Machine learning training apparatus and method of operation thereof
By performing boundary checks and trend tests, or F-tests and T-tests, on new data, the suitability of the model is determined, which solves the problem of increased error caused by low correlation between new data features and generalized features, and improves the reliability of the model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- LG ENERGY SOLUTION LTD
- Filing Date
- 2021-12-08
- Publication Date
- 2026-07-31
AI Technical Summary
When machine learning models have a low correlation between new data features and generalized features, the error increases, leading to a decrease in model reliability.
New data is collected through the data management unit, features are extracted by the data analysis unit, and boundary checks and trend tests or F-tests and T-tests are performed by the determination unit to determine whether the new data is suitable for application to machine learning models.
This reduces the error of the machine learning model and improves its reliability.
Smart Images
Figure CN116249993B_ABST
Abstract
Description
Technical Field
[0001] Cross-references to related applications
[0002] This application claims priority to Korean Patent Application No. 10-2020-0182652, filed on December 23, 2020, the entire contents of which are incorporated herein by reference for all purposes. Technical Field
[0004] The embodiments disclosed herein relate to a machine learning training apparatus and its operation method. Background Technology
[0005] Machine learning refers to techniques used to train computers to predict specific outcomes. Typically, predicting outcomes using machine learning involves preparing training data to train the machine and computer in a way suitable for the problem, validating the model using validation data, and using the validated model to predict outcomes.
[0006] Machine learning models are generated using training data selected according to specific criteria. When attempting to predict outcomes by feeding new data into a machine learning model generated based on the selected data, the model's error increases due to the discrepancies between the features to be generalized by machine learning and the features of the new data, thus degrading the model's reliability in addition to its accuracy, especially when the correlation between the features to be generalized by machine learning and the features of the new data is low. Summary of the Invention
[0007] Technical issues
[0008] This disclosure provides a machine learning training apparatus and its operating method, which can reduce the error of the machine learning model and improve the reliability of the machine learning model.
[0009] The technical problems of the embodiments disclosed herein are not limited to those described above, and other unmentioned technical problems will be clearly understood by those skilled in the art from the following description.
[0010] Technical solution
[0011] According to one aspect of this disclosure, a machine learning training apparatus is provided, comprising: a data management unit configured to collect new data; a data analysis unit configured to extract features of the data used to generate a machine learning model and features of the new data; and a determination unit configured to determine whether the number of segments of the new data is less than a reference number and to determine whether to apply the new data to the machine learning model by using different methods based on the number of segments of the new data.
[0012] According to an embodiment, when the number of new data fragments is less than the reference number, the determining unit performs boundary checks and trend tests relative to the features of the new data and the features of the data used to generate the machine learning model to determine whether to apply the new data to the machine learning model.
[0013] According to an embodiment, a trend test can determine whether the coefficients of an equation obtained from a graph showing the changes in features of new data over time fall within the range of the coefficients of an equation obtained from a graph showing the changes in features of data used to generate a machine learning model.
[0014] According to an embodiment, when it is determined, as a result of performing a trend check, that the features of the new data fall within the trend range of the features of the data used to generate the machine learning model, and as a result of performing a boundary check, it is determined that the features of the new data fall within the boundary range of the features of the data used to generate the machine learning model, the determining unit can apply the new data to the training of the machine learning model.
[0015] According to an embodiment, when the number of new data fragments is greater than or equal to the reference number, the determining unit can perform F-tests and T-tests relative to the characteristics of the new data and the characteristics of the used data.
[0016] According to an embodiment, when the result value obtained by performing an F-test and the result value obtained by performing a T-test are greater than or equal to a threshold, the determining unit excludes new data from the training of the machine learning model.
[0017] According to an embodiment, when the result value obtained by performing F-test and T-test is less than a threshold, the determining unit can determine whether to apply the new data to the machine learning model by performing boundary checks and trend tests relative to the features of the new data and the features of the data used to generate the machine learning model.
[0018] According to an embodiment, when it is determined, as a result of performing a trend check, that the features of the new data fall within the trend range of the features of the data used to generate the machine learning model, and as a result of performing a boundary check, it is determined that the features of the new data fall within the boundary range of the features of the data used to generate the machine learning model, the determining unit can apply the new data to the training of the machine learning model.
[0019] According to an embodiment, the machine learning training apparatus may further include a machine learning model training unit configured to train a machine learning model by applying new data to the machine learning model based on the determination result of a determination unit.
[0020] According to another aspect of this disclosure, a method for operating a machine learning training apparatus is provided, the method comprising: collecting new data; extracting features of the data used to generate a machine learning model and features of the new data; determining whether the number of segments of the new data is less than a reference number; and determining whether to apply the new data to the machine learning model by using different methods based on the number of segments of the new data.
[0021] According to an embodiment, determining whether to apply new data to a machine learning model based on the number of new data fragments using different methods may include: when the number of new data fragments is less than a reference number, determining whether to apply new data to a machine learning model by performing boundary checks and trend tests relative to the features of the new data and the features of the data used to generate the machine learning model.
[0022] According to an embodiment, determining whether to apply new data to a machine learning model based on the number of new data fragments using different methods may include: performing F-tests and T-tests on the data used to generate the machine learning model and the new data when the number of new data fragments is greater than or equal to the reference number.
[0023] According to an embodiment, determining whether to apply new data to a machine learning model based on the number of new data fragments using different methods may include excluding the new data from the training of the machine learning model when the result value obtained by performing an F-test and the result value obtained by performing a T-test are greater than or equal to a threshold.
[0024] According to an embodiment, determining whether to apply new data to a machine learning model based on the number of new data fragments using different methods may include: when the result value obtained by performing F-tests and T-tests is less than a threshold, determining whether to apply new data to a machine learning model by performing boundary checks and trend tests relative to the features of the new data and the features of the data used to generate the machine learning model.
[0025] Beneficial effects
[0026] By utilizing the machine learning training apparatus and operating method according to the embodiments disclosed herein, the error of the machine learning model can be reduced and the reliability of the machine learning model can be improved. Attached Figure Description
[0027] Figure 1 This is a block diagram of a machine learning training apparatus according to embodiments disclosed herein.
[0028] Figure 2 This is a flowchart illustrating a method of operating a machine learning training apparatus according to embodiments disclosed herein.
[0029] Figure 3This is a flowchart illustrating a method of operating a machine learning training apparatus according to another embodiment disclosed herein.
[0030] Figure 4 It is a graph showing the changes in the characteristics of data or new data used to generate a machine learning model according to embodiments disclosed herein over time.
[0031] Figure 5 The graph illustrates the drift coefficient values of linear and constant terms obtained from a graph showing the changes in features of data used to generate a machine learning model, according to embodiments disclosed herein.
[0032] Figure 6 It is a three-dimensional (3D) graph showing the drift coefficient values obtained from a graph showing the changes in features of the data used to generate a machine learning model, according to embodiments disclosed herein. Detailed Implementation
[0033] In the following, some embodiments disclosed in this document will be described in detail with reference to exemplary accompanying drawings. When adding reference numerals to components in each drawing, it should be noted that the same components are given the same reference numerals, even if they are indicated in different drawings. Furthermore, in describing the embodiments disclosed in this document, detailed descriptions of related known configurations or functions will be omitted if it is determined that such detailed descriptions would impede understanding of the embodiments disclosed in this document.
[0034] To describe the components of the embodiments disclosed herein, terms such as first, second, A, B, (a), (b), etc., may be used. These terms are used only to distinguish one component from another and do not limit the components to their nature, order, sequence, etc. Terms used herein, including technical and scientific terms, have the same meaning as commonly understood by those skilled in the art, provided that these terms are not defined differently. Generally, terms defined in commonly used dictionaries should be interpreted as having the same meaning as in the context of the relevant art and should not be interpreted as having an ideal or exaggerated meaning, unless they are clearly defined in this document.
[0035] Figure 1 This is a block diagram of a machine learning training apparatus according to embodiments disclosed herein.
[0036] refer to Figure 1 The machine learning training apparatus 100 according to the embodiments disclosed herein may include a data management unit 110, a data analysis unit 120, and a determination unit 130.
[0037] The data management unit 110 can manage data used for machine learning models. The data management unit 110 can store data used to generate the machine learning model (hereinafter referred to as "usage data"). The data management unit 110 can also collect new data to be input into the machine learning model. For example, usage data can be defined as values recording the change in charge from the state of discharge of a specific battery cell among a plurality of battery cells to the state of full charge of that specific battery cell, or from the state of full charge to the state of discharge. For example, new data can be defined as values recording the change in charge from the state of discharge of a battery cell different from the specific battery cell among the plurality of battery cells to the state of full charge of that battery cell, or from the state of full charge to the state of discharge.
[0038] Additionally, for example, data can be defined as recording the voltage and / or current changes of a specific battery cell among a plurality of battery cells, and new data can be defined as recording the voltage and / or current changes of a battery cell different from the specific battery cell among the plurality of battery cells. In this document, state of discharge can include a condition where the battery voltage is less than a reference value (e.g., cutoff voltage).
[0039] The data analysis unit 120 can extract features from usage data and new data. In this document, a feature can refer to a single, measurable empirical attribute found in a time series of data arranged at constant time intervals over time. For example, a feature could include battery capacity. In this document, battery capacity can refer to the amount of electricity that can be stored in the battery, and battery capacity data can be calculated from usage data and new data.
[0040] The determining unit 130 can determine whether the number of new data segments is less than a reference number. In this document, the reference number can be defined as a criterion used to determine whether the new data is treated as a group or as a separate data segment. For example, the determining unit 130 can treat the new data as a separate data segment if the number of new data segments is less than 10, and treat the new data as a group if the number of new data segments is greater than or equal to 10.
[0041] The determining unit 130 can determine whether to apply new data to the machine learning model based on the number of new data fragments using different methods. For example, when the number of new data fragments is greater than or equal to the reference number, the determining unit 130 can determine whether the new data is suitable to be input into the machine learning model by comparing between groups. Therefore, by filtering out new data that is not suitable to be input into the machine learning model through comparison between the new data group and the used data group, the number of additional tests used to check whether new data is suitable as data to be individually input into the machine learning model can be reduced.
[0042] That is, the determining unit 130 can perform inter-group comparison tests on the new data and the used data, and perform additional tests on whether data that meets specific conditions is suitable as data to be individually input into the machine learning model. The determining unit 130 can also perform tests to check whether each segment of the new data is suitable as data to be individually input into the machine learning model.
[0043] Unit 130 can determine whether new data is suitable as representative data for the group to be identified by the machine learning model through such a test.
[0044] The determining unit 130 can eliminate inappropriate data from the prediction process using the machine learning model. Therefore, the determining unit 130 can reduce the error of the machine learning model and improve its reliability.
[0045] When the number of new data fragments is less than the reference number, the determination unit 130 can perform boundary checks and trend tests relative to the characteristics of the new data and the characteristics of the used data.
[0046] When the number of new data segments is greater than or equal to the reference number, the determination unit 130 can perform F-tests and T-tests relative to the characteristics of the new data and the characteristics of the used data.
[0047] Figure 2 This is a flowchart illustrating a method of operating a machine learning training apparatus according to embodiments disclosed herein.
[0048] refer to Figure 2 The operation method of the machine learning training apparatus according to the embodiments disclosed herein may include an operation S210 of collecting new data, an operation S220 of extracting features of the used data and the new data, an operation S230 of determining whether the number of segments of the new data is less than a reference number, and an operation S240 of determining whether to apply the new data to the machine learning model by using different methods based on the number of segments of the new data.
[0049] In the following text, reference will be made to Figure 1 Describe operations S210 to S240 in detail.
[0050] In operation S210, the data management unit 110 can collect new data to be input into the machine learning model. For example, data can be defined as recording the change in charge level from the state of discharge of a specific battery cell among multiple battery cells to that specific battery cell's state of full charge, or from a state of full charge to a state of discharge. Similarly, new data can be defined as recording the change in charge level from the state of discharge of a battery cell different from the specific battery cell among the multiple battery cells to that battery cell's state of full charge, or from a state of full charge to a state of discharge.
[0051] Additionally, for example, data can be defined as recording the voltage and / or current changes of a specific battery cell among a plurality of battery cells, and new data can be defined as recording the voltage and / or current changes of a battery cell different from the specific battery cell among the plurality of battery cells. In this document, state of discharge can include a condition where the battery voltage is less than a reference value (e.g., cutoff voltage).
[0052] In operation S220, the data analysis unit 120 can extract features of the used data and new data.
[0053] For example, a feature may include the battery's capacity. In this document, battery capacity may refer to the amount of electricity that can be stored in the battery, and battery capacity data can be calculated from usage data and new data.
[0054] In operation S230, determining unit 130 can determine whether the number of new data segments is less than a reference number. Here, the reference number can be defined as a criterion used to determine whether the new data is treated as a group or as a separate data segment. For example, in operation S230, determining unit 130 can treat the new data as a separate data segment if the number of new data segments is less than 10, and treat the new data as a group if the number of new data segments is greater than or equal to 10.
[0055] In operation S240, the determining unit 130 can determine whether to apply the new data to the machine learning model based on the number of fragments of the new data using different methods.
[0056] For example, in operation S240, when the number of new data segments is less than the reference number, the determining unit 130 can perform boundary checks and trend tests to determine whether the new data is suitable as data to be input into the machine learning model. In operation S240, when the number of new data segments is greater than or equal to the reference number, the determining unit 130 can perform F-tests and T-tests to determine whether the new data is suitable as data to be input into the machine learning model.
[0057] Figure 3 This is a flowchart illustrating a method of operating a machine learning training apparatus according to another embodiment disclosed herein. Figure 4 It is a graph showing how the characteristics of usage data or new data change over time according to the embodiments disclosed herein. Figure 5 It is a graph showing the drift coefficient values of linear and constant terms obtained from a graph showing the changes in the characteristics of the data used, according to the embodiments disclosed herein. Figure 6It is a three-dimensional (3D) graph showing the drift coefficient values obtained from a graph showing the variation of features of the data used, according to embodiments disclosed herein.
[0058] First, refer to Figure 3 The operation method of the machine learning training apparatus according to the embodiments disclosed herein may include: an operation S301 of collecting new data; an operation S302 of extracting features from the used data and the new data; an operation S303 of determining whether the number of segments of the new data is less than a reference number; an operation S304 of determining whether to apply the new data to the machine learning model using different methods based on the number of segments of the new data; an operation S305 of determining whether the number of segments of the new data is less than, greater than, or equal to the reference number; an operation S306 of performing F-tests and T-tests on the used data and the new data when the number of segments of the new data is greater than or equal to the reference number; and an operation S307 of determining whether the result value obtained by performing F-tests and T-tests is less than a threshold. S307. When the number of new data segments is less than the reference number or the result value obtained by performing F-test and T-test is less than the threshold, perform boundary checks and trend tests relative to the characteristics of the new data and the characteristics of the data used. S308. Determine whether the new data meets the boundary range or trend range of the data used as the result of the boundary checks and trend tests. S309. When the new data meets the boundary range and trend range, use the new data for the machine learning model. S310. And when the result value obtained by performing F-test and T-test is greater than the threshold or when the new data fails to meet the boundary range or trend range of the data used, exclude the new data from the training of the machine learning model. S311.
[0059] In the following text, reference will be made to Figure 1 Describe operations S301 to S311 in detail.
[0060] In operation S301, the data management unit 110 can collect new data that will be input into the machine learning model.
[0061] In operation S302, the data analysis unit 120 can extract features of the used data and new data.
[0062] In operation S303, the determining unit 130 can determine whether the number of new data segments is less than the reference number.
[0063] In operation S304, the determining unit 130 can determine whether to apply the new data to the machine learning model based on the number of fragments of the new data using different methods.
[0064] In operation S305, the determining unit 130 can determine whether the number of new data segments is less than, greater than, or equal to the reference number.
[0065] In operation S306, when the number of fragments of new data is greater than or equal to the number of references, determining unit 130 may perform F-tests and T-tests relative to the new data and the used data.
[0066] The F-test is a technique for verifying the statistical significance of the difference in variance between two groups at a statistically significant level.
[0067] Equation 1 can refer to the equation used to obtain the variance difference between two groups through the F-test.
[0068] Equation 1
[0069]
[0070] In Equation 1, 'a' indicates a characteristic of the new data. 'mean(a)' indicates the average of the characteristics of the new data. 'n_a' indicates the number of characteristics of the new data. 'b' indicates the characteristic used in the data. 'mean(b)' indicates the average of the characteristics used in the data. 'n_b' indicates the number of characteristics used in the data.
[0071] The T-test is a technique for verifying the statistical significance of the difference between the means of two groups at a statistically significant level.
[0072] Equation 2 can refer to the equation used to obtain the difference in means between two groups through a T-test.
[0073] Equation 2
[0074]
[0075] In Equation 2, 'a' indicates a characteristic of the new data. 'mean(a)' indicates the mean of the characteristics of the new data. 'n_a' indicates the number of characteristics of the new data. 'var(a)' indicates the variance of the characteristics of the new data. 'b' indicates the characteristic used in the data. 'mean(b)' indicates the mean of the characteristics used in the data. 'n_b' indicates the number of characteristics used in the data. 'var(b)' indicates the variance of the characteristics used in the data.
[0076] Statistical significance can be defined as the method by which a statistician determines whether his or her experimental results are extreme cases that are accidental or unlikely to be accidental. For example, a p-value can be determined to be significant when the p-value, which is the result of an F-test or a T-test, is greater than or equal to a threshold, and a p-value can be determined to be meaningless when the p-value is less than the threshold.
[0077] In operation S307, determining unit 130 can determine whether the result value obtained by performing an F-test and the result value obtained by performing a T-test are less than a threshold. In this document, the threshold can be defined as a criterion used to determine "anomalies" in the output of a machine learning model due to extreme results. That is, the threshold can be defined as a criterion indicating the degree to which data contradicts a specific statistical model. For example, the threshold can be set to, but is not limited to, 0.05, and can be set slightly based on statistical experimental results.
[0078] In operation S308, when the result value obtained by performing the F-test and the result value obtained by performing the T-test are less than the threshold, the determination unit 130 may perform boundary checks and trend tests relative to the characteristics of the new data and the data used.
[0079] In operation S308, when the number of new data segments is less than the reference number, the determining unit 130 may perform boundary checks and trend tests relative to the features of the new data and the features of the data used to generate the machine learning model. Here, the boundary checks may include range checks to determine whether the data exists within a specific range. For example, for each segment of the new data, it may be determined individually whether the minimum and maximum values of the features of the new data fall within the minimum and maximum ranges of the features of the used data. As a result of the boundary checks, when the minimum and maximum values of the features of the new data fall within the minimum and maximum ranges of the features of the used data, the determining unit 130 may determine that the new data satisfies the boundary range of the used data.
[0080] In trend testing, a trend can refer to a gradual and continuous change in a long-term time series of data. That is, a trend can refer to an upward or downward trend in time series data. For example, when the charging, resting, discharging, and resting of a battery is considered a cycle, the gradual change in battery capacity that occurs as the battery cycles repeat can be considered a trend. Trend testing can determine whether the coefficients of the equation obtained from a graph indicating changes in characteristics of new data fall within the range of the coefficients of the equation obtained from a graph indicating changes in characteristics of usage data.
[0081] The determining unit 130 can approximate new data and used data separately by specifying the X-axis as time and the Y-axis as data. The determining unit 130 can derive, by approximating the used data and new data, as follows: Figure 4 The graph shown is shown in the figure.
[0082] refer to Figure 4 This illustrates the characteristics of a battery (i.e., the battery capacity trend) as the battery is cycled repeatedly. Here, the data can be usage data or new data.
[0083] Unit 130 can define the coefficients of the quadratic function equation obtained from the graph as trend coefficients based on an approximation. The quadratic coefficients of the quadratic function equation can be indicated by 'a', the coefficients of the linear terms can be indicated by 'b', and the coefficients of the constant terms can be indicated by 'c'.
[0084] refer to Figure 5 This shows from Figure 4 The graph shown in the figure yields the coefficients of the linear term and the coefficients of the constant term.
[0085] refer to Figure 6 The determining unit 130 can display the trend coefficients a, b, and c of the characteristics of the used data in a 3D curve graph. The determining unit 130 can individually determine whether the trend coefficients of the characteristics of the new data fall within the range of the trend coefficients of the characteristics of the used data shown in the 3D curve graph. As a result of the trend test, when the trend coefficients of the characteristics of the new data fall within the range of the trend coefficients of the characteristics of the used data, the data determining unit 130 can determine that the trend range is satisfied.
[0086] refer to Figure 6 The diagram shows a 3D graph of the trend coefficients of usage data obtained from a graph showing the characteristics of data that appear as the battery cycles repeatedly (i.e., the battery capacity trend).
[0087] Return to reference Figure 3 In operation S309, as a result of boundary checks and trend tests, determining unit 130 can determine whether the new data meets the boundary range and trend range of the data to be used.
[0088] In operation S310, when new data satisfies both the boundary range and the trend range, the determination unit 130 can use the new data for the machine learning model.
[0089] In operation S311, when the result value obtained by performing the F-test and the result value obtained by performing the T-test are greater than or equal to the threshold, the determining unit 130 can exclude new data from the training of the machine learning model.
[0090] In operation S311, when new data fails to meet the boundary range or trend range of the data to be used by performing boundary checks and trend tests, the determination unit 130 can exclude new data from the training of the machine learning model.
[0091] The above description merely illustrates the technical concept of this disclosure, and various modifications and variations will be possible for those skilled in the art without departing from the essential characteristics of this disclosure.
[0092] Therefore, the embodiments disclosed herein are intended to describe, and not limit, the technical spirit of this disclosure, and the scope of the technical spirit of this disclosure is not limited by these embodiments. The scope of protection of this disclosure should be interpreted by the appended claims, and all technical spirit within the same scope should be understood to be included within the scope of this disclosure.
Claims
1. A machine learning training apparatus for training a machine learning model, the machine learning model being used to predict battery cells, comprising: A data management unit, configured to collect new data; A data analysis unit, configured to extract features from the data used to generate a machine learning model and features from the new data, The data used to generate the machine learning model is defined as the values recording the changes in charge, voltage, and / or current of the first battery cell between a discharge state and a fully charged state. The new data is defined as the values of charge, voltage, and / or current changes recorded by a second battery cell, different from the first battery cell, between a discharged state and a fully charged state. The features include the capacity of a single battery cell; and A determining unit is configured to determine whether the number of fragments of the new data is less than a reference number, determine the association between the features of the data used to generate the machine learning model and the features of the new data by using different methods based on the number of fragments of the new data, and determine whether to apply the new data to the machine learning model based on the determined association.
2. The machine learning training apparatus according to claim 1, wherein, When the number of fragments of the new data is less than the reference number, the determining unit performs boundary checks and trend tests relative to the features of the new data and the features of the data used to generate the machine learning model to determine whether to apply the new data to the machine learning model.
3. The machine learning training apparatus according to claim 2, wherein, The trend test determines whether the coefficients of the equation obtained from a graph showing the changes in the features of the new data over time fall within the range of the coefficients of the equation obtained from a graph showing the changes in the features of the data used to generate the machine learning model.
4. The machine learning training apparatus according to claim 2, wherein, When the new data features are determined to fall within the trend range of the features of the data used to generate the machine learning model as a result of performing the trend check, and the new data features are determined to fall within the boundary range of the features of the data used to generate the machine learning model as a result of performing the boundary check, the determining unit applies the new data to the training of the machine learning model.
5. The machine learning training apparatus according to claim 1, wherein, When the number of fragments of the new data is greater than or equal to the reference number, the determining unit performs F-tests and T-tests on the data used to generate the machine learning model and the new data.
6. The machine learning training apparatus according to claim 5, wherein, When the result value obtained by performing the F-test and the result value obtained by performing the T-test are greater than or equal to the threshold, the determining unit excludes the new data from the training of the machine learning model.
7. The machine learning training apparatus according to claim 5, wherein, When the result value obtained by performing the F-test and the T-test is less than the threshold, the determining unit determines whether to apply the new data to the machine learning model by performing boundary checks and trend tests relative to the features of the new data and the features of the data used to generate the machine learning model.
8. The machine learning training apparatus according to claim 7, wherein, When the new data features are determined to fall within the trend range of the features of the data used to generate the machine learning model as a result of performing the trend check, and the new data features are determined to fall within the boundary range of the features of the data used to generate the machine learning model as a result of performing the boundary check, the determining unit applies the new data to the training of the machine learning model.
9. The machine learning training apparatus according to claim 1, further comprising a machine learning model training unit, the machine learning model training unit being configured to train the machine learning model by applying the new data to the machine learning model based on the determination result of the determining unit.
10. A method for operating a machine learning training device for training a machine learning model, the machine learning model being used to predict battery cells, the method comprising: Collect new data; Extract features from the data used to generate the machine learning model and features from the new data. The data used to generate the machine learning model is defined as the values recording the changes in charge, voltage, and / or current of the first battery cell between a discharge state and a fully charged state. The new data is defined as the values of charge, voltage, and / or current changes recorded by a second battery cell, different from the first battery cell, between a discharged state and a fully charged state. The features include the capacity of a single battery cell; Determine whether the number of fragments in the new data is less than the reference number; The association between the features of the data used to generate the machine learning model and the features of the new data is determined by using different methods based on the number of fragments of the new data. The determination of whether to apply the new data to the machine learning model is based on the identified associations.
11. The operating method according to claim 10, wherein, Determining whether to apply the new data to the machine learning model based on the number of fragments of the new data using different methods includes: when the number of fragments of the new data is less than the reference number, determining whether to apply the new data to the machine learning model by performing boundary checks and trend tests relative to the features of the new data and the features of the data used to generate the machine learning model.
12. The operating method according to claim 10, wherein, Determining whether to apply the new data to the machine learning model based on the number of fragments of the new data using different methods includes: performing F-tests and T-tests on the data used to generate the machine learning model and the new data when the number of fragments of the new data is greater than or equal to the reference number.
13. The operating method according to claim 12, wherein, Determining whether to apply the new data to the machine learning model based on the number of fragments of the new data using different methods includes: excluding the new data from the training of the machine learning model when the result value obtained by performing the F-test or the result value obtained by performing the T-test is greater than or equal to a threshold.
14. The operating method according to claim 12, wherein, Determining whether to apply the new data to the machine learning model based on the number of fragments of the new data using different methods includes: when the result value obtained by performing the F-test and the T-test is less than a threshold, determining whether to apply the new data to the machine learning model by performing boundary checks and trend tests relative to the features of the new data and the features of the data used to generate the machine learning model.