, from the input terminal 4 via the communication circuit 21. The fourth configuration information includes the third type ID and the serial number of the third component.

[0073] The generation unit 222 generates a first individual ID (fourth identification information) that identifies each third component individually based on the fourth configuration information acquired by the acquisition unit 221. The generation unit 222 also assigns the first individual ID to the third component by associating the generated first individual ID with the fourth configuration information and storing it in the identification information database 231.

[0074] The acquisition unit 221 acquires fifth configuration information, which indicates the configuration specific to each individual second component, from the input terminal 4 via the communication circuit 21. The fifth configuration information includes fourth identification information and the serial number of the second component.

[0075] The generation unit 222 generates a second individual ID (fifth identification information) that identifies each second component individually based on the fifth configuration information acquired by the acquisition unit 221. The generation unit 222 also assigns the second individual ID to the second component by associating the generated second individual ID with the fifth configuration information and storing it in the identification information database 231.

[0076] The output unit 223 transmits the identification information (first type ID to third type ID, as well as the first individual ID and second individual ID) generated by the generation unit 222 to the input terminal 4 using the communication circuit 21.

[0077] Memory 23 consists of a non-volatile, rewritable storage device such as a solid-state drive or a hard disk drive. Memory 23 stores the identification information database 231 and the operation database 232.

[0078] The identification information database 231 is a database for managing the identification information assigned to the battery 3. The identification information database 231 stores the identification information assigned to each component in association with the configuration information of each component. Specifically, the identification information database 231 stores the first type ID in association with the first configuration information, the second type ID in association with the second configuration information, the third type ID in association with the third configuration information, the first individual ID in association with the fourth configuration information, and the second individual ID in association with the fifth configuration information.

[0079] The operational database 232 stores operational data for each battery 3. Specifically, the operational database 232 stores the pack type ID, timestamp, current, voltage, temperature, and status (SOC, SOH, and signs of failure, etc.) in association with each other.

[0080] Figure 4 is a block diagram showing an example of the configuration of battery 3 in Embodiment 1 of this disclosure. Battery 3 includes a battery management device 31 (an example of an estimation device) and a battery unit 32.

[0081] The battery management device 31 is a device that manages the battery 3, such as estimating the state of the battery 3 (SOC, SOH, and signs of failure). The battery management device 31 includes a sensor 33, a processor 34, a memory 35, and a communication circuit 36. The sensor 33 includes a current sensor, a voltage sensor, and a temperature sensor. The current sensor measures the current flowing through the battery 3. The voltage sensor measures the voltage of the battery 3. The temperature sensor detects the temperature of the battery 3.

[0082] The processor 34 is composed of, for example, a central processing unit and includes a generation unit 341, an acquisition unit 342, an estimation unit 343, and an output unit 344. The generation unit 341 generates battery operation data at a predetermined sampling period. For example, the generation unit 341 can generate operation data by acquiring current, voltage, and temperature from the sensor 33 and the state of the battery 3 estimated by the estimation unit 343 from the estimation unit 343.

[0083] The acquisition unit 342 acquires the operational data generated by the generation unit 341. The acquisition unit 342 also acquires the trained model transmitted from the assignment device 2 via the vehicle communication device 5 and the communication circuit 36 ​​and stores it in the memory 35. The acquisition unit 342 also acquires the pack individual ID, described later, that the assignment device 2 assigns to the battery 3 via the vehicle communication device 5 and the communication circuit 36 ​​and stores it in the memory 35.

[0084] The estimation unit 343 inputs the operational data acquired by the acquisition unit 342 into a trained model to estimate the state of the battery. Here, the estimation unit 343 only needs to input information other than the state of the battery 3 (for example, current, voltage, and temperature) from the operational data into the trained model. If the estimation unit 343 does not have a trained model, it can calculate the state of the battery 3 by inputting the current, voltage, and temperature into a predetermined calculation formula.

[0085] The output unit 344 inputs the operational data generated by the generation unit 341 to the communication circuit 36 ​​in order to transmit it to the application device 2 at a predetermined sampling period. As a result, the operational data is transmitted to the application device 2 via the communication circuit 36 ​​and the vehicle communication device 5, and the operational data is accumulated in the application device 2. The output unit 344 inputs the status information indicating the state of the battery 3 estimated by the estimation unit 343 to the communication circuit 36 ​​in order to display it on a display device (not shown). The display device may be a display installed in the vehicle in which the battery 3 is installed, or it may be a user terminal held by the user (driver) of the vehicle.

[0086] Memory 35 is composed of a rewritable semiconductor memory, such as flash memory, and stores the trained model. Memory 35 stores the BMS type ID, which is identification information pre-assigned to the battery management device 31. Memory 35 also stores the pack individual ID assigned by the assignment device 2.

[0087] The communication circuit 36 ​​is a communication circuit that connects the battery 3 to an in-vehicle network 38 such as CAN (Controller Area Network). The communication circuit 36 ​​is connected to the vehicle communication device 5 via the in-vehicle network 38.

[0088] The battery unit 32 consists of the battery pack 51 shown in Figure 5.

[0089] The vehicle communication device 5 is a communication circuit provided in a vehicle equipped with a battery 3. The vehicle communication device 5 connects the battery 3 to the network NT. For example, the vehicle communication device 5 may connect to the network NT by connecting to a user terminal owned by the vehicle user via a wireless communication channel such as BLE. Alternatively, the vehicle communication device 5 may connect to the network NT without going through a user terminal.

[0090] The above describes the configuration of the information processing system. Next, the operation of the information processing system will be explained. Figure 6 is a flowchart showing an example of the process by which the assignment device 2 generates identification information in Embodiment 1 of this disclosure.

[0091] First, let's explain the configuration information A1-A5 and B1-B4 used in this flowchart.

[0092] Configuration information A1 is configuration information for each type of cell 54. Specifically, configuration information A1 includes characteristic data for cell 54. Furthermore, configuration information A1 may also include manufacturer information, model number, and capacity of cell 54. Configuration information A1 is an example of first configuration information.

[0093] Configuration information A2 is the configuration information for each type of block 53. Configuration information A2 includes the cell type ID (first type ID) and the number of cells 54 that make up block 53. Furthermore, configuration information A2 may also include the manufacturer information and model number of block 53. Configuration information A2 is an example of second configuration information.

[0094] Configuration information A3 is configuration information for each type of module 52. Configuration information A3 includes the block type ID (second type ID) and number of blocks 53 that make up module 52. Furthermore, configuration information A3 may also include the number of probes, manufacturer information, and model number of blocks 53 that make up module 52. Configuration information A3 is an example of second configuration information. The number of probes is the number of probes of the temperature sensor attached to block 53.

[0095] Configuration information A4 is configuration information for each type of battery management device 31. Configuration information A4 includes manufacturer information, model number, number of sensors 33, type information of the sensors 33, and BMS type ID for the battery management device 31. Configuration information A4 does not include configuration information for components of other levels. The BMS type ID is identification information that identifies the battery management device 31 by type.

[0096] Configuration information A5 is configuration information for each type of battery pack 51. Configuration information A5 includes the module type ID (second type ID), number, and connection information of the modules 52 that make up the battery pack 51. The connection information is information indicating the connection configuration of the modules 52. The connection information includes, for example, the number of modules 52 connected in series and the number of modules 52 connected in parallel. Furthermore, configuration information A5 may also include manufacturer information, model number, rated capacity, rated discharge output, rated charge output, initial pack rating FCC, initial pack rating DC resistance, operating log measurement recording cycle, operating log update cycle, idle log update cycle, learned model update confirmation cycle, battery pack installed parts information, and the BMS type ID of the battery management device 31. Configuration information A5 is an example of third configuration information.

[0097] Configuration information B1 is configuration information that indicates the configuration specific to each individual battery management device 31. Configuration information B1 includes the BMS type ID, manufacturing lot number, manufacturing number, and manufacturing date and time of the battery management device 31.

[0098] Configuration information B2 is configuration information that indicates the configuration specific to each individual battery pack 51. Configuration information B2 includes the pack type ID (third type ID) and serial number of the battery pack 51. Configuration information B2 may further include the BMS individual ID of the battery management device 31, the manufacturing lot number of the battery pack 51, the manufacturing date and time of the battery pack 51, the update confirmation cycle of the learned model, the initial measured FCC of the battery pack 51, and the initial measured DC resistance of the battery pack 51. Configuration information B2 is an example of fourth configuration information.

[0099] Configuration information B3 is configuration information that indicates the configuration specific to each individual module 52. Configuration information B3 includes the module type ID (second type ID) and serial number of module 52. Furthermore, configuration information B3 may also include the pack individual ID (first individual ID) and pack type ID (third type ID) of the battery pack 51, and the manufacturing lot number and manufacturing date and time of module 52. Configuration information B3 is an example of fifth configuration information.

[0100] Configuration information B4 is configuration information that indicates the configuration of an individual block 53. Configuration information B4 includes the block type ID (second type ID) and serial number of block 53. Furthermore, configuration information B4 may also include the module individual ID (second individual ID), module type ID, manufacturing lot number, and manufacturing date and time. Configuration information B4 is an example of the fifth configuration information.

[0101] Based on the above, Figure 6 will now be explained. In step S101, the input terminal 4 accepts the input of configuration information A1. Here, the user of the input terminal 4 should input the configuration information A1 while referring to the specifications table of the battery 3, according to the input form displayed on the input terminal 4. The same applies to the input of the following configuration information. This user is, for example, a user who manages the operation of a vehicle equipped with battery 3.

[0102] In step S102, the input terminal 4 transmits configuration information A1 to the assigning device 2. In step S103, the assigning device 2 receives the configuration information A1.

[0103] In step S104, the assignment device 2 generates cell type IDs that identify cells 54 by type based on the configuration information A1. The cell type IDs are stored in the identification information database 231 in association with the configuration information A1.

[0104] In step S105, the assignment device 2 transmits the cell type ID to the input terminal 4. If the cell type ID for the corresponding cell type 54 has already been generated, the assignment device 2 should transmit the generated cell type ID to the input terminal 4. The same applies to the type ID and individual ID described below.

[0105] In step S106, input terminal 4 receives the cell type ID. In step S107, input terminal 4 accepts the input of configuration information A2. In step S108, input terminal 4 transmits the configuration information A2 to the assignment device 2.

[0106] In step S109, the assigning device 2 receives configuration information A2. In step S110, the assigning device 2 generates block type IDs that identify blocks 53 by type based on the configuration information A2. The block type IDs are stored in the identification information database 231 in association with the configuration information A2.

[0107] In step S111, the assignment device 2 transmits the cell type ID to the input terminal 4. In step S112, the input terminal 4 receives the block type ID.

[0108] In step S113, the input terminal 4 receives the configuration information A3. In step S114, the input terminal 4 transmits the configuration information A3 to the assignment device 2. In step S115, the assignment device 2 receives the configuration information A3.

[0109] In step S116, the assignment device 2 generates a module type ID that identifies the module 52 by type based on the configuration information A3. In step S117, the assignment device 2 transmits the module type ID to the input terminal 4. In step S118, the input terminal 4 receives the module type ID.

[0110] In step S119, the input terminal 4 receives the configuration information A4. In step S120, the input terminal 4 transmits the configuration information A4 to the assignment device 2. In step S121, the assignment device 2 receives the configuration information A4.

[0111] In step S122, the assigning device 2 generates a BMS type ID that identifies the battery management device 31 by type based on the configuration information A4. In step S123, the assigning device 2 transmits the BMS type ID to the input terminal 4. In step S124, the input terminal 4 receives the BMS type ID.

[0112] In step S125, the input terminal 4 receives the configuration information A5. In step S126, the input terminal 4 transmits the configuration information A5 to the assignment device 2. In step S127, the assignment device 2 receives the configuration information A5.

[0113] In step S128, the assigning device 2 generates a pack type ID that identifies the battery pack 51 by type based on the configuration information A5. In step S129, the assigning device 2 transmits the pack type ID to the input terminal 4. In step S130, the input terminal 4 receives the pack type ID.

[0114] Figure 7 is a flowchart that continues from Figure 6. In step S201, input terminal 4 receives the configuration information B1. In step S202, input terminal 4 transmits the configuration information B1 to the assignment device 2. In step S203, assignment device 2 receives the configuration information B1.

[0115] In step S204, the assignment device 2 generates a BMS individual ID that identifies each battery management device 31 individually based on the configuration information B1. In step S205, the assignment device 2 transmits the BMS individual ID to the input terminal 4. In step S206, the input terminal 4 receives the BMS individual ID.

[0116] In step S207, the input terminal 4 receives the configuration information B2. In step S208, the input terminal 4 transmits the configuration information B2 to the assignment device 2. In step S209, the assignment device 2 receives the configuration information B2.

[0117] In step S210, the assigning device 2 generates a pack individual ID that identifies each battery pack 51 individually based on the configuration information B2. In step S211, the assigning device 2 transmits the pack individual ID to the input terminal 4. In step S212, the input terminal 4 receives the pack individual ID.

[0118] In step S213, the input terminal 4 receives the configuration information B3. In step S214, the input terminal 4 transmits the configuration information B3 to the assignment device 2. In step S215, the assignment device 2 receives the configuration information B3.

[0119] In step S216, the assignment device 2 generates a module individual ID that identifies each module 52 individually based on the configuration information B3. In step S217, the assignment device 2 transmits the module individual ID to the input terminal 4. In step S218, the input terminal 4 receives the module individual ID.

[0120] In step S219, the input terminal 4 receives the configuration information B4. In step S220, the input terminal 4 transmits the configuration information B4 to the assignment device 2. In step S221, the assignment device 2 receives the configuration information B4.

[0121] In step S222, the assignment device 2 generates a block individual ID that identifies each block 53 individually based on the configuration information B4. In step S223, the assignment device 2 transmits the block individual ID to the input terminal 4. In step S224, the input terminal 4 receives the block individual ID.

[0122] Thus, the type ID is generated sequentially from the lower-level components, and the individual ID is generated sequentially from the higher-level components.

[0123] Next, the generation of a trained model will be explained. Figure 8 is a flowchart showing an example of the process by which the generation device 1 generates a trained model in Embodiment 1 of this disclosure. In step S301, the generation device 1 generates information requesting the acquisition of operational data. The acquisition request information includes information specifying the hierarchy of the identification information (hierarchy specification information) and information specifying the range of the timestamp of the operational data (period specification information). The hierarchy of the identification information refers to cell type ID, block type ID, module type ID, pack type ID, pack individual ID, module individual ID, and block individual ID. The hierarchy specification information is information that specifies one of these hierarchies. The hierarchy of the identification information is assumed to be, in descending order from highest to lowest, cell type ID, block type ID, module type ID, pack type ID, pack individual ID, module individual ID, and block individual ID. This order is determined depending on the amount of operational data that can be collected. However, this is merely a prediction, and it is not guaranteed that the amount of operational data will actually increase in this order.

[0124] In step S302, the generating device 1 transmits the acquisition request information to the assigning device 2. In step S303, the assigning device 2 receives the acquisition request information.

[0125] In step S304, the assignment device 2 reads operational data from the operational database 232 that corresponds to the hierarchy indicated by the hierarchy specification information and the period indicated by the period specification information included in the acquisition request information. For example, if a hierarchy of pack type ID is specified, the assignment device 2 reads operational data corresponding to that hierarchy from the operational database 232 for each pack type ID. "Reading operational data for each pack type ID" means, for example, if there are M pack type IDs, reading M sets of operational data in association with the pack type IDs.

[0126] In step S305, the data assignment device 2 transmits the operation data to the generation device 1. In step S306, the generation device 1 receives the operation data.

[0127] In step S307, the generation device 1 adds feature quantities and training data to the operational data. Feature quantities include, for example, current, voltage, and temperature. Training data includes, for example, the state of the battery 3.

[0128] In step S308, the generation device 1 generates trained models by training the operational data for each piece of identification information. For example, if the operational data received in step S306 consists of M data groups, M trained models are generated by training each of the M data groups individually.

[0129] In step S309, the generator 1 calculates the accuracy of the trained model. Here, the accuracy is a value that decreases as the training error increases. The training error is the error of the trained model's estimate relative to the true value, and for example, the root mean square error (RMSE) or mean square error (MSE) is used. If M trained models have been generated, the average accuracy of the M trained models is calculated.

[0130] In step S310, the generation device 1 determines whether the accuracy has improved compared to the accuracy of the previously generated trained model.

[0131] If the accuracy has improved (YES in step S310), the generator 1 decides to adopt the trained model generated this time as the training target (step S311). On the other hand, if the accuracy is lower than that of the previously generated trained model (NO in step S310), the generator 1 terminates the process. In this case, the previously generated trained model will be adopted as the training target. The previously generated trained model is a trained model generated using operational data with a different hierarchy of identification information compared to the trained model generated this time.

[0132] In step S312, the generation device 1 transmits the trained model it has generated to the assignment device 2. In step S313, the assignment device 2 receives the trained model. In step S314, the assignment device 2 saves the trained model to a predetermined storage location in memory 13. As a result, a trained model with a higher accuracy level is saved to the assignment device.

[0133] Figure 9 shows an example of the process when the battery 3 uploads operational data in Embodiment 1 of this disclosure. In step S401, the battery 3 reads the pack individual ID from the memory 35 and sends an authentication request including the pack individual ID to the granting device 2.

[0134] In step S411, the assigning device 2 receives an authentication request. In step S412, the assigning device 2 determines whether the pack individual ID is valid or not. Here, the assigning device 2 determines that the pack individual ID is valid if it is registered in the identification information database 231. On the other hand, if the pack individual ID is not registered in the identification information database 231, it determines that the pack individual ID is invalid and sends a response to the battery 3 indicating that authentication failed.

[0135] Step S413 generates token information and sends it to battery 3. Token information is generated upon successful authentication and is necessary for battery 3 to communicate with the granting device 2.

[0136] In step S402, battery 3 receives token information. In step S403, battery 3 generates operational data, generates a packet containing the operational data and token information, and transmits the generated packet to the granting device 2.

[0137] In step S414, the granting device 2 receives a packet. In step S415, the granting device 2 verifies the packet. For example, if the received packet does not contain token information, the granting device 2 discards the received packet. Alternatively, if the received packet does not contain a pack individual ID, the granting device 2 determines it is a format error and discards the received packet.

[0138] In step S416, the operational data contained in the packets that have been determined to be appropriate through verification is stored in the operational database 232. Thereafter, the battery 3 generates operational data at a predetermined sampling rate and transmits packets containing the operational data, token information, and pack individual ID to the assignment device 2.

[0139] Figure 10 is a flowchart showing an example of the process when the application device 2 downloads a trained model to the battery 3 in Embodiment 1 of this disclosure. In step S501, the battery 3 sends a confirmation request to the application device 2. The confirmation request includes the pack individual ID of the battery 3 itself, the version information of the trained model currently installed in the battery 3, and the name of the trained model. The name of the trained model is, for example, the type ID or individual ID corresponding to the operational data used for training the trained model. The confirmation request is sent periodically according to the update confirmation cycle of the trained model included in the configuration information A5.

[0140] In step S511, the assigning device 2 receives a confirmation request. In step S512, it checks whether a successor version of the trained model exists. For example, if the version information included in the confirmation request is not the latest version information corresponding to the name of the trained model, the assigning device 2 determines that a successor version of the trained model exists. On the other hand, if the version information included in the confirmation request is the latest version information corresponding to the name of the trained model, the assigning device 2 determines that there is no successor version of the trained model.

[0141] If it is determined that a successor version of the trained model exists (YES in step S512), the assigning device 2 transmits storage location information indicating the storage location of the successor version of the trained model to the battery 3 (step S513). The storage location information can be, for example, a URL (Uniform Resource Locator). On the other hand, if it is determined that there is no successor version of the trained model (NO in step S512), the process ends.

[0142] In step S502, battery 3 receives storage location information. In step S503, battery 3 sends a request to send the trained model to the storage location indicated by the storage location information. In step S514, the assigning device 2 receives the request to send. In step S515, the assigning device 2 sends the successor version of the trained model to battery 3.

[0143] In step S504, battery 3 receives the trained model. In step S505, battery 3 installs the trained model. As a result, the latest trained model is installed on battery 3.

[0144] Next, the utilization phase of the trained model will be described. Figure 11 is a flowchart showing an example of the processing in the utilization phase of the trained model in Embodiment 1 of this disclosure. In step S601, the acquisition unit 342 of the battery management device 31 acquires the operating data generated by the generation unit 341. In step S602, the estimation unit 343 of the battery management device 31 estimates the state of the battery 3 by inputting the operating data into the trained model. For example, feature quantities (current, voltage, and temperature) that constitute the operating data are input into the trained model.

[0145] In step S603, the output unit 344 of the battery management device 31 transmits status information indicating the estimated state to the display device. In step S611, the display device receives the status information. In step S612, the display device displays the status information. This presents the user with the status of the battery 3 (e.g., SOC, SOH, or signs of failure).

[0146] As described above, according to the information processing system of Embodiment 1, the type ID is assigned in a way that allows identification of the type and number of components in the lower hierarchy. Therefore, it is possible to acquire operational data that is the same type and number as the components in the lower hierarchy. As a result, it is possible to generate a trained model using operational data of batteries 3 that have the same type and number of components in the lower hierarchy, and to generate a trained model that takes into account the differences in the types and numbers of components that make up battery 3. As a result, it is possible to generate a trained model that can estimate the state of battery 3 with high accuracy.

[0147] Furthermore, because the battery state is estimated using such a pre-trained model, the battery state can be estimated with high accuracy.

[0148] Furthermore, since second identification information is generated that identifies the second component by type using second component information including first identification information and quantity of the first component, it is possible to easily obtain operating data for batteries 3 that are the same type and quantity as the lower-level component, the first component.

[0149] (Embodiment 2) Embodiment 2 explores a hierarchy of identification information with a low training cost or training error for a trained model. In Embodiment 2, the same reference numerals are used for components identical to those in Embodiment 1, and their descriptions are omitted. Figure 12 is a block diagram showing an example of the configuration of the generation device 1A in Embodiment 2 of this disclosure.

[0150] For example, if a trained model is generated for each pack type ID, a number of trained models will be generated corresponding to the number of battery pack types 51, resulting in a huge number of trained models. As a result, the resources of the generation device 1A will be strained, and the training cost will increase.

[0151] Therefore, in Embodiment 2, a hierarchy of identification information with a low learning cost is explored. As mentioned above, the hierarchy of identification information includes cell type ID, block type ID, module type ID, pack type ID, pack individual ID, module individual ID, and block individual ID, with the hierarchy increasing in this order.

[0152] As the hierarchy of identification information increases, the scope of the operational data expands, and the amount of operational data used as training data increases, which is an advantage. Also, as the hierarchy of identification information increases, the number of trained models generated decreases, which is an advantage as it reduces the burden on resources. On the other hand, as the hierarchy of identification information increases, the specialization decreases, making it more difficult to reflect the characteristics of individual batteries 3, and the accuracy of the trained models decreases, which is a disadvantage.

[0153] The lower the hierarchy of the identification information, the greater the specialization, allowing for the generation of a trained model that better reflects the characteristics of each individual battery 3. Furthermore, if the amount of operational data used for training is sufficient, it is possible to generate a highly accurate trained model.

[0154] On the other hand, a disadvantage of lowering the hierarchy of identification information is that the scope of operational data becomes narrower, and the amount of operational data used for training decreases. Another disadvantage is that the number of trained models to be trained increases, leading to increased resource burden. The configuration of Embodiment 2 will be described in detail below. In the following description, the hierarchy of the first, second, and third components of the battery 3 will not be considered. Furthermore, the identification information for identifying the first component will be called the first identification information, the identification information for identifying the second component will be called the second identification information, and the identification information for identifying the third component will be called the third identification information.

[0155] The processor 12A of the generation device 1A includes an acquisition unit 121, a generation unit 122A, and an output unit 123A. The generation unit 122A generates a first trained model corresponding to each first identification information based on the operation data corresponding to each first identification information. The generation unit 122A calculates the training cost or training error of the calculated first trained model. For example, the training cost or training error of the first trained model is the average value of the training cost or the average value of the training error of each first trained model.

[0156] If the training cost or training error of the first trained model is greater than a threshold, the generation unit 122A generates a second trained model corresponding to each second identification information based on the operational data corresponding to each second identification information.

[0157] The generation unit 122A determines the second trained model as the trained model to be trained if the accuracy of the second trained model is higher than the reference accuracy. The accuracy of the second trained model is determined by the average of the accuracy of each second trained model.

[0158] On the other hand, the generation unit 122A determines the first trained model as the trained model to be trained if the training cost or training error of the first trained model is below a threshold. If the accuracy of the second trained model is lower than the reference accuracy, the generation unit 122A generates a third trained model corresponding to each third identification information based on the operational data corresponding to each third identification information.

[0159] The training cost is calculated based on at least one of the following: the number of trained models generated, the amount of operational data used to generate the trained models, and the processing load of processor 12A when the trained models were generated. For example, the training cost is calculated using the following formula.

[0160] Learning cost = A1 · Number of models + A2 · Amount of data + A3 · Processing load The processing load is measured, for example, by the cumulative time when the processor load rate exceeds a baseline load rate.

[0161] Figure 13 is a flowchart showing a first example of the processing of the generation device in Embodiment 2 of this disclosure. This flowchart is executed periodically, for example. In the first example, the learning cost is evaluated to determine the hierarchy of the identification information. In this flowchart, SOC is used as the state of battery 3 estimated by the trained model.

[0162] In step S701, the generation unit 122A generates a trained model M(n) corresponding to the identification information of hierarchy n. For example, the default hierarchy n is the hierarchy of pack type IDs. Specifically, the generation unit 122A obtains operational data corresponding to each piece of identification information belonging to hierarchy n from the operational database 232, and generates a trained model M(n) corresponding to each piece of identification information by training the obtained operational data for each piece of identification information.

[0163] In step S702, the generation unit 122A calculates the training cost of the trained model M(n). In step S703, the generation unit 122A determines whether the training cost of the trained model M(n) is greater than a threshold. Here, the generation unit 122A can compare the average training cost of each trained model M(n) with the threshold.

[0164] If the learning cost is greater than the threshold (YES in step S703), the generation unit 122A generates a trained model M(n-1) corresponding to the identification information of hierarchy n-1, which is one level higher than hierarchy n (step S704). For example, if a trained model M(n) corresponding to the pack type ID has been generated in step S701, then a trained model M(n-1) corresponding to the module type ID, which is one level higher than the pack type ID, will be generated. Specifically, the generation unit 122A obtains operational data corresponding to each identification information belonging to hierarchy n-1 from the operational database 232, and trains the obtained operational data for each identification information to generate a trained model M(n-1) for hierarchy n-1 corresponding to each identification information. Here, one level higher was selected, but two or more levels higher may also be selected. For example, the hierarchy of the highest cell type ID may be selected.

[0165] In step S705, the generation unit 122A calculates the accuracy of the trained model M(n-1). In step S706, the generation unit 122A determines whether the accuracy is equal to or greater than the reference accuracy. Here, the generation unit 122A simply compares the average value of the accuracy of each trained model M(n-1) with the reference accuracy. The accuracy decreases as the learning error increases. For example, the root mean square error (RMSE) or mean square error (MSE) may be used as the learning error.

[0166] If the accuracy is equal to or greater than the reference accuracy (YES in step S706), the generation unit 122A determines that hierarchical level n-1 will be the target for learning (step S708). As a result, the trained model is generated and updated using the operational data corresponding to the identification information of hierarchical level n-1.

[0167] In step S703, if the learning cost is below a threshold (NO in step S703), the generation unit 122A decides to use hierarchical level n as the learning target (step S707). As a result, the trained model is generated and updated using the operational data corresponding to the identification information of hierarchical level n.

[0168] In step S706, if the accuracy is less than the reference accuracy (NO in step S706), the generation unit 122A determines another layer n and returns to step S701. The other layer n is determined to be one level higher than layer n-1 determined in step S704. If a suitable trained model cannot be obtained even after reaching the highest layer, the generation unit 122A may determine the lowest layer in step S709.

[0169] Thus, according to the first example, if the training cost of a pre-trained model is greater than the threshold, a pre-trained model of a higher tier is generated. If the training cost of that pre-trained model is below the threshold and its accuracy is above the reference accuracy, then that higher tier is selected as the target for training. Therefore, it is possible to search for a tier in which a pre-trained model can be obtained whose training cost is less than the threshold and whose accuracy is above the reference accuracy. Note that the first example is an algorithm that ascends the tiers so that the training cost is below the threshold, making it suitable for searching for the tier of pre-trained models used to estimate SOC.

[0170] Figure 14 is a flowchart showing a second example of the processing of the generation device in Embodiment 2 of this disclosure. In the second example, the learning error is evaluated to determine the hierarchy of the identification information. In this flowchart, SOH is adopted as the state of battery 3 estimated by the trained model.

[0171] In the second example, the hierarchy of identification information is as follows: block type ID, module type ID, and pack type ID, with this order being the highest. Furthermore, the hierarchy of pack individual ID, module individual ID, and block individual ID is the same as the hierarchy of pack type ID, module type ID, and block type ID, respectively. In addition, within the same hierarchy, the type ID is given a higher rank than the individual ID.

[0172] In step S801, the generation unit 122A generates a trained model M(n) corresponding to the type ID of hierarchy n. Specifically, the generation unit 122A obtains operational data corresponding to each type ID belonging to hierarchy n from the operational database 232, and trains the obtained operational data for each type ID to generate a trained model M(n) corresponding to each type ID.

[0173] In step S802, the generation unit 122A calculates the learning error of the trained model M(n). Details of the learning error are described above. In step S803, the generation unit 122A determines whether the learning error of the trained model M(n) is greater than a threshold. Here, the generation unit 122A simply compares the average value of the learning errors of each trained model M(n) with the threshold.

[0174] If the learning error is greater than the threshold (YES in step S803), the generation unit 122A generates a trained model M'(n) corresponding to the individual ID of hierarchical level n (step S804). For example, if a trained model M(n) corresponding to the pack type ID has been generated in step S801, a trained model M'(n) corresponding to the pack individual ID will be generated. Specifically, the generation unit 122A obtains operational data corresponding to each individual ID belonging to hierarchical level n from the operational database 232, and generates a trained model M'(n) for hierarchical level n corresponding to each individual ID by learning the obtained operational data for each individual ID.

[0175] In step S805, the generation unit 122A calculates the accuracy of the trained model M'(n). In step S806, the generation unit 122A determines whether the accuracy is equal to or greater than the reference accuracy. Here, the generation unit 122A simply compares the average accuracy of each trained model M'(n) with the reference accuracy. Details of the accuracy are described above.

[0176] If the accuracy is equal to or greater than the reference accuracy (YES in step S806), the generation unit 122A determines the individual IDs of hierarchical level n to be used as the learning targets (step S808). As a result, the trained model is subsequently generated and updated using the operational data corresponding to the individual IDs of hierarchical level n.

[0177] In step S803, if the learning error is below a threshold (NO in step S803), the generation unit 122A determines that hierarchical level n will be the target for learning (step S807). As a result, a trained model is generated and updated using the operational data corresponding to the type ID of hierarchical level n.

[0178] In step S806, if the accuracy is less than the reference accuracy (NO in step S806), the generation unit 122A determines another hierarchy n (step S809) and returns the process to step S801. The other hierarchy n may be one level higher or one level lower than the hierarchy n determined in step S801.

[0179] Thus, according to the second example, if the training error of a trained model is greater than the threshold, a trained model of a lower hierarchical level is generated. If the training error of that trained model is below the threshold and its accuracy is above the reference accuracy, then that lower hierarchical level is selected as the target for training. Therefore, it is possible to search for a hierarchical level in which a trained model can be obtained with a training cost less than the threshold and accuracy above the reference accuracy. In the second example, since an algorithm is used to descend the hierarchical level so that the training error is below the threshold, it is suitable for searching for a hierarchical level of trained models that estimate SOH or signs of failure.

[0180] (modified version) This disclosure may generate a trained model corresponding to individual cell IDs. [Industrial applicability]

[0181] This disclosure is useful in a technique for generating a trained model that can accurately estimate the state of a battery.

Claims

1. A method for manufacturing a trained model of a battery, in a generation device that generates a trained model of a battery including a plurality of hierarchically configured components, The processor of the aforementioned generation device One or more pieces of identification information assigned to a component at a certain level among the aforementioned plurality of components are obtained. The operating data of the battery corresponding to each identification information is acquired. By learning the acquired operational data for each of the one or more identification pieces, a trained model corresponding to each identification piece for estimating the state of the battery is generated. Output the generated trained model, In the above generation, Based on the operational data corresponding to one or more first identification pieces of the first layer, a first trained model corresponding to each first identification piece is generated. The training cost of the first pre-trained model is calculated, If the calculated learning cost is greater than the threshold, a second trained model corresponding to each second identification information is generated based on the operational data corresponding to one or more second identification information in the second layer higher than the first layer. Manufacturing method.

2. A method for manufacturing a trained model of a battery, in a generation device that generates a trained model of a battery including a plurality of hierarchically configured components, The processor of the aforementioned generation device One or more pieces of identification information assigned to a component at a certain level among the aforementioned plurality of components are obtained. The operating data of the battery corresponding to each identification information is acquired. By learning the acquired operational data for each of the one or more identification pieces, a trained model corresponding to each identification piece for estimating the state of the battery is generated. Output the generated trained model, In the above generation, Based on the operational data corresponding to one or more first identification pieces of the first layer, a first trained model corresponding to each first identification piece is generated. The learning error of the first trained model is calculated, If the calculated learning error is greater than a threshold, a second trained model corresponding to each second identification information is generated based on the operational data corresponding to one or more second identification information in the second layer lower than the first layer. Manufacturing method.

3. Each piece of identification information is assigned in a way that allows for the identification of the type and number of components in the lower hierarchy. The manufacturing method according to claim 1.

4. The plurality of components include a first component and a second component which is at a different hierarchical level from the first component. The one or more first identification pieces of information mentioned above are information that identifies the first component, The one or more second identification pieces mentioned above are pieces of information that identify the second component. The manufacturing method according to claim 1 or 2.

5. In the generation process, if the training cost of the first trained model is less than or equal to the threshold, the first trained model is determined to be the trained model to be trained. The manufacturing method according to claim 1.

6. In the generation process described above, if the accuracy of the second trained model is lower than the reference accuracy, a third trained model corresponding to each third identification information is generated based on the operational data corresponding to one or more third identification information of a third layer different from the first and second layers. The manufacturing method according to any one of claims 1 to 5.

7. In the generation process described above, if the accuracy of the second trained model is higher than the reference accuracy, the second trained model is determined to be the trained model to be trained. The manufacturing method according to any one of claims 1 to 6.

8. The training cost is calculated based on at least one of the following: the number of trained models generated, the amount of operational data used to generate the trained models, and the processing load of the processor when generating the trained models. The manufacturing method according to claim 1.

9. The aforementioned one or more identification pieces of information include one or more type identification pieces of information that identify each component by type, The manufacturing method according to any one of claims 1 to 8.

10. Each identification piece of information includes one or more individual identification pieces of information that identify each component individually. The manufacturing method according to any one of claims 1 to 9.

11. The aforementioned plurality of components include a cell, a block containing the cell, a module containing the block, and a battery pack containing the module. The manufacturing method according to any one of claims 1 to 10.

12. A generator that generates a trained model of a battery including multiple components arranged hierarchically, The generation device comprises a processor, The aforementioned processor, One or more pieces of identification information assigned to a component at a certain level among the aforementioned plurality of components are obtained. Obtain operational data corresponding to each identification piece of information. By learning the acquired operational data for each of the one or more identification pieces, the trained model corresponding to each identification piece for estimating the state of the battery is generated. The process is executed to output the generated trained model, In the above generation, Based on the operational data corresponding to one or more first identification pieces of the first layer, a first trained model corresponding to each first identification piece is generated. The training cost of the first pre-trained model is calculated, If the calculated learning cost is greater than the threshold, a second trained model corresponding to each second identification information is generated based on the operational data corresponding to one or more second identification information in the second layer higher than the first layer. generator.

13. A generation device for generating a trained model of a battery including a plurality of hierarchically configured components, The generation device comprises a processor, The aforementioned processor, One or more pieces of identification information assigned to a component at a certain level among the aforementioned plurality of components are obtained. Obtain operational data corresponding to each identification piece of information. By learning the acquired operational data for each of the one or more identification pieces, the trained model corresponding to each identification piece for estimating the state of the battery is generated. The process is executed to output the generated trained model, In the above generation, Based on the operational data corresponding to one or more first identification pieces of the first layer, a first trained model corresponding to each first identification piece is generated. The learning error of the first trained model is calculated, If the calculated learning error is greater than a threshold, a second trained model corresponding to each second identification information is generated based on the operational data corresponding to one or more second identification information in the second layer lower than the first layer. generator.

14. An estimation device for estimating the state of a battery, which includes multiple components arranged hierarchically, The estimation device comprises a processor, The aforementioned processor, The operating data of the aforementioned battery is acquired, The aforementioned operating data is input into the trained model to estimate the state of the battery. Execute a process that outputs state information indicating the estimated state, The trained model is a model generated by training the operational data corresponding to one or more identification pieces assigned to a component of a certain hierarchical level among the multiple components, for each of the one or more identification pieces. In generating the aforementioned model, Based on the operational data corresponding to one or more first identification pieces of the first layer, a first trained model corresponding to each first identification piece is generated. The training cost of the first pre-trained model is calculated, If the calculated learning cost is greater than the threshold, a second trained model corresponding to each second identification information is generated based on the operational data corresponding to one or more second identification information in the second layer higher than the first layer. Estimation device.

15. An estimation device for estimating the state of a battery including a plurality of hierarchically configured components, The estimation device comprises a processor, The aforementioned processor, The operating data of the aforementioned battery is acquired, The aforementioned operating data is input into the trained model to estimate the state of the battery. Execute a process that outputs state information indicating the estimated state, The trained model is a model generated by training the operational data corresponding to one or more identification pieces assigned to a component of a certain hierarchical level among the multiple components, for each of the one or more identification pieces. In the generation of the aforementioned model, Based on the operational data corresponding to one or more first identification pieces of the first layer, a first trained model corresponding to each first identification piece is generated. The learning error of the first trained model is calculated, If the calculated learning error is greater than a threshold, a second trained model corresponding to each second identification information is generated based on the operational data corresponding to one or more second identification information in the second layer lower than the first layer. Estimation device.

Images