Battery degradation prediction system
The system improves battery degradation prediction accuracy by accumulating vehicle data and selecting datasets based on database status, addressing low estimation accuracy in existing systems.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- TOYOTA JIDOSHA KK
- Filing Date
- 2024-11-12
- Publication Date
- 2026-05-22
AI Technical Summary
Existing battery degradation prediction systems face low estimation accuracy due to variations in learning data.
A battery degradation prediction system that accumulates data from multiple vehicles, selects datasets based on database status, and constructs models using data from similar vehicles to improve prediction accuracy.
Enhances the accuracy of battery degradation prediction by using targeted data sets and minimizing characteristic differences between vehicles, thereby improving model performance.
Smart Images
Figure 2026084969000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of a battery degradation degree prediction system for predicting the degradation degree of a secondary battery.
Background Art
[0002] As this type of system, for example, a system has been proposed that learns a prediction model for estimating an index related to the degradation state of a battery based on learning data and estimates an index related to the degradation state of the battery based on the learned prediction model (see Patent Document 1).
Prior Art Documents
Patent Documents
[0003]
Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0004] In the technique described in Patent Document 1, an index related to the degradation state of a battery is estimated based on a prediction model learned by machine learning. However, depending on the learning data used for machine learning, the estimation accuracy may be low.
[0005] The present invention has been made in view of the above problems, and an object thereof is to provide a battery degradation degree prediction system capable of improving the estimation accuracy.
Means for Solving the Problems
[0006] A battery degradation degree prediction system according to an aspect of the present invention includes a database that accumulates the degradation degree of a driving secondary battery mounted in each of a plurality of vehicles, and a data set selection means that changes a data set used for learning a model for predicting the future degradation degree of the secondary battery in each of the plurality of vehicles according to the accumulation status of the database. [Brief explanation of the drawing]
[0007] [Figure 1] This is a conceptual diagram showing the configuration of a battery degradation prediction system according to an embodiment. [Figure 2] A flowchart showing the operation of the battery degradation prediction system according to the embodiment. [Figure 3] This flowchart shows the method for determining the data used for training. [Figure 4] This is a flowchart showing the model building method. [Modes for carrying out the invention]
[0008] Embodiments of the battery degradation prediction system will be described with reference to Figures 1 to 4. In Figure 1, the battery degradation prediction system 10 includes a database 11, a dataset selection unit 12, and a model learning unit 13. For example, the battery degradation prediction system 10 may be implemented by a server. The battery degradation prediction system 10 may be implemented by a single server or by multiple servers. The server may be a cloud server.
[0009] The battery degradation prediction system 10 is configured to communicate bidirectionally with multiple vehicles V1, V2, ..., Vn via the Internet. Each of the multiple vehicles V1, V2, ..., Vn is equipped with a secondary battery B1, B2, ..., Bn for driving. The multiple vehicles V1, V2, ..., Vn may include at least one of an electric vehicle (Battery Electric Vehicle: BEV), a plug-in hybrid electric vehicle (PHEV), a hybrid electric vehicle (HEV), and a fuel cell electric vehicle (FCEV). The secondary batteries B1, B2, ..., Bn may include at least one of a lithium-ion battery, a nickel-metal hydride battery, and a solid-state battery.
[0010] Each of the multiple vehicles V1, V2, ..., Vn periodically transmits data, including the State of Health (SOH) of the secondary battery, to the battery degradation prediction system 10. This data may include specification information, timestamps, vehicle and battery characteristics, and information regarding the degradation level. For example, specification information may include the Vehicle Identification Number (VIN), battery number, battery type, vehicle type, destination, and design change version. Information regarding the degradation level may include the degradation level, days elapsed since L / O (Line Off), mileage, cumulative parking time, SOC (State of Charge) history and battery temperature history while the ignition is ON, SOC history and battery temperature history while the ignition is OFF, etc. Note that while the ignition is ON, at least one of the following may be the vehicle in motion, in power transmission mode, and charging. While the ignition is OFF, at least one of the following may be the vehicle parked and in transport. Multiple data transmitted from multiple vehicles V1, V2, ..., Vn are stored in database 11. As a result, the degradation levels of secondary batteries B1, B2, ..., Bn are accumulated in database 11.
[0011] The battery degradation prediction system 10 uses data, including the degradation level of secondary batteries, stored in the database 11 to train a model that predicts the future degradation level of secondary batteries. The operation of the battery degradation prediction system 10 will be explained with reference to Figures 2 to 4. For example, multiple regression models, multivariate machine learning models, DNNs (Deep Neural Networks), etc., can be applied to the above model.
[0012] In Figure 2, the dataset selection unit 12 of the battery degradation prediction system 10 determines the dataset from the data stored in the database 11 to be used to train a model that predicts the future degradation of the secondary battery of a particular vehicle (step S10). The process of step S10 will be explained with reference to Figure 3. Hereinafter, the vehicle mentioned above will be assumed to be vehicle V1.
[0013] The dataset selection unit 12 may set the values of thresholds A, B, and C, described later, based on the data about vehicle V1 stored in the database 11. For example, the values of thresholds A, B, and C may change depending on the battery type, vehicle type, destination, and design change version. For example, the dataset selection unit 12 may set the values of thresholds A, B, and C based on a table that defines the relationship between the battery type, vehicle type, destination, and design change version and thresholds A, B, and C. Note that the values of thresholds A, B, and C may also be predetermined fixed values.
[0014] In Figure 3, the dataset selection unit 12 determines whether the number N1 of data points about vehicle V1 stored in the database 11 is greater than the threshold A (step S101). If, in the process of step S101, it is determined that the number N1 is greater than the threshold A (step S101: Yes), the dataset selection unit 12 decides to use only the multiple data points about vehicle V1 to train a model that predicts the future degradation of the secondary battery B1 of vehicle V1 (step S102). In this case, the dataset used for training will include only the multiple data points that include the degradation of the secondary battery B1 of vehicle V1.
[0015] In step S101, if it is determined that the number N1 is less than or equal to threshold A (step S101: No), the dataset selection unit 12 determines whether the total number N2 of data about vehicle V1 and data about other vehicles of the same type as vehicle V1 stored in the database 11 is greater than threshold B (step S103). In step S103, if it is determined that the total number N2 is greater than threshold B (step S103: Yes), the dataset selection unit 12 decides to use the data about vehicle V1 and the data about other vehicles of the same type as vehicle V1 to train a model that predicts the future degradation of the secondary battery B1 of vehicle V1 (step S104). In this case, the dataset used for training will include one or more data points containing the degradation of the secondary battery B1 of vehicle V1 and one or more data points containing the degradation of the secondary batteries of other vehicles of the same type as vehicle V1.
[0016] In step S103, if it is determined that the total number N2 is less than or equal to threshold B (step S103: No), the dataset selection unit 12 determines whether the total number N3 of data about vehicle V1 stored in the database 11 and data about other vehicles equipped with secondary batteries of the same type as vehicle V1's secondary battery B1 is greater than threshold C (step S105). In step S105, if it is determined that the total number N3 is greater than threshold C (step S105: Yes), the dataset selection unit 12 decides to use the data about vehicle V1 and the data about other vehicles equipped with secondary batteries of the same type as vehicle V1's secondary battery B1 to train a model that predicts the future degradation of vehicle V1's secondary battery B1 (step S106). In this case, the dataset used for training will include one or more data points containing the degradation of vehicle V1's secondary battery B1 and one or more data points containing the degradation of secondary batteries of other vehicles equipped with secondary batteries of the same type as vehicle V1's secondary battery B2.
[0017] In step S105, if it is determined that the total number N3 is less than or equal to the threshold C (step S105: No), it is decided to use the data on vehicle V1 stored in the database 11 and the data on other vehicles to train a model that predicts the future degradation of the secondary battery B1 of vehicle V1 (step S107). In this case, the other vehicles may include vehicles of a different type than vehicle V1. Also, the other vehicles may include vehicles equipped with a secondary battery of a different type than the secondary battery B1 of vehicle V1. Furthermore, the dataset selection unit 12 may decide to use, for example, the data on vehicle V1 and the data on other vehicles destined for the same destination as vehicle V1 to train the above model. Alternatively, the dataset selection unit 12 may decide to use, for example, the data on vehicle V1 and the data on other vehicles used for the same purpose (commercial use, etc.) as vehicle V1 to train the above model.
[0018] Returning to FIG. 2, after step S10, the model learning unit 13 of the battery degradation degree prediction system 10 constructs a model (for example, a model for predicting the future degradation degree of the secondary battery B1 of the vehicle V1) using the data set determined by the process of step S10 (step S20). The process of this step S20 will be described with reference to FIG. 4.
[0019] In FIG. 4, the model learning unit 13 excludes abnormal data from the data included in the data set (step S201). With this configuration, the quality of the data can be improved. Examples of abnormal data include data related to vehicles in which the drive secondary battery has been replaced, data related to vehicles in which the degradation characteristics of the drive secondary battery have changed due to software design changes, data with relatively large variations in degradation degree, data with relatively low measurement accuracy of degradation degree due to some measurement troubles, and the like. Note that, for determining whether data is abnormal data, for example, a rule-based determination, a determination using an unsupervised learning algorithm such as the k-means method, or the like may be used.
[0020] Next, the model learning unit 13 sets the true value of the degradation degree (step S202). Note that various existing modes can be applied to the method of setting the true value of the degradation degree. Therefore, the detailed description of the method of setting the true value of the degradation degree is omitted. Next, the model learning unit 13 excludes data for which the true value of the degradation degree cannot be set from the data included in the data set (step S203). With this configuration, the quality of the data can be improved.
[0021] Next, the model learning unit 13 clusters the data included in the data set for each vehicle based on the degradation trend (step S204). In the process of step S204, the model learning unit 13 may sample only the data having properties similar to those of the prediction target (for example, the secondary battery B1 of the vehicle V1).
[0022] Next, the model learning unit 13 constructs a model (for example, a model for predicting the future degradation degree of the secondary battery B1 of the vehicle V1) for predicting the future degradation degree of the secondary battery using the dataset (step S205). Next, the model learning unit 13 evaluates the accuracy of the model constructed in the process of step S205 (step S206). Note that various existing methods can be applied to the method for constructing the model and the method for evaluating the accuracy of the constructed model. Therefore, the detailed description of the method for constructing the model and the method for evaluating the accuracy of the constructed model is omitted.
[0023] Note that when the data included in the dataset includes data related to other vehicles in addition to the data related to one vehicle (for example, vehicle V1), the influence of the data related to other vehicles on the model may be greater than the influence of the data related to one vehicle on the model. Therefore, the model learning unit 13 may perform a treatment of making the weight of the data related to one vehicle larger than the weight of the data related to other vehicles, or perform subsampling processing on the data included in the dataset. For example, the data weighting may be arbitrarily performed during modeling, or the learning may be performed so as to be adjusted in machine learning or statistical models. Regarding the data subsampling, subsampling may be performed such that more data related to other vehicles having characteristics similar to those of one vehicle is left, while less data related to other vehicles having characteristics not similar to those of one vehicle is left. For example, the model learning unit 13 may randomly subsample so that the ratio of the data related to one vehicle (for example, vehicle V1) to the data related to other vehicles is a:b.
[0024] Returning to Figure 1, after step S20, the battery degradation prediction system 10 determines whether the accuracy of the constructed model is greater than a predetermined threshold (step S30). If, in the process of step S30, it is determined that the accuracy of the constructed model is greater than a predetermined threshold (step S30: Yes), the battery degradation prediction system 10 decides to adopt the constructed model (step S40). In this case, the battery degradation prediction system 10 may use the constructed model to predict the future degradation of a secondary battery in a vehicle (for example, secondary battery B1 of vehicle V1).
[0025] In the process of step S30, if it is determined that the accuracy of the constructed model is below a predetermined threshold (step S30: No), the battery degradation prediction system 10 decides not to adopt the constructed model (step S50). In this case, the battery degradation prediction system 10 may repeat the operation shown in the flowchart of Figure 2.
[0026] (Technical effects) If there is a sufficient amount of data on a single vehicle (i.e., data on the degradation level of the secondary battery of a single vehicle), it is desirable to construct a model that predicts future degradation levels using only the data on that single vehicle. This is because if data on other vehicles is used in addition to the data on a single vehicle, differences in the characteristics of one vehicle and other vehicles may affect the model. In this embodiment, in the process of step S101 described above (see Figure 3), if the number N1 is determined to be greater than the threshold A, a dataset containing only multiple data points on a single vehicle (e.g., vehicle V1) is used to train the model. In this case, the constructed model is not affected by differences in the characteristics of one vehicle and other vehicles, so an improvement in the accuracy of degradation level prediction can be expected.
[0027] If the amount of data for a single vehicle is insufficient, the amount of data available for training the model can be increased by using data from other vehicles. However, as mentioned above, differences in the characteristics of one vehicle and other vehicles may affect the model. Therefore, in this embodiment, the data from other vehicles to be used is selected according to the amount of data (see flowchart in Figure 3). By configuring it in this way, it is possible to increase the amount of data while suppressing a decrease in the quality of the data used to train the model. In this case as well, an improvement in the accuracy of predicting the degree of degradation can be expected.
[0028] The embodiments of the invention derived from the above-described embodiments are described below.
[0029] A battery degradation prediction system according to one aspect of the invention comprises a database for accumulating the degradation levels of secondary batteries for driving vehicles installed in each of a plurality of vehicles, and a dataset selection means for changing the dataset used for training a model that predicts the future degradation level of the secondary batteries of each of the plurality of vehicles, according to the accumulation status of the database. In the above embodiment, the "dataset selection unit 12" corresponds to an example of the "dataset selection means".
[0030] In one example of the battery degradation prediction system, the dataset selection means may, depending on the accumulation status of the database, select a dataset containing multiple degradation levels of the secondary battery of one of the multiple vehicles that is stored in the database, to be used as a dataset for training a model that predicts the future degradation level of the secondary battery of one of the multiple vehicles.
[0031] In another example of the battery degradation prediction system, the dataset selection means may, depending on the state of the database, select a dataset containing multiple degradation levels of the secondary batteries of the vehicle and vehicles of the same type as the vehicle, which is stored in the database, to be used to train a model that predicts the future degradation level of the secondary battery of one of the multiple vehicles.
[0032] In another example of the battery degradation prediction system, the dataset selection means may, depending on the state of the database, select a dataset containing the secondary battery of one of the multiple vehicles, and multiple degradation levels of secondary batteries of the same type as the secondary battery of the one vehicle, as a dataset used to train a model that predicts the future degradation of the secondary battery of one of the multiple vehicles.
[0033] The present invention is not limited to the embodiments described above, and can be modified as appropriate without contradicting the gist or idea of the invention as can be read from the claims and specification as a whole. Battery degradation prediction systems with such modifications are also included in the technical scope of the present invention. [Explanation of symbols]
[0034] 10...Battery degradation prediction system, 11...Database, 12...Dataset selection unit, 13...Model learning unit
Claims
1. A database that stores the degree of degradation of the secondary batteries used for propulsion installed in each of multiple vehicles, A dataset selection means for changing the dataset used to train a model that predicts the future degradation of the secondary battery of each of the aforementioned multiple vehicles, according to the accumulation status of the database, A battery degradation prediction system equipped with the following features.
2. The dataset selection means selects a dataset containing multiple degradation levels of the secondary battery of one of the multiple vehicles, stored in the database, as a dataset to be used to train a model that predicts the future degradation level of the secondary battery of one of the multiple vehicles, depending on the storage status of the database. The battery degradation prediction system according to claim 1.
3. The dataset selection means selects a dataset stored in the database that includes multiple degradation levels of the secondary batteries of the vehicle in question and vehicles of the same type as the vehicle in question, as a dataset to be used to train a model that predicts the future degradation level of the secondary battery of one of the multiple vehicles. The battery degradation prediction system according to claim 1.
4. The dataset selection means selects a dataset stored in the database, which includes the secondary battery of one of the vehicles and multiple degradation levels of secondary batteries of the same type as the secondary battery of the vehicle, as a dataset to be used to train a model that predicts the future degradation level of the secondary battery of one of the multiple vehicles. The battery degradation prediction system according to claim 1.